xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision b28624fde638caadd4a89f50c9b7e7da0f98c4d2)
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 DIAGNOSTIC
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(lkp, flags, VI_MTX(vp), ap->a_td, ap->a_file, ap->a_line);
374 			if (lkp == vp->v_vnlock || result != 0)
375 				break;
376 			/*
377 			 * Apparent success, except that the vnode
378 			 * mutated between snapshot file vnode and
379 			 * regular file vnode while this process
380 			 * slept.  The lock currently held is not the
381 			 * right lock.  Release it, and try to get the
382 			 * new lock.
383 			 */
384 			(void) _lockmgr(lkp, LK_RELEASE, VI_MTX(vp), ap->a_td, ap->a_file, ap->a_line);
385 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
386 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
387 			flags &= ~LK_INTERLOCK;
388 		}
389 		break;
390 	default:
391 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
392 	}
393 	return (result);
394 #else
395 	return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
396 #endif
397 }
398 
399 /*
400  * Vnode op for reading.
401  */
402 /* ARGSUSED */
403 static int
404 ffs_read(ap)
405 	struct vop_read_args /* {
406 		struct vnode *a_vp;
407 		struct uio *a_uio;
408 		int a_ioflag;
409 		struct ucred *a_cred;
410 	} */ *ap;
411 {
412 	struct vnode *vp;
413 	struct inode *ip;
414 	struct uio *uio;
415 	struct fs *fs;
416 	struct buf *bp;
417 	ufs_lbn_t lbn, nextlbn;
418 	off_t bytesinfile;
419 	long size, xfersize, blkoffset;
420 	int error, orig_resid;
421 	int seqcount;
422 	int ioflag;
423 
424 	vp = ap->a_vp;
425 	uio = ap->a_uio;
426 	ioflag = ap->a_ioflag;
427 	if (ap->a_ioflag & IO_EXT)
428 #ifdef notyet
429 		return (ffs_extread(vp, uio, ioflag));
430 #else
431 		panic("ffs_read+IO_EXT");
432 #endif
433 #ifdef DIRECTIO
434 	if ((ioflag & IO_DIRECT) != 0) {
435 		int workdone;
436 
437 		error = ffs_rawread(vp, uio, &workdone);
438 		if (error != 0 || workdone != 0)
439 			return error;
440 	}
441 #endif
442 
443 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
444 	ip = VTOI(vp);
445 
446 #ifdef DIAGNOSTIC
447 	if (uio->uio_rw != UIO_READ)
448 		panic("ffs_read: mode");
449 
450 	if (vp->v_type == VLNK) {
451 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
452 			panic("ffs_read: short symlink");
453 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
454 		panic("ffs_read: type %d",  vp->v_type);
455 #endif
456 	orig_resid = uio->uio_resid;
457 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
458 	if (orig_resid == 0)
459 		return (0);
460 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
461 	fs = ip->i_fs;
462 	if (uio->uio_offset < ip->i_size &&
463 	    uio->uio_offset >= fs->fs_maxfilesize)
464 		return (EOVERFLOW);
465 
466 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
467 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
468 			break;
469 		lbn = lblkno(fs, uio->uio_offset);
470 		nextlbn = lbn + 1;
471 
472 		/*
473 		 * size of buffer.  The buffer representing the
474 		 * end of the file is rounded up to the size of
475 		 * the block type ( fragment or full block,
476 		 * depending ).
477 		 */
478 		size = blksize(fs, ip, lbn);
479 		blkoffset = blkoff(fs, uio->uio_offset);
480 
481 		/*
482 		 * The amount we want to transfer in this iteration is
483 		 * one FS block less the amount of the data before
484 		 * our startpoint (duh!)
485 		 */
486 		xfersize = fs->fs_bsize - blkoffset;
487 
488 		/*
489 		 * But if we actually want less than the block,
490 		 * or the file doesn't have a whole block more of data,
491 		 * then use the lesser number.
492 		 */
493 		if (uio->uio_resid < xfersize)
494 			xfersize = uio->uio_resid;
495 		if (bytesinfile < xfersize)
496 			xfersize = bytesinfile;
497 
498 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
499 			/*
500 			 * Don't do readahead if this is the end of the file.
501 			 */
502 			error = bread(vp, lbn, size, NOCRED, &bp);
503 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
504 			/*
505 			 * Otherwise if we are allowed to cluster,
506 			 * grab as much as we can.
507 			 *
508 			 * XXX  This may not be a win if we are not
509 			 * doing sequential access.
510 			 */
511 			error = cluster_read(vp, ip->i_size, lbn,
512 				size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
513 		} else if (seqcount > 1) {
514 			/*
515 			 * If we are NOT allowed to cluster, then
516 			 * if we appear to be acting sequentially,
517 			 * fire off a request for a readahead
518 			 * as well as a read. Note that the 4th and 5th
519 			 * arguments point to arrays of the size specified in
520 			 * the 6th argument.
521 			 */
522 			int nextsize = blksize(fs, ip, nextlbn);
523 			error = breadn(vp, lbn,
524 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
525 		} else {
526 			/*
527 			 * Failing all of the above, just read what the
528 			 * user asked for. Interestingly, the same as
529 			 * the first option above.
530 			 */
531 			error = bread(vp, lbn, size, NOCRED, &bp);
532 		}
533 		if (error) {
534 			brelse(bp);
535 			bp = NULL;
536 			break;
537 		}
538 
539 		/*
540 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
541 		 * will cause us to attempt to release the buffer later on
542 		 * and will cause the buffer cache to attempt to free the
543 		 * underlying pages.
544 		 */
545 		if (ioflag & IO_DIRECT)
546 			bp->b_flags |= B_DIRECT;
547 
548 		/*
549 		 * We should only get non-zero b_resid when an I/O error
550 		 * has occurred, which should cause us to break above.
551 		 * However, if the short read did not cause an error,
552 		 * then we want to ensure that we do not uiomove bad
553 		 * or uninitialized data.
554 		 */
555 		size -= bp->b_resid;
556 		if (size < xfersize) {
557 			if (size == 0)
558 				break;
559 			xfersize = size;
560 		}
561 
562 		error = uiomove((char *)bp->b_data + blkoffset,
563 		    (int)xfersize, uio);
564 		if (error)
565 			break;
566 
567 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
568 		   (LIST_EMPTY(&bp->b_dep))) {
569 			/*
570 			 * If there are no dependencies, and it's VMIO,
571 			 * then we don't need the buf, mark it available
572 			 * for freeing. The VM has the data.
573 			 */
574 			bp->b_flags |= B_RELBUF;
575 			brelse(bp);
576 		} else {
577 			/*
578 			 * Otherwise let whoever
579 			 * made the request take care of
580 			 * freeing it. We just queue
581 			 * it onto another list.
582 			 */
583 			bqrelse(bp);
584 		}
585 	}
586 
587 	/*
588 	 * This can only happen in the case of an error
589 	 * because the loop above resets bp to NULL on each iteration
590 	 * and on normal completion has not set a new value into it.
591 	 * so it must have come from a 'break' statement
592 	 */
593 	if (bp != NULL) {
594 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
595 		   (LIST_EMPTY(&bp->b_dep))) {
596 			bp->b_flags |= B_RELBUF;
597 			brelse(bp);
598 		} else {
599 			bqrelse(bp);
600 		}
601 	}
602 
603 	if ((error == 0 || uio->uio_resid != orig_resid) &&
604 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0) {
605 		VI_LOCK(vp);
606 		ip->i_flag |= IN_ACCESS;
607 		VI_UNLOCK(vp);
608 	}
609 	return (error);
610 }
611 
612 /*
613  * Vnode op for writing.
614  */
615 static int
616 ffs_write(ap)
617 	struct vop_write_args /* {
618 		struct vnode *a_vp;
619 		struct uio *a_uio;
620 		int a_ioflag;
621 		struct ucred *a_cred;
622 	} */ *ap;
623 {
624 	struct vnode *vp;
625 	struct uio *uio;
626 	struct inode *ip;
627 	struct fs *fs;
628 	struct buf *bp;
629 	struct thread *td;
630 	ufs_lbn_t lbn;
631 	off_t osize;
632 	int seqcount;
633 	int blkoffset, error, flags, ioflag, resid, size, xfersize;
634 
635 	vp = ap->a_vp;
636 	uio = ap->a_uio;
637 	ioflag = ap->a_ioflag;
638 	if (ap->a_ioflag & IO_EXT)
639 #ifdef notyet
640 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
641 #else
642 		panic("ffs_write+IO_EXT");
643 #endif
644 
645 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
646 	ip = VTOI(vp);
647 
648 #ifdef DIAGNOSTIC
649 	if (uio->uio_rw != UIO_WRITE)
650 		panic("ffs_write: mode");
651 #endif
652 
653 	switch (vp->v_type) {
654 	case VREG:
655 		if (ioflag & IO_APPEND)
656 			uio->uio_offset = ip->i_size;
657 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
658 			return (EPERM);
659 		/* FALLTHROUGH */
660 	case VLNK:
661 		break;
662 	case VDIR:
663 		panic("ffs_write: dir write");
664 		break;
665 	default:
666 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
667 			(int)uio->uio_offset,
668 			(int)uio->uio_resid
669 		);
670 	}
671 
672 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
673 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
674 	fs = ip->i_fs;
675 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
676 		return (EFBIG);
677 	/*
678 	 * Maybe this should be above the vnode op call, but so long as
679 	 * file servers have no limits, I don't think it matters.
680 	 */
681 	td = uio->uio_td;
682 	if (vp->v_type == VREG && td != NULL) {
683 		PROC_LOCK(td->td_proc);
684 		if (uio->uio_offset + uio->uio_resid >
685 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
686 			psignal(td->td_proc, SIGXFSZ);
687 			PROC_UNLOCK(td->td_proc);
688 			return (EFBIG);
689 		}
690 		PROC_UNLOCK(td->td_proc);
691 	}
692 
693 	resid = uio->uio_resid;
694 	osize = ip->i_size;
695 	if (seqcount > BA_SEQMAX)
696 		flags = BA_SEQMAX << BA_SEQSHIFT;
697 	else
698 		flags = seqcount << BA_SEQSHIFT;
699 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
700 		flags |= IO_SYNC;
701 
702 	for (error = 0; uio->uio_resid > 0;) {
703 		lbn = lblkno(fs, uio->uio_offset);
704 		blkoffset = blkoff(fs, uio->uio_offset);
705 		xfersize = fs->fs_bsize - blkoffset;
706 		if (uio->uio_resid < xfersize)
707 			xfersize = uio->uio_resid;
708 		if (uio->uio_offset + xfersize > ip->i_size)
709 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
710 
711                 /*
712 		 * We must perform a read-before-write if the transfer size
713 		 * does not cover the entire buffer.
714                  */
715 		if (fs->fs_bsize > xfersize)
716 			flags |= BA_CLRBUF;
717 		else
718 			flags &= ~BA_CLRBUF;
719 /* XXX is uio->uio_offset the right thing here? */
720 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
721 		    ap->a_cred, flags, &bp);
722 		if (error != 0)
723 			break;
724 		/*
725 		 * If the buffer is not valid we have to clear out any
726 		 * garbage data from the pages instantiated for the buffer.
727 		 * If we do not, a failed uiomove() during a write can leave
728 		 * the prior contents of the pages exposed to a userland
729 		 * mmap().  XXX deal with uiomove() errors a better way.
730 		 */
731 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
732 			vfs_bio_clrbuf(bp);
733 		if (ioflag & IO_DIRECT)
734 			bp->b_flags |= B_DIRECT;
735 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
736 			bp->b_flags |= B_NOCACHE;
737 
738 		if (uio->uio_offset + xfersize > ip->i_size) {
739 			ip->i_size = uio->uio_offset + xfersize;
740 			DIP_SET(ip, i_size, ip->i_size);
741 		}
742 
743 		size = blksize(fs, ip, lbn) - bp->b_resid;
744 		if (size < xfersize)
745 			xfersize = size;
746 
747 		error =
748 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
749 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
750 		   (LIST_EMPTY(&bp->b_dep))) {
751 			bp->b_flags |= B_RELBUF;
752 		}
753 
754 		/*
755 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
756 		 * if we have a severe page deficiency write the buffer
757 		 * asynchronously.  Otherwise try to cluster, and if that
758 		 * doesn't do it then either do an async write (if O_DIRECT),
759 		 * or a delayed write (if not).
760 		 */
761 		if (ioflag & IO_SYNC) {
762 			(void)bwrite(bp);
763 		} else if (vm_page_count_severe() ||
764 			    buf_dirty_count_severe() ||
765 			    (ioflag & IO_ASYNC)) {
766 			bp->b_flags |= B_CLUSTEROK;
767 			bawrite(bp);
768 		} else if (xfersize + blkoffset == fs->fs_bsize) {
769 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
770 				bp->b_flags |= B_CLUSTEROK;
771 				cluster_write(vp, bp, ip->i_size, seqcount);
772 			} else {
773 				bawrite(bp);
774 			}
775 		} else if (ioflag & IO_DIRECT) {
776 			bp->b_flags |= B_CLUSTEROK;
777 			bawrite(bp);
778 		} else {
779 			bp->b_flags |= B_CLUSTEROK;
780 			bdwrite(bp);
781 		}
782 		if (error || xfersize == 0)
783 			break;
784 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
785 	}
786 	/*
787 	 * If we successfully wrote any data, and we are not the superuser
788 	 * we clear the setuid and setgid bits as a precaution against
789 	 * tampering.
790 	 */
791 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
792 	    ap->a_cred) {
793 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
794 			ip->i_mode &= ~(ISUID | ISGID);
795 			DIP_SET(ip, i_mode, ip->i_mode);
796 		}
797 	}
798 	if (error) {
799 		if (ioflag & IO_UNIT) {
800 			(void)ffs_truncate(vp, osize,
801 			    IO_NORMAL | (ioflag & IO_SYNC),
802 			    ap->a_cred, uio->uio_td);
803 			uio->uio_offset -= resid - uio->uio_resid;
804 			uio->uio_resid = resid;
805 		}
806 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
807 		error = ffs_update(vp, 1);
808 	return (error);
809 }
810 
811 /*
812  * get page routine
813  */
814 static int
815 ffs_getpages(ap)
816 	struct vop_getpages_args *ap;
817 {
818 	int i;
819 	vm_page_t mreq;
820 	int pcount;
821 
822 	pcount = round_page(ap->a_count) / PAGE_SIZE;
823 	mreq = ap->a_m[ap->a_reqpage];
824 
825 	/*
826 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
827 	 * then the entire page is valid.  Since the page may be mapped,
828 	 * user programs might reference data beyond the actual end of file
829 	 * occuring within the page.  We have to zero that data.
830 	 */
831 	VM_OBJECT_LOCK(mreq->object);
832 	if (mreq->valid) {
833 		if (mreq->valid != VM_PAGE_BITS_ALL)
834 			vm_page_zero_invalid(mreq, TRUE);
835 		vm_page_lock_queues();
836 		for (i = 0; i < pcount; i++) {
837 			if (i != ap->a_reqpage) {
838 				vm_page_free(ap->a_m[i]);
839 			}
840 		}
841 		vm_page_unlock_queues();
842 		VM_OBJECT_UNLOCK(mreq->object);
843 		return VM_PAGER_OK;
844 	}
845 	VM_OBJECT_UNLOCK(mreq->object);
846 
847 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
848 					    ap->a_count,
849 					    ap->a_reqpage);
850 }
851 
852 
853 /*
854  * Extended attribute area reading.
855  */
856 static int
857 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
858 {
859 	struct inode *ip;
860 	struct ufs2_dinode *dp;
861 	struct fs *fs;
862 	struct buf *bp;
863 	ufs_lbn_t lbn, nextlbn;
864 	off_t bytesinfile;
865 	long size, xfersize, blkoffset;
866 	int error, orig_resid;
867 
868 	ip = VTOI(vp);
869 	fs = ip->i_fs;
870 	dp = ip->i_din2;
871 
872 #ifdef DIAGNOSTIC
873 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
874 		panic("ffs_extread: mode");
875 
876 #endif
877 	orig_resid = uio->uio_resid;
878 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
879 	if (orig_resid == 0)
880 		return (0);
881 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
882 
883 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
884 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
885 			break;
886 		lbn = lblkno(fs, uio->uio_offset);
887 		nextlbn = lbn + 1;
888 
889 		/*
890 		 * size of buffer.  The buffer representing the
891 		 * end of the file is rounded up to the size of
892 		 * the block type ( fragment or full block,
893 		 * depending ).
894 		 */
895 		size = sblksize(fs, dp->di_extsize, lbn);
896 		blkoffset = blkoff(fs, uio->uio_offset);
897 
898 		/*
899 		 * The amount we want to transfer in this iteration is
900 		 * one FS block less the amount of the data before
901 		 * our startpoint (duh!)
902 		 */
903 		xfersize = fs->fs_bsize - blkoffset;
904 
905 		/*
906 		 * But if we actually want less than the block,
907 		 * or the file doesn't have a whole block more of data,
908 		 * then use the lesser number.
909 		 */
910 		if (uio->uio_resid < xfersize)
911 			xfersize = uio->uio_resid;
912 		if (bytesinfile < xfersize)
913 			xfersize = bytesinfile;
914 
915 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
916 			/*
917 			 * Don't do readahead if this is the end of the info.
918 			 */
919 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
920 		} else {
921 			/*
922 			 * If we have a second block, then
923 			 * fire off a request for a readahead
924 			 * as well as a read. Note that the 4th and 5th
925 			 * arguments point to arrays of the size specified in
926 			 * the 6th argument.
927 			 */
928 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
929 
930 			nextlbn = -1 - nextlbn;
931 			error = breadn(vp, -1 - lbn,
932 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
933 		}
934 		if (error) {
935 			brelse(bp);
936 			bp = NULL;
937 			break;
938 		}
939 
940 		/*
941 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
942 		 * will cause us to attempt to release the buffer later on
943 		 * and will cause the buffer cache to attempt to free the
944 		 * underlying pages.
945 		 */
946 		if (ioflag & IO_DIRECT)
947 			bp->b_flags |= B_DIRECT;
948 
949 		/*
950 		 * We should only get non-zero b_resid when an I/O error
951 		 * has occurred, which should cause us to break above.
952 		 * However, if the short read did not cause an error,
953 		 * then we want to ensure that we do not uiomove bad
954 		 * or uninitialized data.
955 		 */
956 		size -= bp->b_resid;
957 		if (size < xfersize) {
958 			if (size == 0)
959 				break;
960 			xfersize = size;
961 		}
962 
963 		error = uiomove((char *)bp->b_data + blkoffset,
964 					(int)xfersize, uio);
965 		if (error)
966 			break;
967 
968 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
969 		   (LIST_EMPTY(&bp->b_dep))) {
970 			/*
971 			 * If there are no dependencies, and it's VMIO,
972 			 * then we don't need the buf, mark it available
973 			 * for freeing. The VM has the data.
974 			 */
975 			bp->b_flags |= B_RELBUF;
976 			brelse(bp);
977 		} else {
978 			/*
979 			 * Otherwise let whoever
980 			 * made the request take care of
981 			 * freeing it. We just queue
982 			 * it onto another list.
983 			 */
984 			bqrelse(bp);
985 		}
986 	}
987 
988 	/*
989 	 * This can only happen in the case of an error
990 	 * because the loop above resets bp to NULL on each iteration
991 	 * and on normal completion has not set a new value into it.
992 	 * so it must have come from a 'break' statement
993 	 */
994 	if (bp != NULL) {
995 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
996 		   (LIST_EMPTY(&bp->b_dep))) {
997 			bp->b_flags |= B_RELBUF;
998 			brelse(bp);
999 		} else {
1000 			bqrelse(bp);
1001 		}
1002 	}
1003 
1004 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1005 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0) {
1006 		VI_LOCK(vp);
1007 		ip->i_flag |= IN_ACCESS;
1008 		VI_UNLOCK(vp);
1009 	}
1010 	return (error);
1011 }
1012 
1013 /*
1014  * Extended attribute area writing.
1015  */
1016 static int
1017 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1018 {
1019 	struct inode *ip;
1020 	struct ufs2_dinode *dp;
1021 	struct fs *fs;
1022 	struct buf *bp;
1023 	ufs_lbn_t lbn;
1024 	off_t osize;
1025 	int blkoffset, error, flags, resid, size, xfersize;
1026 
1027 	ip = VTOI(vp);
1028 	fs = ip->i_fs;
1029 	dp = ip->i_din2;
1030 
1031 	KASSERT(!(ip->i_flag & IN_SPACECOUNTED), ("inode %u: inode is dead",
1032 	    ip->i_number));
1033 
1034 #ifdef DIAGNOSTIC
1035 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1036 		panic("ffs_extwrite: mode");
1037 #endif
1038 
1039 	if (ioflag & IO_APPEND)
1040 		uio->uio_offset = dp->di_extsize;
1041 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1042 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1043 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1044 		return (EFBIG);
1045 
1046 	resid = uio->uio_resid;
1047 	osize = dp->di_extsize;
1048 	flags = IO_EXT;
1049 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1050 		flags |= IO_SYNC;
1051 
1052 	for (error = 0; uio->uio_resid > 0;) {
1053 		lbn = lblkno(fs, uio->uio_offset);
1054 		blkoffset = blkoff(fs, uio->uio_offset);
1055 		xfersize = fs->fs_bsize - blkoffset;
1056 		if (uio->uio_resid < xfersize)
1057 			xfersize = uio->uio_resid;
1058 
1059                 /*
1060 		 * We must perform a read-before-write if the transfer size
1061 		 * does not cover the entire buffer.
1062                  */
1063 		if (fs->fs_bsize > xfersize)
1064 			flags |= BA_CLRBUF;
1065 		else
1066 			flags &= ~BA_CLRBUF;
1067 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1068 		    ucred, flags, &bp);
1069 		if (error != 0)
1070 			break;
1071 		/*
1072 		 * If the buffer is not valid we have to clear out any
1073 		 * garbage data from the pages instantiated for the buffer.
1074 		 * If we do not, a failed uiomove() during a write can leave
1075 		 * the prior contents of the pages exposed to a userland
1076 		 * mmap().  XXX deal with uiomove() errors a better way.
1077 		 */
1078 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1079 			vfs_bio_clrbuf(bp);
1080 		if (ioflag & IO_DIRECT)
1081 			bp->b_flags |= B_DIRECT;
1082 
1083 		if (uio->uio_offset + xfersize > dp->di_extsize)
1084 			dp->di_extsize = uio->uio_offset + xfersize;
1085 
1086 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1087 		if (size < xfersize)
1088 			xfersize = size;
1089 
1090 		error =
1091 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1092 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1093 		   (LIST_EMPTY(&bp->b_dep))) {
1094 			bp->b_flags |= B_RELBUF;
1095 		}
1096 
1097 		/*
1098 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1099 		 * if we have a severe page deficiency write the buffer
1100 		 * asynchronously.  Otherwise try to cluster, and if that
1101 		 * doesn't do it then either do an async write (if O_DIRECT),
1102 		 * or a delayed write (if not).
1103 		 */
1104 		if (ioflag & IO_SYNC) {
1105 			(void)bwrite(bp);
1106 		} else if (vm_page_count_severe() ||
1107 			    buf_dirty_count_severe() ||
1108 			    xfersize + blkoffset == fs->fs_bsize ||
1109 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1110 			bawrite(bp);
1111 		else
1112 			bdwrite(bp);
1113 		if (error || xfersize == 0)
1114 			break;
1115 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1116 	}
1117 	/*
1118 	 * If we successfully wrote any data, and we are not the superuser
1119 	 * we clear the setuid and setgid bits as a precaution against
1120 	 * tampering.
1121 	 */
1122 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1123 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1124 			ip->i_mode &= ~(ISUID | ISGID);
1125 			dp->di_mode = ip->i_mode;
1126 		}
1127 	}
1128 	if (error) {
1129 		if (ioflag & IO_UNIT) {
1130 			(void)ffs_truncate(vp, osize,
1131 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1132 			uio->uio_offset -= resid - uio->uio_resid;
1133 			uio->uio_resid = resid;
1134 		}
1135 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1136 		error = ffs_update(vp, 1);
1137 	return (error);
1138 }
1139 
1140 
1141 /*
1142  * Vnode operating to retrieve a named extended attribute.
1143  *
1144  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1145  * the length of the EA, and possibly the pointer to the entry and to the data.
1146  */
1147 static int
1148 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1149 {
1150 	u_char *p, *pe, *pn, *p0;
1151 	int eapad1, eapad2, ealength, ealen, nlen;
1152 	uint32_t ul;
1153 
1154 	pe = ptr + length;
1155 	nlen = strlen(name);
1156 
1157 	for (p = ptr; p < pe; p = pn) {
1158 		p0 = p;
1159 		bcopy(p, &ul, sizeof(ul));
1160 		pn = p + ul;
1161 		/* make sure this entry is complete */
1162 		if (pn > pe)
1163 			break;
1164 		p += sizeof(uint32_t);
1165 		if (*p != nspace)
1166 			continue;
1167 		p++;
1168 		eapad2 = *p++;
1169 		if (*p != nlen)
1170 			continue;
1171 		p++;
1172 		if (bcmp(p, name, nlen))
1173 			continue;
1174 		ealength = sizeof(uint32_t) + 3 + nlen;
1175 		eapad1 = 8 - (ealength % 8);
1176 		if (eapad1 == 8)
1177 			eapad1 = 0;
1178 		ealength += eapad1;
1179 		ealen = ul - ealength - eapad2;
1180 		p += nlen + eapad1;
1181 		if (eap != NULL)
1182 			*eap = p0;
1183 		if (eac != NULL)
1184 			*eac = p;
1185 		return (ealen);
1186 	}
1187 	return(-1);
1188 }
1189 
1190 static int
1191 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1192 {
1193 	struct inode *ip;
1194 	struct ufs2_dinode *dp;
1195 	struct fs *fs;
1196 	struct uio luio;
1197 	struct iovec liovec;
1198 	int easize, error;
1199 	u_char *eae;
1200 
1201 	ip = VTOI(vp);
1202 	fs = ip->i_fs;
1203 	dp = ip->i_din2;
1204 	easize = dp->di_extsize;
1205 	if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1206 		return (EFBIG);
1207 
1208 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1209 
1210 	liovec.iov_base = eae;
1211 	liovec.iov_len = easize;
1212 	luio.uio_iov = &liovec;
1213 	luio.uio_iovcnt = 1;
1214 	luio.uio_offset = 0;
1215 	luio.uio_resid = easize;
1216 	luio.uio_segflg = UIO_SYSSPACE;
1217 	luio.uio_rw = UIO_READ;
1218 	luio.uio_td = td;
1219 
1220 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1221 	if (error) {
1222 		free(eae, M_TEMP);
1223 		return(error);
1224 	}
1225 	*p = eae;
1226 	return (0);
1227 }
1228 
1229 static int
1230 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1231 {
1232 	struct inode *ip;
1233 	struct ufs2_dinode *dp;
1234 	int error;
1235 
1236 	ip = VTOI(vp);
1237 
1238 	if (ip->i_ea_area != NULL)
1239 		return (EBUSY);
1240 	dp = ip->i_din2;
1241 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1242 	if (error)
1243 		return (error);
1244 	ip->i_ea_len = dp->di_extsize;
1245 	ip->i_ea_error = 0;
1246 	return (0);
1247 }
1248 
1249 /*
1250  * Vnode extattr transaction commit/abort
1251  */
1252 static int
1253 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1254 {
1255 	struct inode *ip;
1256 	struct uio luio;
1257 	struct iovec liovec;
1258 	int error;
1259 	struct ufs2_dinode *dp;
1260 
1261 	ip = VTOI(vp);
1262 	if (ip->i_ea_area == NULL)
1263 		return (EINVAL);
1264 	dp = ip->i_din2;
1265 	error = ip->i_ea_error;
1266 	if (commit && error == 0) {
1267 		if (cred == NOCRED)
1268 			cred =  vp->v_mount->mnt_cred;
1269 		liovec.iov_base = ip->i_ea_area;
1270 		liovec.iov_len = ip->i_ea_len;
1271 		luio.uio_iov = &liovec;
1272 		luio.uio_iovcnt = 1;
1273 		luio.uio_offset = 0;
1274 		luio.uio_resid = ip->i_ea_len;
1275 		luio.uio_segflg = UIO_SYSSPACE;
1276 		luio.uio_rw = UIO_WRITE;
1277 		luio.uio_td = td;
1278 		/* XXX: I'm not happy about truncating to zero size */
1279 		if (ip->i_ea_len < dp->di_extsize)
1280 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1281 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1282 	}
1283 	free(ip->i_ea_area, M_TEMP);
1284 	ip->i_ea_area = NULL;
1285 	ip->i_ea_len = 0;
1286 	ip->i_ea_error = 0;
1287 	return (error);
1288 }
1289 
1290 /*
1291  * Vnode extattr strategy routine for fifos.
1292  *
1293  * We need to check for a read or write of the external attributes.
1294  * Otherwise we just fall through and do the usual thing.
1295  */
1296 static int
1297 ffsext_strategy(struct vop_strategy_args *ap)
1298 /*
1299 struct vop_strategy_args {
1300 	struct vnodeop_desc *a_desc;
1301 	struct vnode *a_vp;
1302 	struct buf *a_bp;
1303 };
1304 */
1305 {
1306 	struct vnode *vp;
1307 	daddr_t lbn;
1308 
1309 	vp = ap->a_vp;
1310 	lbn = ap->a_bp->b_lblkno;
1311 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1312 	    lbn < 0 && lbn >= -NXADDR)
1313 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1314 	if (vp->v_type == VFIFO)
1315 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1316 	panic("spec nodes went here");
1317 }
1318 
1319 /*
1320  * Vnode extattr transaction commit/abort
1321  */
1322 static int
1323 ffs_openextattr(struct vop_openextattr_args *ap)
1324 /*
1325 struct vop_openextattr_args {
1326 	struct vnodeop_desc *a_desc;
1327 	struct vnode *a_vp;
1328 	IN struct ucred *a_cred;
1329 	IN struct thread *a_td;
1330 };
1331 */
1332 {
1333 	struct inode *ip;
1334 	struct fs *fs;
1335 
1336 	ip = VTOI(ap->a_vp);
1337 	fs = ip->i_fs;
1338 
1339 	if (ap->a_vp->v_type == VCHR)
1340 		return (EOPNOTSUPP);
1341 
1342 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1343 }
1344 
1345 
1346 /*
1347  * Vnode extattr transaction commit/abort
1348  */
1349 static int
1350 ffs_closeextattr(struct vop_closeextattr_args *ap)
1351 /*
1352 struct vop_closeextattr_args {
1353 	struct vnodeop_desc *a_desc;
1354 	struct vnode *a_vp;
1355 	int a_commit;
1356 	IN struct ucred *a_cred;
1357 	IN struct thread *a_td;
1358 };
1359 */
1360 {
1361 	struct inode *ip;
1362 	struct fs *fs;
1363 
1364 	ip = VTOI(ap->a_vp);
1365 	fs = ip->i_fs;
1366 
1367 	if (ap->a_vp->v_type == VCHR)
1368 		return (EOPNOTSUPP);
1369 
1370 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1371 		return (EROFS);
1372 
1373 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1374 }
1375 
1376 /*
1377  * Vnode operation to remove a named attribute.
1378  */
1379 static int
1380 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1381 /*
1382 vop_deleteextattr {
1383 	IN struct vnode *a_vp;
1384 	IN int a_attrnamespace;
1385 	IN const char *a_name;
1386 	IN struct ucred *a_cred;
1387 	IN struct thread *a_td;
1388 };
1389 */
1390 {
1391 	struct inode *ip;
1392 	struct fs *fs;
1393 	uint32_t ealength, ul;
1394 	int ealen, olen, eapad1, eapad2, error, i, easize;
1395 	u_char *eae, *p;
1396 	int stand_alone;
1397 
1398 	ip = VTOI(ap->a_vp);
1399 	fs = ip->i_fs;
1400 
1401 	if (ap->a_vp->v_type == VCHR)
1402 		return (EOPNOTSUPP);
1403 
1404 	if (strlen(ap->a_name) == 0)
1405 		return (EINVAL);
1406 
1407 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1408 		return (EROFS);
1409 
1410 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1411 	    ap->a_cred, ap->a_td, IWRITE);
1412 	if (error) {
1413 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1414 			ip->i_ea_error = error;
1415 		return (error);
1416 	}
1417 
1418 	if (ip->i_ea_area == NULL) {
1419 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1420 		if (error)
1421 			return (error);
1422 		stand_alone = 1;
1423 	} else {
1424 		stand_alone = 0;
1425 	}
1426 
1427 	ealength = eapad1 = ealen = eapad2 = 0;
1428 
1429 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1430 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1431 	easize = ip->i_ea_len;
1432 
1433 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1434 	    &p, NULL);
1435 	if (olen == -1) {
1436 		/* delete but nonexistent */
1437 		free(eae, M_TEMP);
1438 		if (stand_alone)
1439 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1440 		return(ENOATTR);
1441 	}
1442 	bcopy(p, &ul, sizeof ul);
1443 	i = p - eae + ul;
1444 	if (ul != ealength) {
1445 		bcopy(p + ul, p + ealength, easize - i);
1446 		easize += (ealength - ul);
1447 	}
1448 	if (easize > NXADDR * fs->fs_bsize) {
1449 		free(eae, M_TEMP);
1450 		if (stand_alone)
1451 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1452 		else if (ip->i_ea_error == 0)
1453 			ip->i_ea_error = ENOSPC;
1454 		return(ENOSPC);
1455 	}
1456 	p = ip->i_ea_area;
1457 	ip->i_ea_area = eae;
1458 	ip->i_ea_len = easize;
1459 	free(p, M_TEMP);
1460 	if (stand_alone)
1461 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1462 	return(error);
1463 }
1464 
1465 /*
1466  * Vnode operation to retrieve a named extended attribute.
1467  */
1468 static int
1469 ffs_getextattr(struct vop_getextattr_args *ap)
1470 /*
1471 vop_getextattr {
1472 	IN struct vnode *a_vp;
1473 	IN int a_attrnamespace;
1474 	IN const char *a_name;
1475 	INOUT struct uio *a_uio;
1476 	OUT size_t *a_size;
1477 	IN struct ucred *a_cred;
1478 	IN struct thread *a_td;
1479 };
1480 */
1481 {
1482 	struct inode *ip;
1483 	struct fs *fs;
1484 	u_char *eae, *p;
1485 	unsigned easize;
1486 	int error, ealen, stand_alone;
1487 
1488 	ip = VTOI(ap->a_vp);
1489 	fs = ip->i_fs;
1490 
1491 	if (ap->a_vp->v_type == VCHR)
1492 		return (EOPNOTSUPP);
1493 
1494 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1495 	    ap->a_cred, ap->a_td, IREAD);
1496 	if (error)
1497 		return (error);
1498 
1499 	if (ip->i_ea_area == NULL) {
1500 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1501 		if (error)
1502 			return (error);
1503 		stand_alone = 1;
1504 	} else {
1505 		stand_alone = 0;
1506 	}
1507 	eae = ip->i_ea_area;
1508 	easize = ip->i_ea_len;
1509 
1510 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1511 	    NULL, &p);
1512 	if (ealen >= 0) {
1513 		error = 0;
1514 		if (ap->a_size != NULL)
1515 			*ap->a_size = ealen;
1516 		else if (ap->a_uio != NULL)
1517 			error = uiomove(p, ealen, ap->a_uio);
1518 	} else
1519 		error = ENOATTR;
1520 	if (stand_alone)
1521 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1522 	return(error);
1523 }
1524 
1525 /*
1526  * Vnode operation to retrieve extended attributes on a vnode.
1527  */
1528 static int
1529 ffs_listextattr(struct vop_listextattr_args *ap)
1530 /*
1531 vop_listextattr {
1532 	IN struct vnode *a_vp;
1533 	IN int a_attrnamespace;
1534 	INOUT struct uio *a_uio;
1535 	OUT size_t *a_size;
1536 	IN struct ucred *a_cred;
1537 	IN struct thread *a_td;
1538 };
1539 */
1540 {
1541 	struct inode *ip;
1542 	struct fs *fs;
1543 	u_char *eae, *p, *pe, *pn;
1544 	unsigned easize;
1545 	uint32_t ul;
1546 	int error, ealen, stand_alone;
1547 
1548 	ip = VTOI(ap->a_vp);
1549 	fs = ip->i_fs;
1550 
1551 	if (ap->a_vp->v_type == VCHR)
1552 		return (EOPNOTSUPP);
1553 
1554 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1555 	    ap->a_cred, ap->a_td, IREAD);
1556 	if (error)
1557 		return (error);
1558 
1559 	if (ip->i_ea_area == NULL) {
1560 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1561 		if (error)
1562 			return (error);
1563 		stand_alone = 1;
1564 	} else {
1565 		stand_alone = 0;
1566 	}
1567 	eae = ip->i_ea_area;
1568 	easize = ip->i_ea_len;
1569 
1570 	error = 0;
1571 	if (ap->a_size != NULL)
1572 		*ap->a_size = 0;
1573 	pe = eae + easize;
1574 	for(p = eae; error == 0 && p < pe; p = pn) {
1575 		bcopy(p, &ul, sizeof(ul));
1576 		pn = p + ul;
1577 		if (pn > pe)
1578 			break;
1579 		p += sizeof(ul);
1580 		if (*p++ != ap->a_attrnamespace)
1581 			continue;
1582 		p++;	/* pad2 */
1583 		ealen = *p;
1584 		if (ap->a_size != NULL) {
1585 			*ap->a_size += ealen + 1;
1586 		} else if (ap->a_uio != NULL) {
1587 			error = uiomove(p, ealen + 1, ap->a_uio);
1588 		}
1589 	}
1590 	if (stand_alone)
1591 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1592 	return(error);
1593 }
1594 
1595 /*
1596  * Vnode operation to set a named attribute.
1597  */
1598 static int
1599 ffs_setextattr(struct vop_setextattr_args *ap)
1600 /*
1601 vop_setextattr {
1602 	IN struct vnode *a_vp;
1603 	IN int a_attrnamespace;
1604 	IN const char *a_name;
1605 	INOUT struct uio *a_uio;
1606 	IN struct ucred *a_cred;
1607 	IN struct thread *a_td;
1608 };
1609 */
1610 {
1611 	struct inode *ip;
1612 	struct fs *fs;
1613 	uint32_t ealength, ul;
1614 	int ealen, olen, eapad1, eapad2, error, i, easize;
1615 	u_char *eae, *p;
1616 	int stand_alone;
1617 
1618 	ip = VTOI(ap->a_vp);
1619 	fs = ip->i_fs;
1620 
1621 	if (ap->a_vp->v_type == VCHR)
1622 		return (EOPNOTSUPP);
1623 
1624 	if (strlen(ap->a_name) == 0)
1625 		return (EINVAL);
1626 
1627 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1628 	if (ap->a_uio == NULL)
1629 		return (EOPNOTSUPP);
1630 
1631 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1632 		return (EROFS);
1633 
1634 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1635 	    ap->a_cred, ap->a_td, IWRITE);
1636 	if (error) {
1637 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1638 			ip->i_ea_error = error;
1639 		return (error);
1640 	}
1641 
1642 	if (ip->i_ea_area == NULL) {
1643 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1644 		if (error)
1645 			return (error);
1646 		stand_alone = 1;
1647 	} else {
1648 		stand_alone = 0;
1649 	}
1650 
1651 	ealen = ap->a_uio->uio_resid;
1652 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1653 	eapad1 = 8 - (ealength % 8);
1654 	if (eapad1 == 8)
1655 		eapad1 = 0;
1656 	eapad2 = 8 - (ealen % 8);
1657 	if (eapad2 == 8)
1658 		eapad2 = 0;
1659 	ealength += eapad1 + ealen + eapad2;
1660 
1661 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1662 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1663 	easize = ip->i_ea_len;
1664 
1665 	olen = ffs_findextattr(eae, easize,
1666 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1667         if (olen == -1) {
1668 		/* new, append at end */
1669 		p = eae + easize;
1670 		easize += ealength;
1671 	} else {
1672 		bcopy(p, &ul, sizeof ul);
1673 		i = p - eae + ul;
1674 		if (ul != ealength) {
1675 			bcopy(p + ul, p + ealength, easize - i);
1676 			easize += (ealength - ul);
1677 		}
1678 	}
1679 	if (easize > NXADDR * fs->fs_bsize) {
1680 		free(eae, M_TEMP);
1681 		if (stand_alone)
1682 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1683 		else if (ip->i_ea_error == 0)
1684 			ip->i_ea_error = ENOSPC;
1685 		return(ENOSPC);
1686 	}
1687 	bcopy(&ealength, p, sizeof(ealength));
1688 	p += sizeof(ealength);
1689 	*p++ = ap->a_attrnamespace;
1690 	*p++ = eapad2;
1691 	*p++ = strlen(ap->a_name);
1692 	strcpy(p, ap->a_name);
1693 	p += strlen(ap->a_name);
1694 	bzero(p, eapad1);
1695 	p += eapad1;
1696 	error = uiomove(p, ealen, ap->a_uio);
1697 	if (error) {
1698 		free(eae, M_TEMP);
1699 		if (stand_alone)
1700 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1701 		else if (ip->i_ea_error == 0)
1702 			ip->i_ea_error = error;
1703 		return(error);
1704 	}
1705 	p += ealen;
1706 	bzero(p, eapad2);
1707 
1708 	p = ip->i_ea_area;
1709 	ip->i_ea_area = eae;
1710 	ip->i_ea_len = easize;
1711 	free(p, M_TEMP);
1712 	if (stand_alone)
1713 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1714 	return(error);
1715 }
1716 
1717 /*
1718  * Vnode pointer to File handle
1719  */
1720 static int
1721 ffs_vptofh(struct vop_vptofh_args *ap)
1722 /*
1723 vop_vptofh {
1724 	IN struct vnode *a_vp;
1725 	IN struct fid *a_fhp;
1726 };
1727 */
1728 {
1729 	struct inode *ip;
1730 	struct ufid *ufhp;
1731 
1732 	ip = VTOI(ap->a_vp);
1733 	ufhp = (struct ufid *)ap->a_fhp;
1734 	ufhp->ufid_len = sizeof(struct ufid);
1735 	ufhp->ufid_ino = ip->i_number;
1736 	ufhp->ufid_gen = ip->i_gen;
1737 	return (0);
1738 }
1739