xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 87d5d10d7d7af72e5529b1cd684f8da41664183b)
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/rwlock.h>
79 #include <sys/stat.h>
80 #include <sys/vmmeter.h>
81 #include <sys/vnode.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_param.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vnode_pager.h>
90 
91 #include <ufs/ufs/extattr.h>
92 #include <ufs/ufs/quota.h>
93 #include <ufs/ufs/inode.h>
94 #include <ufs/ufs/ufs_extern.h>
95 #include <ufs/ufs/ufsmount.h>
96 
97 #include <ufs/ffs/fs.h>
98 #include <ufs/ffs/ffs_extern.h>
99 #include "opt_directio.h"
100 #include "opt_ffs.h"
101 
102 #ifdef DIRECTIO
103 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
104 #endif
105 static vop_fsync_t	ffs_fsync;
106 static vop_fdatasync_t	ffs_fdatasync;
107 static vop_lock1_t	ffs_lock;
108 static vop_read_t	ffs_read;
109 static vop_write_t	ffs_write;
110 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
111 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
112 		    struct ucred *cred);
113 static vop_strategy_t	ffsext_strategy;
114 static vop_closeextattr_t	ffs_closeextattr;
115 static vop_deleteextattr_t	ffs_deleteextattr;
116 static vop_getextattr_t	ffs_getextattr;
117 static vop_listextattr_t	ffs_listextattr;
118 static vop_openextattr_t	ffs_openextattr;
119 static vop_setextattr_t	ffs_setextattr;
120 static vop_vptofh_t	ffs_vptofh;
121 
122 
123 /* Global vfs data structures for ufs. */
124 struct vop_vector ffs_vnodeops1 = {
125 	.vop_default =		&ufs_vnodeops,
126 	.vop_fsync =		ffs_fsync,
127 	.vop_fdatasync =	ffs_fdatasync,
128 	.vop_getpages =		vnode_pager_local_getpages,
129 	.vop_getpages_async =	vnode_pager_local_getpages_async,
130 	.vop_lock1 =		ffs_lock,
131 	.vop_read =		ffs_read,
132 	.vop_reallocblks =	ffs_reallocblks,
133 	.vop_write =		ffs_write,
134 	.vop_vptofh =		ffs_vptofh,
135 };
136 
137 struct vop_vector ffs_fifoops1 = {
138 	.vop_default =		&ufs_fifoops,
139 	.vop_fsync =		ffs_fsync,
140 	.vop_fdatasync =	ffs_fdatasync,
141 	.vop_reallocblks =	ffs_reallocblks, /* XXX: really ??? */
142 	.vop_vptofh =		ffs_vptofh,
143 };
144 
145 /* Global vfs data structures for ufs. */
146 struct vop_vector ffs_vnodeops2 = {
147 	.vop_default =		&ufs_vnodeops,
148 	.vop_fsync =		ffs_fsync,
149 	.vop_fdatasync =	ffs_fdatasync,
150 	.vop_getpages =		vnode_pager_local_getpages,
151 	.vop_getpages_async =	vnode_pager_local_getpages_async,
152 	.vop_lock1 =		ffs_lock,
153 	.vop_read =		ffs_read,
154 	.vop_reallocblks =	ffs_reallocblks,
155 	.vop_write =		ffs_write,
156 	.vop_closeextattr =	ffs_closeextattr,
157 	.vop_deleteextattr =	ffs_deleteextattr,
158 	.vop_getextattr =	ffs_getextattr,
159 	.vop_listextattr =	ffs_listextattr,
160 	.vop_openextattr =	ffs_openextattr,
161 	.vop_setextattr =	ffs_setextattr,
162 	.vop_vptofh =		ffs_vptofh,
163 };
164 
165 struct vop_vector ffs_fifoops2 = {
166 	.vop_default =		&ufs_fifoops,
167 	.vop_fsync =		ffs_fsync,
168 	.vop_fdatasync =	ffs_fdatasync,
169 	.vop_lock1 =		ffs_lock,
170 	.vop_reallocblks =	ffs_reallocblks,
171 	.vop_strategy =		ffsext_strategy,
172 	.vop_closeextattr =	ffs_closeextattr,
173 	.vop_deleteextattr =	ffs_deleteextattr,
174 	.vop_getextattr =	ffs_getextattr,
175 	.vop_listextattr =	ffs_listextattr,
176 	.vop_openextattr =	ffs_openextattr,
177 	.vop_setextattr =	ffs_setextattr,
178 	.vop_vptofh =		ffs_vptofh,
179 };
180 
181 /*
182  * Synch an open file.
183  */
184 /* ARGSUSED */
185 static int
186 ffs_fsync(struct vop_fsync_args *ap)
187 {
188 	struct vnode *vp;
189 	struct bufobj *bo;
190 	int error;
191 
192 	vp = ap->a_vp;
193 	bo = &vp->v_bufobj;
194 retry:
195 	error = ffs_syncvnode(vp, ap->a_waitfor, 0);
196 	if (error)
197 		return (error);
198 	if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
199 		error = softdep_fsync(vp);
200 		if (error)
201 			return (error);
202 
203 		/*
204 		 * The softdep_fsync() function may drop vp lock,
205 		 * allowing for dirty buffers to reappear on the
206 		 * bo_dirty list. Recheck and resync as needed.
207 		 */
208 		BO_LOCK(bo);
209 		if ((vp->v_type == VREG || vp->v_type == VDIR) &&
210 		    (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
211 			BO_UNLOCK(bo);
212 			goto retry;
213 		}
214 		BO_UNLOCK(bo);
215 	}
216 	return (0);
217 }
218 
219 int
220 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
221 {
222 	struct inode *ip;
223 	struct bufobj *bo;
224 	struct buf *bp, *nbp;
225 	ufs_lbn_t lbn;
226 	int error, passes;
227 	bool still_dirty, wait;
228 
229 	ip = VTOI(vp);
230 	ip->i_flag &= ~IN_NEEDSYNC;
231 	bo = &vp->v_bufobj;
232 
233 	/*
234 	 * When doing MNT_WAIT we must first flush all dependencies
235 	 * on the inode.
236 	 */
237 	if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
238 	    (error = softdep_sync_metadata(vp)) != 0)
239 		return (error);
240 
241 	/*
242 	 * Flush all dirty buffers associated with a vnode.
243 	 */
244 	error = 0;
245 	passes = 0;
246 	wait = false;	/* Always do an async pass first. */
247 	lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1));
248 	BO_LOCK(bo);
249 loop:
250 	TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
251 		bp->b_vflags &= ~BV_SCANNED;
252 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
253 		/*
254 		 * Reasons to skip this buffer: it has already been considered
255 		 * on this pass, the buffer has dependencies that will cause
256 		 * it to be redirtied and it has not already been deferred,
257 		 * or it is already being written.
258 		 */
259 		if ((bp->b_vflags & BV_SCANNED) != 0)
260 			continue;
261 		bp->b_vflags |= BV_SCANNED;
262 		/*
263 		 * Flush indirects in order, if requested.
264 		 *
265 		 * Note that if only datasync is requested, we can
266 		 * skip indirect blocks when softupdates are not
267 		 * active.  Otherwise we must flush them with data,
268 		 * since dependencies prevent data block writes.
269 		 */
270 		if (waitfor == MNT_WAIT && bp->b_lblkno <= -NDADDR &&
271 		    (lbn_level(bp->b_lblkno) >= passes ||
272 		    ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp))))
273 			continue;
274 		if (bp->b_lblkno > lbn)
275 			panic("ffs_syncvnode: syncing truncated data.");
276 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) {
277 			BO_UNLOCK(bo);
278 		} else if (wait) {
279 			if (BUF_LOCK(bp,
280 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
281 			    BO_LOCKPTR(bo)) != 0) {
282 				bp->b_vflags &= ~BV_SCANNED;
283 				goto next;
284 			}
285 		} else
286 			continue;
287 		if ((bp->b_flags & B_DELWRI) == 0)
288 			panic("ffs_fsync: not dirty");
289 		/*
290 		 * Check for dependencies and potentially complete them.
291 		 */
292 		if (!LIST_EMPTY(&bp->b_dep) &&
293 		    (error = softdep_sync_buf(vp, bp,
294 		    wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
295 			/* I/O error. */
296 			if (error != EBUSY) {
297 				BUF_UNLOCK(bp);
298 				return (error);
299 			}
300 			/* If we deferred once, don't defer again. */
301 		    	if ((bp->b_flags & B_DEFERRED) == 0) {
302 				bp->b_flags |= B_DEFERRED;
303 				BUF_UNLOCK(bp);
304 				goto next;
305 			}
306 		}
307 		if (wait) {
308 			bremfree(bp);
309 			if ((error = bwrite(bp)) != 0)
310 				return (error);
311 		} else if ((bp->b_flags & B_CLUSTEROK)) {
312 			(void) vfs_bio_awrite(bp);
313 		} else {
314 			bremfree(bp);
315 			(void) bawrite(bp);
316 		}
317 next:
318 		/*
319 		 * Since we may have slept during the I/O, we need
320 		 * to start from a known point.
321 		 */
322 		BO_LOCK(bo);
323 		nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
324 	}
325 	if (waitfor != MNT_WAIT) {
326 		BO_UNLOCK(bo);
327 		if ((flags & NO_INO_UPDT) != 0)
328 			return (0);
329 		else
330 			return (ffs_update(vp, 0));
331 	}
332 	/* Drain IO to see if we're done. */
333 	bufobj_wwait(bo, 0, 0);
334 	/*
335 	 * Block devices associated with filesystems may have new I/O
336 	 * requests posted for them even if the vnode is locked, so no
337 	 * amount of trying will get them clean.  We make several passes
338 	 * as a best effort.
339 	 *
340 	 * Regular files may need multiple passes to flush all dependency
341 	 * work as it is possible that we must write once per indirect
342 	 * level, once for the leaf, and once for the inode and each of
343 	 * these will be done with one sync and one async pass.
344 	 */
345 	if (bo->bo_dirty.bv_cnt > 0) {
346 		if ((flags & DATA_ONLY) == 0) {
347 			still_dirty = true;
348 		} else {
349 			/*
350 			 * For data-only sync, dirty indirect buffers
351 			 * are ignored.
352 			 */
353 			still_dirty = false;
354 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
355 				if (bp->b_lblkno > -NDADDR) {
356 					still_dirty = true;
357 					break;
358 				}
359 			}
360 		}
361 
362 		if (still_dirty) {
363 			/* Write the inode after sync passes to flush deps. */
364 			if (wait && DOINGSOFTDEP(vp) &&
365 			    (flags & NO_INO_UPDT) == 0) {
366 				BO_UNLOCK(bo);
367 				ffs_update(vp, 1);
368 				BO_LOCK(bo);
369 			}
370 			/* switch between sync/async. */
371 			wait = !wait;
372 			if (wait || ++passes < NIADDR + 2)
373 				goto loop;
374 #ifdef INVARIANTS
375 			if (!vn_isdisk(vp, NULL))
376 				vn_printf(vp, "ffs_fsync: dirty ");
377 #endif
378 		}
379 	}
380 	BO_UNLOCK(bo);
381 	error = 0;
382 	if ((flags & DATA_ONLY) == 0) {
383 		if ((flags & NO_INO_UPDT) == 0)
384 			error = ffs_update(vp, 1);
385 		if (DOINGSUJ(vp))
386 			softdep_journal_fsync(VTOI(vp));
387 	}
388 	return (error);
389 }
390 
391 static int
392 ffs_fdatasync(struct vop_fdatasync_args *ap)
393 {
394 
395 	return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY));
396 }
397 
398 static int
399 ffs_lock(ap)
400 	struct vop_lock1_args /* {
401 		struct vnode *a_vp;
402 		int a_flags;
403 		struct thread *a_td;
404 		char *file;
405 		int line;
406 	} */ *ap;
407 {
408 #ifndef NO_FFS_SNAPSHOT
409 	struct vnode *vp;
410 	int flags;
411 	struct lock *lkp;
412 	int result;
413 
414 	switch (ap->a_flags & LK_TYPE_MASK) {
415 	case LK_SHARED:
416 	case LK_UPGRADE:
417 	case LK_EXCLUSIVE:
418 		vp = ap->a_vp;
419 		flags = ap->a_flags;
420 		for (;;) {
421 #ifdef DEBUG_VFS_LOCKS
422 			KASSERT(vp->v_holdcnt != 0,
423 			    ("ffs_lock %p: zero hold count", vp));
424 #endif
425 			lkp = vp->v_vnlock;
426 			result = _lockmgr_args(lkp, flags, VI_MTX(vp),
427 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
428 			    ap->a_file, ap->a_line);
429 			if (lkp == vp->v_vnlock || result != 0)
430 				break;
431 			/*
432 			 * Apparent success, except that the vnode
433 			 * mutated between snapshot file vnode and
434 			 * regular file vnode while this process
435 			 * slept.  The lock currently held is not the
436 			 * right lock.  Release it, and try to get the
437 			 * new lock.
438 			 */
439 			(void) _lockmgr_args(lkp, LK_RELEASE, NULL,
440 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
441 			    ap->a_file, ap->a_line);
442 			if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
443 			    (LK_INTERLOCK | LK_NOWAIT))
444 				return (EBUSY);
445 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
446 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
447 			flags &= ~LK_INTERLOCK;
448 		}
449 		break;
450 	default:
451 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
452 	}
453 	return (result);
454 #else
455 	return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
456 #endif
457 }
458 
459 /*
460  * Vnode op for reading.
461  */
462 static int
463 ffs_read(ap)
464 	struct vop_read_args /* {
465 		struct vnode *a_vp;
466 		struct uio *a_uio;
467 		int a_ioflag;
468 		struct ucred *a_cred;
469 	} */ *ap;
470 {
471 	struct vnode *vp;
472 	struct inode *ip;
473 	struct uio *uio;
474 	struct fs *fs;
475 	struct buf *bp;
476 	ufs_lbn_t lbn, nextlbn;
477 	off_t bytesinfile;
478 	long size, xfersize, blkoffset;
479 	ssize_t orig_resid;
480 	int error;
481 	int seqcount;
482 	int ioflag;
483 
484 	vp = ap->a_vp;
485 	uio = ap->a_uio;
486 	ioflag = ap->a_ioflag;
487 	if (ap->a_ioflag & IO_EXT)
488 #ifdef notyet
489 		return (ffs_extread(vp, uio, ioflag));
490 #else
491 		panic("ffs_read+IO_EXT");
492 #endif
493 #ifdef DIRECTIO
494 	if ((ioflag & IO_DIRECT) != 0) {
495 		int workdone;
496 
497 		error = ffs_rawread(vp, uio, &workdone);
498 		if (error != 0 || workdone != 0)
499 			return error;
500 	}
501 #endif
502 
503 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
504 	ip = VTOI(vp);
505 
506 #ifdef INVARIANTS
507 	if (uio->uio_rw != UIO_READ)
508 		panic("ffs_read: mode");
509 
510 	if (vp->v_type == VLNK) {
511 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
512 			panic("ffs_read: short symlink");
513 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
514 		panic("ffs_read: type %d",  vp->v_type);
515 #endif
516 	orig_resid = uio->uio_resid;
517 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
518 	if (orig_resid == 0)
519 		return (0);
520 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
521 	fs = ITOFS(ip);
522 	if (uio->uio_offset < ip->i_size &&
523 	    uio->uio_offset >= fs->fs_maxfilesize)
524 		return (EOVERFLOW);
525 
526 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
527 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
528 			break;
529 		lbn = lblkno(fs, uio->uio_offset);
530 		nextlbn = lbn + 1;
531 
532 		/*
533 		 * size of buffer.  The buffer representing the
534 		 * end of the file is rounded up to the size of
535 		 * the block type ( fragment or full block,
536 		 * depending ).
537 		 */
538 		size = blksize(fs, ip, lbn);
539 		blkoffset = blkoff(fs, uio->uio_offset);
540 
541 		/*
542 		 * The amount we want to transfer in this iteration is
543 		 * one FS block less the amount of the data before
544 		 * our startpoint (duh!)
545 		 */
546 		xfersize = fs->fs_bsize - blkoffset;
547 
548 		/*
549 		 * But if we actually want less than the block,
550 		 * or the file doesn't have a whole block more of data,
551 		 * then use the lesser number.
552 		 */
553 		if (uio->uio_resid < xfersize)
554 			xfersize = uio->uio_resid;
555 		if (bytesinfile < xfersize)
556 			xfersize = bytesinfile;
557 
558 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
559 			/*
560 			 * Don't do readahead if this is the end of the file.
561 			 */
562 			error = bread_gb(vp, lbn, size, NOCRED,
563 			    GB_UNMAPPED, &bp);
564 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
565 			/*
566 			 * Otherwise if we are allowed to cluster,
567 			 * grab as much as we can.
568 			 *
569 			 * XXX  This may not be a win if we are not
570 			 * doing sequential access.
571 			 */
572 			error = cluster_read(vp, ip->i_size, lbn,
573 			    size, NOCRED, blkoffset + uio->uio_resid,
574 			    seqcount, GB_UNMAPPED, &bp);
575 		} else if (seqcount > 1) {
576 			/*
577 			 * If we are NOT allowed to cluster, then
578 			 * if we appear to be acting sequentially,
579 			 * fire off a request for a readahead
580 			 * as well as a read. Note that the 4th and 5th
581 			 * arguments point to arrays of the size specified in
582 			 * the 6th argument.
583 			 */
584 			u_int nextsize = blksize(fs, ip, nextlbn);
585 			error = breadn_flags(vp, lbn, size, &nextlbn,
586 			    &nextsize, 1, NOCRED, GB_UNMAPPED, &bp);
587 		} else {
588 			/*
589 			 * Failing all of the above, just read what the
590 			 * user asked for. Interestingly, the same as
591 			 * the first option above.
592 			 */
593 			error = bread_gb(vp, lbn, size, NOCRED,
594 			    GB_UNMAPPED, &bp);
595 		}
596 		if (error) {
597 			brelse(bp);
598 			bp = NULL;
599 			break;
600 		}
601 
602 		/*
603 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
604 		 * will cause us to attempt to release the buffer later on
605 		 * and will cause the buffer cache to attempt to free the
606 		 * underlying pages.
607 		 */
608 		if (ioflag & IO_DIRECT)
609 			bp->b_flags |= B_DIRECT;
610 
611 		/*
612 		 * We should only get non-zero b_resid when an I/O error
613 		 * has occurred, which should cause us to break above.
614 		 * However, if the short read did not cause an error,
615 		 * then we want to ensure that we do not uiomove bad
616 		 * or uninitialized data.
617 		 */
618 		size -= bp->b_resid;
619 		if (size < xfersize) {
620 			if (size == 0)
621 				break;
622 			xfersize = size;
623 		}
624 
625 		if (buf_mapped(bp)) {
626 			error = vn_io_fault_uiomove((char *)bp->b_data +
627 			    blkoffset, (int)xfersize, uio);
628 		} else {
629 			error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
630 			    (int)xfersize, uio);
631 		}
632 		if (error)
633 			break;
634 
635 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
636 		   (LIST_EMPTY(&bp->b_dep))) {
637 			/*
638 			 * If there are no dependencies, and it's VMIO,
639 			 * then we don't need the buf, mark it available
640 			 * for freeing.  For non-direct VMIO reads, the VM
641 			 * has the data.
642 			 */
643 			bp->b_flags |= B_RELBUF;
644 			brelse(bp);
645 		} else {
646 			/*
647 			 * Otherwise let whoever
648 			 * made the request take care of
649 			 * freeing it. We just queue
650 			 * it onto another list.
651 			 */
652 			bqrelse(bp);
653 		}
654 	}
655 
656 	/*
657 	 * This can only happen in the case of an error
658 	 * because the loop above resets bp to NULL on each iteration
659 	 * and on normal completion has not set a new value into it.
660 	 * so it must have come from a 'break' statement
661 	 */
662 	if (bp != NULL) {
663 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
664 		   (LIST_EMPTY(&bp->b_dep))) {
665 			bp->b_flags |= B_RELBUF;
666 			brelse(bp);
667 		} else {
668 			bqrelse(bp);
669 		}
670 	}
671 
672 	if ((error == 0 || uio->uio_resid != orig_resid) &&
673 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0 &&
674 	    (ip->i_flag & IN_ACCESS) == 0) {
675 		VI_LOCK(vp);
676 		ip->i_flag |= IN_ACCESS;
677 		VI_UNLOCK(vp);
678 	}
679 	return (error);
680 }
681 
682 /*
683  * Vnode op for writing.
684  */
685 static int
686 ffs_write(ap)
687 	struct vop_write_args /* {
688 		struct vnode *a_vp;
689 		struct uio *a_uio;
690 		int a_ioflag;
691 		struct ucred *a_cred;
692 	} */ *ap;
693 {
694 	struct vnode *vp;
695 	struct uio *uio;
696 	struct inode *ip;
697 	struct fs *fs;
698 	struct buf *bp;
699 	ufs_lbn_t lbn;
700 	off_t osize;
701 	ssize_t resid;
702 	int seqcount;
703 	int blkoffset, error, flags, ioflag, size, xfersize;
704 
705 	vp = ap->a_vp;
706 	uio = ap->a_uio;
707 	ioflag = ap->a_ioflag;
708 	if (ap->a_ioflag & IO_EXT)
709 #ifdef notyet
710 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
711 #else
712 		panic("ffs_write+IO_EXT");
713 #endif
714 
715 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
716 	ip = VTOI(vp);
717 
718 #ifdef INVARIANTS
719 	if (uio->uio_rw != UIO_WRITE)
720 		panic("ffs_write: mode");
721 #endif
722 
723 	switch (vp->v_type) {
724 	case VREG:
725 		if (ioflag & IO_APPEND)
726 			uio->uio_offset = ip->i_size;
727 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
728 			return (EPERM);
729 		/* FALLTHROUGH */
730 	case VLNK:
731 		break;
732 	case VDIR:
733 		panic("ffs_write: dir write");
734 		break;
735 	default:
736 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
737 			(int)uio->uio_offset,
738 			(int)uio->uio_resid
739 		);
740 	}
741 
742 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
743 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
744 	fs = ITOFS(ip);
745 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
746 		return (EFBIG);
747 	/*
748 	 * Maybe this should be above the vnode op call, but so long as
749 	 * file servers have no limits, I don't think it matters.
750 	 */
751 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
752 		return (EFBIG);
753 
754 	resid = uio->uio_resid;
755 	osize = ip->i_size;
756 	if (seqcount > BA_SEQMAX)
757 		flags = BA_SEQMAX << BA_SEQSHIFT;
758 	else
759 		flags = seqcount << BA_SEQSHIFT;
760 	if (ioflag & IO_SYNC)
761 		flags |= IO_SYNC;
762 	flags |= BA_UNMAPPED;
763 
764 	for (error = 0; uio->uio_resid > 0;) {
765 		lbn = lblkno(fs, uio->uio_offset);
766 		blkoffset = blkoff(fs, uio->uio_offset);
767 		xfersize = fs->fs_bsize - blkoffset;
768 		if (uio->uio_resid < xfersize)
769 			xfersize = uio->uio_resid;
770 		if (uio->uio_offset + xfersize > ip->i_size)
771 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
772 
773 		/*
774 		 * We must perform a read-before-write if the transfer size
775 		 * does not cover the entire buffer.
776 		 */
777 		if (fs->fs_bsize > xfersize)
778 			flags |= BA_CLRBUF;
779 		else
780 			flags &= ~BA_CLRBUF;
781 /* XXX is uio->uio_offset the right thing here? */
782 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
783 		    ap->a_cred, flags, &bp);
784 		if (error != 0) {
785 			vnode_pager_setsize(vp, ip->i_size);
786 			break;
787 		}
788 		if (ioflag & IO_DIRECT)
789 			bp->b_flags |= B_DIRECT;
790 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
791 			bp->b_flags |= B_NOCACHE;
792 
793 		if (uio->uio_offset + xfersize > ip->i_size) {
794 			ip->i_size = uio->uio_offset + xfersize;
795 			DIP_SET(ip, i_size, ip->i_size);
796 		}
797 
798 		size = blksize(fs, ip, lbn) - bp->b_resid;
799 		if (size < xfersize)
800 			xfersize = size;
801 
802 		if (buf_mapped(bp)) {
803 			error = vn_io_fault_uiomove((char *)bp->b_data +
804 			    blkoffset, (int)xfersize, uio);
805 		} else {
806 			error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
807 			    (int)xfersize, uio);
808 		}
809 		/*
810 		 * If the buffer is not already filled and we encounter an
811 		 * error while trying to fill it, we have to clear out any
812 		 * garbage data from the pages instantiated for the buffer.
813 		 * If we do not, a failed uiomove() during a write can leave
814 		 * the prior contents of the pages exposed to a userland mmap.
815 		 *
816 		 * Note that we need only clear buffers with a transfer size
817 		 * equal to the block size because buffers with a shorter
818 		 * transfer size were cleared above by the call to UFS_BALLOC()
819 		 * with the BA_CLRBUF flag set.
820 		 *
821 		 * If the source region for uiomove identically mmaps the
822 		 * buffer, uiomove() performed the NOP copy, and the buffer
823 		 * content remains valid because the page fault handler
824 		 * validated the pages.
825 		 */
826 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
827 		    fs->fs_bsize == xfersize)
828 			vfs_bio_clrbuf(bp);
829 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
830 		   (LIST_EMPTY(&bp->b_dep))) {
831 			bp->b_flags |= B_RELBUF;
832 		}
833 
834 		/*
835 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
836 		 * if we have a severe page deficiency write the buffer
837 		 * asynchronously.  Otherwise try to cluster, and if that
838 		 * doesn't do it then either do an async write (if O_DIRECT),
839 		 * or a delayed write (if not).
840 		 */
841 		if (ioflag & IO_SYNC) {
842 			(void)bwrite(bp);
843 		} else if (vm_page_count_severe() ||
844 			    buf_dirty_count_severe() ||
845 			    (ioflag & IO_ASYNC)) {
846 			bp->b_flags |= B_CLUSTEROK;
847 			bawrite(bp);
848 		} else if (xfersize + blkoffset == fs->fs_bsize) {
849 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
850 				bp->b_flags |= B_CLUSTEROK;
851 				cluster_write(vp, bp, ip->i_size, seqcount,
852 				    GB_UNMAPPED);
853 			} else {
854 				bawrite(bp);
855 			}
856 		} else if (ioflag & IO_DIRECT) {
857 			bp->b_flags |= B_CLUSTEROK;
858 			bawrite(bp);
859 		} else {
860 			bp->b_flags |= B_CLUSTEROK;
861 			bdwrite(bp);
862 		}
863 		if (error || xfersize == 0)
864 			break;
865 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
866 	}
867 	/*
868 	 * If we successfully wrote any data, and we are not the superuser
869 	 * we clear the setuid and setgid bits as a precaution against
870 	 * tampering.
871 	 */
872 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
873 	    ap->a_cred) {
874 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
875 			ip->i_mode &= ~(ISUID | ISGID);
876 			DIP_SET(ip, i_mode, ip->i_mode);
877 		}
878 	}
879 	if (error) {
880 		if (ioflag & IO_UNIT) {
881 			(void)ffs_truncate(vp, osize,
882 			    IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
883 			uio->uio_offset -= resid - uio->uio_resid;
884 			uio->uio_resid = resid;
885 		}
886 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
887 		error = ffs_update(vp, 1);
888 	return (error);
889 }
890 
891 /*
892  * Extended attribute area reading.
893  */
894 static int
895 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
896 {
897 	struct inode *ip;
898 	struct ufs2_dinode *dp;
899 	struct fs *fs;
900 	struct buf *bp;
901 	ufs_lbn_t lbn, nextlbn;
902 	off_t bytesinfile;
903 	long size, xfersize, blkoffset;
904 	ssize_t orig_resid;
905 	int error;
906 
907 	ip = VTOI(vp);
908 	fs = ITOFS(ip);
909 	dp = ip->i_din2;
910 
911 #ifdef INVARIANTS
912 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
913 		panic("ffs_extread: mode");
914 
915 #endif
916 	orig_resid = uio->uio_resid;
917 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
918 	if (orig_resid == 0)
919 		return (0);
920 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
921 
922 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
923 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
924 			break;
925 		lbn = lblkno(fs, uio->uio_offset);
926 		nextlbn = lbn + 1;
927 
928 		/*
929 		 * size of buffer.  The buffer representing the
930 		 * end of the file is rounded up to the size of
931 		 * the block type ( fragment or full block,
932 		 * depending ).
933 		 */
934 		size = sblksize(fs, dp->di_extsize, lbn);
935 		blkoffset = blkoff(fs, uio->uio_offset);
936 
937 		/*
938 		 * The amount we want to transfer in this iteration is
939 		 * one FS block less the amount of the data before
940 		 * our startpoint (duh!)
941 		 */
942 		xfersize = fs->fs_bsize - blkoffset;
943 
944 		/*
945 		 * But if we actually want less than the block,
946 		 * or the file doesn't have a whole block more of data,
947 		 * then use the lesser number.
948 		 */
949 		if (uio->uio_resid < xfersize)
950 			xfersize = uio->uio_resid;
951 		if (bytesinfile < xfersize)
952 			xfersize = bytesinfile;
953 
954 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
955 			/*
956 			 * Don't do readahead if this is the end of the info.
957 			 */
958 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
959 		} else {
960 			/*
961 			 * If we have a second block, then
962 			 * fire off a request for a readahead
963 			 * as well as a read. Note that the 4th and 5th
964 			 * arguments point to arrays of the size specified in
965 			 * the 6th argument.
966 			 */
967 			u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
968 
969 			nextlbn = -1 - nextlbn;
970 			error = breadn(vp, -1 - lbn,
971 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
972 		}
973 		if (error) {
974 			brelse(bp);
975 			bp = NULL;
976 			break;
977 		}
978 
979 		/*
980 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
981 		 * will cause us to attempt to release the buffer later on
982 		 * and will cause the buffer cache to attempt to free the
983 		 * underlying pages.
984 		 */
985 		if (ioflag & IO_DIRECT)
986 			bp->b_flags |= B_DIRECT;
987 
988 		/*
989 		 * We should only get non-zero b_resid when an I/O error
990 		 * has occurred, which should cause us to break above.
991 		 * However, if the short read did not cause an error,
992 		 * then we want to ensure that we do not uiomove bad
993 		 * or uninitialized data.
994 		 */
995 		size -= bp->b_resid;
996 		if (size < xfersize) {
997 			if (size == 0)
998 				break;
999 			xfersize = size;
1000 		}
1001 
1002 		error = uiomove((char *)bp->b_data + blkoffset,
1003 					(int)xfersize, uio);
1004 		if (error)
1005 			break;
1006 
1007 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1008 		   (LIST_EMPTY(&bp->b_dep))) {
1009 			/*
1010 			 * If there are no dependencies, and it's VMIO,
1011 			 * then we don't need the buf, mark it available
1012 			 * for freeing.  For non-direct VMIO reads, the VM
1013 			 * has the data.
1014 			 */
1015 			bp->b_flags |= B_RELBUF;
1016 			brelse(bp);
1017 		} else {
1018 			/*
1019 			 * Otherwise let whoever
1020 			 * made the request take care of
1021 			 * freeing it. We just queue
1022 			 * it onto another list.
1023 			 */
1024 			bqrelse(bp);
1025 		}
1026 	}
1027 
1028 	/*
1029 	 * This can only happen in the case of an error
1030 	 * because the loop above resets bp to NULL on each iteration
1031 	 * and on normal completion has not set a new value into it.
1032 	 * so it must have come from a 'break' statement
1033 	 */
1034 	if (bp != NULL) {
1035 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1036 		   (LIST_EMPTY(&bp->b_dep))) {
1037 			bp->b_flags |= B_RELBUF;
1038 			brelse(bp);
1039 		} else {
1040 			bqrelse(bp);
1041 		}
1042 	}
1043 	return (error);
1044 }
1045 
1046 /*
1047  * Extended attribute area writing.
1048  */
1049 static int
1050 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1051 {
1052 	struct inode *ip;
1053 	struct ufs2_dinode *dp;
1054 	struct fs *fs;
1055 	struct buf *bp;
1056 	ufs_lbn_t lbn;
1057 	off_t osize;
1058 	ssize_t resid;
1059 	int blkoffset, error, flags, size, xfersize;
1060 
1061 	ip = VTOI(vp);
1062 	fs = ITOFS(ip);
1063 	dp = ip->i_din2;
1064 
1065 #ifdef INVARIANTS
1066 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1067 		panic("ffs_extwrite: mode");
1068 #endif
1069 
1070 	if (ioflag & IO_APPEND)
1071 		uio->uio_offset = dp->di_extsize;
1072 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1073 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1074 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1075 		return (EFBIG);
1076 
1077 	resid = uio->uio_resid;
1078 	osize = dp->di_extsize;
1079 	flags = IO_EXT;
1080 	if (ioflag & IO_SYNC)
1081 		flags |= IO_SYNC;
1082 
1083 	for (error = 0; uio->uio_resid > 0;) {
1084 		lbn = lblkno(fs, uio->uio_offset);
1085 		blkoffset = blkoff(fs, uio->uio_offset);
1086 		xfersize = fs->fs_bsize - blkoffset;
1087 		if (uio->uio_resid < xfersize)
1088 			xfersize = uio->uio_resid;
1089 
1090 		/*
1091 		 * We must perform a read-before-write if the transfer size
1092 		 * does not cover the entire buffer.
1093 		 */
1094 		if (fs->fs_bsize > xfersize)
1095 			flags |= BA_CLRBUF;
1096 		else
1097 			flags &= ~BA_CLRBUF;
1098 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1099 		    ucred, flags, &bp);
1100 		if (error != 0)
1101 			break;
1102 		/*
1103 		 * If the buffer is not valid we have to clear out any
1104 		 * garbage data from the pages instantiated for the buffer.
1105 		 * If we do not, a failed uiomove() during a write can leave
1106 		 * the prior contents of the pages exposed to a userland
1107 		 * mmap().  XXX deal with uiomove() errors a better way.
1108 		 */
1109 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1110 			vfs_bio_clrbuf(bp);
1111 		if (ioflag & IO_DIRECT)
1112 			bp->b_flags |= B_DIRECT;
1113 
1114 		if (uio->uio_offset + xfersize > dp->di_extsize)
1115 			dp->di_extsize = uio->uio_offset + xfersize;
1116 
1117 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1118 		if (size < xfersize)
1119 			xfersize = size;
1120 
1121 		error =
1122 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1123 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1124 		   (LIST_EMPTY(&bp->b_dep))) {
1125 			bp->b_flags |= B_RELBUF;
1126 		}
1127 
1128 		/*
1129 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1130 		 * if we have a severe page deficiency write the buffer
1131 		 * asynchronously.  Otherwise try to cluster, and if that
1132 		 * doesn't do it then either do an async write (if O_DIRECT),
1133 		 * or a delayed write (if not).
1134 		 */
1135 		if (ioflag & IO_SYNC) {
1136 			(void)bwrite(bp);
1137 		} else if (vm_page_count_severe() ||
1138 			    buf_dirty_count_severe() ||
1139 			    xfersize + blkoffset == fs->fs_bsize ||
1140 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1141 			bawrite(bp);
1142 		else
1143 			bdwrite(bp);
1144 		if (error || xfersize == 0)
1145 			break;
1146 		ip->i_flag |= IN_CHANGE;
1147 	}
1148 	/*
1149 	 * If we successfully wrote any data, and we are not the superuser
1150 	 * we clear the setuid and setgid bits as a precaution against
1151 	 * tampering.
1152 	 */
1153 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1154 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1155 			ip->i_mode &= ~(ISUID | ISGID);
1156 			dp->di_mode = ip->i_mode;
1157 		}
1158 	}
1159 	if (error) {
1160 		if (ioflag & IO_UNIT) {
1161 			(void)ffs_truncate(vp, osize,
1162 			    IO_EXT | (ioflag&IO_SYNC), ucred);
1163 			uio->uio_offset -= resid - uio->uio_resid;
1164 			uio->uio_resid = resid;
1165 		}
1166 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1167 		error = ffs_update(vp, 1);
1168 	return (error);
1169 }
1170 
1171 
1172 /*
1173  * Vnode operating to retrieve a named extended attribute.
1174  *
1175  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1176  * the length of the EA, and possibly the pointer to the entry and to the data.
1177  */
1178 static int
1179 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1180 {
1181 	u_char *p, *pe, *pn, *p0;
1182 	int eapad1, eapad2, ealength, ealen, nlen;
1183 	uint32_t ul;
1184 
1185 	pe = ptr + length;
1186 	nlen = strlen(name);
1187 
1188 	for (p = ptr; p < pe; p = pn) {
1189 		p0 = p;
1190 		bcopy(p, &ul, sizeof(ul));
1191 		pn = p + ul;
1192 		/* make sure this entry is complete */
1193 		if (pn > pe)
1194 			break;
1195 		p += sizeof(uint32_t);
1196 		if (*p != nspace)
1197 			continue;
1198 		p++;
1199 		eapad2 = *p++;
1200 		if (*p != nlen)
1201 			continue;
1202 		p++;
1203 		if (bcmp(p, name, nlen))
1204 			continue;
1205 		ealength = sizeof(uint32_t) + 3 + nlen;
1206 		eapad1 = 8 - (ealength % 8);
1207 		if (eapad1 == 8)
1208 			eapad1 = 0;
1209 		ealength += eapad1;
1210 		ealen = ul - ealength - eapad2;
1211 		p += nlen + eapad1;
1212 		if (eap != NULL)
1213 			*eap = p0;
1214 		if (eac != NULL)
1215 			*eac = p;
1216 		return (ealen);
1217 	}
1218 	return(-1);
1219 }
1220 
1221 static int
1222 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1223 {
1224 	struct inode *ip;
1225 	struct ufs2_dinode *dp;
1226 	struct fs *fs;
1227 	struct uio luio;
1228 	struct iovec liovec;
1229 	u_int easize;
1230 	int error;
1231 	u_char *eae;
1232 
1233 	ip = VTOI(vp);
1234 	fs = ITOFS(ip);
1235 	dp = ip->i_din2;
1236 	easize = dp->di_extsize;
1237 	if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1238 		return (EFBIG);
1239 
1240 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1241 
1242 	liovec.iov_base = eae;
1243 	liovec.iov_len = easize;
1244 	luio.uio_iov = &liovec;
1245 	luio.uio_iovcnt = 1;
1246 	luio.uio_offset = 0;
1247 	luio.uio_resid = easize;
1248 	luio.uio_segflg = UIO_SYSSPACE;
1249 	luio.uio_rw = UIO_READ;
1250 	luio.uio_td = td;
1251 
1252 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1253 	if (error) {
1254 		free(eae, M_TEMP);
1255 		return(error);
1256 	}
1257 	*p = eae;
1258 	return (0);
1259 }
1260 
1261 static void
1262 ffs_lock_ea(struct vnode *vp)
1263 {
1264 	struct inode *ip;
1265 
1266 	ip = VTOI(vp);
1267 	VI_LOCK(vp);
1268 	while (ip->i_flag & IN_EA_LOCKED) {
1269 		ip->i_flag |= IN_EA_LOCKWAIT;
1270 		msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1271 		    0);
1272 	}
1273 	ip->i_flag |= IN_EA_LOCKED;
1274 	VI_UNLOCK(vp);
1275 }
1276 
1277 static void
1278 ffs_unlock_ea(struct vnode *vp)
1279 {
1280 	struct inode *ip;
1281 
1282 	ip = VTOI(vp);
1283 	VI_LOCK(vp);
1284 	if (ip->i_flag & IN_EA_LOCKWAIT)
1285 		wakeup(&ip->i_ea_refs);
1286 	ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1287 	VI_UNLOCK(vp);
1288 }
1289 
1290 static int
1291 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1292 {
1293 	struct inode *ip;
1294 	struct ufs2_dinode *dp;
1295 	int error;
1296 
1297 	ip = VTOI(vp);
1298 
1299 	ffs_lock_ea(vp);
1300 	if (ip->i_ea_area != NULL) {
1301 		ip->i_ea_refs++;
1302 		ffs_unlock_ea(vp);
1303 		return (0);
1304 	}
1305 	dp = ip->i_din2;
1306 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1307 	if (error) {
1308 		ffs_unlock_ea(vp);
1309 		return (error);
1310 	}
1311 	ip->i_ea_len = dp->di_extsize;
1312 	ip->i_ea_error = 0;
1313 	ip->i_ea_refs++;
1314 	ffs_unlock_ea(vp);
1315 	return (0);
1316 }
1317 
1318 /*
1319  * Vnode extattr transaction commit/abort
1320  */
1321 static int
1322 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1323 {
1324 	struct inode *ip;
1325 	struct uio luio;
1326 	struct iovec liovec;
1327 	int error;
1328 	struct ufs2_dinode *dp;
1329 
1330 	ip = VTOI(vp);
1331 
1332 	ffs_lock_ea(vp);
1333 	if (ip->i_ea_area == NULL) {
1334 		ffs_unlock_ea(vp);
1335 		return (EINVAL);
1336 	}
1337 	dp = ip->i_din2;
1338 	error = ip->i_ea_error;
1339 	if (commit && error == 0) {
1340 		ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1341 		if (cred == NOCRED)
1342 			cred =  vp->v_mount->mnt_cred;
1343 		liovec.iov_base = ip->i_ea_area;
1344 		liovec.iov_len = ip->i_ea_len;
1345 		luio.uio_iov = &liovec;
1346 		luio.uio_iovcnt = 1;
1347 		luio.uio_offset = 0;
1348 		luio.uio_resid = ip->i_ea_len;
1349 		luio.uio_segflg = UIO_SYSSPACE;
1350 		luio.uio_rw = UIO_WRITE;
1351 		luio.uio_td = td;
1352 		/* XXX: I'm not happy about truncating to zero size */
1353 		if (ip->i_ea_len < dp->di_extsize)
1354 			error = ffs_truncate(vp, 0, IO_EXT, cred);
1355 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1356 	}
1357 	if (--ip->i_ea_refs == 0) {
1358 		free(ip->i_ea_area, M_TEMP);
1359 		ip->i_ea_area = NULL;
1360 		ip->i_ea_len = 0;
1361 		ip->i_ea_error = 0;
1362 	}
1363 	ffs_unlock_ea(vp);
1364 	return (error);
1365 }
1366 
1367 /*
1368  * Vnode extattr strategy routine for fifos.
1369  *
1370  * We need to check for a read or write of the external attributes.
1371  * Otherwise we just fall through and do the usual thing.
1372  */
1373 static int
1374 ffsext_strategy(struct vop_strategy_args *ap)
1375 /*
1376 struct vop_strategy_args {
1377 	struct vnodeop_desc *a_desc;
1378 	struct vnode *a_vp;
1379 	struct buf *a_bp;
1380 };
1381 */
1382 {
1383 	struct vnode *vp;
1384 	daddr_t lbn;
1385 
1386 	vp = ap->a_vp;
1387 	lbn = ap->a_bp->b_lblkno;
1388 	if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -NXADDR)
1389 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1390 	if (vp->v_type == VFIFO)
1391 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1392 	panic("spec nodes went here");
1393 }
1394 
1395 /*
1396  * Vnode extattr transaction commit/abort
1397  */
1398 static int
1399 ffs_openextattr(struct vop_openextattr_args *ap)
1400 /*
1401 struct vop_openextattr_args {
1402 	struct vnodeop_desc *a_desc;
1403 	struct vnode *a_vp;
1404 	IN struct ucred *a_cred;
1405 	IN struct thread *a_td;
1406 };
1407 */
1408 {
1409 
1410 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1411 		return (EOPNOTSUPP);
1412 
1413 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1414 }
1415 
1416 
1417 /*
1418  * Vnode extattr transaction commit/abort
1419  */
1420 static int
1421 ffs_closeextattr(struct vop_closeextattr_args *ap)
1422 /*
1423 struct vop_closeextattr_args {
1424 	struct vnodeop_desc *a_desc;
1425 	struct vnode *a_vp;
1426 	int a_commit;
1427 	IN struct ucred *a_cred;
1428 	IN struct thread *a_td;
1429 };
1430 */
1431 {
1432 
1433 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1434 		return (EOPNOTSUPP);
1435 
1436 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1437 		return (EROFS);
1438 
1439 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1440 }
1441 
1442 /*
1443  * Vnode operation to remove a named attribute.
1444  */
1445 static int
1446 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1447 /*
1448 vop_deleteextattr {
1449 	IN struct vnode *a_vp;
1450 	IN int a_attrnamespace;
1451 	IN const char *a_name;
1452 	IN struct ucred *a_cred;
1453 	IN struct thread *a_td;
1454 };
1455 */
1456 {
1457 	struct inode *ip;
1458 	struct fs *fs;
1459 	uint32_t ealength, ul;
1460 	int ealen, olen, eapad1, eapad2, error, i, easize;
1461 	u_char *eae, *p;
1462 
1463 	ip = VTOI(ap->a_vp);
1464 	fs = ITOFS(ip);
1465 
1466 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1467 		return (EOPNOTSUPP);
1468 
1469 	if (strlen(ap->a_name) == 0)
1470 		return (EINVAL);
1471 
1472 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1473 		return (EROFS);
1474 
1475 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1476 	    ap->a_cred, ap->a_td, VWRITE);
1477 	if (error) {
1478 
1479 		/*
1480 		 * ffs_lock_ea is not needed there, because the vnode
1481 		 * must be exclusively locked.
1482 		 */
1483 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1484 			ip->i_ea_error = error;
1485 		return (error);
1486 	}
1487 
1488 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1489 	if (error)
1490 		return (error);
1491 
1492 	ealength = eapad1 = ealen = eapad2 = 0;
1493 
1494 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1495 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1496 	easize = ip->i_ea_len;
1497 
1498 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1499 	    &p, NULL);
1500 	if (olen == -1) {
1501 		/* delete but nonexistent */
1502 		free(eae, M_TEMP);
1503 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1504 		return(ENOATTR);
1505 	}
1506 	bcopy(p, &ul, sizeof ul);
1507 	i = p - eae + ul;
1508 	if (ul != ealength) {
1509 		bcopy(p + ul, p + ealength, easize - i);
1510 		easize += (ealength - ul);
1511 	}
1512 	if (easize > NXADDR * fs->fs_bsize) {
1513 		free(eae, M_TEMP);
1514 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1515 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1516 			ip->i_ea_error = ENOSPC;
1517 		return(ENOSPC);
1518 	}
1519 	p = ip->i_ea_area;
1520 	ip->i_ea_area = eae;
1521 	ip->i_ea_len = easize;
1522 	free(p, M_TEMP);
1523 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1524 	return(error);
1525 }
1526 
1527 /*
1528  * Vnode operation to retrieve a named extended attribute.
1529  */
1530 static int
1531 ffs_getextattr(struct vop_getextattr_args *ap)
1532 /*
1533 vop_getextattr {
1534 	IN struct vnode *a_vp;
1535 	IN int a_attrnamespace;
1536 	IN const char *a_name;
1537 	INOUT struct uio *a_uio;
1538 	OUT size_t *a_size;
1539 	IN struct ucred *a_cred;
1540 	IN struct thread *a_td;
1541 };
1542 */
1543 {
1544 	struct inode *ip;
1545 	u_char *eae, *p;
1546 	unsigned easize;
1547 	int error, ealen;
1548 
1549 	ip = VTOI(ap->a_vp);
1550 
1551 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1552 		return (EOPNOTSUPP);
1553 
1554 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1555 	    ap->a_cred, ap->a_td, VREAD);
1556 	if (error)
1557 		return (error);
1558 
1559 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1560 	if (error)
1561 		return (error);
1562 
1563 	eae = ip->i_ea_area;
1564 	easize = ip->i_ea_len;
1565 
1566 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1567 	    NULL, &p);
1568 	if (ealen >= 0) {
1569 		error = 0;
1570 		if (ap->a_size != NULL)
1571 			*ap->a_size = ealen;
1572 		else if (ap->a_uio != NULL)
1573 			error = uiomove(p, ealen, ap->a_uio);
1574 	} else
1575 		error = ENOATTR;
1576 
1577 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1578 	return(error);
1579 }
1580 
1581 /*
1582  * Vnode operation to retrieve extended attributes on a vnode.
1583  */
1584 static int
1585 ffs_listextattr(struct vop_listextattr_args *ap)
1586 /*
1587 vop_listextattr {
1588 	IN struct vnode *a_vp;
1589 	IN int a_attrnamespace;
1590 	INOUT struct uio *a_uio;
1591 	OUT size_t *a_size;
1592 	IN struct ucred *a_cred;
1593 	IN struct thread *a_td;
1594 };
1595 */
1596 {
1597 	struct inode *ip;
1598 	u_char *eae, *p, *pe, *pn;
1599 	unsigned easize;
1600 	uint32_t ul;
1601 	int error, ealen;
1602 
1603 	ip = VTOI(ap->a_vp);
1604 
1605 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1606 		return (EOPNOTSUPP);
1607 
1608 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1609 	    ap->a_cred, ap->a_td, VREAD);
1610 	if (error)
1611 		return (error);
1612 
1613 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1614 	if (error)
1615 		return (error);
1616 	eae = ip->i_ea_area;
1617 	easize = ip->i_ea_len;
1618 
1619 	error = 0;
1620 	if (ap->a_size != NULL)
1621 		*ap->a_size = 0;
1622 	pe = eae + easize;
1623 	for(p = eae; error == 0 && p < pe; p = pn) {
1624 		bcopy(p, &ul, sizeof(ul));
1625 		pn = p + ul;
1626 		if (pn > pe)
1627 			break;
1628 		p += sizeof(ul);
1629 		if (*p++ != ap->a_attrnamespace)
1630 			continue;
1631 		p++;	/* pad2 */
1632 		ealen = *p;
1633 		if (ap->a_size != NULL) {
1634 			*ap->a_size += ealen + 1;
1635 		} else if (ap->a_uio != NULL) {
1636 			error = uiomove(p, ealen + 1, ap->a_uio);
1637 		}
1638 	}
1639 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1640 	return(error);
1641 }
1642 
1643 /*
1644  * Vnode operation to set a named attribute.
1645  */
1646 static int
1647 ffs_setextattr(struct vop_setextattr_args *ap)
1648 /*
1649 vop_setextattr {
1650 	IN struct vnode *a_vp;
1651 	IN int a_attrnamespace;
1652 	IN const char *a_name;
1653 	INOUT struct uio *a_uio;
1654 	IN struct ucred *a_cred;
1655 	IN struct thread *a_td;
1656 };
1657 */
1658 {
1659 	struct inode *ip;
1660 	struct fs *fs;
1661 	uint32_t ealength, ul;
1662 	ssize_t ealen;
1663 	int olen, eapad1, eapad2, error, i, easize;
1664 	u_char *eae, *p;
1665 
1666 	ip = VTOI(ap->a_vp);
1667 	fs = ITOFS(ip);
1668 
1669 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1670 		return (EOPNOTSUPP);
1671 
1672 	if (strlen(ap->a_name) == 0)
1673 		return (EINVAL);
1674 
1675 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1676 	if (ap->a_uio == NULL)
1677 		return (EOPNOTSUPP);
1678 
1679 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1680 		return (EROFS);
1681 
1682 	ealen = ap->a_uio->uio_resid;
1683 	if (ealen < 0 || ealen > lblktosize(fs, NXADDR))
1684 		return (EINVAL);
1685 
1686 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1687 	    ap->a_cred, ap->a_td, VWRITE);
1688 	if (error) {
1689 
1690 		/*
1691 		 * ffs_lock_ea is not needed there, because the vnode
1692 		 * must be exclusively locked.
1693 		 */
1694 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1695 			ip->i_ea_error = error;
1696 		return (error);
1697 	}
1698 
1699 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1700 	if (error)
1701 		return (error);
1702 
1703 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1704 	eapad1 = 8 - (ealength % 8);
1705 	if (eapad1 == 8)
1706 		eapad1 = 0;
1707 	eapad2 = 8 - (ealen % 8);
1708 	if (eapad2 == 8)
1709 		eapad2 = 0;
1710 	ealength += eapad1 + ealen + eapad2;
1711 
1712 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1713 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1714 	easize = ip->i_ea_len;
1715 
1716 	olen = ffs_findextattr(eae, easize,
1717 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1718         if (olen == -1) {
1719 		/* new, append at end */
1720 		p = eae + easize;
1721 		easize += ealength;
1722 	} else {
1723 		bcopy(p, &ul, sizeof ul);
1724 		i = p - eae + ul;
1725 		if (ul != ealength) {
1726 			bcopy(p + ul, p + ealength, easize - i);
1727 			easize += (ealength - ul);
1728 		}
1729 	}
1730 	if (easize > lblktosize(fs, NXADDR)) {
1731 		free(eae, M_TEMP);
1732 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1733 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1734 			ip->i_ea_error = ENOSPC;
1735 		return(ENOSPC);
1736 	}
1737 	bcopy(&ealength, p, sizeof(ealength));
1738 	p += sizeof(ealength);
1739 	*p++ = ap->a_attrnamespace;
1740 	*p++ = eapad2;
1741 	*p++ = strlen(ap->a_name);
1742 	strcpy(p, ap->a_name);
1743 	p += strlen(ap->a_name);
1744 	bzero(p, eapad1);
1745 	p += eapad1;
1746 	error = uiomove(p, ealen, ap->a_uio);
1747 	if (error) {
1748 		free(eae, M_TEMP);
1749 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1750 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1751 			ip->i_ea_error = error;
1752 		return(error);
1753 	}
1754 	p += ealen;
1755 	bzero(p, eapad2);
1756 
1757 	p = ip->i_ea_area;
1758 	ip->i_ea_area = eae;
1759 	ip->i_ea_len = easize;
1760 	free(p, M_TEMP);
1761 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1762 	return(error);
1763 }
1764 
1765 /*
1766  * Vnode pointer to File handle
1767  */
1768 static int
1769 ffs_vptofh(struct vop_vptofh_args *ap)
1770 /*
1771 vop_vptofh {
1772 	IN struct vnode *a_vp;
1773 	IN struct fid *a_fhp;
1774 };
1775 */
1776 {
1777 	struct inode *ip;
1778 	struct ufid *ufhp;
1779 
1780 	ip = VTOI(ap->a_vp);
1781 	ufhp = (struct ufid *)ap->a_fhp;
1782 	ufhp->ufid_len = sizeof(struct ufid);
1783 	ufhp->ufid_ino = ip->i_number;
1784 	ufhp->ufid_gen = ip->i_gen;
1785 	return (0);
1786 }
1787