xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision a0dd79dbdf917a8fbe2762d668f05a7c9f682b22)
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/stat.h>
79 #include <sys/vmmeter.h>
80 #include <sys/vnode.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_extern.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vnode_pager.h>
88 
89 #include <ufs/ufs/extattr.h>
90 #include <ufs/ufs/quota.h>
91 #include <ufs/ufs/inode.h>
92 #include <ufs/ufs/ufs_extern.h>
93 #include <ufs/ufs/ufsmount.h>
94 
95 #include <ufs/ffs/fs.h>
96 #include <ufs/ffs/ffs_extern.h>
97 #include "opt_directio.h"
98 #include "opt_ffs.h"
99 
100 #ifdef DIRECTIO
101 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
102 #endif
103 static vop_fsync_t	ffs_fsync;
104 static vop_lock1_t	ffs_lock;
105 static vop_getpages_t	ffs_getpages;
106 static vop_read_t	ffs_read;
107 static vop_write_t	ffs_write;
108 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
109 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
110 		    struct ucred *cred);
111 static vop_strategy_t	ffsext_strategy;
112 static vop_closeextattr_t	ffs_closeextattr;
113 static vop_deleteextattr_t	ffs_deleteextattr;
114 static vop_getextattr_t	ffs_getextattr;
115 static vop_listextattr_t	ffs_listextattr;
116 static vop_openextattr_t	ffs_openextattr;
117 static vop_setextattr_t	ffs_setextattr;
118 static vop_vptofh_t	ffs_vptofh;
119 
120 
121 /* Global vfs data structures for ufs. */
122 struct vop_vector ffs_vnodeops1 = {
123 	.vop_default =		&ufs_vnodeops,
124 	.vop_fsync =		ffs_fsync,
125 	.vop_getpages =		ffs_getpages,
126 	.vop_lock1 =		ffs_lock,
127 	.vop_read =		ffs_read,
128 	.vop_reallocblks =	ffs_reallocblks,
129 	.vop_write =		ffs_write,
130 	.vop_vptofh =		ffs_vptofh,
131 };
132 
133 struct vop_vector ffs_fifoops1 = {
134 	.vop_default =		&ufs_fifoops,
135 	.vop_fsync =		ffs_fsync,
136 	.vop_reallocblks =	ffs_reallocblks, /* XXX: really ??? */
137 	.vop_vptofh =		ffs_vptofh,
138 };
139 
140 /* Global vfs data structures for ufs. */
141 struct vop_vector ffs_vnodeops2 = {
142 	.vop_default =		&ufs_vnodeops,
143 	.vop_fsync =		ffs_fsync,
144 	.vop_getpages =		ffs_getpages,
145 	.vop_lock1 =		ffs_lock,
146 	.vop_read =		ffs_read,
147 	.vop_reallocblks =	ffs_reallocblks,
148 	.vop_write =		ffs_write,
149 	.vop_closeextattr =	ffs_closeextattr,
150 	.vop_deleteextattr =	ffs_deleteextattr,
151 	.vop_getextattr =	ffs_getextattr,
152 	.vop_listextattr =	ffs_listextattr,
153 	.vop_openextattr =	ffs_openextattr,
154 	.vop_setextattr =	ffs_setextattr,
155 	.vop_vptofh =		ffs_vptofh,
156 };
157 
158 struct vop_vector ffs_fifoops2 = {
159 	.vop_default =		&ufs_fifoops,
160 	.vop_fsync =		ffs_fsync,
161 	.vop_lock1 =		ffs_lock,
162 	.vop_reallocblks =	ffs_reallocblks,
163 	.vop_strategy =		ffsext_strategy,
164 	.vop_closeextattr =	ffs_closeextattr,
165 	.vop_deleteextattr =	ffs_deleteextattr,
166 	.vop_getextattr =	ffs_getextattr,
167 	.vop_listextattr =	ffs_listextattr,
168 	.vop_openextattr =	ffs_openextattr,
169 	.vop_setextattr =	ffs_setextattr,
170 	.vop_vptofh =		ffs_vptofh,
171 };
172 
173 /*
174  * Synch an open file.
175  */
176 /* ARGSUSED */
177 static int
178 ffs_fsync(struct vop_fsync_args *ap)
179 {
180 	struct vnode *vp;
181 	struct bufobj *bo;
182 	int error;
183 
184 	vp = ap->a_vp;
185 	bo = &vp->v_bufobj;
186 retry:
187 	error = ffs_syncvnode(vp, ap->a_waitfor);
188 	if (error)
189 		return (error);
190 	if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
191 		error = softdep_fsync(vp);
192 		if (error)
193 			return (error);
194 
195 		/*
196 		 * The softdep_fsync() function may drop vp lock,
197 		 * allowing for dirty buffers to reappear on the
198 		 * bo_dirty list. Recheck and resync as needed.
199 		 */
200 		BO_LOCK(bo);
201 		if (vp->v_type == VREG && (bo->bo_numoutput > 0 ||
202 		    bo->bo_dirty.bv_cnt > 0)) {
203 			BO_UNLOCK(bo);
204 			goto retry;
205 		}
206 		BO_UNLOCK(bo);
207 	}
208 	return (0);
209 }
210 
211 int
212 ffs_syncvnode(struct vnode *vp, int waitfor)
213 {
214 	struct inode *ip;
215 	struct bufobj *bo;
216 	struct buf *bp;
217 	struct buf *nbp;
218 	ufs_lbn_t lbn;
219 	int error, wait, passes;
220 
221 	ip = VTOI(vp);
222 	ip->i_flag &= ~IN_NEEDSYNC;
223 	bo = &vp->v_bufobj;
224 
225 	/*
226 	 * When doing MNT_WAIT we must first flush all dependencies
227 	 * on the inode.
228 	 */
229 	if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
230 	    (error = softdep_sync_metadata(vp)) != 0)
231 		return (error);
232 
233 	/*
234 	 * Flush all dirty buffers associated with a vnode.
235 	 */
236 	error = 0;
237 	passes = 0;
238 	wait = 0;	/* Always do an async pass first. */
239 	lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
240 	BO_LOCK(bo);
241 loop:
242 	TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
243 		bp->b_vflags &= ~BV_SCANNED;
244 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
245 		/*
246 		 * Reasons to skip this buffer: it has already been considered
247 		 * on this pass, the buffer has dependencies that will cause
248 		 * it to be redirtied and it has not already been deferred,
249 		 * or it is already being written.
250 		 */
251 		if ((bp->b_vflags & BV_SCANNED) != 0)
252 			continue;
253 		bp->b_vflags |= BV_SCANNED;
254 		/* Flush indirects in order. */
255 		if (waitfor == MNT_WAIT && bp->b_lblkno <= -NDADDR &&
256 		    lbn_level(bp->b_lblkno) >= passes)
257 			continue;
258 		if (bp->b_lblkno > lbn)
259 			panic("ffs_syncvnode: syncing truncated data.");
260 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
261 			continue;
262 		BO_UNLOCK(bo);
263 		if ((bp->b_flags & B_DELWRI) == 0)
264 			panic("ffs_fsync: not dirty");
265 		/*
266 		 * Check for dependencies and potentially complete them.
267 		 */
268 		if (!LIST_EMPTY(&bp->b_dep) &&
269 		    (error = softdep_sync_buf(vp, bp,
270 		    wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
271 			/* I/O error. */
272 			if (error != EBUSY) {
273 				BUF_UNLOCK(bp);
274 				return (error);
275 			}
276 			/* If we deferred once, don't defer again. */
277 		    	if ((bp->b_flags & B_DEFERRED) == 0) {
278 				bp->b_flags |= B_DEFERRED;
279 				BUF_UNLOCK(bp);
280 				goto next;
281 			}
282 		}
283 		if (wait) {
284 			bremfree(bp);
285 			if ((error = bwrite(bp)) != 0)
286 				return (error);
287 		} else if ((bp->b_flags & B_CLUSTEROK)) {
288 			(void) vfs_bio_awrite(bp);
289 		} else {
290 			bremfree(bp);
291 			(void) bawrite(bp);
292 		}
293 next:
294 		/*
295 		 * Since we may have slept during the I/O, we need
296 		 * to start from a known point.
297 		 */
298 		BO_LOCK(bo);
299 		nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
300 	}
301 	if (waitfor != MNT_WAIT) {
302 		BO_UNLOCK(bo);
303 		return (ffs_update(vp, waitfor));
304 	}
305 	/* Drain IO to see if we're done. */
306 	bufobj_wwait(bo, 0, 0);
307 	/*
308 	 * Block devices associated with filesystems may have new I/O
309 	 * requests posted for them even if the vnode is locked, so no
310 	 * amount of trying will get them clean.  We make several passes
311 	 * as a best effort.
312 	 *
313 	 * Regular files may need multiple passes to flush all dependency
314 	 * work as it is possible that we must write once per indirect
315 	 * level, once for the leaf, and once for the inode and each of
316 	 * these will be done with one sync and one async pass.
317 	 */
318 	if (bo->bo_dirty.bv_cnt > 0) {
319 		/* Write the inode after sync passes to flush deps. */
320 		if (wait && DOINGSOFTDEP(vp)) {
321 			BO_UNLOCK(bo);
322 			ffs_update(vp, MNT_WAIT);
323 			BO_LOCK(bo);
324 		}
325 		/* switch between sync/async. */
326 		wait = !wait;
327 		if (wait == 1 || ++passes < NIADDR + 2)
328 			goto loop;
329 #ifdef INVARIANTS
330 		if (!vn_isdisk(vp, NULL))
331 			vprint("ffs_fsync: dirty", vp);
332 #endif
333 	}
334 	BO_UNLOCK(bo);
335 	error = ffs_update(vp, MNT_WAIT);
336 	if (DOINGSUJ(vp))
337 		softdep_journal_fsync(VTOI(vp));
338 	return (error);
339 }
340 
341 static int
342 ffs_lock(ap)
343 	struct vop_lock1_args /* {
344 		struct vnode *a_vp;
345 		int a_flags;
346 		struct thread *a_td;
347 		char *file;
348 		int line;
349 	} */ *ap;
350 {
351 #ifndef NO_FFS_SNAPSHOT
352 	struct vnode *vp;
353 	int flags;
354 	struct lock *lkp;
355 	int result;
356 
357 	switch (ap->a_flags & LK_TYPE_MASK) {
358 	case LK_SHARED:
359 	case LK_UPGRADE:
360 	case LK_EXCLUSIVE:
361 		vp = ap->a_vp;
362 		flags = ap->a_flags;
363 		for (;;) {
364 #ifdef DEBUG_VFS_LOCKS
365 			KASSERT(vp->v_holdcnt != 0,
366 			    ("ffs_lock %p: zero hold count", vp));
367 #endif
368 			lkp = vp->v_vnlock;
369 			result = _lockmgr_args(lkp, flags, VI_MTX(vp),
370 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
371 			    ap->a_file, ap->a_line);
372 			if (lkp == vp->v_vnlock || result != 0)
373 				break;
374 			/*
375 			 * Apparent success, except that the vnode
376 			 * mutated between snapshot file vnode and
377 			 * regular file vnode while this process
378 			 * slept.  The lock currently held is not the
379 			 * right lock.  Release it, and try to get the
380 			 * new lock.
381 			 */
382 			(void) _lockmgr_args(lkp, LK_RELEASE, NULL,
383 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
384 			    ap->a_file, ap->a_line);
385 			if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
386 			    (LK_INTERLOCK | LK_NOWAIT))
387 				return (EBUSY);
388 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
389 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
390 			flags &= ~LK_INTERLOCK;
391 		}
392 		break;
393 	default:
394 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
395 	}
396 	return (result);
397 #else
398 	return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
399 #endif
400 }
401 
402 /*
403  * Vnode op for reading.
404  */
405 /* ARGSUSED */
406 static int
407 ffs_read(ap)
408 	struct vop_read_args /* {
409 		struct vnode *a_vp;
410 		struct uio *a_uio;
411 		int a_ioflag;
412 		struct ucred *a_cred;
413 	} */ *ap;
414 {
415 	struct vnode *vp;
416 	struct inode *ip;
417 	struct uio *uio;
418 	struct fs *fs;
419 	struct buf *bp;
420 	ufs_lbn_t lbn, nextlbn;
421 	off_t bytesinfile;
422 	long size, xfersize, blkoffset;
423 	ssize_t orig_resid;
424 	int error;
425 	int seqcount;
426 	int ioflag;
427 
428 	vp = ap->a_vp;
429 	uio = ap->a_uio;
430 	ioflag = ap->a_ioflag;
431 	if (ap->a_ioflag & IO_EXT)
432 #ifdef notyet
433 		return (ffs_extread(vp, uio, ioflag));
434 #else
435 		panic("ffs_read+IO_EXT");
436 #endif
437 #ifdef DIRECTIO
438 	if ((ioflag & IO_DIRECT) != 0) {
439 		int workdone;
440 
441 		error = ffs_rawread(vp, uio, &workdone);
442 		if (error != 0 || workdone != 0)
443 			return error;
444 	}
445 #endif
446 
447 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
448 	ip = VTOI(vp);
449 
450 #ifdef INVARIANTS
451 	if (uio->uio_rw != UIO_READ)
452 		panic("ffs_read: mode");
453 
454 	if (vp->v_type == VLNK) {
455 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
456 			panic("ffs_read: short symlink");
457 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
458 		panic("ffs_read: type %d",  vp->v_type);
459 #endif
460 	orig_resid = uio->uio_resid;
461 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
462 	if (orig_resid == 0)
463 		return (0);
464 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
465 	fs = ip->i_fs;
466 	if (uio->uio_offset < ip->i_size &&
467 	    uio->uio_offset >= fs->fs_maxfilesize)
468 		return (EOVERFLOW);
469 
470 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
471 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
472 			break;
473 		lbn = lblkno(fs, uio->uio_offset);
474 		nextlbn = lbn + 1;
475 
476 		/*
477 		 * size of buffer.  The buffer representing the
478 		 * end of the file is rounded up to the size of
479 		 * the block type ( fragment or full block,
480 		 * depending ).
481 		 */
482 		size = blksize(fs, ip, lbn);
483 		blkoffset = blkoff(fs, uio->uio_offset);
484 
485 		/*
486 		 * The amount we want to transfer in this iteration is
487 		 * one FS block less the amount of the data before
488 		 * our startpoint (duh!)
489 		 */
490 		xfersize = fs->fs_bsize - blkoffset;
491 
492 		/*
493 		 * But if we actually want less than the block,
494 		 * or the file doesn't have a whole block more of data,
495 		 * then use the lesser number.
496 		 */
497 		if (uio->uio_resid < xfersize)
498 			xfersize = uio->uio_resid;
499 		if (bytesinfile < xfersize)
500 			xfersize = bytesinfile;
501 
502 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
503 			/*
504 			 * Don't do readahead if this is the end of the file.
505 			 */
506 			error = bread(vp, lbn, size, NOCRED, &bp);
507 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
508 			/*
509 			 * Otherwise if we are allowed to cluster,
510 			 * grab as much as we can.
511 			 *
512 			 * XXX  This may not be a win if we are not
513 			 * doing sequential access.
514 			 */
515 			error = cluster_read(vp, ip->i_size, lbn,
516 				size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
517 		} else if (seqcount > 1) {
518 			/*
519 			 * If we are NOT allowed to cluster, then
520 			 * if we appear to be acting sequentially,
521 			 * fire off a request for a readahead
522 			 * as well as a read. Note that the 4th and 5th
523 			 * arguments point to arrays of the size specified in
524 			 * the 6th argument.
525 			 */
526 			int nextsize = blksize(fs, ip, nextlbn);
527 			error = breadn(vp, lbn,
528 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
529 		} else {
530 			/*
531 			 * Failing all of the above, just read what the
532 			 * user asked for. Interestingly, the same as
533 			 * the first option above.
534 			 */
535 			error = bread(vp, lbn, size, NOCRED, &bp);
536 		}
537 		if (error) {
538 			brelse(bp);
539 			bp = NULL;
540 			break;
541 		}
542 
543 		/*
544 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
545 		 * will cause us to attempt to release the buffer later on
546 		 * and will cause the buffer cache to attempt to free the
547 		 * underlying pages.
548 		 */
549 		if (ioflag & IO_DIRECT)
550 			bp->b_flags |= B_DIRECT;
551 
552 		/*
553 		 * We should only get non-zero b_resid when an I/O error
554 		 * has occurred, which should cause us to break above.
555 		 * However, if the short read did not cause an error,
556 		 * then we want to ensure that we do not uiomove bad
557 		 * or uninitialized data.
558 		 */
559 		size -= bp->b_resid;
560 		if (size < xfersize) {
561 			if (size == 0)
562 				break;
563 			xfersize = size;
564 		}
565 
566 		error = uiomove((char *)bp->b_data + blkoffset,
567 		    (int)xfersize, uio);
568 		if (error)
569 			break;
570 
571 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
572 		   (LIST_EMPTY(&bp->b_dep))) {
573 			/*
574 			 * If there are no dependencies, and it's VMIO,
575 			 * then we don't need the buf, mark it available
576 			 * for freeing.  For non-direct VMIO reads, the VM
577 			 * has the data.
578 			 */
579 			bp->b_flags |= B_RELBUF;
580 			brelse(bp);
581 		} else {
582 			/*
583 			 * Otherwise let whoever
584 			 * made the request take care of
585 			 * freeing it. We just queue
586 			 * it onto another list.
587 			 */
588 			bqrelse(bp);
589 		}
590 	}
591 
592 	/*
593 	 * This can only happen in the case of an error
594 	 * because the loop above resets bp to NULL on each iteration
595 	 * and on normal completion has not set a new value into it.
596 	 * so it must have come from a 'break' statement
597 	 */
598 	if (bp != NULL) {
599 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
600 		   (LIST_EMPTY(&bp->b_dep))) {
601 			bp->b_flags |= B_RELBUF;
602 			brelse(bp);
603 		} else {
604 			bqrelse(bp);
605 		}
606 	}
607 
608 	if ((error == 0 || uio->uio_resid != orig_resid) &&
609 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
610 	    (ip->i_flag & IN_ACCESS) == 0) {
611 		VI_LOCK(vp);
612 		ip->i_flag |= IN_ACCESS;
613 		VI_UNLOCK(vp);
614 	}
615 	return (error);
616 }
617 
618 /*
619  * Vnode op for writing.
620  */
621 static int
622 ffs_write(ap)
623 	struct vop_write_args /* {
624 		struct vnode *a_vp;
625 		struct uio *a_uio;
626 		int a_ioflag;
627 		struct ucred *a_cred;
628 	} */ *ap;
629 {
630 	struct vnode *vp;
631 	struct uio *uio;
632 	struct inode *ip;
633 	struct fs *fs;
634 	struct buf *bp;
635 	ufs_lbn_t lbn;
636 	off_t osize;
637 	ssize_t resid;
638 	int seqcount;
639 	int blkoffset, error, flags, ioflag, size, xfersize;
640 
641 	vp = ap->a_vp;
642 	uio = ap->a_uio;
643 	ioflag = ap->a_ioflag;
644 	if (ap->a_ioflag & IO_EXT)
645 #ifdef notyet
646 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
647 #else
648 		panic("ffs_write+IO_EXT");
649 #endif
650 
651 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
652 	ip = VTOI(vp);
653 
654 #ifdef INVARIANTS
655 	if (uio->uio_rw != UIO_WRITE)
656 		panic("ffs_write: mode");
657 #endif
658 
659 	switch (vp->v_type) {
660 	case VREG:
661 		if (ioflag & IO_APPEND)
662 			uio->uio_offset = ip->i_size;
663 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
664 			return (EPERM);
665 		/* FALLTHROUGH */
666 	case VLNK:
667 		break;
668 	case VDIR:
669 		panic("ffs_write: dir write");
670 		break;
671 	default:
672 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
673 			(int)uio->uio_offset,
674 			(int)uio->uio_resid
675 		);
676 	}
677 
678 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
679 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
680 	fs = ip->i_fs;
681 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
682 		return (EFBIG);
683 	/*
684 	 * Maybe this should be above the vnode op call, but so long as
685 	 * file servers have no limits, I don't think it matters.
686 	 */
687 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
688 		return (EFBIG);
689 
690 	resid = uio->uio_resid;
691 	osize = ip->i_size;
692 	if (seqcount > BA_SEQMAX)
693 		flags = BA_SEQMAX << BA_SEQSHIFT;
694 	else
695 		flags = seqcount << BA_SEQSHIFT;
696 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
697 		flags |= IO_SYNC;
698 
699 	for (error = 0; uio->uio_resid > 0;) {
700 		lbn = lblkno(fs, uio->uio_offset);
701 		blkoffset = blkoff(fs, uio->uio_offset);
702 		xfersize = fs->fs_bsize - blkoffset;
703 		if (uio->uio_resid < xfersize)
704 			xfersize = uio->uio_resid;
705 		if (uio->uio_offset + xfersize > ip->i_size)
706 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
707 
708                 /*
709 		 * We must perform a read-before-write if the transfer size
710 		 * does not cover the entire buffer.
711                  */
712 		if (fs->fs_bsize > xfersize)
713 			flags |= BA_CLRBUF;
714 		else
715 			flags &= ~BA_CLRBUF;
716 /* XXX is uio->uio_offset the right thing here? */
717 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
718 		    ap->a_cred, flags, &bp);
719 		if (error != 0) {
720 			vnode_pager_setsize(vp, ip->i_size);
721 			break;
722 		}
723 		if (ioflag & IO_DIRECT)
724 			bp->b_flags |= B_DIRECT;
725 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
726 			bp->b_flags |= B_NOCACHE;
727 
728 		if (uio->uio_offset + xfersize > ip->i_size) {
729 			ip->i_size = uio->uio_offset + xfersize;
730 			DIP_SET(ip, i_size, ip->i_size);
731 		}
732 
733 		size = blksize(fs, ip, lbn) - bp->b_resid;
734 		if (size < xfersize)
735 			xfersize = size;
736 
737 		error =
738 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
739 		/*
740 		 * If the buffer is not already filled and we encounter an
741 		 * error while trying to fill it, we have to clear out any
742 		 * garbage data from the pages instantiated for the buffer.
743 		 * If we do not, a failed uiomove() during a write can leave
744 		 * the prior contents of the pages exposed to a userland mmap.
745 		 *
746 		 * Note that we need only clear buffers with a transfer size
747 		 * equal to the block size because buffers with a shorter
748 		 * transfer size were cleared above by the call to UFS_BALLOC()
749 		 * with the BA_CLRBUF flag set.
750 		 *
751 		 * If the source region for uiomove identically mmaps the
752 		 * buffer, uiomove() performed the NOP copy, and the buffer
753 		 * content remains valid because the page fault handler
754 		 * validated the pages.
755 		 */
756 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
757 		    fs->fs_bsize == xfersize)
758 			vfs_bio_clrbuf(bp);
759 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
760 		   (LIST_EMPTY(&bp->b_dep))) {
761 			bp->b_flags |= B_RELBUF;
762 		}
763 
764 		/*
765 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
766 		 * if we have a severe page deficiency write the buffer
767 		 * asynchronously.  Otherwise try to cluster, and if that
768 		 * doesn't do it then either do an async write (if O_DIRECT),
769 		 * or a delayed write (if not).
770 		 */
771 		if (ioflag & IO_SYNC) {
772 			(void)bwrite(bp);
773 		} else if (vm_page_count_severe() ||
774 			    buf_dirty_count_severe() ||
775 			    (ioflag & IO_ASYNC)) {
776 			bp->b_flags |= B_CLUSTEROK;
777 			bawrite(bp);
778 		} else if (xfersize + blkoffset == fs->fs_bsize) {
779 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
780 				bp->b_flags |= B_CLUSTEROK;
781 				cluster_write(vp, bp, ip->i_size, seqcount);
782 			} else {
783 				bawrite(bp);
784 			}
785 		} else if (ioflag & IO_DIRECT) {
786 			bp->b_flags |= B_CLUSTEROK;
787 			bawrite(bp);
788 		} else {
789 			bp->b_flags |= B_CLUSTEROK;
790 			bdwrite(bp);
791 		}
792 		if (error || xfersize == 0)
793 			break;
794 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
795 	}
796 	/*
797 	 * If we successfully wrote any data, and we are not the superuser
798 	 * we clear the setuid and setgid bits as a precaution against
799 	 * tampering.
800 	 */
801 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
802 	    ap->a_cred) {
803 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
804 			ip->i_mode &= ~(ISUID | ISGID);
805 			DIP_SET(ip, i_mode, ip->i_mode);
806 		}
807 	}
808 	if (error) {
809 		if (ioflag & IO_UNIT) {
810 			(void)ffs_truncate(vp, osize,
811 			    IO_NORMAL | (ioflag & IO_SYNC),
812 			    ap->a_cred, uio->uio_td);
813 			uio->uio_offset -= resid - uio->uio_resid;
814 			uio->uio_resid = resid;
815 		}
816 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
817 		error = ffs_update(vp, 1);
818 	return (error);
819 }
820 
821 /*
822  * get page routine
823  */
824 static int
825 ffs_getpages(ap)
826 	struct vop_getpages_args *ap;
827 {
828 	int i;
829 	vm_page_t mreq;
830 	int pcount;
831 
832 	pcount = round_page(ap->a_count) / PAGE_SIZE;
833 	mreq = ap->a_m[ap->a_reqpage];
834 
835 	/*
836 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
837 	 * then the entire page is valid.  Since the page may be mapped,
838 	 * user programs might reference data beyond the actual end of file
839 	 * occuring within the page.  We have to zero that data.
840 	 */
841 	VM_OBJECT_LOCK(mreq->object);
842 	if (mreq->valid) {
843 		if (mreq->valid != VM_PAGE_BITS_ALL)
844 			vm_page_zero_invalid(mreq, TRUE);
845 		for (i = 0; i < pcount; i++) {
846 			if (i != ap->a_reqpage) {
847 				vm_page_lock(ap->a_m[i]);
848 				vm_page_free(ap->a_m[i]);
849 				vm_page_unlock(ap->a_m[i]);
850 			}
851 		}
852 		VM_OBJECT_UNLOCK(mreq->object);
853 		return VM_PAGER_OK;
854 	}
855 	VM_OBJECT_UNLOCK(mreq->object);
856 
857 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
858 					    ap->a_count,
859 					    ap->a_reqpage);
860 }
861 
862 
863 /*
864  * Extended attribute area reading.
865  */
866 static int
867 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
868 {
869 	struct inode *ip;
870 	struct ufs2_dinode *dp;
871 	struct fs *fs;
872 	struct buf *bp;
873 	ufs_lbn_t lbn, nextlbn;
874 	off_t bytesinfile;
875 	long size, xfersize, blkoffset;
876 	ssize_t orig_resid;
877 	int error;
878 
879 	ip = VTOI(vp);
880 	fs = ip->i_fs;
881 	dp = ip->i_din2;
882 
883 #ifdef INVARIANTS
884 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
885 		panic("ffs_extread: mode");
886 
887 #endif
888 	orig_resid = uio->uio_resid;
889 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
890 	if (orig_resid == 0)
891 		return (0);
892 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
893 
894 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
895 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
896 			break;
897 		lbn = lblkno(fs, uio->uio_offset);
898 		nextlbn = lbn + 1;
899 
900 		/*
901 		 * size of buffer.  The buffer representing the
902 		 * end of the file is rounded up to the size of
903 		 * the block type ( fragment or full block,
904 		 * depending ).
905 		 */
906 		size = sblksize(fs, dp->di_extsize, lbn);
907 		blkoffset = blkoff(fs, uio->uio_offset);
908 
909 		/*
910 		 * The amount we want to transfer in this iteration is
911 		 * one FS block less the amount of the data before
912 		 * our startpoint (duh!)
913 		 */
914 		xfersize = fs->fs_bsize - blkoffset;
915 
916 		/*
917 		 * But if we actually want less than the block,
918 		 * or the file doesn't have a whole block more of data,
919 		 * then use the lesser number.
920 		 */
921 		if (uio->uio_resid < xfersize)
922 			xfersize = uio->uio_resid;
923 		if (bytesinfile < xfersize)
924 			xfersize = bytesinfile;
925 
926 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
927 			/*
928 			 * Don't do readahead if this is the end of the info.
929 			 */
930 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
931 		} else {
932 			/*
933 			 * If we have a second block, then
934 			 * fire off a request for a readahead
935 			 * as well as a read. Note that the 4th and 5th
936 			 * arguments point to arrays of the size specified in
937 			 * the 6th argument.
938 			 */
939 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
940 
941 			nextlbn = -1 - nextlbn;
942 			error = breadn(vp, -1 - lbn,
943 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
944 		}
945 		if (error) {
946 			brelse(bp);
947 			bp = NULL;
948 			break;
949 		}
950 
951 		/*
952 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
953 		 * will cause us to attempt to release the buffer later on
954 		 * and will cause the buffer cache to attempt to free the
955 		 * underlying pages.
956 		 */
957 		if (ioflag & IO_DIRECT)
958 			bp->b_flags |= B_DIRECT;
959 
960 		/*
961 		 * We should only get non-zero b_resid when an I/O error
962 		 * has occurred, which should cause us to break above.
963 		 * However, if the short read did not cause an error,
964 		 * then we want to ensure that we do not uiomove bad
965 		 * or uninitialized data.
966 		 */
967 		size -= bp->b_resid;
968 		if (size < xfersize) {
969 			if (size == 0)
970 				break;
971 			xfersize = size;
972 		}
973 
974 		error = uiomove((char *)bp->b_data + blkoffset,
975 					(int)xfersize, uio);
976 		if (error)
977 			break;
978 
979 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
980 		   (LIST_EMPTY(&bp->b_dep))) {
981 			/*
982 			 * If there are no dependencies, and it's VMIO,
983 			 * then we don't need the buf, mark it available
984 			 * for freeing.  For non-direct VMIO reads, the VM
985 			 * has the data.
986 			 */
987 			bp->b_flags |= B_RELBUF;
988 			brelse(bp);
989 		} else {
990 			/*
991 			 * Otherwise let whoever
992 			 * made the request take care of
993 			 * freeing it. We just queue
994 			 * it onto another list.
995 			 */
996 			bqrelse(bp);
997 		}
998 	}
999 
1000 	/*
1001 	 * This can only happen in the case of an error
1002 	 * because the loop above resets bp to NULL on each iteration
1003 	 * and on normal completion has not set a new value into it.
1004 	 * so it must have come from a 'break' statement
1005 	 */
1006 	if (bp != NULL) {
1007 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1008 		   (LIST_EMPTY(&bp->b_dep))) {
1009 			bp->b_flags |= B_RELBUF;
1010 			brelse(bp);
1011 		} else {
1012 			bqrelse(bp);
1013 		}
1014 	}
1015 	return (error);
1016 }
1017 
1018 /*
1019  * Extended attribute area writing.
1020  */
1021 static int
1022 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1023 {
1024 	struct inode *ip;
1025 	struct ufs2_dinode *dp;
1026 	struct fs *fs;
1027 	struct buf *bp;
1028 	ufs_lbn_t lbn;
1029 	off_t osize;
1030 	ssize_t resid;
1031 	int blkoffset, error, flags, size, xfersize;
1032 
1033 	ip = VTOI(vp);
1034 	fs = ip->i_fs;
1035 	dp = ip->i_din2;
1036 
1037 #ifdef INVARIANTS
1038 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1039 		panic("ffs_extwrite: mode");
1040 #endif
1041 
1042 	if (ioflag & IO_APPEND)
1043 		uio->uio_offset = dp->di_extsize;
1044 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1045 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1046 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1047 		return (EFBIG);
1048 
1049 	resid = uio->uio_resid;
1050 	osize = dp->di_extsize;
1051 	flags = IO_EXT;
1052 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1053 		flags |= IO_SYNC;
1054 
1055 	for (error = 0; uio->uio_resid > 0;) {
1056 		lbn = lblkno(fs, uio->uio_offset);
1057 		blkoffset = blkoff(fs, uio->uio_offset);
1058 		xfersize = fs->fs_bsize - blkoffset;
1059 		if (uio->uio_resid < xfersize)
1060 			xfersize = uio->uio_resid;
1061 
1062 		/*
1063 		 * We must perform a read-before-write if the transfer size
1064 		 * does not cover the entire buffer.
1065                  */
1066 		if (fs->fs_bsize > xfersize)
1067 			flags |= BA_CLRBUF;
1068 		else
1069 			flags &= ~BA_CLRBUF;
1070 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1071 		    ucred, flags, &bp);
1072 		if (error != 0)
1073 			break;
1074 		/*
1075 		 * If the buffer is not valid we have to clear out any
1076 		 * garbage data from the pages instantiated for the buffer.
1077 		 * If we do not, a failed uiomove() during a write can leave
1078 		 * the prior contents of the pages exposed to a userland
1079 		 * mmap().  XXX deal with uiomove() errors a better way.
1080 		 */
1081 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1082 			vfs_bio_clrbuf(bp);
1083 		if (ioflag & IO_DIRECT)
1084 			bp->b_flags |= B_DIRECT;
1085 
1086 		if (uio->uio_offset + xfersize > dp->di_extsize)
1087 			dp->di_extsize = uio->uio_offset + xfersize;
1088 
1089 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1090 		if (size < xfersize)
1091 			xfersize = size;
1092 
1093 		error =
1094 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1095 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1096 		   (LIST_EMPTY(&bp->b_dep))) {
1097 			bp->b_flags |= B_RELBUF;
1098 		}
1099 
1100 		/*
1101 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1102 		 * if we have a severe page deficiency write the buffer
1103 		 * asynchronously.  Otherwise try to cluster, and if that
1104 		 * doesn't do it then either do an async write (if O_DIRECT),
1105 		 * or a delayed write (if not).
1106 		 */
1107 		if (ioflag & IO_SYNC) {
1108 			(void)bwrite(bp);
1109 		} else if (vm_page_count_severe() ||
1110 			    buf_dirty_count_severe() ||
1111 			    xfersize + blkoffset == fs->fs_bsize ||
1112 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1113 			bawrite(bp);
1114 		else
1115 			bdwrite(bp);
1116 		if (error || xfersize == 0)
1117 			break;
1118 		ip->i_flag |= IN_CHANGE;
1119 	}
1120 	/*
1121 	 * If we successfully wrote any data, and we are not the superuser
1122 	 * we clear the setuid and setgid bits as a precaution against
1123 	 * tampering.
1124 	 */
1125 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1126 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1127 			ip->i_mode &= ~(ISUID | ISGID);
1128 			dp->di_mode = ip->i_mode;
1129 		}
1130 	}
1131 	if (error) {
1132 		if (ioflag & IO_UNIT) {
1133 			(void)ffs_truncate(vp, osize,
1134 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1135 			uio->uio_offset -= resid - uio->uio_resid;
1136 			uio->uio_resid = resid;
1137 		}
1138 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1139 		error = ffs_update(vp, 1);
1140 	return (error);
1141 }
1142 
1143 
1144 /*
1145  * Vnode operating to retrieve a named extended attribute.
1146  *
1147  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1148  * the length of the EA, and possibly the pointer to the entry and to the data.
1149  */
1150 static int
1151 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1152 {
1153 	u_char *p, *pe, *pn, *p0;
1154 	int eapad1, eapad2, ealength, ealen, nlen;
1155 	uint32_t ul;
1156 
1157 	pe = ptr + length;
1158 	nlen = strlen(name);
1159 
1160 	for (p = ptr; p < pe; p = pn) {
1161 		p0 = p;
1162 		bcopy(p, &ul, sizeof(ul));
1163 		pn = p + ul;
1164 		/* make sure this entry is complete */
1165 		if (pn > pe)
1166 			break;
1167 		p += sizeof(uint32_t);
1168 		if (*p != nspace)
1169 			continue;
1170 		p++;
1171 		eapad2 = *p++;
1172 		if (*p != nlen)
1173 			continue;
1174 		p++;
1175 		if (bcmp(p, name, nlen))
1176 			continue;
1177 		ealength = sizeof(uint32_t) + 3 + nlen;
1178 		eapad1 = 8 - (ealength % 8);
1179 		if (eapad1 == 8)
1180 			eapad1 = 0;
1181 		ealength += eapad1;
1182 		ealen = ul - ealength - eapad2;
1183 		p += nlen + eapad1;
1184 		if (eap != NULL)
1185 			*eap = p0;
1186 		if (eac != NULL)
1187 			*eac = p;
1188 		return (ealen);
1189 	}
1190 	return(-1);
1191 }
1192 
1193 static int
1194 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1195 {
1196 	struct inode *ip;
1197 	struct ufs2_dinode *dp;
1198 	struct fs *fs;
1199 	struct uio luio;
1200 	struct iovec liovec;
1201 	int easize, error;
1202 	u_char *eae;
1203 
1204 	ip = VTOI(vp);
1205 	fs = ip->i_fs;
1206 	dp = ip->i_din2;
1207 	easize = dp->di_extsize;
1208 	if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1209 		return (EFBIG);
1210 
1211 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1212 
1213 	liovec.iov_base = eae;
1214 	liovec.iov_len = easize;
1215 	luio.uio_iov = &liovec;
1216 	luio.uio_iovcnt = 1;
1217 	luio.uio_offset = 0;
1218 	luio.uio_resid = easize;
1219 	luio.uio_segflg = UIO_SYSSPACE;
1220 	luio.uio_rw = UIO_READ;
1221 	luio.uio_td = td;
1222 
1223 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1224 	if (error) {
1225 		free(eae, M_TEMP);
1226 		return(error);
1227 	}
1228 	*p = eae;
1229 	return (0);
1230 }
1231 
1232 static void
1233 ffs_lock_ea(struct vnode *vp)
1234 {
1235 	struct inode *ip;
1236 
1237 	ip = VTOI(vp);
1238 	VI_LOCK(vp);
1239 	while (ip->i_flag & IN_EA_LOCKED) {
1240 		ip->i_flag |= IN_EA_LOCKWAIT;
1241 		msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1242 		    0);
1243 	}
1244 	ip->i_flag |= IN_EA_LOCKED;
1245 	VI_UNLOCK(vp);
1246 }
1247 
1248 static void
1249 ffs_unlock_ea(struct vnode *vp)
1250 {
1251 	struct inode *ip;
1252 
1253 	ip = VTOI(vp);
1254 	VI_LOCK(vp);
1255 	if (ip->i_flag & IN_EA_LOCKWAIT)
1256 		wakeup(&ip->i_ea_refs);
1257 	ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1258 	VI_UNLOCK(vp);
1259 }
1260 
1261 static int
1262 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1263 {
1264 	struct inode *ip;
1265 	struct ufs2_dinode *dp;
1266 	int error;
1267 
1268 	ip = VTOI(vp);
1269 
1270 	ffs_lock_ea(vp);
1271 	if (ip->i_ea_area != NULL) {
1272 		ip->i_ea_refs++;
1273 		ffs_unlock_ea(vp);
1274 		return (0);
1275 	}
1276 	dp = ip->i_din2;
1277 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1278 	if (error) {
1279 		ffs_unlock_ea(vp);
1280 		return (error);
1281 	}
1282 	ip->i_ea_len = dp->di_extsize;
1283 	ip->i_ea_error = 0;
1284 	ip->i_ea_refs++;
1285 	ffs_unlock_ea(vp);
1286 	return (0);
1287 }
1288 
1289 /*
1290  * Vnode extattr transaction commit/abort
1291  */
1292 static int
1293 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1294 {
1295 	struct inode *ip;
1296 	struct uio luio;
1297 	struct iovec liovec;
1298 	int error;
1299 	struct ufs2_dinode *dp;
1300 
1301 	ip = VTOI(vp);
1302 
1303 	ffs_lock_ea(vp);
1304 	if (ip->i_ea_area == NULL) {
1305 		ffs_unlock_ea(vp);
1306 		return (EINVAL);
1307 	}
1308 	dp = ip->i_din2;
1309 	error = ip->i_ea_error;
1310 	if (commit && error == 0) {
1311 		ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1312 		if (cred == NOCRED)
1313 			cred =  vp->v_mount->mnt_cred;
1314 		liovec.iov_base = ip->i_ea_area;
1315 		liovec.iov_len = ip->i_ea_len;
1316 		luio.uio_iov = &liovec;
1317 		luio.uio_iovcnt = 1;
1318 		luio.uio_offset = 0;
1319 		luio.uio_resid = ip->i_ea_len;
1320 		luio.uio_segflg = UIO_SYSSPACE;
1321 		luio.uio_rw = UIO_WRITE;
1322 		luio.uio_td = td;
1323 		/* XXX: I'm not happy about truncating to zero size */
1324 		if (ip->i_ea_len < dp->di_extsize)
1325 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1326 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1327 	}
1328 	if (--ip->i_ea_refs == 0) {
1329 		free(ip->i_ea_area, M_TEMP);
1330 		ip->i_ea_area = NULL;
1331 		ip->i_ea_len = 0;
1332 		ip->i_ea_error = 0;
1333 	}
1334 	ffs_unlock_ea(vp);
1335 	return (error);
1336 }
1337 
1338 /*
1339  * Vnode extattr strategy routine for fifos.
1340  *
1341  * We need to check for a read or write of the external attributes.
1342  * Otherwise we just fall through and do the usual thing.
1343  */
1344 static int
1345 ffsext_strategy(struct vop_strategy_args *ap)
1346 /*
1347 struct vop_strategy_args {
1348 	struct vnodeop_desc *a_desc;
1349 	struct vnode *a_vp;
1350 	struct buf *a_bp;
1351 };
1352 */
1353 {
1354 	struct vnode *vp;
1355 	daddr_t lbn;
1356 
1357 	vp = ap->a_vp;
1358 	lbn = ap->a_bp->b_lblkno;
1359 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1360 	    lbn < 0 && lbn >= -NXADDR)
1361 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1362 	if (vp->v_type == VFIFO)
1363 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1364 	panic("spec nodes went here");
1365 }
1366 
1367 /*
1368  * Vnode extattr transaction commit/abort
1369  */
1370 static int
1371 ffs_openextattr(struct vop_openextattr_args *ap)
1372 /*
1373 struct vop_openextattr_args {
1374 	struct vnodeop_desc *a_desc;
1375 	struct vnode *a_vp;
1376 	IN struct ucred *a_cred;
1377 	IN struct thread *a_td;
1378 };
1379 */
1380 {
1381 	struct inode *ip;
1382 	struct fs *fs;
1383 
1384 	ip = VTOI(ap->a_vp);
1385 	fs = ip->i_fs;
1386 
1387 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1388 		return (EOPNOTSUPP);
1389 
1390 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1391 }
1392 
1393 
1394 /*
1395  * Vnode extattr transaction commit/abort
1396  */
1397 static int
1398 ffs_closeextattr(struct vop_closeextattr_args *ap)
1399 /*
1400 struct vop_closeextattr_args {
1401 	struct vnodeop_desc *a_desc;
1402 	struct vnode *a_vp;
1403 	int a_commit;
1404 	IN struct ucred *a_cred;
1405 	IN struct thread *a_td;
1406 };
1407 */
1408 {
1409 	struct inode *ip;
1410 	struct fs *fs;
1411 
1412 	ip = VTOI(ap->a_vp);
1413 	fs = ip->i_fs;
1414 
1415 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1416 		return (EOPNOTSUPP);
1417 
1418 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1419 		return (EROFS);
1420 
1421 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1422 }
1423 
1424 /*
1425  * Vnode operation to remove a named attribute.
1426  */
1427 static int
1428 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1429 /*
1430 vop_deleteextattr {
1431 	IN struct vnode *a_vp;
1432 	IN int a_attrnamespace;
1433 	IN const char *a_name;
1434 	IN struct ucred *a_cred;
1435 	IN struct thread *a_td;
1436 };
1437 */
1438 {
1439 	struct inode *ip;
1440 	struct fs *fs;
1441 	uint32_t ealength, ul;
1442 	int ealen, olen, eapad1, eapad2, error, i, easize;
1443 	u_char *eae, *p;
1444 
1445 	ip = VTOI(ap->a_vp);
1446 	fs = ip->i_fs;
1447 
1448 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1449 		return (EOPNOTSUPP);
1450 
1451 	if (strlen(ap->a_name) == 0)
1452 		return (EINVAL);
1453 
1454 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1455 		return (EROFS);
1456 
1457 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1458 	    ap->a_cred, ap->a_td, VWRITE);
1459 	if (error) {
1460 
1461 		/*
1462 		 * ffs_lock_ea is not needed there, because the vnode
1463 		 * must be exclusively locked.
1464 		 */
1465 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1466 			ip->i_ea_error = error;
1467 		return (error);
1468 	}
1469 
1470 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1471 	if (error)
1472 		return (error);
1473 
1474 	ealength = eapad1 = ealen = eapad2 = 0;
1475 
1476 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1477 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1478 	easize = ip->i_ea_len;
1479 
1480 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1481 	    &p, NULL);
1482 	if (olen == -1) {
1483 		/* delete but nonexistent */
1484 		free(eae, M_TEMP);
1485 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1486 		return(ENOATTR);
1487 	}
1488 	bcopy(p, &ul, sizeof ul);
1489 	i = p - eae + ul;
1490 	if (ul != ealength) {
1491 		bcopy(p + ul, p + ealength, easize - i);
1492 		easize += (ealength - ul);
1493 	}
1494 	if (easize > NXADDR * fs->fs_bsize) {
1495 		free(eae, M_TEMP);
1496 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1497 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1498 			ip->i_ea_error = ENOSPC;
1499 		return(ENOSPC);
1500 	}
1501 	p = ip->i_ea_area;
1502 	ip->i_ea_area = eae;
1503 	ip->i_ea_len = easize;
1504 	free(p, M_TEMP);
1505 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1506 	return(error);
1507 }
1508 
1509 /*
1510  * Vnode operation to retrieve a named extended attribute.
1511  */
1512 static int
1513 ffs_getextattr(struct vop_getextattr_args *ap)
1514 /*
1515 vop_getextattr {
1516 	IN struct vnode *a_vp;
1517 	IN int a_attrnamespace;
1518 	IN const char *a_name;
1519 	INOUT struct uio *a_uio;
1520 	OUT size_t *a_size;
1521 	IN struct ucred *a_cred;
1522 	IN struct thread *a_td;
1523 };
1524 */
1525 {
1526 	struct inode *ip;
1527 	struct fs *fs;
1528 	u_char *eae, *p;
1529 	unsigned easize;
1530 	int error, ealen;
1531 
1532 	ip = VTOI(ap->a_vp);
1533 	fs = ip->i_fs;
1534 
1535 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1536 		return (EOPNOTSUPP);
1537 
1538 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1539 	    ap->a_cred, ap->a_td, VREAD);
1540 	if (error)
1541 		return (error);
1542 
1543 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1544 	if (error)
1545 		return (error);
1546 
1547 	eae = ip->i_ea_area;
1548 	easize = ip->i_ea_len;
1549 
1550 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1551 	    NULL, &p);
1552 	if (ealen >= 0) {
1553 		error = 0;
1554 		if (ap->a_size != NULL)
1555 			*ap->a_size = ealen;
1556 		else if (ap->a_uio != NULL)
1557 			error = uiomove(p, ealen, ap->a_uio);
1558 	} else
1559 		error = ENOATTR;
1560 
1561 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1562 	return(error);
1563 }
1564 
1565 /*
1566  * Vnode operation to retrieve extended attributes on a vnode.
1567  */
1568 static int
1569 ffs_listextattr(struct vop_listextattr_args *ap)
1570 /*
1571 vop_listextattr {
1572 	IN struct vnode *a_vp;
1573 	IN int a_attrnamespace;
1574 	INOUT struct uio *a_uio;
1575 	OUT size_t *a_size;
1576 	IN struct ucred *a_cred;
1577 	IN struct thread *a_td;
1578 };
1579 */
1580 {
1581 	struct inode *ip;
1582 	struct fs *fs;
1583 	u_char *eae, *p, *pe, *pn;
1584 	unsigned easize;
1585 	uint32_t ul;
1586 	int error, ealen;
1587 
1588 	ip = VTOI(ap->a_vp);
1589 	fs = ip->i_fs;
1590 
1591 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1592 		return (EOPNOTSUPP);
1593 
1594 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1595 	    ap->a_cred, ap->a_td, VREAD);
1596 	if (error)
1597 		return (error);
1598 
1599 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1600 	if (error)
1601 		return (error);
1602 	eae = ip->i_ea_area;
1603 	easize = ip->i_ea_len;
1604 
1605 	error = 0;
1606 	if (ap->a_size != NULL)
1607 		*ap->a_size = 0;
1608 	pe = eae + easize;
1609 	for(p = eae; error == 0 && p < pe; p = pn) {
1610 		bcopy(p, &ul, sizeof(ul));
1611 		pn = p + ul;
1612 		if (pn > pe)
1613 			break;
1614 		p += sizeof(ul);
1615 		if (*p++ != ap->a_attrnamespace)
1616 			continue;
1617 		p++;	/* pad2 */
1618 		ealen = *p;
1619 		if (ap->a_size != NULL) {
1620 			*ap->a_size += ealen + 1;
1621 		} else if (ap->a_uio != NULL) {
1622 			error = uiomove(p, ealen + 1, ap->a_uio);
1623 		}
1624 	}
1625 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1626 	return(error);
1627 }
1628 
1629 /*
1630  * Vnode operation to set a named attribute.
1631  */
1632 static int
1633 ffs_setextattr(struct vop_setextattr_args *ap)
1634 /*
1635 vop_setextattr {
1636 	IN struct vnode *a_vp;
1637 	IN int a_attrnamespace;
1638 	IN const char *a_name;
1639 	INOUT struct uio *a_uio;
1640 	IN struct ucred *a_cred;
1641 	IN struct thread *a_td;
1642 };
1643 */
1644 {
1645 	struct inode *ip;
1646 	struct fs *fs;
1647 	uint32_t ealength, ul;
1648 	int ealen, olen, eapad1, eapad2, error, i, easize;
1649 	u_char *eae, *p;
1650 
1651 	ip = VTOI(ap->a_vp);
1652 	fs = ip->i_fs;
1653 
1654 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1655 		return (EOPNOTSUPP);
1656 
1657 	if (strlen(ap->a_name) == 0)
1658 		return (EINVAL);
1659 
1660 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1661 	if (ap->a_uio == NULL)
1662 		return (EOPNOTSUPP);
1663 
1664 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1665 		return (EROFS);
1666 
1667 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1668 	    ap->a_cred, ap->a_td, VWRITE);
1669 	if (error) {
1670 
1671 		/*
1672 		 * ffs_lock_ea is not needed there, because the vnode
1673 		 * must be exclusively locked.
1674 		 */
1675 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1676 			ip->i_ea_error = error;
1677 		return (error);
1678 	}
1679 
1680 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1681 	if (error)
1682 		return (error);
1683 
1684 	ealen = ap->a_uio->uio_resid;
1685 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1686 	eapad1 = 8 - (ealength % 8);
1687 	if (eapad1 == 8)
1688 		eapad1 = 0;
1689 	eapad2 = 8 - (ealen % 8);
1690 	if (eapad2 == 8)
1691 		eapad2 = 0;
1692 	ealength += eapad1 + ealen + eapad2;
1693 
1694 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1695 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1696 	easize = ip->i_ea_len;
1697 
1698 	olen = ffs_findextattr(eae, easize,
1699 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1700         if (olen == -1) {
1701 		/* new, append at end */
1702 		p = eae + easize;
1703 		easize += ealength;
1704 	} else {
1705 		bcopy(p, &ul, sizeof ul);
1706 		i = p - eae + ul;
1707 		if (ul != ealength) {
1708 			bcopy(p + ul, p + ealength, easize - i);
1709 			easize += (ealength - ul);
1710 		}
1711 	}
1712 	if (easize > NXADDR * fs->fs_bsize) {
1713 		free(eae, M_TEMP);
1714 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1715 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1716 			ip->i_ea_error = ENOSPC;
1717 		return(ENOSPC);
1718 	}
1719 	bcopy(&ealength, p, sizeof(ealength));
1720 	p += sizeof(ealength);
1721 	*p++ = ap->a_attrnamespace;
1722 	*p++ = eapad2;
1723 	*p++ = strlen(ap->a_name);
1724 	strcpy(p, ap->a_name);
1725 	p += strlen(ap->a_name);
1726 	bzero(p, eapad1);
1727 	p += eapad1;
1728 	error = uiomove(p, ealen, ap->a_uio);
1729 	if (error) {
1730 		free(eae, M_TEMP);
1731 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1732 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1733 			ip->i_ea_error = error;
1734 		return(error);
1735 	}
1736 	p += ealen;
1737 	bzero(p, eapad2);
1738 
1739 	p = ip->i_ea_area;
1740 	ip->i_ea_area = eae;
1741 	ip->i_ea_len = easize;
1742 	free(p, M_TEMP);
1743 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1744 	return(error);
1745 }
1746 
1747 /*
1748  * Vnode pointer to File handle
1749  */
1750 static int
1751 ffs_vptofh(struct vop_vptofh_args *ap)
1752 /*
1753 vop_vptofh {
1754 	IN struct vnode *a_vp;
1755 	IN struct fid *a_fhp;
1756 };
1757 */
1758 {
1759 	struct inode *ip;
1760 	struct ufid *ufhp;
1761 
1762 	ip = VTOI(ap->a_vp);
1763 	ufhp = (struct ufid *)ap->a_fhp;
1764 	ufhp->ufid_len = sizeof(struct ufid);
1765 	ufhp->ufid_ino = ip->i_number;
1766 	ufhp->ufid_gen = ip->i_gen;
1767 	return (0);
1768 }
1769