xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision acd3428b7d3e94cef0e1881c868cb4b131d4ff41)
1 /*-
2  * Copyright (c) 2002, 2003 Networks Associates Technology, Inc.
3  * All rights reserved.
4  *
5  * This software was developed for the FreeBSD Project by Marshall
6  * Kirk McKusick and Network Associates Laboratories, the Security
7  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
8  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
9  * research program
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * Copyright (c) 1982, 1986, 1989, 1993
33  *	The Regents of the University of California.  All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 4. Neither the name of the University nor the names of its contributors
44  *    may be used to endorse or promote products derived from this software
45  *    without specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
57  * SUCH DAMAGE.
58  *
59  *	from: @(#)ufs_readwrite.c	8.11 (Berkeley) 5/8/95
60  * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
61  *	@(#)ffs_vnops.c	8.15 (Berkeley) 5/14/95
62  */
63 
64 #include <sys/cdefs.h>
65 __FBSDID("$FreeBSD$");
66 
67 #include <sys/param.h>
68 #include <sys/bio.h>
69 #include <sys/systm.h>
70 #include <sys/buf.h>
71 #include <sys/conf.h>
72 #include <sys/extattr.h>
73 #include <sys/kernel.h>
74 #include <sys/limits.h>
75 #include <sys/malloc.h>
76 #include <sys/mount.h>
77 #include <sys/priv.h>
78 #include <sys/proc.h>
79 #include <sys/resourcevar.h>
80 #include <sys/signalvar.h>
81 #include <sys/stat.h>
82 #include <sys/vmmeter.h>
83 #include <sys/vnode.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pager.h>
90 #include <vm/vnode_pager.h>
91 
92 #include <ufs/ufs/extattr.h>
93 #include <ufs/ufs/quota.h>
94 #include <ufs/ufs/inode.h>
95 #include <ufs/ufs/ufs_extern.h>
96 #include <ufs/ufs/ufsmount.h>
97 
98 #include <ufs/ffs/fs.h>
99 #include <ufs/ffs/ffs_extern.h>
100 #include "opt_directio.h"
101 #include "opt_ffs.h"
102 
103 #ifdef DIRECTIO
104 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
105 #endif
106 static vop_fsync_t	ffs_fsync;
107 static vop_lock_t	ffs_lock;
108 static vop_getpages_t	ffs_getpages;
109 static vop_read_t	ffs_read;
110 static vop_write_t	ffs_write;
111 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
112 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
113 		    struct ucred *cred);
114 static vop_strategy_t	ffsext_strategy;
115 static vop_closeextattr_t	ffs_closeextattr;
116 static vop_deleteextattr_t	ffs_deleteextattr;
117 static vop_getextattr_t	ffs_getextattr;
118 static vop_listextattr_t	ffs_listextattr;
119 static vop_openextattr_t	ffs_openextattr;
120 static vop_setextattr_t	ffs_setextattr;
121 
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_getpages =		ffs_getpages,
128 	.vop_lock =		ffs_lock,
129 	.vop_read =		ffs_read,
130 	.vop_reallocblks =	ffs_reallocblks,
131 	.vop_write =		ffs_write,
132 };
133 
134 struct vop_vector ffs_fifoops1 = {
135 	.vop_default =		&ufs_fifoops,
136 	.vop_fsync =		ffs_fsync,
137 	.vop_reallocblks =	ffs_reallocblks, /* XXX: really ??? */
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_lock =		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 };
156 
157 struct vop_vector ffs_fifoops2 = {
158 	.vop_default =		&ufs_fifoops,
159 	.vop_fsync =		ffs_fsync,
160 	.vop_lock =		ffs_lock,
161 	.vop_reallocblks =	ffs_reallocblks,
162 	.vop_strategy =		ffsext_strategy,
163 	.vop_closeextattr =	ffs_closeextattr,
164 	.vop_deleteextattr =	ffs_deleteextattr,
165 	.vop_getextattr =	ffs_getextattr,
166 	.vop_listextattr =	ffs_listextattr,
167 	.vop_openextattr =	ffs_openextattr,
168 	.vop_setextattr =	ffs_setextattr,
169 };
170 
171 /*
172  * Synch an open file.
173  */
174 /* ARGSUSED */
175 static int
176 ffs_fsync(struct vop_fsync_args *ap)
177 {
178 	int error;
179 
180 	error = ffs_syncvnode(ap->a_vp, ap->a_waitfor);
181 	if (error)
182 		return (error);
183 	if (ap->a_waitfor == MNT_WAIT &&
184 	    (ap->a_vp->v_mount->mnt_flag & MNT_SOFTDEP))
185                 error = softdep_fsync(ap->a_vp);
186 	return (error);
187 }
188 
189 int
190 ffs_syncvnode(struct vnode *vp, int waitfor)
191 {
192 	struct inode *ip = VTOI(vp);
193 	struct buf *bp;
194 	struct buf *nbp;
195 	int s, error, wait, passes, skipmeta;
196 	ufs_lbn_t lbn;
197 
198 	wait = (waitfor == MNT_WAIT);
199 	lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
200 
201 	/*
202 	 * Flush all dirty buffers associated with a vnode.
203 	 */
204 	passes = NIADDR + 1;
205 	skipmeta = 0;
206 	if (wait)
207 		skipmeta = 1;
208 	s = splbio();
209 	VI_LOCK(vp);
210 loop:
211 	TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs)
212 		bp->b_vflags &= ~BV_SCANNED;
213 	TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
214 		/*
215 		 * Reasons to skip this buffer: it has already been considered
216 		 * on this pass, this pass is the first time through on a
217 		 * synchronous flush request and the buffer being considered
218 		 * is metadata, the buffer has dependencies that will cause
219 		 * it to be redirtied and it has not already been deferred,
220 		 * or it is already being written.
221 		 */
222 		if ((bp->b_vflags & BV_SCANNED) != 0)
223 			continue;
224 		bp->b_vflags |= BV_SCANNED;
225 		if ((skipmeta == 1 && bp->b_lblkno < 0))
226 			continue;
227 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
228 			continue;
229 		VI_UNLOCK(vp);
230 		if (!wait && LIST_FIRST(&bp->b_dep) != NULL &&
231 		    (bp->b_flags & B_DEFERRED) == 0 &&
232 		    buf_countdeps(bp, 0)) {
233 			bp->b_flags |= B_DEFERRED;
234 			BUF_UNLOCK(bp);
235 			VI_LOCK(vp);
236 			continue;
237 		}
238 		if ((bp->b_flags & B_DELWRI) == 0)
239 			panic("ffs_fsync: not dirty");
240 		/*
241 		 * If this is a synchronous flush request, or it is not a
242 		 * file or device, start the write on this buffer immediatly.
243 		 */
244 		if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
245 
246 			/*
247 			 * On our final pass through, do all I/O synchronously
248 			 * so that we can find out if our flush is failing
249 			 * because of write errors.
250 			 */
251 			if (passes > 0 || !wait) {
252 				if ((bp->b_flags & B_CLUSTEROK) && !wait) {
253 					(void) vfs_bio_awrite(bp);
254 				} else {
255 					bremfree(bp);
256 					splx(s);
257 					(void) bawrite(bp);
258 					s = splbio();
259 				}
260 			} else {
261 				bremfree(bp);
262 				splx(s);
263 				if ((error = bwrite(bp)) != 0)
264 					return (error);
265 				s = splbio();
266 			}
267 		} else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
268 			/*
269 			 * If the buffer is for data that has been truncated
270 			 * off the file, then throw it away.
271 			 */
272 			bremfree(bp);
273 			bp->b_flags |= B_INVAL | B_NOCACHE;
274 			splx(s);
275 			brelse(bp);
276 			s = splbio();
277 		} else
278 			vfs_bio_awrite(bp);
279 
280 		/*
281 		 * Since we may have slept during the I/O, we need
282 		 * to start from a known point.
283 		 */
284 		VI_LOCK(vp);
285 		nbp = TAILQ_FIRST(&vp->v_bufobj.bo_dirty.bv_hd);
286 	}
287 	/*
288 	 * If we were asked to do this synchronously, then go back for
289 	 * another pass, this time doing the metadata.
290 	 */
291 	if (skipmeta) {
292 		skipmeta = 0;
293 		goto loop;
294 	}
295 
296 	if (wait) {
297 		bufobj_wwait(&vp->v_bufobj, 3, 0);
298 		VI_UNLOCK(vp);
299 
300 		/*
301 		 * Ensure that any filesystem metatdata associated
302 		 * with the vnode has been written.
303 		 */
304 		splx(s);
305 		if ((error = softdep_sync_metadata(vp)) != 0)
306 			return (error);
307 		s = splbio();
308 
309 		VI_LOCK(vp);
310 		if (vp->v_bufobj.bo_dirty.bv_cnt > 0) {
311 			/*
312 			 * Block devices associated with filesystems may
313 			 * have new I/O requests posted for them even if
314 			 * the vnode is locked, so no amount of trying will
315 			 * get them clean. Thus we give block devices a
316 			 * good effort, then just give up. For all other file
317 			 * types, go around and try again until it is clean.
318 			 */
319 			if (passes > 0) {
320 				passes -= 1;
321 				goto loop;
322 			}
323 #ifdef DIAGNOSTIC
324 			if (!vn_isdisk(vp, NULL))
325 				vprint("ffs_fsync: dirty", vp);
326 #endif
327 		}
328 	}
329 	VI_UNLOCK(vp);
330 	splx(s);
331 	return (ffs_update(vp, wait));
332 }
333 
334 static int
335 ffs_lock(ap)
336 	struct vop_lock_args /* {
337 		struct vnode *a_vp;
338 		int a_flags;
339 		struct thread *a_td;
340 	} */ *ap;
341 {
342 #ifndef NO_FFS_SNAPSHOT
343 	struct vnode *vp;
344 	int flags;
345 	struct lock *lkp;
346 	int result;
347 
348 	switch (ap->a_flags & LK_TYPE_MASK) {
349 	case LK_SHARED:
350 	case LK_UPGRADE:
351 	case LK_EXCLUSIVE:
352 		vp = ap->a_vp;
353 		flags = ap->a_flags;
354 		for (;;) {
355 			/*
356 			 * vnode interlock must be held to ensure that
357 			 * the possibly external lock isn't freed,
358 			 * e.g. when mutating from snapshot file vnode
359 			 * to regular file vnode.
360 			 */
361 			if ((flags & LK_INTERLOCK) == 0) {
362 				VI_LOCK(vp);
363 				flags |= LK_INTERLOCK;
364 			}
365 			lkp = vp->v_vnlock;
366 			result = lockmgr(lkp, flags, VI_MTX(vp), ap->a_td);
367 			if (lkp == vp->v_vnlock || result != 0)
368 				break;
369 			/*
370 			 * Apparent success, except that the vnode
371 			 * mutated between snapshot file vnode and
372 			 * regular file vnode while this process
373 			 * slept.  The lock currently held is not the
374 			 * right lock.  Release it, and try to get the
375 			 * new lock.
376 			 */
377 			(void) lockmgr(lkp, LK_RELEASE, VI_MTX(vp), ap->a_td);
378 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
379 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
380 			flags &= ~LK_INTERLOCK;
381 		}
382 		break;
383 	default:
384 		result = VOP_LOCK_APV(&ufs_vnodeops, ap);
385 	}
386 	return (result);
387 #else
388 	return (VOP_LOCK_APV(&ufs_vnodeops, ap));
389 #endif
390 }
391 
392 /*
393  * Vnode op for reading.
394  */
395 /* ARGSUSED */
396 static int
397 ffs_read(ap)
398 	struct vop_read_args /* {
399 		struct vnode *a_vp;
400 		struct uio *a_uio;
401 		int a_ioflag;
402 		struct ucred *a_cred;
403 	} */ *ap;
404 {
405 	struct vnode *vp;
406 	struct inode *ip;
407 	struct uio *uio;
408 	struct fs *fs;
409 	struct buf *bp;
410 	ufs_lbn_t lbn, nextlbn;
411 	off_t bytesinfile;
412 	long size, xfersize, blkoffset;
413 	int error, orig_resid;
414 	int seqcount;
415 	int ioflag;
416 
417 	vp = ap->a_vp;
418 	uio = ap->a_uio;
419 	ioflag = ap->a_ioflag;
420 	if (ap->a_ioflag & IO_EXT)
421 #ifdef notyet
422 		return (ffs_extread(vp, uio, ioflag));
423 #else
424 		panic("ffs_read+IO_EXT");
425 #endif
426 #ifdef DIRECTIO
427 	if ((ioflag & IO_DIRECT) != 0) {
428 		int workdone;
429 
430 		error = ffs_rawread(vp, uio, &workdone);
431 		if (error != 0 || workdone != 0)
432 			return error;
433 	}
434 #endif
435 
436 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
437 	ip = VTOI(vp);
438 
439 #ifdef DIAGNOSTIC
440 	if (uio->uio_rw != UIO_READ)
441 		panic("ffs_read: mode");
442 
443 	if (vp->v_type == VLNK) {
444 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
445 			panic("ffs_read: short symlink");
446 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
447 		panic("ffs_read: type %d",  vp->v_type);
448 #endif
449 	orig_resid = uio->uio_resid;
450 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
451 	if (orig_resid == 0)
452 		return (0);
453 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
454 	fs = ip->i_fs;
455 	if (uio->uio_offset < ip->i_size &&
456 	    uio->uio_offset >= fs->fs_maxfilesize)
457 		return (EOVERFLOW);
458 
459 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
460 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
461 			break;
462 		lbn = lblkno(fs, uio->uio_offset);
463 		nextlbn = lbn + 1;
464 
465 		/*
466 		 * size of buffer.  The buffer representing the
467 		 * end of the file is rounded up to the size of
468 		 * the block type ( fragment or full block,
469 		 * depending ).
470 		 */
471 		size = blksize(fs, ip, lbn);
472 		blkoffset = blkoff(fs, uio->uio_offset);
473 
474 		/*
475 		 * The amount we want to transfer in this iteration is
476 		 * one FS block less the amount of the data before
477 		 * our startpoint (duh!)
478 		 */
479 		xfersize = fs->fs_bsize - blkoffset;
480 
481 		/*
482 		 * But if we actually want less than the block,
483 		 * or the file doesn't have a whole block more of data,
484 		 * then use the lesser number.
485 		 */
486 		if (uio->uio_resid < xfersize)
487 			xfersize = uio->uio_resid;
488 		if (bytesinfile < xfersize)
489 			xfersize = bytesinfile;
490 
491 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
492 			/*
493 			 * Don't do readahead if this is the end of the file.
494 			 */
495 			error = bread(vp, lbn, size, NOCRED, &bp);
496 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
497 			/*
498 			 * Otherwise if we are allowed to cluster,
499 			 * grab as much as we can.
500 			 *
501 			 * XXX  This may not be a win if we are not
502 			 * doing sequential access.
503 			 */
504 			error = cluster_read(vp, ip->i_size, lbn,
505 				size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
506 		} else if (seqcount > 1) {
507 			/*
508 			 * If we are NOT allowed to cluster, then
509 			 * if we appear to be acting sequentially,
510 			 * fire off a request for a readahead
511 			 * as well as a read. Note that the 4th and 5th
512 			 * arguments point to arrays of the size specified in
513 			 * the 6th argument.
514 			 */
515 			int nextsize = blksize(fs, ip, nextlbn);
516 			error = breadn(vp, lbn,
517 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
518 		} else {
519 			/*
520 			 * Failing all of the above, just read what the
521 			 * user asked for. Interestingly, the same as
522 			 * the first option above.
523 			 */
524 			error = bread(vp, lbn, size, NOCRED, &bp);
525 		}
526 		if (error) {
527 			brelse(bp);
528 			bp = NULL;
529 			break;
530 		}
531 
532 		/*
533 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
534 		 * will cause us to attempt to release the buffer later on
535 		 * and will cause the buffer cache to attempt to free the
536 		 * underlying pages.
537 		 */
538 		if (ioflag & IO_DIRECT)
539 			bp->b_flags |= B_DIRECT;
540 
541 		/*
542 		 * We should only get non-zero b_resid when an I/O error
543 		 * has occurred, which should cause us to break above.
544 		 * However, if the short read did not cause an error,
545 		 * then we want to ensure that we do not uiomove bad
546 		 * or uninitialized data.
547 		 */
548 		size -= bp->b_resid;
549 		if (size < xfersize) {
550 			if (size == 0)
551 				break;
552 			xfersize = size;
553 		}
554 
555 		error = uiomove((char *)bp->b_data + blkoffset,
556 		    (int)xfersize, uio);
557 		if (error)
558 			break;
559 
560 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
561 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
562 			/*
563 			 * If there are no dependencies, and it's VMIO,
564 			 * then we don't need the buf, mark it available
565 			 * for freeing. The VM has the data.
566 			 */
567 			bp->b_flags |= B_RELBUF;
568 			brelse(bp);
569 		} else {
570 			/*
571 			 * Otherwise let whoever
572 			 * made the request take care of
573 			 * freeing it. We just queue
574 			 * it onto another list.
575 			 */
576 			bqrelse(bp);
577 		}
578 	}
579 
580 	/*
581 	 * This can only happen in the case of an error
582 	 * because the loop above resets bp to NULL on each iteration
583 	 * and on normal completion has not set a new value into it.
584 	 * so it must have come from a 'break' statement
585 	 */
586 	if (bp != NULL) {
587 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
588 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
589 			bp->b_flags |= B_RELBUF;
590 			brelse(bp);
591 		} else {
592 			bqrelse(bp);
593 		}
594 	}
595 
596 	if ((error == 0 || uio->uio_resid != orig_resid) &&
597 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0) {
598 		VI_LOCK(vp);
599 		ip->i_flag |= IN_ACCESS;
600 		VI_UNLOCK(vp);
601 	}
602 	return (error);
603 }
604 
605 /*
606  * Vnode op for writing.
607  */
608 static int
609 ffs_write(ap)
610 	struct vop_write_args /* {
611 		struct vnode *a_vp;
612 		struct uio *a_uio;
613 		int a_ioflag;
614 		struct ucred *a_cred;
615 	} */ *ap;
616 {
617 	struct vnode *vp;
618 	struct uio *uio;
619 	struct inode *ip;
620 	struct fs *fs;
621 	struct buf *bp;
622 	struct thread *td;
623 	ufs_lbn_t lbn;
624 	off_t osize;
625 	int seqcount;
626 	int blkoffset, error, flags, ioflag, resid, size, xfersize;
627 
628 	vp = ap->a_vp;
629 	uio = ap->a_uio;
630 	ioflag = ap->a_ioflag;
631 	if (ap->a_ioflag & IO_EXT)
632 #ifdef notyet
633 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
634 #else
635 		panic("ffs_write+IO_EXT");
636 #endif
637 
638 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
639 	ip = VTOI(vp);
640 
641 #ifdef DIAGNOSTIC
642 	if (uio->uio_rw != UIO_WRITE)
643 		panic("ffs_write: mode");
644 #endif
645 
646 	switch (vp->v_type) {
647 	case VREG:
648 		if (ioflag & IO_APPEND)
649 			uio->uio_offset = ip->i_size;
650 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
651 			return (EPERM);
652 		/* FALLTHROUGH */
653 	case VLNK:
654 		break;
655 	case VDIR:
656 		panic("ffs_write: dir write");
657 		break;
658 	default:
659 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
660 			(int)uio->uio_offset,
661 			(int)uio->uio_resid
662 		);
663 	}
664 
665 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
666 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
667 	fs = ip->i_fs;
668 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
669 		return (EFBIG);
670 	/*
671 	 * Maybe this should be above the vnode op call, but so long as
672 	 * file servers have no limits, I don't think it matters.
673 	 */
674 	td = uio->uio_td;
675 	if (vp->v_type == VREG && td != NULL) {
676 		PROC_LOCK(td->td_proc);
677 		if (uio->uio_offset + uio->uio_resid >
678 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
679 			psignal(td->td_proc, SIGXFSZ);
680 			PROC_UNLOCK(td->td_proc);
681 			return (EFBIG);
682 		}
683 		PROC_UNLOCK(td->td_proc);
684 	}
685 
686 	resid = uio->uio_resid;
687 	osize = ip->i_size;
688 	if (seqcount > BA_SEQMAX)
689 		flags = BA_SEQMAX << BA_SEQSHIFT;
690 	else
691 		flags = seqcount << BA_SEQSHIFT;
692 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
693 		flags |= IO_SYNC;
694 
695 	for (error = 0; uio->uio_resid > 0;) {
696 		lbn = lblkno(fs, uio->uio_offset);
697 		blkoffset = blkoff(fs, uio->uio_offset);
698 		xfersize = fs->fs_bsize - blkoffset;
699 		if (uio->uio_resid < xfersize)
700 			xfersize = uio->uio_resid;
701 		if (uio->uio_offset + xfersize > ip->i_size)
702 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
703 
704                 /*
705 		 * We must perform a read-before-write if the transfer size
706 		 * does not cover the entire buffer.
707                  */
708 		if (fs->fs_bsize > xfersize)
709 			flags |= BA_CLRBUF;
710 		else
711 			flags &= ~BA_CLRBUF;
712 /* XXX is uio->uio_offset the right thing here? */
713 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
714 		    ap->a_cred, flags, &bp);
715 		if (error != 0)
716 			break;
717 		/*
718 		 * If the buffer is not valid we have to clear out any
719 		 * garbage data from the pages instantiated for the buffer.
720 		 * If we do not, a failed uiomove() during a write can leave
721 		 * the prior contents of the pages exposed to a userland
722 		 * mmap().  XXX deal with uiomove() errors a better way.
723 		 */
724 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
725 			vfs_bio_clrbuf(bp);
726 		if (ioflag & IO_DIRECT)
727 			bp->b_flags |= B_DIRECT;
728 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
729 			bp->b_flags |= B_NOCACHE;
730 
731 		if (uio->uio_offset + xfersize > ip->i_size) {
732 			ip->i_size = uio->uio_offset + xfersize;
733 			DIP_SET(ip, i_size, ip->i_size);
734 		}
735 
736 		size = blksize(fs, ip, lbn) - bp->b_resid;
737 		if (size < xfersize)
738 			xfersize = size;
739 
740 		error =
741 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
742 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
743 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
744 			bp->b_flags |= B_RELBUF;
745 		}
746 
747 		/*
748 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
749 		 * if we have a severe page deficiency write the buffer
750 		 * asynchronously.  Otherwise try to cluster, and if that
751 		 * doesn't do it then either do an async write (if O_DIRECT),
752 		 * or a delayed write (if not).
753 		 */
754 		if (ioflag & IO_SYNC) {
755 			(void)bwrite(bp);
756 		} else if (vm_page_count_severe() ||
757 			    buf_dirty_count_severe() ||
758 			    (ioflag & IO_ASYNC)) {
759 			bp->b_flags |= B_CLUSTEROK;
760 			bawrite(bp);
761 		} else if (xfersize + blkoffset == fs->fs_bsize) {
762 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
763 				bp->b_flags |= B_CLUSTEROK;
764 				cluster_write(vp, bp, ip->i_size, seqcount);
765 			} else {
766 				bawrite(bp);
767 			}
768 		} else if (ioflag & IO_DIRECT) {
769 			bp->b_flags |= B_CLUSTEROK;
770 			bawrite(bp);
771 		} else {
772 			bp->b_flags |= B_CLUSTEROK;
773 			bdwrite(bp);
774 		}
775 		if (error || xfersize == 0)
776 			break;
777 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
778 	}
779 	/*
780 	 * If we successfully wrote any data, and we are not the superuser
781 	 * we clear the setuid and setgid bits as a precaution against
782 	 * tampering.
783 	 */
784 	if (resid > uio->uio_resid && ap->a_cred &&
785 	    priv_check_cred(ap->a_cred, PRIV_VFS_CLEARSUGID,
786 	    SUSER_ALLOWJAIL)) {
787 		ip->i_mode &= ~(ISUID | ISGID);
788 		DIP_SET(ip, i_mode, ip->i_mode);
789 	}
790 	if (error) {
791 		if (ioflag & IO_UNIT) {
792 			(void)ffs_truncate(vp, osize,
793 			    IO_NORMAL | (ioflag & IO_SYNC),
794 			    ap->a_cred, uio->uio_td);
795 			uio->uio_offset -= resid - uio->uio_resid;
796 			uio->uio_resid = resid;
797 		}
798 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
799 		error = ffs_update(vp, 1);
800 	return (error);
801 }
802 
803 /*
804  * get page routine
805  */
806 static int
807 ffs_getpages(ap)
808 	struct vop_getpages_args *ap;
809 {
810 	int i;
811 	vm_page_t mreq;
812 	int pcount;
813 
814 	pcount = round_page(ap->a_count) / PAGE_SIZE;
815 	mreq = ap->a_m[ap->a_reqpage];
816 
817 	/*
818 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
819 	 * then the entire page is valid.  Since the page may be mapped,
820 	 * user programs might reference data beyond the actual end of file
821 	 * occuring within the page.  We have to zero that data.
822 	 */
823 	VM_OBJECT_LOCK(mreq->object);
824 	if (mreq->valid) {
825 		if (mreq->valid != VM_PAGE_BITS_ALL)
826 			vm_page_zero_invalid(mreq, TRUE);
827 		vm_page_lock_queues();
828 		for (i = 0; i < pcount; i++) {
829 			if (i != ap->a_reqpage) {
830 				vm_page_free(ap->a_m[i]);
831 			}
832 		}
833 		vm_page_unlock_queues();
834 		VM_OBJECT_UNLOCK(mreq->object);
835 		return VM_PAGER_OK;
836 	}
837 	VM_OBJECT_UNLOCK(mreq->object);
838 
839 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
840 					    ap->a_count,
841 					    ap->a_reqpage);
842 }
843 
844 
845 /*
846  * Extended attribute area reading.
847  */
848 static int
849 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
850 {
851 	struct inode *ip;
852 	struct ufs2_dinode *dp;
853 	struct fs *fs;
854 	struct buf *bp;
855 	ufs_lbn_t lbn, nextlbn;
856 	off_t bytesinfile;
857 	long size, xfersize, blkoffset;
858 	int error, orig_resid;
859 
860 	ip = VTOI(vp);
861 	fs = ip->i_fs;
862 	dp = ip->i_din2;
863 
864 #ifdef DIAGNOSTIC
865 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
866 		panic("ffs_extread: mode");
867 
868 #endif
869 	orig_resid = uio->uio_resid;
870 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
871 	if (orig_resid == 0)
872 		return (0);
873 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
874 
875 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
876 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
877 			break;
878 		lbn = lblkno(fs, uio->uio_offset);
879 		nextlbn = lbn + 1;
880 
881 		/*
882 		 * size of buffer.  The buffer representing the
883 		 * end of the file is rounded up to the size of
884 		 * the block type ( fragment or full block,
885 		 * depending ).
886 		 */
887 		size = sblksize(fs, dp->di_extsize, lbn);
888 		blkoffset = blkoff(fs, uio->uio_offset);
889 
890 		/*
891 		 * The amount we want to transfer in this iteration is
892 		 * one FS block less the amount of the data before
893 		 * our startpoint (duh!)
894 		 */
895 		xfersize = fs->fs_bsize - blkoffset;
896 
897 		/*
898 		 * But if we actually want less than the block,
899 		 * or the file doesn't have a whole block more of data,
900 		 * then use the lesser number.
901 		 */
902 		if (uio->uio_resid < xfersize)
903 			xfersize = uio->uio_resid;
904 		if (bytesinfile < xfersize)
905 			xfersize = bytesinfile;
906 
907 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
908 			/*
909 			 * Don't do readahead if this is the end of the info.
910 			 */
911 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
912 		} else {
913 			/*
914 			 * If we have a second block, then
915 			 * fire off a request for a readahead
916 			 * as well as a read. Note that the 4th and 5th
917 			 * arguments point to arrays of the size specified in
918 			 * the 6th argument.
919 			 */
920 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
921 
922 			nextlbn = -1 - nextlbn;
923 			error = breadn(vp, -1 - lbn,
924 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
925 		}
926 		if (error) {
927 			brelse(bp);
928 			bp = NULL;
929 			break;
930 		}
931 
932 		/*
933 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
934 		 * will cause us to attempt to release the buffer later on
935 		 * and will cause the buffer cache to attempt to free the
936 		 * underlying pages.
937 		 */
938 		if (ioflag & IO_DIRECT)
939 			bp->b_flags |= B_DIRECT;
940 
941 		/*
942 		 * We should only get non-zero b_resid when an I/O error
943 		 * has occurred, which should cause us to break above.
944 		 * However, if the short read did not cause an error,
945 		 * then we want to ensure that we do not uiomove bad
946 		 * or uninitialized data.
947 		 */
948 		size -= bp->b_resid;
949 		if (size < xfersize) {
950 			if (size == 0)
951 				break;
952 			xfersize = size;
953 		}
954 
955 		error = uiomove((char *)bp->b_data + blkoffset,
956 					(int)xfersize, uio);
957 		if (error)
958 			break;
959 
960 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
961 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
962 			/*
963 			 * If there are no dependencies, and it's VMIO,
964 			 * then we don't need the buf, mark it available
965 			 * for freeing. The VM has the data.
966 			 */
967 			bp->b_flags |= B_RELBUF;
968 			brelse(bp);
969 		} else {
970 			/*
971 			 * Otherwise let whoever
972 			 * made the request take care of
973 			 * freeing it. We just queue
974 			 * it onto another list.
975 			 */
976 			bqrelse(bp);
977 		}
978 	}
979 
980 	/*
981 	 * This can only happen in the case of an error
982 	 * because the loop above resets bp to NULL on each iteration
983 	 * and on normal completion has not set a new value into it.
984 	 * so it must have come from a 'break' statement
985 	 */
986 	if (bp != NULL) {
987 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
988 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
989 			bp->b_flags |= B_RELBUF;
990 			brelse(bp);
991 		} else {
992 			bqrelse(bp);
993 		}
994 	}
995 
996 	if ((error == 0 || uio->uio_resid != orig_resid) &&
997 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0) {
998 		VI_LOCK(vp);
999 		ip->i_flag |= IN_ACCESS;
1000 		VI_UNLOCK(vp);
1001 	}
1002 	return (error);
1003 }
1004 
1005 /*
1006  * Extended attribute area writing.
1007  */
1008 static int
1009 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1010 {
1011 	struct inode *ip;
1012 	struct ufs2_dinode *dp;
1013 	struct fs *fs;
1014 	struct buf *bp;
1015 	ufs_lbn_t lbn;
1016 	off_t osize;
1017 	int blkoffset, error, flags, resid, size, xfersize;
1018 
1019 	ip = VTOI(vp);
1020 	fs = ip->i_fs;
1021 	dp = ip->i_din2;
1022 
1023 #ifdef DIAGNOSTIC
1024 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1025 		panic("ffs_extwrite: mode");
1026 #endif
1027 
1028 	if (ioflag & IO_APPEND)
1029 		uio->uio_offset = dp->di_extsize;
1030 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1031 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1032 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1033 		return (EFBIG);
1034 
1035 	resid = uio->uio_resid;
1036 	osize = dp->di_extsize;
1037 	flags = IO_EXT;
1038 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1039 		flags |= IO_SYNC;
1040 
1041 	for (error = 0; uio->uio_resid > 0;) {
1042 		lbn = lblkno(fs, uio->uio_offset);
1043 		blkoffset = blkoff(fs, uio->uio_offset);
1044 		xfersize = fs->fs_bsize - blkoffset;
1045 		if (uio->uio_resid < xfersize)
1046 			xfersize = uio->uio_resid;
1047 
1048                 /*
1049 		 * We must perform a read-before-write if the transfer size
1050 		 * does not cover the entire buffer.
1051                  */
1052 		if (fs->fs_bsize > xfersize)
1053 			flags |= BA_CLRBUF;
1054 		else
1055 			flags &= ~BA_CLRBUF;
1056 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1057 		    ucred, flags, &bp);
1058 		if (error != 0)
1059 			break;
1060 		/*
1061 		 * If the buffer is not valid we have to clear out any
1062 		 * garbage data from the pages instantiated for the buffer.
1063 		 * If we do not, a failed uiomove() during a write can leave
1064 		 * the prior contents of the pages exposed to a userland
1065 		 * mmap().  XXX deal with uiomove() errors a better way.
1066 		 */
1067 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1068 			vfs_bio_clrbuf(bp);
1069 		if (ioflag & IO_DIRECT)
1070 			bp->b_flags |= B_DIRECT;
1071 
1072 		if (uio->uio_offset + xfersize > dp->di_extsize)
1073 			dp->di_extsize = uio->uio_offset + xfersize;
1074 
1075 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1076 		if (size < xfersize)
1077 			xfersize = size;
1078 
1079 		error =
1080 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1081 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1082 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1083 			bp->b_flags |= B_RELBUF;
1084 		}
1085 
1086 		/*
1087 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1088 		 * if we have a severe page deficiency write the buffer
1089 		 * asynchronously.  Otherwise try to cluster, and if that
1090 		 * doesn't do it then either do an async write (if O_DIRECT),
1091 		 * or a delayed write (if not).
1092 		 */
1093 		if (ioflag & IO_SYNC) {
1094 			(void)bwrite(bp);
1095 		} else if (vm_page_count_severe() ||
1096 			    buf_dirty_count_severe() ||
1097 			    xfersize + blkoffset == fs->fs_bsize ||
1098 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1099 			bawrite(bp);
1100 		else
1101 			bdwrite(bp);
1102 		if (error || xfersize == 0)
1103 			break;
1104 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1105 	}
1106 	/*
1107 	 * If we successfully wrote any data, and we are not the superuser
1108 	 * we clear the setuid and setgid bits as a precaution against
1109 	 * tampering.
1110 	 */
1111 	if (resid > uio->uio_resid && ucred &&
1112 	    priv_check_cred(ucred, PRIV_VFS_CLEARSUGID, SUSER_ALLOWJAIL)) {
1113 		ip->i_mode &= ~(ISUID | ISGID);
1114 		dp->di_mode = ip->i_mode;
1115 	}
1116 	if (error) {
1117 		if (ioflag & IO_UNIT) {
1118 			(void)ffs_truncate(vp, osize,
1119 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1120 			uio->uio_offset -= resid - uio->uio_resid;
1121 			uio->uio_resid = resid;
1122 		}
1123 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1124 		error = ffs_update(vp, 1);
1125 	return (error);
1126 }
1127 
1128 
1129 /*
1130  * Vnode operating to retrieve a named extended attribute.
1131  *
1132  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1133  * the length of the EA, and possibly the pointer to the entry and to the data.
1134  */
1135 static int
1136 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1137 {
1138 	u_char *p, *pe, *pn, *p0;
1139 	int eapad1, eapad2, ealength, ealen, nlen;
1140 	uint32_t ul;
1141 
1142 	pe = ptr + length;
1143 	nlen = strlen(name);
1144 
1145 	for (p = ptr; p < pe; p = pn) {
1146 		p0 = p;
1147 		bcopy(p, &ul, sizeof(ul));
1148 		pn = p + ul;
1149 		/* make sure this entry is complete */
1150 		if (pn > pe)
1151 			break;
1152 		p += sizeof(uint32_t);
1153 		if (*p != nspace)
1154 			continue;
1155 		p++;
1156 		eapad2 = *p++;
1157 		if (*p != nlen)
1158 			continue;
1159 		p++;
1160 		if (bcmp(p, name, nlen))
1161 			continue;
1162 		ealength = sizeof(uint32_t) + 3 + nlen;
1163 		eapad1 = 8 - (ealength % 8);
1164 		if (eapad1 == 8)
1165 			eapad1 = 0;
1166 		ealength += eapad1;
1167 		ealen = ul - ealength - eapad2;
1168 		p += nlen + eapad1;
1169 		if (eap != NULL)
1170 			*eap = p0;
1171 		if (eac != NULL)
1172 			*eac = p;
1173 		return (ealen);
1174 	}
1175 	return(-1);
1176 }
1177 
1178 static int
1179 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1180 {
1181 	struct inode *ip;
1182 	struct ufs2_dinode *dp;
1183 	struct uio luio;
1184 	struct iovec liovec;
1185 	int easize, error;
1186 	u_char *eae;
1187 
1188 	ip = VTOI(vp);
1189 	dp = ip->i_din2;
1190 	easize = dp->di_extsize;
1191 
1192 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1193 
1194 	liovec.iov_base = eae;
1195 	liovec.iov_len = easize;
1196 	luio.uio_iov = &liovec;
1197 	luio.uio_iovcnt = 1;
1198 	luio.uio_offset = 0;
1199 	luio.uio_resid = easize;
1200 	luio.uio_segflg = UIO_SYSSPACE;
1201 	luio.uio_rw = UIO_READ;
1202 	luio.uio_td = td;
1203 
1204 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1205 	if (error) {
1206 		free(eae, M_TEMP);
1207 		return(error);
1208 	}
1209 	*p = eae;
1210 	return (0);
1211 }
1212 
1213 static int
1214 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1215 {
1216 	struct inode *ip;
1217 	struct ufs2_dinode *dp;
1218 	int error;
1219 
1220 	ip = VTOI(vp);
1221 
1222 	if (ip->i_ea_area != NULL)
1223 		return (EBUSY);
1224 	dp = ip->i_din2;
1225 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1226 	if (error)
1227 		return (error);
1228 	ip->i_ea_len = dp->di_extsize;
1229 	ip->i_ea_error = 0;
1230 	return (0);
1231 }
1232 
1233 /*
1234  * Vnode extattr transaction commit/abort
1235  */
1236 static int
1237 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1238 {
1239 	struct inode *ip;
1240 	struct uio luio;
1241 	struct iovec liovec;
1242 	int error;
1243 	struct ufs2_dinode *dp;
1244 
1245 	ip = VTOI(vp);
1246 	if (ip->i_ea_area == NULL)
1247 		return (EINVAL);
1248 	dp = ip->i_din2;
1249 	error = ip->i_ea_error;
1250 	if (commit && error == 0) {
1251 		if (cred == NOCRED)
1252 			cred =  vp->v_mount->mnt_cred;
1253 		liovec.iov_base = ip->i_ea_area;
1254 		liovec.iov_len = ip->i_ea_len;
1255 		luio.uio_iov = &liovec;
1256 		luio.uio_iovcnt = 1;
1257 		luio.uio_offset = 0;
1258 		luio.uio_resid = ip->i_ea_len;
1259 		luio.uio_segflg = UIO_SYSSPACE;
1260 		luio.uio_rw = UIO_WRITE;
1261 		luio.uio_td = td;
1262 		/* XXX: I'm not happy about truncating to zero size */
1263 		if (ip->i_ea_len < dp->di_extsize)
1264 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1265 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1266 	}
1267 	free(ip->i_ea_area, M_TEMP);
1268 	ip->i_ea_area = NULL;
1269 	ip->i_ea_len = 0;
1270 	ip->i_ea_error = 0;
1271 	return (error);
1272 }
1273 
1274 /*
1275  * Vnode extattr strategy routine for fifos.
1276  *
1277  * We need to check for a read or write of the external attributes.
1278  * Otherwise we just fall through and do the usual thing.
1279  */
1280 static int
1281 ffsext_strategy(struct vop_strategy_args *ap)
1282 /*
1283 struct vop_strategy_args {
1284 	struct vnodeop_desc *a_desc;
1285 	struct vnode *a_vp;
1286 	struct buf *a_bp;
1287 };
1288 */
1289 {
1290 	struct vnode *vp;
1291 	daddr_t lbn;
1292 
1293 	vp = ap->a_vp;
1294 	lbn = ap->a_bp->b_lblkno;
1295 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1296 	    lbn < 0 && lbn >= -NXADDR)
1297 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1298 	if (vp->v_type == VFIFO)
1299 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1300 	panic("spec nodes went here");
1301 }
1302 
1303 /*
1304  * Vnode extattr transaction commit/abort
1305  */
1306 static int
1307 ffs_openextattr(struct vop_openextattr_args *ap)
1308 /*
1309 struct vop_openextattr_args {
1310 	struct vnodeop_desc *a_desc;
1311 	struct vnode *a_vp;
1312 	IN struct ucred *a_cred;
1313 	IN struct thread *a_td;
1314 };
1315 */
1316 {
1317 	struct inode *ip;
1318 	struct fs *fs;
1319 
1320 	ip = VTOI(ap->a_vp);
1321 	fs = ip->i_fs;
1322 
1323 	if (ap->a_vp->v_type == VCHR)
1324 		return (EOPNOTSUPP);
1325 
1326 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1327 }
1328 
1329 
1330 /*
1331  * Vnode extattr transaction commit/abort
1332  */
1333 static int
1334 ffs_closeextattr(struct vop_closeextattr_args *ap)
1335 /*
1336 struct vop_closeextattr_args {
1337 	struct vnodeop_desc *a_desc;
1338 	struct vnode *a_vp;
1339 	int a_commit;
1340 	IN struct ucred *a_cred;
1341 	IN struct thread *a_td;
1342 };
1343 */
1344 {
1345 	struct inode *ip;
1346 	struct fs *fs;
1347 
1348 	ip = VTOI(ap->a_vp);
1349 	fs = ip->i_fs;
1350 
1351 	if (ap->a_vp->v_type == VCHR)
1352 		return (EOPNOTSUPP);
1353 
1354 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1355 }
1356 
1357 /*
1358  * Vnode operation to remove a named attribute.
1359  */
1360 static int
1361 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1362 /*
1363 vop_deleteextattr {
1364 	IN struct vnode *a_vp;
1365 	IN int a_attrnamespace;
1366 	IN const char *a_name;
1367 	IN struct ucred *a_cred;
1368 	IN struct thread *a_td;
1369 };
1370 */
1371 {
1372 	struct inode *ip;
1373 	struct fs *fs;
1374 	uint32_t ealength, ul;
1375 	int ealen, olen, eapad1, eapad2, error, i, easize;
1376 	u_char *eae, *p;
1377 	int stand_alone;
1378 
1379 	ip = VTOI(ap->a_vp);
1380 	fs = ip->i_fs;
1381 
1382 	if (ap->a_vp->v_type == VCHR)
1383 		return (EOPNOTSUPP);
1384 
1385 	if (strlen(ap->a_name) == 0)
1386 		return (EINVAL);
1387 
1388 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1389 	    ap->a_cred, ap->a_td, IWRITE);
1390 	if (error) {
1391 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1392 			ip->i_ea_error = error;
1393 		return (error);
1394 	}
1395 
1396 	if (ip->i_ea_area == NULL) {
1397 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1398 		if (error)
1399 			return (error);
1400 		stand_alone = 1;
1401 	} else {
1402 		stand_alone = 0;
1403 	}
1404 
1405 	ealength = eapad1 = ealen = eapad2 = 0;
1406 
1407 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1408 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1409 	easize = ip->i_ea_len;
1410 
1411 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1412 	    &p, NULL);
1413 	if (olen == -1) {
1414 		/* delete but nonexistent */
1415 		free(eae, M_TEMP);
1416 		if (stand_alone)
1417 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1418 		return(ENOATTR);
1419 	}
1420 	bcopy(p, &ul, sizeof ul);
1421 	i = p - eae + ul;
1422 	if (ul != ealength) {
1423 		bcopy(p + ul, p + ealength, easize - i);
1424 		easize += (ealength - ul);
1425 	}
1426 	if (easize > NXADDR * fs->fs_bsize) {
1427 		free(eae, M_TEMP);
1428 		if (stand_alone)
1429 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1430 		else if (ip->i_ea_error == 0)
1431 			ip->i_ea_error = ENOSPC;
1432 		return(ENOSPC);
1433 	}
1434 	p = ip->i_ea_area;
1435 	ip->i_ea_area = eae;
1436 	ip->i_ea_len = easize;
1437 	free(p, M_TEMP);
1438 	if (stand_alone)
1439 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1440 	return(error);
1441 }
1442 
1443 /*
1444  * Vnode operation to retrieve a named extended attribute.
1445  */
1446 static int
1447 ffs_getextattr(struct vop_getextattr_args *ap)
1448 /*
1449 vop_getextattr {
1450 	IN struct vnode *a_vp;
1451 	IN int a_attrnamespace;
1452 	IN const char *a_name;
1453 	INOUT struct uio *a_uio;
1454 	OUT size_t *a_size;
1455 	IN struct ucred *a_cred;
1456 	IN struct thread *a_td;
1457 };
1458 */
1459 {
1460 	struct inode *ip;
1461 	struct fs *fs;
1462 	u_char *eae, *p;
1463 	unsigned easize;
1464 	int error, ealen, stand_alone;
1465 
1466 	ip = VTOI(ap->a_vp);
1467 	fs = ip->i_fs;
1468 
1469 	if (ap->a_vp->v_type == VCHR)
1470 		return (EOPNOTSUPP);
1471 
1472 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1473 	    ap->a_cred, ap->a_td, IREAD);
1474 	if (error)
1475 		return (error);
1476 
1477 	if (ip->i_ea_area == NULL) {
1478 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1479 		if (error)
1480 			return (error);
1481 		stand_alone = 1;
1482 	} else {
1483 		stand_alone = 0;
1484 	}
1485 	eae = ip->i_ea_area;
1486 	easize = ip->i_ea_len;
1487 
1488 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1489 	    NULL, &p);
1490 	if (ealen >= 0) {
1491 		error = 0;
1492 		if (ap->a_size != NULL)
1493 			*ap->a_size = ealen;
1494 		else if (ap->a_uio != NULL)
1495 			error = uiomove(p, ealen, ap->a_uio);
1496 	} else
1497 		error = ENOATTR;
1498 	if (stand_alone)
1499 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1500 	return(error);
1501 }
1502 
1503 /*
1504  * Vnode operation to retrieve extended attributes on a vnode.
1505  */
1506 static int
1507 ffs_listextattr(struct vop_listextattr_args *ap)
1508 /*
1509 vop_listextattr {
1510 	IN struct vnode *a_vp;
1511 	IN int a_attrnamespace;
1512 	INOUT struct uio *a_uio;
1513 	OUT size_t *a_size;
1514 	IN struct ucred *a_cred;
1515 	IN struct thread *a_td;
1516 };
1517 */
1518 {
1519 	struct inode *ip;
1520 	struct fs *fs;
1521 	u_char *eae, *p, *pe, *pn;
1522 	unsigned easize;
1523 	uint32_t ul;
1524 	int error, ealen, stand_alone;
1525 
1526 	ip = VTOI(ap->a_vp);
1527 	fs = ip->i_fs;
1528 
1529 	if (ap->a_vp->v_type == VCHR)
1530 		return (EOPNOTSUPP);
1531 
1532 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1533 	    ap->a_cred, ap->a_td, IREAD);
1534 	if (error)
1535 		return (error);
1536 
1537 	if (ip->i_ea_area == NULL) {
1538 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1539 		if (error)
1540 			return (error);
1541 		stand_alone = 1;
1542 	} else {
1543 		stand_alone = 0;
1544 	}
1545 	eae = ip->i_ea_area;
1546 	easize = ip->i_ea_len;
1547 
1548 	error = 0;
1549 	if (ap->a_size != NULL)
1550 		*ap->a_size = 0;
1551 	pe = eae + easize;
1552 	for(p = eae; error == 0 && p < pe; p = pn) {
1553 		bcopy(p, &ul, sizeof(ul));
1554 		pn = p + ul;
1555 		if (pn > pe)
1556 			break;
1557 		p += sizeof(ul);
1558 		if (*p++ != ap->a_attrnamespace)
1559 			continue;
1560 		p++;	/* pad2 */
1561 		ealen = *p;
1562 		if (ap->a_size != NULL) {
1563 			*ap->a_size += ealen + 1;
1564 		} else if (ap->a_uio != NULL) {
1565 			error = uiomove(p, ealen + 1, ap->a_uio);
1566 		}
1567 	}
1568 	if (stand_alone)
1569 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1570 	return(error);
1571 }
1572 
1573 /*
1574  * Vnode operation to set a named attribute.
1575  */
1576 static int
1577 ffs_setextattr(struct vop_setextattr_args *ap)
1578 /*
1579 vop_setextattr {
1580 	IN struct vnode *a_vp;
1581 	IN int a_attrnamespace;
1582 	IN const char *a_name;
1583 	INOUT struct uio *a_uio;
1584 	IN struct ucred *a_cred;
1585 	IN struct thread *a_td;
1586 };
1587 */
1588 {
1589 	struct inode *ip;
1590 	struct fs *fs;
1591 	uint32_t ealength, ul;
1592 	int ealen, olen, eapad1, eapad2, error, i, easize;
1593 	u_char *eae, *p;
1594 	int stand_alone;
1595 
1596 	ip = VTOI(ap->a_vp);
1597 	fs = ip->i_fs;
1598 
1599 	if (ap->a_vp->v_type == VCHR)
1600 		return (EOPNOTSUPP);
1601 
1602 	if (strlen(ap->a_name) == 0)
1603 		return (EINVAL);
1604 
1605 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1606 	if (ap->a_uio == NULL)
1607 		return (EOPNOTSUPP);
1608 
1609 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1610 	    ap->a_cred, ap->a_td, IWRITE);
1611 	if (error) {
1612 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1613 			ip->i_ea_error = error;
1614 		return (error);
1615 	}
1616 
1617 	if (ip->i_ea_area == NULL) {
1618 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1619 		if (error)
1620 			return (error);
1621 		stand_alone = 1;
1622 	} else {
1623 		stand_alone = 0;
1624 	}
1625 
1626 	ealen = ap->a_uio->uio_resid;
1627 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1628 	eapad1 = 8 - (ealength % 8);
1629 	if (eapad1 == 8)
1630 		eapad1 = 0;
1631 	eapad2 = 8 - (ealen % 8);
1632 	if (eapad2 == 8)
1633 		eapad2 = 0;
1634 	ealength += eapad1 + ealen + eapad2;
1635 
1636 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1637 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1638 	easize = ip->i_ea_len;
1639 
1640 	olen = ffs_findextattr(eae, easize,
1641 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1642         if (olen == -1) {
1643 		/* new, append at end */
1644 		p = eae + easize;
1645 		easize += ealength;
1646 	} else {
1647 		bcopy(p, &ul, sizeof ul);
1648 		i = p - eae + ul;
1649 		if (ul != ealength) {
1650 			bcopy(p + ul, p + ealength, easize - i);
1651 			easize += (ealength - ul);
1652 		}
1653 	}
1654 	if (easize > NXADDR * fs->fs_bsize) {
1655 		free(eae, M_TEMP);
1656 		if (stand_alone)
1657 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1658 		else if (ip->i_ea_error == 0)
1659 			ip->i_ea_error = ENOSPC;
1660 		return(ENOSPC);
1661 	}
1662 	bcopy(&ealength, p, sizeof(ealength));
1663 	p += sizeof(ealength);
1664 	*p++ = ap->a_attrnamespace;
1665 	*p++ = eapad2;
1666 	*p++ = strlen(ap->a_name);
1667 	strcpy(p, ap->a_name);
1668 	p += strlen(ap->a_name);
1669 	bzero(p, eapad1);
1670 	p += eapad1;
1671 	error = uiomove(p, ealen, ap->a_uio);
1672 	if (error) {
1673 		free(eae, M_TEMP);
1674 		if (stand_alone)
1675 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1676 		else if (ip->i_ea_error == 0)
1677 			ip->i_ea_error = error;
1678 		return(error);
1679 	}
1680 	p += ealen;
1681 	bzero(p, eapad2);
1682 
1683 	p = ip->i_ea_area;
1684 	ip->i_ea_area = eae;
1685 	ip->i_ea_len = easize;
1686 	free(p, M_TEMP);
1687 	if (stand_alone)
1688 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1689 	return(error);
1690 }
1691