xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 6b806d21d144c25f4fad714e1c0cf780f5e27d7e)
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/proc.h>
78 #include <sys/resourcevar.h>
79 #include <sys/signalvar.h>
80 #include <sys/stat.h>
81 #include <sys/vmmeter.h>
82 #include <sys/vnode.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vnode_pager.h>
90 
91 #include <ufs/ufs/extattr.h>
92 #include <ufs/ufs/quota.h>
93 #include <ufs/ufs/inode.h>
94 #include <ufs/ufs/ufs_extern.h>
95 #include <ufs/ufs/ufsmount.h>
96 
97 #include <ufs/ffs/fs.h>
98 #include <ufs/ffs/ffs_extern.h>
99 #include "opt_directio.h"
100 #include "opt_ffs.h"
101 
102 #ifdef DIRECTIO
103 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
104 #endif
105 static vop_fsync_t	ffs_fsync;
106 static vop_lock_t	ffs_lock;
107 static vop_getpages_t	ffs_getpages;
108 static vop_read_t	ffs_read;
109 static vop_write_t	ffs_write;
110 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
111 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
112 		    struct ucred *cred);
113 static vop_strategy_t	ffsext_strategy;
114 static vop_closeextattr_t	ffs_closeextattr;
115 static vop_deleteextattr_t	ffs_deleteextattr;
116 static vop_getextattr_t	ffs_getextattr;
117 static vop_listextattr_t	ffs_listextattr;
118 static vop_openextattr_t	ffs_openextattr;
119 static vop_setextattr_t	ffs_setextattr;
120 
121 
122 /* Global vfs data structures for ufs. */
123 struct vop_vector ffs_vnodeops1 = {
124 	.vop_default =		&ufs_vnodeops,
125 	.vop_fsync =		ffs_fsync,
126 	.vop_getpages =		ffs_getpages,
127 	.vop_lock =		ffs_lock,
128 	.vop_read =		ffs_read,
129 	.vop_reallocblks =	ffs_reallocblks,
130 	.vop_write =		ffs_write,
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 };
138 
139 /* Global vfs data structures for ufs. */
140 struct vop_vector ffs_vnodeops2 = {
141 	.vop_default =		&ufs_vnodeops,
142 	.vop_fsync =		ffs_fsync,
143 	.vop_getpages =		ffs_getpages,
144 	.vop_lock =		ffs_lock,
145 	.vop_read =		ffs_read,
146 	.vop_reallocblks =	ffs_reallocblks,
147 	.vop_write =		ffs_write,
148 	.vop_closeextattr =	ffs_closeextattr,
149 	.vop_deleteextattr =	ffs_deleteextattr,
150 	.vop_getextattr =	ffs_getextattr,
151 	.vop_listextattr =	ffs_listextattr,
152 	.vop_openextattr =	ffs_openextattr,
153 	.vop_setextattr =	ffs_setextattr,
154 };
155 
156 struct vop_vector ffs_fifoops2 = {
157 	.vop_default =		&ufs_fifoops,
158 	.vop_fsync =		ffs_fsync,
159 	.vop_lock =		ffs_lock,
160 	.vop_reallocblks =	ffs_reallocblks,
161 	.vop_strategy =		ffsext_strategy,
162 	.vop_closeextattr =	ffs_closeextattr,
163 	.vop_deleteextattr =	ffs_deleteextattr,
164 	.vop_getextattr =	ffs_getextattr,
165 	.vop_listextattr =	ffs_listextattr,
166 	.vop_openextattr =	ffs_openextattr,
167 	.vop_setextattr =	ffs_setextattr,
168 };
169 
170 /*
171  * Synch an open file.
172  */
173 /* ARGSUSED */
174 static int
175 ffs_fsync(struct vop_fsync_args *ap)
176 {
177 	int error;
178 
179 	error = ffs_syncvnode(ap->a_vp, ap->a_waitfor);
180 	if (error)
181 		return (error);
182 	if (ap->a_waitfor == MNT_WAIT &&
183 	    (ap->a_vp->v_mount->mnt_flag & MNT_SOFTDEP))
184                 error = softdep_fsync(ap->a_vp);
185 	return (error);
186 }
187 
188 int
189 ffs_syncvnode(struct vnode *vp, int waitfor)
190 {
191 	struct inode *ip = VTOI(vp);
192 	struct buf *bp;
193 	struct buf *nbp;
194 	int s, error, wait, passes, skipmeta;
195 	ufs_lbn_t lbn;
196 
197 	wait = (waitfor == MNT_WAIT);
198 	lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
199 
200 	/*
201 	 * Flush all dirty buffers associated with a vnode.
202 	 */
203 	passes = NIADDR + 1;
204 	skipmeta = 0;
205 	if (wait)
206 		skipmeta = 1;
207 	s = splbio();
208 	VI_LOCK(vp);
209 loop:
210 	TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs)
211 		bp->b_vflags &= ~BV_SCANNED;
212 	TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
213 		/*
214 		 * Reasons to skip this buffer: it has already been considered
215 		 * on this pass, this pass is the first time through on a
216 		 * synchronous flush request and the buffer being considered
217 		 * is metadata, the buffer has dependencies that will cause
218 		 * it to be redirtied and it has not already been deferred,
219 		 * or it is already being written.
220 		 */
221 		if ((bp->b_vflags & BV_SCANNED) != 0)
222 			continue;
223 		bp->b_vflags |= BV_SCANNED;
224 		if ((skipmeta == 1 && bp->b_lblkno < 0))
225 			continue;
226 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
227 			continue;
228 		VI_UNLOCK(vp);
229 		if (!wait && LIST_FIRST(&bp->b_dep) != NULL &&
230 		    (bp->b_flags & B_DEFERRED) == 0 &&
231 		    buf_countdeps(bp, 0)) {
232 			bp->b_flags |= B_DEFERRED;
233 			BUF_UNLOCK(bp);
234 			VI_LOCK(vp);
235 			continue;
236 		}
237 		if ((bp->b_flags & B_DELWRI) == 0)
238 			panic("ffs_fsync: not dirty");
239 		/*
240 		 * If this is a synchronous flush request, or it is not a
241 		 * file or device, start the write on this buffer immediatly.
242 		 */
243 		if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
244 
245 			/*
246 			 * On our final pass through, do all I/O synchronously
247 			 * so that we can find out if our flush is failing
248 			 * because of write errors.
249 			 */
250 			if (passes > 0 || !wait) {
251 				if ((bp->b_flags & B_CLUSTEROK) && !wait) {
252 					(void) vfs_bio_awrite(bp);
253 				} else {
254 					bremfree(bp);
255 					splx(s);
256 					(void) bawrite(bp);
257 					s = splbio();
258 				}
259 			} else {
260 				bremfree(bp);
261 				splx(s);
262 				if ((error = bwrite(bp)) != 0)
263 					return (error);
264 				s = splbio();
265 			}
266 		} else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
267 			/*
268 			 * If the buffer is for data that has been truncated
269 			 * off the file, then throw it away.
270 			 */
271 			bremfree(bp);
272 			bp->b_flags |= B_INVAL | B_NOCACHE;
273 			splx(s);
274 			brelse(bp);
275 			s = splbio();
276 		} else
277 			vfs_bio_awrite(bp);
278 
279 		/*
280 		 * Since we may have slept during the I/O, we need
281 		 * to start from a known point.
282 		 */
283 		VI_LOCK(vp);
284 		nbp = TAILQ_FIRST(&vp->v_bufobj.bo_dirty.bv_hd);
285 	}
286 	/*
287 	 * If we were asked to do this synchronously, then go back for
288 	 * another pass, this time doing the metadata.
289 	 */
290 	if (skipmeta) {
291 		skipmeta = 0;
292 		goto loop;
293 	}
294 
295 	if (wait) {
296 		bufobj_wwait(&vp->v_bufobj, 3, 0);
297 		VI_UNLOCK(vp);
298 
299 		/*
300 		 * Ensure that any filesystem metatdata associated
301 		 * with the vnode has been written.
302 		 */
303 		splx(s);
304 		if ((error = softdep_sync_metadata(vp)) != 0)
305 			return (error);
306 		s = splbio();
307 
308 		VI_LOCK(vp);
309 		if (vp->v_bufobj.bo_dirty.bv_cnt > 0) {
310 			/*
311 			 * Block devices associated with filesystems may
312 			 * have new I/O requests posted for them even if
313 			 * the vnode is locked, so no amount of trying will
314 			 * get them clean. Thus we give block devices a
315 			 * good effort, then just give up. For all other file
316 			 * types, go around and try again until it is clean.
317 			 */
318 			if (passes > 0) {
319 				passes -= 1;
320 				goto loop;
321 			}
322 #ifdef DIAGNOSTIC
323 			if (!vn_isdisk(vp, NULL))
324 				vprint("ffs_fsync: dirty", vp);
325 #endif
326 		}
327 	}
328 	VI_UNLOCK(vp);
329 	splx(s);
330 	return (ffs_update(vp, wait));
331 }
332 
333 /*
334  * Snapshots require all lock requests to be exclusive.
335  */
336 static int
337 ffs_lock(ap)
338 	struct vop_lock_args /* {
339 		struct vnode *a_vp;
340 		int a_flags;
341 		struct thread *a_td;
342 	} */ *ap;
343 {
344 	struct vnode *vp = ap->a_vp;
345 
346 	if ((VTOI(vp)->i_flags & SF_SNAPSHOT) &&
347 	    ((ap->a_flags & LK_TYPE_MASK) == LK_SHARED)) {
348 		ap->a_flags &= ~LK_TYPE_MASK;
349 		ap->a_flags |= LK_EXCLUSIVE;
350 	}
351 	return (VOP_LOCK_APV(&ufs_vnodeops, ap));
352 }
353 
354 /*
355  * Vnode op for reading.
356  */
357 /* ARGSUSED */
358 static int
359 ffs_read(ap)
360 	struct vop_read_args /* {
361 		struct vnode *a_vp;
362 		struct uio *a_uio;
363 		int a_ioflag;
364 		struct ucred *a_cred;
365 	} */ *ap;
366 {
367 	struct vnode *vp;
368 	struct inode *ip;
369 	struct uio *uio;
370 	struct fs *fs;
371 	struct buf *bp;
372 	ufs_lbn_t lbn, nextlbn;
373 	off_t bytesinfile;
374 	long size, xfersize, blkoffset;
375 	int error, orig_resid;
376 	int seqcount;
377 	int ioflag;
378 
379 	vp = ap->a_vp;
380 	uio = ap->a_uio;
381 	ioflag = ap->a_ioflag;
382 	if (ap->a_ioflag & IO_EXT)
383 #ifdef notyet
384 		return (ffs_extread(vp, uio, ioflag));
385 #else
386 		panic("ffs_read+IO_EXT");
387 #endif
388 #ifdef DIRECTIO
389 	if ((ioflag & IO_DIRECT) != 0) {
390 		int workdone;
391 
392 		error = ffs_rawread(vp, uio, &workdone);
393 		if (error != 0 || workdone != 0)
394 			return error;
395 	}
396 #endif
397 
398 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
399 	ip = VTOI(vp);
400 
401 #ifdef DIAGNOSTIC
402 	if (uio->uio_rw != UIO_READ)
403 		panic("ffs_read: mode");
404 
405 	if (vp->v_type == VLNK) {
406 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
407 			panic("ffs_read: short symlink");
408 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
409 		panic("ffs_read: type %d",  vp->v_type);
410 #endif
411 	orig_resid = uio->uio_resid;
412 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
413 	if (orig_resid == 0)
414 		return (0);
415 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
416 	fs = ip->i_fs;
417 	if (uio->uio_offset < ip->i_size &&
418 	    uio->uio_offset >= fs->fs_maxfilesize)
419 		return (EOVERFLOW);
420 
421 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
422 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
423 			break;
424 		lbn = lblkno(fs, uio->uio_offset);
425 		nextlbn = lbn + 1;
426 
427 		/*
428 		 * size of buffer.  The buffer representing the
429 		 * end of the file is rounded up to the size of
430 		 * the block type ( fragment or full block,
431 		 * depending ).
432 		 */
433 		size = blksize(fs, ip, lbn);
434 		blkoffset = blkoff(fs, uio->uio_offset);
435 
436 		/*
437 		 * The amount we want to transfer in this iteration is
438 		 * one FS block less the amount of the data before
439 		 * our startpoint (duh!)
440 		 */
441 		xfersize = fs->fs_bsize - blkoffset;
442 
443 		/*
444 		 * But if we actually want less than the block,
445 		 * or the file doesn't have a whole block more of data,
446 		 * then use the lesser number.
447 		 */
448 		if (uio->uio_resid < xfersize)
449 			xfersize = uio->uio_resid;
450 		if (bytesinfile < xfersize)
451 			xfersize = bytesinfile;
452 
453 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
454 			/*
455 			 * Don't do readahead if this is the end of the file.
456 			 */
457 			error = bread(vp, lbn, size, NOCRED, &bp);
458 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
459 			/*
460 			 * Otherwise if we are allowed to cluster,
461 			 * grab as much as we can.
462 			 *
463 			 * XXX  This may not be a win if we are not
464 			 * doing sequential access.
465 			 */
466 			error = cluster_read(vp, ip->i_size, lbn,
467 				size, NOCRED, uio->uio_resid, seqcount, &bp);
468 		} else if (seqcount > 1) {
469 			/*
470 			 * If we are NOT allowed to cluster, then
471 			 * if we appear to be acting sequentially,
472 			 * fire off a request for a readahead
473 			 * as well as a read. Note that the 4th and 5th
474 			 * arguments point to arrays of the size specified in
475 			 * the 6th argument.
476 			 */
477 			int nextsize = blksize(fs, ip, nextlbn);
478 			error = breadn(vp, lbn,
479 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
480 		} else {
481 			/*
482 			 * Failing all of the above, just read what the
483 			 * user asked for. Interestingly, the same as
484 			 * the first option above.
485 			 */
486 			error = bread(vp, lbn, size, NOCRED, &bp);
487 		}
488 		if (error) {
489 			brelse(bp);
490 			bp = NULL;
491 			break;
492 		}
493 
494 		/*
495 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
496 		 * will cause us to attempt to release the buffer later on
497 		 * and will cause the buffer cache to attempt to free the
498 		 * underlying pages.
499 		 */
500 		if (ioflag & IO_DIRECT)
501 			bp->b_flags |= B_DIRECT;
502 
503 		/*
504 		 * We should only get non-zero b_resid when an I/O error
505 		 * has occurred, which should cause us to break above.
506 		 * However, if the short read did not cause an error,
507 		 * then we want to ensure that we do not uiomove bad
508 		 * or uninitialized data.
509 		 */
510 		size -= bp->b_resid;
511 		if (size < xfersize) {
512 			if (size == 0)
513 				break;
514 			xfersize = size;
515 		}
516 
517 		error = uiomove((char *)bp->b_data + blkoffset,
518 		    (int)xfersize, uio);
519 		if (error)
520 			break;
521 
522 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
523 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
524 			/*
525 			 * If there are no dependencies, and it's VMIO,
526 			 * then we don't need the buf, mark it available
527 			 * for freeing. The VM has the data.
528 			 */
529 			bp->b_flags |= B_RELBUF;
530 			brelse(bp);
531 		} else {
532 			/*
533 			 * Otherwise let whoever
534 			 * made the request take care of
535 			 * freeing it. We just queue
536 			 * it onto another list.
537 			 */
538 			bqrelse(bp);
539 		}
540 	}
541 
542 	/*
543 	 * This can only happen in the case of an error
544 	 * because the loop above resets bp to NULL on each iteration
545 	 * and on normal completion has not set a new value into it.
546 	 * so it must have come from a 'break' statement
547 	 */
548 	if (bp != NULL) {
549 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
550 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
551 			bp->b_flags |= B_RELBUF;
552 			brelse(bp);
553 		} else {
554 			bqrelse(bp);
555 		}
556 	}
557 
558 	if ((error == 0 || uio->uio_resid != orig_resid) &&
559 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
560 		ip->i_flag |= IN_ACCESS;
561 	return (error);
562 }
563 
564 /*
565  * Vnode op for writing.
566  */
567 static int
568 ffs_write(ap)
569 	struct vop_write_args /* {
570 		struct vnode *a_vp;
571 		struct uio *a_uio;
572 		int a_ioflag;
573 		struct ucred *a_cred;
574 	} */ *ap;
575 {
576 	struct vnode *vp;
577 	struct uio *uio;
578 	struct inode *ip;
579 	struct fs *fs;
580 	struct buf *bp;
581 	struct thread *td;
582 	ufs_lbn_t lbn;
583 	off_t osize;
584 	int seqcount;
585 	int blkoffset, error, extended, flags, ioflag, resid, size, xfersize;
586 
587 	vp = ap->a_vp;
588 	uio = ap->a_uio;
589 	ioflag = ap->a_ioflag;
590 	if (ap->a_ioflag & IO_EXT)
591 #ifdef notyet
592 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
593 #else
594 		panic("ffs_write+IO_EXT");
595 #endif
596 
597 	extended = 0;
598 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
599 	ip = VTOI(vp);
600 
601 #ifdef DIAGNOSTIC
602 	if (uio->uio_rw != UIO_WRITE)
603 		panic("ffs_write: mode");
604 #endif
605 
606 	switch (vp->v_type) {
607 	case VREG:
608 		if (ioflag & IO_APPEND)
609 			uio->uio_offset = ip->i_size;
610 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
611 			return (EPERM);
612 		/* FALLTHROUGH */
613 	case VLNK:
614 		break;
615 	case VDIR:
616 		panic("ffs_write: dir write");
617 		break;
618 	default:
619 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
620 			(int)uio->uio_offset,
621 			(int)uio->uio_resid
622 		);
623 	}
624 
625 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
626 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
627 	fs = ip->i_fs;
628 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
629 		return (EFBIG);
630 	/*
631 	 * Maybe this should be above the vnode op call, but so long as
632 	 * file servers have no limits, I don't think it matters.
633 	 */
634 	td = uio->uio_td;
635 	if (vp->v_type == VREG && td != NULL) {
636 		PROC_LOCK(td->td_proc);
637 		if (uio->uio_offset + uio->uio_resid >
638 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
639 			psignal(td->td_proc, SIGXFSZ);
640 			PROC_UNLOCK(td->td_proc);
641 			return (EFBIG);
642 		}
643 		PROC_UNLOCK(td->td_proc);
644 	}
645 
646 	resid = uio->uio_resid;
647 	osize = ip->i_size;
648 	if (seqcount > BA_SEQMAX)
649 		flags = BA_SEQMAX << BA_SEQSHIFT;
650 	else
651 		flags = seqcount << BA_SEQSHIFT;
652 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
653 		flags |= IO_SYNC;
654 
655 	for (error = 0; uio->uio_resid > 0;) {
656 		lbn = lblkno(fs, uio->uio_offset);
657 		blkoffset = blkoff(fs, uio->uio_offset);
658 		xfersize = fs->fs_bsize - blkoffset;
659 		if (uio->uio_resid < xfersize)
660 			xfersize = uio->uio_resid;
661 		if (uio->uio_offset + xfersize > ip->i_size)
662 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
663 
664                 /*
665 		 * We must perform a read-before-write if the transfer size
666 		 * does not cover the entire buffer.
667                  */
668 		if (fs->fs_bsize > xfersize)
669 			flags |= BA_CLRBUF;
670 		else
671 			flags &= ~BA_CLRBUF;
672 /* XXX is uio->uio_offset the right thing here? */
673 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
674 		    ap->a_cred, flags, &bp);
675 		if (error != 0)
676 			break;
677 		/*
678 		 * If the buffer is not valid we have to clear out any
679 		 * garbage data from the pages instantiated for the buffer.
680 		 * If we do not, a failed uiomove() during a write can leave
681 		 * the prior contents of the pages exposed to a userland
682 		 * mmap().  XXX deal with uiomove() errors a better way.
683 		 */
684 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
685 			vfs_bio_clrbuf(bp);
686 		if (ioflag & IO_DIRECT)
687 			bp->b_flags |= B_DIRECT;
688 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
689 			bp->b_flags |= B_NOCACHE;
690 
691 		if (uio->uio_offset + xfersize > ip->i_size) {
692 			ip->i_size = uio->uio_offset + xfersize;
693 			DIP_SET(ip, i_size, ip->i_size);
694 			extended = 1;
695 		}
696 
697 		size = blksize(fs, ip, lbn) - bp->b_resid;
698 		if (size < xfersize)
699 			xfersize = size;
700 
701 		error =
702 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
703 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
704 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
705 			bp->b_flags |= B_RELBUF;
706 		}
707 
708 		/*
709 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
710 		 * if we have a severe page deficiency write the buffer
711 		 * asynchronously.  Otherwise try to cluster, and if that
712 		 * doesn't do it then either do an async write (if O_DIRECT),
713 		 * or a delayed write (if not).
714 		 */
715 		if (ioflag & IO_SYNC) {
716 			(void)bwrite(bp);
717 		} else if (vm_page_count_severe() ||
718 			    buf_dirty_count_severe() ||
719 			    (ioflag & IO_ASYNC)) {
720 			bp->b_flags |= B_CLUSTEROK;
721 			bawrite(bp);
722 		} else if (xfersize + blkoffset == fs->fs_bsize) {
723 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
724 				bp->b_flags |= B_CLUSTEROK;
725 				cluster_write(vp, bp, ip->i_size, seqcount);
726 			} else {
727 				bawrite(bp);
728 			}
729 		} else if (ioflag & IO_DIRECT) {
730 			bp->b_flags |= B_CLUSTEROK;
731 			bawrite(bp);
732 		} else {
733 			bp->b_flags |= B_CLUSTEROK;
734 			bdwrite(bp);
735 		}
736 		if (error || xfersize == 0)
737 			break;
738 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
739 	}
740 	/*
741 	 * If we successfully wrote any data, and we are not the superuser
742 	 * we clear the setuid and setgid bits as a precaution against
743 	 * tampering.
744 	 */
745 	if (resid > uio->uio_resid && ap->a_cred &&
746 	    suser_cred(ap->a_cred, SUSER_ALLOWJAIL)) {
747 		ip->i_mode &= ~(ISUID | ISGID);
748 		DIP_SET(ip, i_mode, ip->i_mode);
749 	}
750 	if (resid > uio->uio_resid)
751 		VN_KNOTE_UNLOCKED(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
752 	if (error) {
753 		if (ioflag & IO_UNIT) {
754 			(void)ffs_truncate(vp, osize,
755 			    IO_NORMAL | (ioflag & IO_SYNC),
756 			    ap->a_cred, uio->uio_td);
757 			uio->uio_offset -= resid - uio->uio_resid;
758 			uio->uio_resid = resid;
759 		}
760 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
761 		error = ffs_update(vp, 1);
762 	return (error);
763 }
764 
765 /*
766  * get page routine
767  */
768 static int
769 ffs_getpages(ap)
770 	struct vop_getpages_args *ap;
771 {
772 	int i;
773 	vm_page_t mreq;
774 	int pcount;
775 
776 	pcount = round_page(ap->a_count) / PAGE_SIZE;
777 	mreq = ap->a_m[ap->a_reqpage];
778 
779 	/*
780 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
781 	 * then the entire page is valid.  Since the page may be mapped,
782 	 * user programs might reference data beyond the actual end of file
783 	 * occuring within the page.  We have to zero that data.
784 	 */
785 	VM_OBJECT_LOCK(mreq->object);
786 	if (mreq->valid) {
787 		if (mreq->valid != VM_PAGE_BITS_ALL)
788 			vm_page_zero_invalid(mreq, TRUE);
789 		vm_page_lock_queues();
790 		for (i = 0; i < pcount; i++) {
791 			if (i != ap->a_reqpage) {
792 				vm_page_free(ap->a_m[i]);
793 			}
794 		}
795 		vm_page_unlock_queues();
796 		VM_OBJECT_UNLOCK(mreq->object);
797 		return VM_PAGER_OK;
798 	}
799 	VM_OBJECT_UNLOCK(mreq->object);
800 
801 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
802 					    ap->a_count,
803 					    ap->a_reqpage);
804 }
805 
806 
807 /*
808  * Extended attribute area reading.
809  */
810 static int
811 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
812 {
813 	struct inode *ip;
814 	struct ufs2_dinode *dp;
815 	struct fs *fs;
816 	struct buf *bp;
817 	ufs_lbn_t lbn, nextlbn;
818 	off_t bytesinfile;
819 	long size, xfersize, blkoffset;
820 	int error, orig_resid;
821 
822 	ip = VTOI(vp);
823 	fs = ip->i_fs;
824 	dp = ip->i_din2;
825 
826 #ifdef DIAGNOSTIC
827 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
828 		panic("ffs_extread: mode");
829 
830 #endif
831 	orig_resid = uio->uio_resid;
832 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
833 	if (orig_resid == 0)
834 		return (0);
835 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
836 
837 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
838 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
839 			break;
840 		lbn = lblkno(fs, uio->uio_offset);
841 		nextlbn = lbn + 1;
842 
843 		/*
844 		 * size of buffer.  The buffer representing the
845 		 * end of the file is rounded up to the size of
846 		 * the block type ( fragment or full block,
847 		 * depending ).
848 		 */
849 		size = sblksize(fs, dp->di_extsize, lbn);
850 		blkoffset = blkoff(fs, uio->uio_offset);
851 
852 		/*
853 		 * The amount we want to transfer in this iteration is
854 		 * one FS block less the amount of the data before
855 		 * our startpoint (duh!)
856 		 */
857 		xfersize = fs->fs_bsize - blkoffset;
858 
859 		/*
860 		 * But if we actually want less than the block,
861 		 * or the file doesn't have a whole block more of data,
862 		 * then use the lesser number.
863 		 */
864 		if (uio->uio_resid < xfersize)
865 			xfersize = uio->uio_resid;
866 		if (bytesinfile < xfersize)
867 			xfersize = bytesinfile;
868 
869 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
870 			/*
871 			 * Don't do readahead if this is the end of the info.
872 			 */
873 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
874 		} else {
875 			/*
876 			 * If we have a second block, then
877 			 * fire off a request for a readahead
878 			 * as well as a read. Note that the 4th and 5th
879 			 * arguments point to arrays of the size specified in
880 			 * the 6th argument.
881 			 */
882 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
883 
884 			nextlbn = -1 - nextlbn;
885 			error = breadn(vp, -1 - lbn,
886 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
887 		}
888 		if (error) {
889 			brelse(bp);
890 			bp = NULL;
891 			break;
892 		}
893 
894 		/*
895 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
896 		 * will cause us to attempt to release the buffer later on
897 		 * and will cause the buffer cache to attempt to free the
898 		 * underlying pages.
899 		 */
900 		if (ioflag & IO_DIRECT)
901 			bp->b_flags |= B_DIRECT;
902 
903 		/*
904 		 * We should only get non-zero b_resid when an I/O error
905 		 * has occurred, which should cause us to break above.
906 		 * However, if the short read did not cause an error,
907 		 * then we want to ensure that we do not uiomove bad
908 		 * or uninitialized data.
909 		 */
910 		size -= bp->b_resid;
911 		if (size < xfersize) {
912 			if (size == 0)
913 				break;
914 			xfersize = size;
915 		}
916 
917 		error = uiomove((char *)bp->b_data + blkoffset,
918 					(int)xfersize, uio);
919 		if (error)
920 			break;
921 
922 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
923 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
924 			/*
925 			 * If there are no dependencies, and it's VMIO,
926 			 * then we don't need the buf, mark it available
927 			 * for freeing. The VM has the data.
928 			 */
929 			bp->b_flags |= B_RELBUF;
930 			brelse(bp);
931 		} else {
932 			/*
933 			 * Otherwise let whoever
934 			 * made the request take care of
935 			 * freeing it. We just queue
936 			 * it onto another list.
937 			 */
938 			bqrelse(bp);
939 		}
940 	}
941 
942 	/*
943 	 * This can only happen in the case of an error
944 	 * because the loop above resets bp to NULL on each iteration
945 	 * and on normal completion has not set a new value into it.
946 	 * so it must have come from a 'break' statement
947 	 */
948 	if (bp != NULL) {
949 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
950 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
951 			bp->b_flags |= B_RELBUF;
952 			brelse(bp);
953 		} else {
954 			bqrelse(bp);
955 		}
956 	}
957 
958 	if ((error == 0 || uio->uio_resid != orig_resid) &&
959 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
960 		ip->i_flag |= IN_ACCESS;
961 	return (error);
962 }
963 
964 /*
965  * Extended attribute area writing.
966  */
967 static int
968 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
969 {
970 	struct inode *ip;
971 	struct ufs2_dinode *dp;
972 	struct fs *fs;
973 	struct buf *bp;
974 	ufs_lbn_t lbn;
975 	off_t osize;
976 	int blkoffset, error, flags, resid, size, xfersize;
977 
978 	ip = VTOI(vp);
979 	fs = ip->i_fs;
980 	dp = ip->i_din2;
981 
982 #ifdef DIAGNOSTIC
983 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
984 		panic("ffs_extwrite: mode");
985 #endif
986 
987 	if (ioflag & IO_APPEND)
988 		uio->uio_offset = dp->di_extsize;
989 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
990 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
991 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
992 		return (EFBIG);
993 
994 	resid = uio->uio_resid;
995 	osize = dp->di_extsize;
996 	flags = IO_EXT;
997 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
998 		flags |= IO_SYNC;
999 
1000 	for (error = 0; uio->uio_resid > 0;) {
1001 		lbn = lblkno(fs, uio->uio_offset);
1002 		blkoffset = blkoff(fs, uio->uio_offset);
1003 		xfersize = fs->fs_bsize - blkoffset;
1004 		if (uio->uio_resid < xfersize)
1005 			xfersize = uio->uio_resid;
1006 
1007                 /*
1008 		 * We must perform a read-before-write if the transfer size
1009 		 * does not cover the entire buffer.
1010                  */
1011 		if (fs->fs_bsize > xfersize)
1012 			flags |= BA_CLRBUF;
1013 		else
1014 			flags &= ~BA_CLRBUF;
1015 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1016 		    ucred, flags, &bp);
1017 		if (error != 0)
1018 			break;
1019 		/*
1020 		 * If the buffer is not valid we have to clear out any
1021 		 * garbage data from the pages instantiated for the buffer.
1022 		 * If we do not, a failed uiomove() during a write can leave
1023 		 * the prior contents of the pages exposed to a userland
1024 		 * mmap().  XXX deal with uiomove() errors a better way.
1025 		 */
1026 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1027 			vfs_bio_clrbuf(bp);
1028 		if (ioflag & IO_DIRECT)
1029 			bp->b_flags |= B_DIRECT;
1030 
1031 		if (uio->uio_offset + xfersize > dp->di_extsize)
1032 			dp->di_extsize = uio->uio_offset + xfersize;
1033 
1034 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1035 		if (size < xfersize)
1036 			xfersize = size;
1037 
1038 		error =
1039 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1040 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1041 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1042 			bp->b_flags |= B_RELBUF;
1043 		}
1044 
1045 		/*
1046 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1047 		 * if we have a severe page deficiency write the buffer
1048 		 * asynchronously.  Otherwise try to cluster, and if that
1049 		 * doesn't do it then either do an async write (if O_DIRECT),
1050 		 * or a delayed write (if not).
1051 		 */
1052 		if (ioflag & IO_SYNC) {
1053 			(void)bwrite(bp);
1054 		} else if (vm_page_count_severe() ||
1055 			    buf_dirty_count_severe() ||
1056 			    xfersize + blkoffset == fs->fs_bsize ||
1057 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1058 			bawrite(bp);
1059 		else
1060 			bdwrite(bp);
1061 		if (error || xfersize == 0)
1062 			break;
1063 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1064 	}
1065 	/*
1066 	 * If we successfully wrote any data, and we are not the superuser
1067 	 * we clear the setuid and setgid bits as a precaution against
1068 	 * tampering.
1069 	 */
1070 	if (resid > uio->uio_resid && ucred &&
1071 	    suser_cred(ucred, SUSER_ALLOWJAIL)) {
1072 		ip->i_mode &= ~(ISUID | ISGID);
1073 		dp->di_mode = ip->i_mode;
1074 	}
1075 	if (error) {
1076 		if (ioflag & IO_UNIT) {
1077 			(void)ffs_truncate(vp, osize,
1078 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1079 			uio->uio_offset -= resid - uio->uio_resid;
1080 			uio->uio_resid = resid;
1081 		}
1082 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1083 		error = ffs_update(vp, 1);
1084 	return (error);
1085 }
1086 
1087 
1088 /*
1089  * Vnode operating to retrieve a named extended attribute.
1090  *
1091  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1092  * the length of the EA, and possibly the pointer to the entry and to the data.
1093  */
1094 static int
1095 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1096 {
1097 	u_char *p, *pe, *pn, *p0;
1098 	int eapad1, eapad2, ealength, ealen, nlen;
1099 	uint32_t ul;
1100 
1101 	pe = ptr + length;
1102 	nlen = strlen(name);
1103 
1104 	for (p = ptr; p < pe; p = pn) {
1105 		p0 = p;
1106 		bcopy(p, &ul, sizeof(ul));
1107 		pn = p + ul;
1108 		/* make sure this entry is complete */
1109 		if (pn > pe)
1110 			break;
1111 		p += sizeof(uint32_t);
1112 		if (*p != nspace)
1113 			continue;
1114 		p++;
1115 		eapad2 = *p++;
1116 		if (*p != nlen)
1117 			continue;
1118 		p++;
1119 		if (bcmp(p, name, nlen))
1120 			continue;
1121 		ealength = sizeof(uint32_t) + 3 + nlen;
1122 		eapad1 = 8 - (ealength % 8);
1123 		if (eapad1 == 8)
1124 			eapad1 = 0;
1125 		ealength += eapad1;
1126 		ealen = ul - ealength - eapad2;
1127 		p += nlen + eapad1;
1128 		if (eap != NULL)
1129 			*eap = p0;
1130 		if (eac != NULL)
1131 			*eac = p;
1132 		return (ealen);
1133 	}
1134 	return(-1);
1135 }
1136 
1137 static int
1138 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1139 {
1140 	struct inode *ip;
1141 	struct ufs2_dinode *dp;
1142 	struct uio luio;
1143 	struct iovec liovec;
1144 	int easize, error;
1145 	u_char *eae;
1146 
1147 	ip = VTOI(vp);
1148 	dp = ip->i_din2;
1149 	easize = dp->di_extsize;
1150 
1151 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1152 
1153 	liovec.iov_base = eae;
1154 	liovec.iov_len = easize;
1155 	luio.uio_iov = &liovec;
1156 	luio.uio_iovcnt = 1;
1157 	luio.uio_offset = 0;
1158 	luio.uio_resid = easize;
1159 	luio.uio_segflg = UIO_SYSSPACE;
1160 	luio.uio_rw = UIO_READ;
1161 	luio.uio_td = td;
1162 
1163 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1164 	if (error) {
1165 		free(eae, M_TEMP);
1166 		return(error);
1167 	}
1168 	*p = eae;
1169 	return (0);
1170 }
1171 
1172 static int
1173 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1174 {
1175 	struct inode *ip;
1176 	struct ufs2_dinode *dp;
1177 	int error;
1178 
1179 	ip = VTOI(vp);
1180 
1181 	if (ip->i_ea_area != NULL)
1182 		return (EBUSY);
1183 	dp = ip->i_din2;
1184 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1185 	if (error)
1186 		return (error);
1187 	ip->i_ea_len = dp->di_extsize;
1188 	ip->i_ea_error = 0;
1189 	return (0);
1190 }
1191 
1192 /*
1193  * Vnode extattr transaction commit/abort
1194  */
1195 static int
1196 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1197 {
1198 	struct inode *ip;
1199 	struct uio luio;
1200 	struct iovec liovec;
1201 	int error;
1202 	struct ufs2_dinode *dp;
1203 
1204 	ip = VTOI(vp);
1205 	if (ip->i_ea_area == NULL)
1206 		return (EINVAL);
1207 	dp = ip->i_din2;
1208 	error = ip->i_ea_error;
1209 	if (commit && error == 0) {
1210 		if (cred == NOCRED)
1211 			cred =  vp->v_mount->mnt_cred;
1212 		liovec.iov_base = ip->i_ea_area;
1213 		liovec.iov_len = ip->i_ea_len;
1214 		luio.uio_iov = &liovec;
1215 		luio.uio_iovcnt = 1;
1216 		luio.uio_offset = 0;
1217 		luio.uio_resid = ip->i_ea_len;
1218 		luio.uio_segflg = UIO_SYSSPACE;
1219 		luio.uio_rw = UIO_WRITE;
1220 		luio.uio_td = td;
1221 		/* XXX: I'm not happy about truncating to zero size */
1222 		if (ip->i_ea_len < dp->di_extsize)
1223 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1224 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1225 	}
1226 	free(ip->i_ea_area, M_TEMP);
1227 	ip->i_ea_area = NULL;
1228 	ip->i_ea_len = 0;
1229 	ip->i_ea_error = 0;
1230 	return (error);
1231 }
1232 
1233 /*
1234  * Vnode extattr strategy routine for fifos.
1235  *
1236  * We need to check for a read or write of the external attributes.
1237  * Otherwise we just fall through and do the usual thing.
1238  */
1239 static int
1240 ffsext_strategy(struct vop_strategy_args *ap)
1241 /*
1242 struct vop_strategy_args {
1243 	struct vnodeop_desc *a_desc;
1244 	struct vnode *a_vp;
1245 	struct buf *a_bp;
1246 };
1247 */
1248 {
1249 	struct vnode *vp;
1250 	daddr_t lbn;
1251 
1252 	vp = ap->a_vp;
1253 	lbn = ap->a_bp->b_lblkno;
1254 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1255 	    lbn < 0 && lbn >= -NXADDR)
1256 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1257 	if (vp->v_type == VFIFO)
1258 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1259 	panic("spec nodes went here");
1260 }
1261 
1262 /*
1263  * Vnode extattr transaction commit/abort
1264  */
1265 static int
1266 ffs_openextattr(struct vop_openextattr_args *ap)
1267 /*
1268 struct vop_openextattr_args {
1269 	struct vnodeop_desc *a_desc;
1270 	struct vnode *a_vp;
1271 	IN struct ucred *a_cred;
1272 	IN struct thread *a_td;
1273 };
1274 */
1275 {
1276 	struct inode *ip;
1277 	struct fs *fs;
1278 
1279 	ip = VTOI(ap->a_vp);
1280 	fs = ip->i_fs;
1281 
1282 	if (ap->a_vp->v_type == VCHR)
1283 		return (EOPNOTSUPP);
1284 
1285 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1286 }
1287 
1288 
1289 /*
1290  * Vnode extattr transaction commit/abort
1291  */
1292 static int
1293 ffs_closeextattr(struct vop_closeextattr_args *ap)
1294 /*
1295 struct vop_closeextattr_args {
1296 	struct vnodeop_desc *a_desc;
1297 	struct vnode *a_vp;
1298 	int a_commit;
1299 	IN struct ucred *a_cred;
1300 	IN struct thread *a_td;
1301 };
1302 */
1303 {
1304 	struct inode *ip;
1305 	struct fs *fs;
1306 
1307 	ip = VTOI(ap->a_vp);
1308 	fs = ip->i_fs;
1309 
1310 	if (ap->a_vp->v_type == VCHR)
1311 		return (EOPNOTSUPP);
1312 
1313 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1314 }
1315 
1316 /*
1317  * Vnode operation to remove a named attribute.
1318  */
1319 static int
1320 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1321 /*
1322 vop_deleteextattr {
1323 	IN struct vnode *a_vp;
1324 	IN int a_attrnamespace;
1325 	IN const char *a_name;
1326 	IN struct ucred *a_cred;
1327 	IN struct thread *a_td;
1328 };
1329 */
1330 {
1331 	struct inode *ip;
1332 	struct fs *fs;
1333 	uint32_t ealength, ul;
1334 	int ealen, olen, eapad1, eapad2, error, i, easize;
1335 	u_char *eae, *p;
1336 	int stand_alone;
1337 
1338 	ip = VTOI(ap->a_vp);
1339 	fs = ip->i_fs;
1340 
1341 	if (ap->a_vp->v_type == VCHR)
1342 		return (EOPNOTSUPP);
1343 
1344 	if (strlen(ap->a_name) == 0)
1345 		return (EINVAL);
1346 
1347 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1348 	    ap->a_cred, ap->a_td, IWRITE);
1349 	if (error) {
1350 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1351 			ip->i_ea_error = error;
1352 		return (error);
1353 	}
1354 
1355 	if (ip->i_ea_area == NULL) {
1356 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1357 		if (error)
1358 			return (error);
1359 		stand_alone = 1;
1360 	} else {
1361 		stand_alone = 0;
1362 	}
1363 
1364 	ealength = eapad1 = ealen = eapad2 = 0;
1365 
1366 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1367 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1368 	easize = ip->i_ea_len;
1369 
1370 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1371 	    &p, NULL);
1372 	if (olen == -1) {
1373 		/* delete but nonexistent */
1374 		free(eae, M_TEMP);
1375 		if (stand_alone)
1376 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1377 		return(ENOATTR);
1378 	}
1379 	bcopy(p, &ul, sizeof ul);
1380 	i = p - eae + ul;
1381 	if (ul != ealength) {
1382 		bcopy(p + ul, p + ealength, easize - i);
1383 		easize += (ealength - ul);
1384 	}
1385 	if (easize > NXADDR * fs->fs_bsize) {
1386 		free(eae, M_TEMP);
1387 		if (stand_alone)
1388 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1389 		else if (ip->i_ea_error == 0)
1390 			ip->i_ea_error = ENOSPC;
1391 		return(ENOSPC);
1392 	}
1393 	p = ip->i_ea_area;
1394 	ip->i_ea_area = eae;
1395 	ip->i_ea_len = easize;
1396 	free(p, M_TEMP);
1397 	if (stand_alone)
1398 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1399 	return(error);
1400 }
1401 
1402 /*
1403  * Vnode operation to retrieve a named extended attribute.
1404  */
1405 static int
1406 ffs_getextattr(struct vop_getextattr_args *ap)
1407 /*
1408 vop_getextattr {
1409 	IN struct vnode *a_vp;
1410 	IN int a_attrnamespace;
1411 	IN const char *a_name;
1412 	INOUT struct uio *a_uio;
1413 	OUT size_t *a_size;
1414 	IN struct ucred *a_cred;
1415 	IN struct thread *a_td;
1416 };
1417 */
1418 {
1419 	struct inode *ip;
1420 	struct fs *fs;
1421 	u_char *eae, *p;
1422 	unsigned easize;
1423 	int error, ealen, stand_alone;
1424 
1425 	ip = VTOI(ap->a_vp);
1426 	fs = ip->i_fs;
1427 
1428 	if (ap->a_vp->v_type == VCHR)
1429 		return (EOPNOTSUPP);
1430 
1431 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1432 	    ap->a_cred, ap->a_td, IREAD);
1433 	if (error)
1434 		return (error);
1435 
1436 	if (ip->i_ea_area == NULL) {
1437 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1438 		if (error)
1439 			return (error);
1440 		stand_alone = 1;
1441 	} else {
1442 		stand_alone = 0;
1443 	}
1444 	eae = ip->i_ea_area;
1445 	easize = ip->i_ea_len;
1446 
1447 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1448 	    NULL, &p);
1449 	if (ealen >= 0) {
1450 		error = 0;
1451 		if (ap->a_size != NULL)
1452 			*ap->a_size = ealen;
1453 		else if (ap->a_uio != NULL)
1454 			error = uiomove(p, ealen, ap->a_uio);
1455 	} else
1456 		error = ENOATTR;
1457 	if (stand_alone)
1458 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1459 	return(error);
1460 }
1461 
1462 /*
1463  * Vnode operation to retrieve extended attributes on a vnode.
1464  */
1465 static int
1466 ffs_listextattr(struct vop_listextattr_args *ap)
1467 /*
1468 vop_listextattr {
1469 	IN struct vnode *a_vp;
1470 	IN int a_attrnamespace;
1471 	INOUT struct uio *a_uio;
1472 	OUT size_t *a_size;
1473 	IN struct ucred *a_cred;
1474 	IN struct thread *a_td;
1475 };
1476 */
1477 {
1478 	struct inode *ip;
1479 	struct fs *fs;
1480 	u_char *eae, *p, *pe, *pn;
1481 	unsigned easize;
1482 	uint32_t ul;
1483 	int error, ealen, stand_alone;
1484 
1485 	ip = VTOI(ap->a_vp);
1486 	fs = ip->i_fs;
1487 
1488 	if (ap->a_vp->v_type == VCHR)
1489 		return (EOPNOTSUPP);
1490 
1491 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1492 	    ap->a_cred, ap->a_td, IREAD);
1493 	if (error)
1494 		return (error);
1495 
1496 	if (ip->i_ea_area == NULL) {
1497 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1498 		if (error)
1499 			return (error);
1500 		stand_alone = 1;
1501 	} else {
1502 		stand_alone = 0;
1503 	}
1504 	eae = ip->i_ea_area;
1505 	easize = ip->i_ea_len;
1506 
1507 	error = 0;
1508 	if (ap->a_size != NULL)
1509 		*ap->a_size = 0;
1510 	pe = eae + easize;
1511 	for(p = eae; error == 0 && p < pe; p = pn) {
1512 		bcopy(p, &ul, sizeof(ul));
1513 		pn = p + ul;
1514 		if (pn > pe)
1515 			break;
1516 		p += sizeof(ul);
1517 		if (*p++ != ap->a_attrnamespace)
1518 			continue;
1519 		p++;	/* pad2 */
1520 		ealen = *p;
1521 		if (ap->a_size != NULL) {
1522 			*ap->a_size += ealen + 1;
1523 		} else if (ap->a_uio != NULL) {
1524 			error = uiomove(p, ealen + 1, ap->a_uio);
1525 		}
1526 	}
1527 	if (stand_alone)
1528 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1529 	return(error);
1530 }
1531 
1532 /*
1533  * Vnode operation to set a named attribute.
1534  */
1535 static int
1536 ffs_setextattr(struct vop_setextattr_args *ap)
1537 /*
1538 vop_setextattr {
1539 	IN struct vnode *a_vp;
1540 	IN int a_attrnamespace;
1541 	IN const char *a_name;
1542 	INOUT struct uio *a_uio;
1543 	IN struct ucred *a_cred;
1544 	IN struct thread *a_td;
1545 };
1546 */
1547 {
1548 	struct inode *ip;
1549 	struct fs *fs;
1550 	uint32_t ealength, ul;
1551 	int ealen, olen, eapad1, eapad2, error, i, easize;
1552 	u_char *eae, *p;
1553 	int stand_alone;
1554 
1555 	ip = VTOI(ap->a_vp);
1556 	fs = ip->i_fs;
1557 
1558 	if (ap->a_vp->v_type == VCHR)
1559 		return (EOPNOTSUPP);
1560 
1561 	if (strlen(ap->a_name) == 0)
1562 		return (EINVAL);
1563 
1564 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1565 	if (ap->a_uio == NULL)
1566 		return (EOPNOTSUPP);
1567 
1568 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1569 	    ap->a_cred, ap->a_td, IWRITE);
1570 	if (error) {
1571 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1572 			ip->i_ea_error = error;
1573 		return (error);
1574 	}
1575 
1576 	if (ip->i_ea_area == NULL) {
1577 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1578 		if (error)
1579 			return (error);
1580 		stand_alone = 1;
1581 	} else {
1582 		stand_alone = 0;
1583 	}
1584 
1585 	ealen = ap->a_uio->uio_resid;
1586 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1587 	eapad1 = 8 - (ealength % 8);
1588 	if (eapad1 == 8)
1589 		eapad1 = 0;
1590 	eapad2 = 8 - (ealen % 8);
1591 	if (eapad2 == 8)
1592 		eapad2 = 0;
1593 	ealength += eapad1 + ealen + eapad2;
1594 
1595 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1596 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1597 	easize = ip->i_ea_len;
1598 
1599 	olen = ffs_findextattr(eae, easize,
1600 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1601         if (olen == -1) {
1602 		/* new, append at end */
1603 		p = eae + easize;
1604 		easize += ealength;
1605 	} else {
1606 		bcopy(p, &ul, sizeof ul);
1607 		i = p - eae + ul;
1608 		if (ul != ealength) {
1609 			bcopy(p + ul, p + ealength, easize - i);
1610 			easize += (ealength - ul);
1611 		}
1612 	}
1613 	if (easize > NXADDR * fs->fs_bsize) {
1614 		free(eae, M_TEMP);
1615 		if (stand_alone)
1616 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1617 		else if (ip->i_ea_error == 0)
1618 			ip->i_ea_error = ENOSPC;
1619 		return(ENOSPC);
1620 	}
1621 	bcopy(&ealength, p, sizeof(ealength));
1622 	p += sizeof(ealength);
1623 	*p++ = ap->a_attrnamespace;
1624 	*p++ = eapad2;
1625 	*p++ = strlen(ap->a_name);
1626 	strcpy(p, ap->a_name);
1627 	p += strlen(ap->a_name);
1628 	bzero(p, eapad1);
1629 	p += eapad1;
1630 	error = uiomove(p, ealen, ap->a_uio);
1631 	if (error) {
1632 		free(eae, M_TEMP);
1633 		if (stand_alone)
1634 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1635 		else if (ip->i_ea_error == 0)
1636 			ip->i_ea_error = error;
1637 		return(error);
1638 	}
1639 	p += ealen;
1640 	bzero(p, eapad2);
1641 
1642 	p = ip->i_ea_area;
1643 	ip->i_ea_area = eae;
1644 	ip->i_ea_len = easize;
1645 	free(p, M_TEMP);
1646 	if (stand_alone)
1647 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1648 	return(error);
1649 }
1650