xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 74bf4e164ba5851606a27d4feff27717452583e5)
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  * Copyright (c) 1982, 1986, 1989, 1993
12  *	The Regents of the University of California.  All rights reserved.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	from: @(#)ufs_readwrite.c	8.11 (Berkeley) 5/8/95
39  * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
40  *	@(#)ffs_vnops.c	8.15 (Berkeley) 5/14/95
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include <sys/param.h>
47 #include <sys/bio.h>
48 #include <sys/systm.h>
49 #include <sys/buf.h>
50 #include <sys/conf.h>
51 #include <sys/extattr.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
54 #include <sys/malloc.h>
55 #include <sys/mount.h>
56 #include <sys/proc.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/stat.h>
60 #include <sys/vmmeter.h>
61 #include <sys/vnode.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_extern.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_pager.h>
68 #include <vm/vnode_pager.h>
69 
70 #include <ufs/ufs/extattr.h>
71 #include <ufs/ufs/quota.h>
72 #include <ufs/ufs/inode.h>
73 #include <ufs/ufs/ufs_extern.h>
74 #include <ufs/ufs/ufsmount.h>
75 
76 #include <ufs/ffs/fs.h>
77 #include <ufs/ffs/ffs_extern.h>
78 #include "opt_directio.h"
79 
80 #ifdef DIRECTIO
81 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
82 #endif
83 static int	ffs_fsync(struct vop_fsync_args *);
84 static int	ffs_getpages(struct vop_getpages_args *);
85 static int	ffs_read(struct vop_read_args *);
86 static int	ffs_write(struct vop_write_args *);
87 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
88 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
89 		    struct ucred *cred);
90 static int	ffsext_strategy(struct vop_strategy_args *);
91 static int	ffs_closeextattr(struct vop_closeextattr_args *);
92 static int	ffs_deleteextattr(struct vop_deleteextattr_args *);
93 static int	ffs_getextattr(struct vop_getextattr_args *);
94 static int	ffs_listextattr(struct vop_listextattr_args *);
95 static int	ffs_openextattr(struct vop_openextattr_args *);
96 static int	ffs_setextattr(struct vop_setextattr_args *);
97 
98 
99 /* Global vfs data structures for ufs. */
100 vop_t **ffs_vnodeop_p;
101 static struct vnodeopv_entry_desc ffs_vnodeop_entries[] = {
102 	{ &vop_default_desc,		(vop_t *) ufs_vnoperate },
103 	{ &vop_fsync_desc,		(vop_t *) ffs_fsync },
104 	{ &vop_getpages_desc,		(vop_t *) ffs_getpages },
105 	{ &vop_read_desc,		(vop_t *) ffs_read },
106 	{ &vop_reallocblks_desc,	(vop_t *) ffs_reallocblks },
107 	{ &vop_write_desc,		(vop_t *) ffs_write },
108 	{ &vop_closeextattr_desc,	(vop_t *) ffs_closeextattr },
109 	{ &vop_deleteextattr_desc,	(vop_t *) ffs_deleteextattr },
110 	{ &vop_getextattr_desc,		(vop_t *) ffs_getextattr },
111 	{ &vop_listextattr_desc,	(vop_t *) ffs_listextattr },
112 	{ &vop_openextattr_desc,	(vop_t *) ffs_openextattr },
113 	{ &vop_setextattr_desc,		(vop_t *) ffs_setextattr },
114 	{ NULL, NULL }
115 };
116 static struct vnodeopv_desc ffs_vnodeop_opv_desc =
117 	{ &ffs_vnodeop_p, ffs_vnodeop_entries };
118 
119 vop_t **ffs_specop_p;
120 static struct vnodeopv_entry_desc ffs_specop_entries[] = {
121 	{ &vop_default_desc,		(vop_t *) ufs_vnoperatespec },
122 	{ &vop_fsync_desc,		(vop_t *) ffs_fsync },
123 	{ &vop_reallocblks_desc,	(vop_t *) ffs_reallocblks },
124 	{ &vop_strategy_desc,		(vop_t *) ffsext_strategy },
125 	{ &vop_closeextattr_desc,	(vop_t *) ffs_closeextattr },
126 	{ &vop_deleteextattr_desc,	(vop_t *) ffs_deleteextattr },
127 	{ &vop_getextattr_desc,		(vop_t *) ffs_getextattr },
128 	{ &vop_listextattr_desc,	(vop_t *) ffs_listextattr },
129 	{ &vop_openextattr_desc,	(vop_t *) ffs_openextattr },
130 	{ &vop_setextattr_desc,		(vop_t *) ffs_setextattr },
131 	{ NULL, NULL }
132 };
133 static struct vnodeopv_desc ffs_specop_opv_desc =
134 	{ &ffs_specop_p, ffs_specop_entries };
135 
136 vop_t **ffs_fifoop_p;
137 static struct vnodeopv_entry_desc ffs_fifoop_entries[] = {
138 	{ &vop_default_desc,		(vop_t *) ufs_vnoperatefifo },
139 	{ &vop_fsync_desc,		(vop_t *) ffs_fsync },
140 	{ &vop_reallocblks_desc,	(vop_t *) ffs_reallocblks },
141 	{ &vop_strategy_desc,		(vop_t *) ffsext_strategy },
142 	{ &vop_closeextattr_desc,	(vop_t *) ffs_closeextattr },
143 	{ &vop_deleteextattr_desc,	(vop_t *) ffs_deleteextattr },
144 	{ &vop_getextattr_desc,		(vop_t *) ffs_getextattr },
145 	{ &vop_listextattr_desc,	(vop_t *) ffs_listextattr },
146 	{ &vop_openextattr_desc,	(vop_t *) ffs_openextattr },
147 	{ &vop_setextattr_desc,		(vop_t *) ffs_setextattr },
148 	{ NULL, NULL }
149 };
150 static struct vnodeopv_desc ffs_fifoop_opv_desc =
151 	{ &ffs_fifoop_p, ffs_fifoop_entries };
152 
153 VNODEOP_SET(ffs_vnodeop_opv_desc);
154 VNODEOP_SET(ffs_specop_opv_desc);
155 VNODEOP_SET(ffs_fifoop_opv_desc);
156 
157 /*
158  * Synch an open file.
159  */
160 /* ARGSUSED */
161 static int
162 ffs_fsync(ap)
163 	struct vop_fsync_args /* {
164 		struct vnode *a_vp;
165 		struct ucred *a_cred;
166 		int a_waitfor;
167 		struct thread *a_td;
168 	} */ *ap;
169 {
170 	struct vnode *vp = ap->a_vp;
171 	struct inode *ip = VTOI(vp);
172 	struct buf *bp;
173 	struct buf *nbp;
174 	int s, error, wait, passes, skipmeta;
175 	ufs_lbn_t lbn;
176 
177 	wait = (ap->a_waitfor == MNT_WAIT);
178 	if (vn_isdisk(vp, NULL)) {
179 		lbn = INT_MAX;
180 		if (vp->v_rdev->si_mountpoint != NULL &&
181 		    (vp->v_rdev->si_mountpoint->mnt_flag & MNT_SOFTDEP))
182 			softdep_fsync_mountdev(vp);
183 	} else {
184 		lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
185 	}
186 
187 	/*
188 	 * Flush all dirty buffers associated with a vnode.
189 	 */
190 	passes = NIADDR + 1;
191 	skipmeta = 0;
192 	if (wait)
193 		skipmeta = 1;
194 	s = splbio();
195 	VI_LOCK(vp);
196 loop:
197 	TAILQ_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs)
198 		bp->b_vflags &= ~BV_SCANNED;
199 	for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
200 		nbp = TAILQ_NEXT(bp, b_vnbufs);
201 		/*
202 		 * Reasons to skip this buffer: it has already been considered
203 		 * on this pass, this pass is the first time through on a
204 		 * synchronous flush request and the buffer being considered
205 		 * is metadata, the buffer has dependencies that will cause
206 		 * it to be redirtied and it has not already been deferred,
207 		 * or it is already being written.
208 		 */
209 		if ((bp->b_vflags & BV_SCANNED) != 0)
210 			continue;
211 		bp->b_vflags |= BV_SCANNED;
212 		if ((skipmeta == 1 && bp->b_lblkno < 0))
213 			continue;
214 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
215 			continue;
216 		if (!wait && LIST_FIRST(&bp->b_dep) != NULL &&
217 		    (bp->b_flags & B_DEFERRED) == 0 &&
218 		    buf_countdeps(bp, 0)) {
219 			bp->b_flags |= B_DEFERRED;
220 			BUF_UNLOCK(bp);
221 			continue;
222 		}
223 		VI_UNLOCK(vp);
224 		if ((bp->b_flags & B_DELWRI) == 0)
225 			panic("ffs_fsync: not dirty");
226 		if (vp != bp->b_vp)
227 			panic("ffs_fsync: vp != vp->b_vp");
228 		/*
229 		 * If this is a synchronous flush request, or it is not a
230 		 * file or device, start the write on this buffer immediatly.
231 		 */
232 		if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
233 
234 			/*
235 			 * On our final pass through, do all I/O synchronously
236 			 * so that we can find out if our flush is failing
237 			 * because of write errors.
238 			 */
239 			if (passes > 0 || !wait) {
240 				if ((bp->b_flags & B_CLUSTEROK) && !wait) {
241 					(void) vfs_bio_awrite(bp);
242 				} else {
243 					bremfree(bp);
244 					splx(s);
245 					(void) bawrite(bp);
246 					s = splbio();
247 				}
248 			} else {
249 				bremfree(bp);
250 				splx(s);
251 				if ((error = bwrite(bp)) != 0)
252 					return (error);
253 				s = splbio();
254 			}
255 		} else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
256 			/*
257 			 * If the buffer is for data that has been truncated
258 			 * off the file, then throw it away.
259 			 */
260 			bremfree(bp);
261 			bp->b_flags |= B_INVAL | B_NOCACHE;
262 			splx(s);
263 			brelse(bp);
264 			s = splbio();
265 		} else
266 			vfs_bio_awrite(bp);
267 
268 		/*
269 		 * Since we may have slept during the I/O, we need
270 		 * to start from a known point.
271 		 */
272 		VI_LOCK(vp);
273 		nbp = TAILQ_FIRST(&vp->v_dirtyblkhd);
274 	}
275 	/*
276 	 * If we were asked to do this synchronously, then go back for
277 	 * another pass, this time doing the metadata.
278 	 */
279 	if (skipmeta) {
280 		skipmeta = 0;
281 		goto loop;
282 	}
283 
284 	if (wait) {
285 		while (vp->v_numoutput) {
286 			vp->v_iflag |= VI_BWAIT;
287 			msleep((caddr_t)&vp->v_numoutput, VI_MTX(vp),
288 			    PRIBIO + 4, "ffsfsn", 0);
289   		}
290 		VI_UNLOCK(vp);
291 
292 		/*
293 		 * Ensure that any filesystem metatdata associated
294 		 * with the vnode has been written.
295 		 */
296 		splx(s);
297 		if ((error = softdep_sync_metadata(ap)) != 0)
298 			return (error);
299 		s = splbio();
300 
301 		VI_LOCK(vp);
302 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
303 			/*
304 			 * Block devices associated with filesystems may
305 			 * have new I/O requests posted for them even if
306 			 * the vnode is locked, so no amount of trying will
307 			 * get them clean. Thus we give block devices a
308 			 * good effort, then just give up. For all other file
309 			 * types, go around and try again until it is clean.
310 			 */
311 			if (passes > 0) {
312 				passes -= 1;
313 				goto loop;
314 			}
315 #ifdef DIAGNOSTIC
316 			if (!vn_isdisk(vp, NULL))
317 				vprint("ffs_fsync: dirty", vp);
318 #endif
319 		}
320 	}
321 	VI_UNLOCK(vp);
322 	splx(s);
323 	return (UFS_UPDATE(vp, wait));
324 }
325 
326 
327 /*
328  * Vnode op for reading.
329  */
330 /* ARGSUSED */
331 static int
332 ffs_read(ap)
333 	struct vop_read_args /* {
334 		struct vnode *a_vp;
335 		struct uio *a_uio;
336 		int a_ioflag;
337 		struct ucred *a_cred;
338 	} */ *ap;
339 {
340 	struct vnode *vp;
341 	struct inode *ip;
342 	struct uio *uio;
343 	struct fs *fs;
344 	struct buf *bp;
345 	ufs_lbn_t lbn, nextlbn;
346 	off_t bytesinfile;
347 	long size, xfersize, blkoffset;
348 	int error, orig_resid;
349 	int seqcount;
350 	int ioflag;
351 
352 	vp = ap->a_vp;
353 	uio = ap->a_uio;
354 	ioflag = ap->a_ioflag;
355 	if (ap->a_ioflag & IO_EXT)
356 #ifdef notyet
357 		return (ffs_extread(vp, uio, ioflag));
358 #else
359 		panic("ffs_read+IO_EXT");
360 #endif
361 #ifdef DIRECTIO
362 	if ((ioflag & IO_DIRECT) != 0) {
363 		int workdone;
364 
365 		error = ffs_rawread(vp, uio, &workdone);
366 		if (error != 0 || workdone != 0)
367 			return error;
368 	}
369 #endif
370 
371 	GIANT_REQUIRED;
372 
373 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
374 	ip = VTOI(vp);
375 
376 #ifdef DIAGNOSTIC
377 	if (uio->uio_rw != UIO_READ)
378 		panic("ffs_read: mode");
379 
380 	if (vp->v_type == VLNK) {
381 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
382 			panic("ffs_read: short symlink");
383 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
384 		panic("ffs_read: type %d",  vp->v_type);
385 #endif
386 	orig_resid = uio->uio_resid;
387 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
388 	if (orig_resid == 0)
389 		return (0);
390 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
391 	fs = ip->i_fs;
392 	if (uio->uio_offset < ip->i_size &&
393 	    uio->uio_offset >= fs->fs_maxfilesize)
394 		return (EOVERFLOW);
395 
396 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
397 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
398 			break;
399 		lbn = lblkno(fs, uio->uio_offset);
400 		nextlbn = lbn + 1;
401 
402 		/*
403 		 * size of buffer.  The buffer representing the
404 		 * end of the file is rounded up to the size of
405 		 * the block type ( fragment or full block,
406 		 * depending ).
407 		 */
408 		size = blksize(fs, ip, lbn);
409 		blkoffset = blkoff(fs, uio->uio_offset);
410 
411 		/*
412 		 * The amount we want to transfer in this iteration is
413 		 * one FS block less the amount of the data before
414 		 * our startpoint (duh!)
415 		 */
416 		xfersize = fs->fs_bsize - blkoffset;
417 
418 		/*
419 		 * But if we actually want less than the block,
420 		 * or the file doesn't have a whole block more of data,
421 		 * then use the lesser number.
422 		 */
423 		if (uio->uio_resid < xfersize)
424 			xfersize = uio->uio_resid;
425 		if (bytesinfile < xfersize)
426 			xfersize = bytesinfile;
427 
428 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
429 			/*
430 			 * Don't do readahead if this is the end of the file.
431 			 */
432 			error = bread(vp, lbn, size, NOCRED, &bp);
433 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
434 			/*
435 			 * Otherwise if we are allowed to cluster,
436 			 * grab as much as we can.
437 			 *
438 			 * XXX  This may not be a win if we are not
439 			 * doing sequential access.
440 			 */
441 			error = cluster_read(vp, ip->i_size, lbn,
442 				size, NOCRED, uio->uio_resid, seqcount, &bp);
443 		} else if (seqcount > 1) {
444 			/*
445 			 * If we are NOT allowed to cluster, then
446 			 * if we appear to be acting sequentially,
447 			 * fire off a request for a readahead
448 			 * as well as a read. Note that the 4th and 5th
449 			 * arguments point to arrays of the size specified in
450 			 * the 6th argument.
451 			 */
452 			int nextsize = blksize(fs, ip, nextlbn);
453 			error = breadn(vp, lbn,
454 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
455 		} else {
456 			/*
457 			 * Failing all of the above, just read what the
458 			 * user asked for. Interestingly, the same as
459 			 * the first option above.
460 			 */
461 			error = bread(vp, lbn, size, NOCRED, &bp);
462 		}
463 		if (error) {
464 			brelse(bp);
465 			bp = NULL;
466 			break;
467 		}
468 
469 		/*
470 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
471 		 * will cause us to attempt to release the buffer later on
472 		 * and will cause the buffer cache to attempt to free the
473 		 * underlying pages.
474 		 */
475 		if (ioflag & IO_DIRECT)
476 			bp->b_flags |= B_DIRECT;
477 
478 		/*
479 		 * We should only get non-zero b_resid when an I/O error
480 		 * has occurred, which should cause us to break above.
481 		 * However, if the short read did not cause an error,
482 		 * then we want to ensure that we do not uiomove bad
483 		 * or uninitialized data.
484 		 */
485 		size -= bp->b_resid;
486 		if (size < xfersize) {
487 			if (size == 0)
488 				break;
489 			xfersize = size;
490 		}
491 
492 		error = uiomove((char *)bp->b_data + blkoffset,
493 		    (int)xfersize, uio);
494 		if (error)
495 			break;
496 
497 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
498 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
499 			/*
500 			 * If there are no dependencies, and it's VMIO,
501 			 * then we don't need the buf, mark it available
502 			 * for freeing. The VM has the data.
503 			 */
504 			bp->b_flags |= B_RELBUF;
505 			brelse(bp);
506 		} else {
507 			/*
508 			 * Otherwise let whoever
509 			 * made the request take care of
510 			 * freeing it. We just queue
511 			 * it onto another list.
512 			 */
513 			bqrelse(bp);
514 		}
515 	}
516 
517 	/*
518 	 * This can only happen in the case of an error
519 	 * because the loop above resets bp to NULL on each iteration
520 	 * and on normal completion has not set a new value into it.
521 	 * so it must have come from a 'break' statement
522 	 */
523 	if (bp != NULL) {
524 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
525 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
526 			bp->b_flags |= B_RELBUF;
527 			brelse(bp);
528 		} else {
529 			bqrelse(bp);
530 		}
531 	}
532 
533 	if ((error == 0 || uio->uio_resid != orig_resid) &&
534 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
535 		ip->i_flag |= IN_ACCESS;
536 	return (error);
537 }
538 
539 /*
540  * Vnode op for writing.
541  */
542 static int
543 ffs_write(ap)
544 	struct vop_write_args /* {
545 		struct vnode *a_vp;
546 		struct uio *a_uio;
547 		int a_ioflag;
548 		struct ucred *a_cred;
549 	} */ *ap;
550 {
551 	struct vnode *vp;
552 	struct uio *uio;
553 	struct inode *ip;
554 	struct fs *fs;
555 	struct buf *bp;
556 	struct thread *td;
557 	ufs_lbn_t lbn;
558 	off_t osize;
559 	int seqcount;
560 	int blkoffset, error, extended, flags, ioflag, resid, size, xfersize;
561 
562 	vp = ap->a_vp;
563 	uio = ap->a_uio;
564 	ioflag = ap->a_ioflag;
565 	if (ap->a_ioflag & IO_EXT)
566 #ifdef notyet
567 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
568 #else
569 		panic("ffs_write+IO_EXT");
570 #endif
571 
572 	GIANT_REQUIRED;
573 
574 	extended = 0;
575 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
576 	ip = VTOI(vp);
577 
578 #ifdef DIAGNOSTIC
579 	if (uio->uio_rw != UIO_WRITE)
580 		panic("ffs_write: mode");
581 #endif
582 
583 	switch (vp->v_type) {
584 	case VREG:
585 		if (ioflag & IO_APPEND)
586 			uio->uio_offset = ip->i_size;
587 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
588 			return (EPERM);
589 		/* FALLTHROUGH */
590 	case VLNK:
591 		break;
592 	case VDIR:
593 		panic("ffs_write: dir write");
594 		break;
595 	default:
596 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
597 			(int)uio->uio_offset,
598 			(int)uio->uio_resid
599 		);
600 	}
601 
602 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
603 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
604 	fs = ip->i_fs;
605 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
606 		return (EFBIG);
607 	/*
608 	 * Maybe this should be above the vnode op call, but so long as
609 	 * file servers have no limits, I don't think it matters.
610 	 */
611 	td = uio->uio_td;
612 	if (vp->v_type == VREG && td != NULL) {
613 		PROC_LOCK(td->td_proc);
614 		if (uio->uio_offset + uio->uio_resid >
615 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
616 			psignal(td->td_proc, SIGXFSZ);
617 			PROC_UNLOCK(td->td_proc);
618 			return (EFBIG);
619 		}
620 		PROC_UNLOCK(td->td_proc);
621 	}
622 
623 	resid = uio->uio_resid;
624 	osize = ip->i_size;
625 	if (seqcount > BA_SEQMAX)
626 		flags = BA_SEQMAX << BA_SEQSHIFT;
627 	else
628 		flags = seqcount << BA_SEQSHIFT;
629 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
630 		flags |= IO_SYNC;
631 
632 	for (error = 0; uio->uio_resid > 0;) {
633 		lbn = lblkno(fs, uio->uio_offset);
634 		blkoffset = blkoff(fs, uio->uio_offset);
635 		xfersize = fs->fs_bsize - blkoffset;
636 		if (uio->uio_resid < xfersize)
637 			xfersize = uio->uio_resid;
638 		if (uio->uio_offset + xfersize > ip->i_size)
639 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
640 
641                 /*
642 		 * We must perform a read-before-write if the transfer size
643 		 * does not cover the entire buffer.
644                  */
645 		if (fs->fs_bsize > xfersize)
646 			flags |= BA_CLRBUF;
647 		else
648 			flags &= ~BA_CLRBUF;
649 /* XXX is uio->uio_offset the right thing here? */
650 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
651 		    ap->a_cred, flags, &bp);
652 		if (error != 0)
653 			break;
654 		/*
655 		 * If the buffer is not valid we have to clear out any
656 		 * garbage data from the pages instantiated for the buffer.
657 		 * If we do not, a failed uiomove() during a write can leave
658 		 * the prior contents of the pages exposed to a userland
659 		 * mmap().  XXX deal with uiomove() errors a better way.
660 		 */
661 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
662 			vfs_bio_clrbuf(bp);
663 		if (ioflag & IO_DIRECT)
664 			bp->b_flags |= B_DIRECT;
665 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
666 			bp->b_flags |= B_NOCACHE;
667 
668 		if (uio->uio_offset + xfersize > ip->i_size) {
669 			ip->i_size = uio->uio_offset + xfersize;
670 			DIP_SET(ip, i_size, ip->i_size);
671 			extended = 1;
672 		}
673 
674 		size = blksize(fs, ip, lbn) - bp->b_resid;
675 		if (size < xfersize)
676 			xfersize = size;
677 
678 		error =
679 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
680 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
681 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
682 			bp->b_flags |= B_RELBUF;
683 		}
684 
685 		/*
686 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
687 		 * if we have a severe page deficiency write the buffer
688 		 * asynchronously.  Otherwise try to cluster, and if that
689 		 * doesn't do it then either do an async write (if O_DIRECT),
690 		 * or a delayed write (if not).
691 		 */
692 		if (ioflag & IO_SYNC) {
693 			(void)bwrite(bp);
694 		} else if (vm_page_count_severe() ||
695 			    buf_dirty_count_severe() ||
696 			    (ioflag & IO_ASYNC)) {
697 			bp->b_flags |= B_CLUSTEROK;
698 			bawrite(bp);
699 		} else if (xfersize + blkoffset == fs->fs_bsize) {
700 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
701 				bp->b_flags |= B_CLUSTEROK;
702 				cluster_write(bp, ip->i_size, seqcount);
703 			} else {
704 				bawrite(bp);
705 			}
706 		} else if (ioflag & IO_DIRECT) {
707 			bp->b_flags |= B_CLUSTEROK;
708 			bawrite(bp);
709 		} else {
710 			bp->b_flags |= B_CLUSTEROK;
711 			bdwrite(bp);
712 		}
713 		if (error || xfersize == 0)
714 			break;
715 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
716 	}
717 	/*
718 	 * If we successfully wrote any data, and we are not the superuser
719 	 * we clear the setuid and setgid bits as a precaution against
720 	 * tampering.
721 	 */
722 	if (resid > uio->uio_resid && ap->a_cred &&
723 	    suser_cred(ap->a_cred, SUSER_ALLOWJAIL)) {
724 		ip->i_mode &= ~(ISUID | ISGID);
725 		DIP_SET(ip, i_mode, ip->i_mode);
726 	}
727 	if (resid > uio->uio_resid)
728 		VN_KNOTE_UNLOCKED(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
729 	if (error) {
730 		if (ioflag & IO_UNIT) {
731 			(void)UFS_TRUNCATE(vp, osize,
732 			    IO_NORMAL | (ioflag & IO_SYNC),
733 			    ap->a_cred, uio->uio_td);
734 			uio->uio_offset -= resid - uio->uio_resid;
735 			uio->uio_resid = resid;
736 		}
737 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
738 		error = UFS_UPDATE(vp, 1);
739 	return (error);
740 }
741 
742 /*
743  * get page routine
744  */
745 static int
746 ffs_getpages(ap)
747 	struct vop_getpages_args *ap;
748 {
749 	off_t foff, physoffset;
750 	int i, size, bsize;
751 	struct vnode *dp, *vp;
752 	vm_object_t obj;
753 	vm_pindex_t pindex;
754 	vm_page_t mreq;
755 	int bbackwards, bforwards;
756 	int pbackwards, pforwards;
757 	int firstpage;
758 	ufs2_daddr_t reqblkno, reqlblkno;
759 	int poff;
760 	int pcount;
761 	int rtval;
762 	int pagesperblock;
763 
764 	GIANT_REQUIRED;
765 
766 	pcount = round_page(ap->a_count) / PAGE_SIZE;
767 	mreq = ap->a_m[ap->a_reqpage];
768 
769 	/*
770 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
771 	 * then the entire page is valid.  Since the page may be mapped,
772 	 * user programs might reference data beyond the actual end of file
773 	 * occuring within the page.  We have to zero that data.
774 	 */
775 	VM_OBJECT_LOCK(mreq->object);
776 	if (mreq->valid) {
777 		if (mreq->valid != VM_PAGE_BITS_ALL)
778 			vm_page_zero_invalid(mreq, TRUE);
779 		vm_page_lock_queues();
780 		for (i = 0; i < pcount; i++) {
781 			if (i != ap->a_reqpage) {
782 				vm_page_free(ap->a_m[i]);
783 			}
784 		}
785 		vm_page_unlock_queues();
786 		VM_OBJECT_UNLOCK(mreq->object);
787 		return VM_PAGER_OK;
788 	}
789 	VM_OBJECT_UNLOCK(mreq->object);
790 	vp = ap->a_vp;
791 	obj = vp->v_object;
792 	bsize = vp->v_mount->mnt_stat.f_iosize;
793 	pindex = mreq->pindex;
794 	foff = IDX_TO_OFF(pindex) /* + ap->a_offset should be zero */;
795 
796 	if (bsize < PAGE_SIZE)
797 		return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
798 						    ap->a_count,
799 						    ap->a_reqpage);
800 
801 	/*
802 	 * foff is the file offset of the required page
803 	 * reqlblkno is the logical block that contains the page
804 	 * poff is the index of the page into the logical block
805 	 */
806 	reqlblkno = foff / bsize;
807 	poff = (foff % bsize) / PAGE_SIZE;
808 
809 	dp = VTOI(vp)->i_devvp;
810 	if (ufs_bmaparray(vp, reqlblkno, &reqblkno, 0, &bforwards, &bbackwards)
811 	    || (reqblkno == -1)) {
812 		VM_OBJECT_LOCK(obj);
813 		vm_page_lock_queues();
814 		for(i = 0; i < pcount; i++) {
815 			if (i != ap->a_reqpage)
816 				vm_page_free(ap->a_m[i]);
817 		}
818 		vm_page_unlock_queues();
819 		if (reqblkno == -1) {
820 			if ((mreq->flags & PG_ZERO) == 0)
821 				pmap_zero_page(mreq);
822 			vm_page_undirty(mreq);
823 			mreq->valid = VM_PAGE_BITS_ALL;
824 			VM_OBJECT_UNLOCK(obj);
825 			return VM_PAGER_OK;
826 		} else {
827 			VM_OBJECT_UNLOCK(obj);
828 			return VM_PAGER_ERROR;
829 		}
830 	}
831 
832 	physoffset = (off_t)reqblkno * DEV_BSIZE + poff * PAGE_SIZE;
833 	pagesperblock = bsize / PAGE_SIZE;
834 	/*
835 	 * find the first page that is contiguous...
836 	 * note that pbackwards is the number of pages that are contiguous
837 	 * backwards.
838 	 */
839 	firstpage = 0;
840 	if (ap->a_count) {
841 		pbackwards = poff + bbackwards * pagesperblock;
842 		if (ap->a_reqpage > pbackwards) {
843 			firstpage = ap->a_reqpage - pbackwards;
844 			VM_OBJECT_LOCK(obj);
845 			vm_page_lock_queues();
846 			for(i=0;i<firstpage;i++)
847 				vm_page_free(ap->a_m[i]);
848 			vm_page_unlock_queues();
849 			VM_OBJECT_UNLOCK(obj);
850 		}
851 
852 	/*
853 	 * pforwards is the number of pages that are contiguous
854 	 * after the current page.
855 	 */
856 		pforwards = (pagesperblock - (poff + 1)) +
857 			bforwards * pagesperblock;
858 		if (pforwards < (pcount - (ap->a_reqpage + 1))) {
859 			VM_OBJECT_LOCK(obj);
860 			vm_page_lock_queues();
861 			for( i = ap->a_reqpage + pforwards + 1; i < pcount; i++)
862 				vm_page_free(ap->a_m[i]);
863 			vm_page_unlock_queues();
864 			VM_OBJECT_UNLOCK(obj);
865 			pcount = ap->a_reqpage + pforwards + 1;
866 		}
867 
868 	/*
869 	 * number of pages for I/O corrected for the non-contig pages at
870 	 * the beginning of the array.
871 	 */
872 		pcount -= firstpage;
873 	}
874 
875 	/*
876 	 * calculate the size of the transfer
877 	 */
878 
879 	size = pcount * PAGE_SIZE;
880 
881 	if ((IDX_TO_OFF(ap->a_m[firstpage]->pindex) + size) >
882 		obj->un_pager.vnp.vnp_size)
883 		size = obj->un_pager.vnp.vnp_size -
884 			IDX_TO_OFF(ap->a_m[firstpage]->pindex);
885 
886 	physoffset -= foff;
887 	rtval = VOP_GETPAGES(dp, &ap->a_m[firstpage], size,
888 		(ap->a_reqpage - firstpage), physoffset);
889 
890 	return (rtval);
891 }
892 
893 /*
894  * Extended attribute area reading.
895  */
896 static int
897 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
898 {
899 	struct inode *ip;
900 	struct ufs2_dinode *dp;
901 	struct fs *fs;
902 	struct buf *bp;
903 	ufs_lbn_t lbn, nextlbn;
904 	off_t bytesinfile;
905 	long size, xfersize, blkoffset;
906 	int error, orig_resid;
907 
908 	GIANT_REQUIRED;
909 
910 	ip = VTOI(vp);
911 	fs = ip->i_fs;
912 	dp = ip->i_din2;
913 
914 #ifdef DIAGNOSTIC
915 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
916 		panic("ffs_extread: mode");
917 
918 #endif
919 	orig_resid = uio->uio_resid;
920 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
921 	if (orig_resid == 0)
922 		return (0);
923 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
924 
925 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
926 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
927 			break;
928 		lbn = lblkno(fs, uio->uio_offset);
929 		nextlbn = lbn + 1;
930 
931 		/*
932 		 * size of buffer.  The buffer representing the
933 		 * end of the file is rounded up to the size of
934 		 * the block type ( fragment or full block,
935 		 * depending ).
936 		 */
937 		size = sblksize(fs, dp->di_extsize, lbn);
938 		blkoffset = blkoff(fs, uio->uio_offset);
939 
940 		/*
941 		 * The amount we want to transfer in this iteration is
942 		 * one FS block less the amount of the data before
943 		 * our startpoint (duh!)
944 		 */
945 		xfersize = fs->fs_bsize - blkoffset;
946 
947 		/*
948 		 * But if we actually want less than the block,
949 		 * or the file doesn't have a whole block more of data,
950 		 * then use the lesser number.
951 		 */
952 		if (uio->uio_resid < xfersize)
953 			xfersize = uio->uio_resid;
954 		if (bytesinfile < xfersize)
955 			xfersize = bytesinfile;
956 
957 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
958 			/*
959 			 * Don't do readahead if this is the end of the info.
960 			 */
961 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
962 		} else {
963 			/*
964 			 * If we have a second block, then
965 			 * fire off a request for a readahead
966 			 * as well as a read. Note that the 4th and 5th
967 			 * arguments point to arrays of the size specified in
968 			 * the 6th argument.
969 			 */
970 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
971 
972 			nextlbn = -1 - nextlbn;
973 			error = breadn(vp, -1 - lbn,
974 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
975 		}
976 		if (error) {
977 			brelse(bp);
978 			bp = NULL;
979 			break;
980 		}
981 
982 		/*
983 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
984 		 * will cause us to attempt to release the buffer later on
985 		 * and will cause the buffer cache to attempt to free the
986 		 * underlying pages.
987 		 */
988 		if (ioflag & IO_DIRECT)
989 			bp->b_flags |= B_DIRECT;
990 
991 		/*
992 		 * We should only get non-zero b_resid when an I/O error
993 		 * has occurred, which should cause us to break above.
994 		 * However, if the short read did not cause an error,
995 		 * then we want to ensure that we do not uiomove bad
996 		 * or uninitialized data.
997 		 */
998 		size -= bp->b_resid;
999 		if (size < xfersize) {
1000 			if (size == 0)
1001 				break;
1002 			xfersize = size;
1003 		}
1004 
1005 		error = uiomove((char *)bp->b_data + blkoffset,
1006 					(int)xfersize, uio);
1007 		if (error)
1008 			break;
1009 
1010 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1011 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1012 			/*
1013 			 * If there are no dependencies, and it's VMIO,
1014 			 * then we don't need the buf, mark it available
1015 			 * for freeing. The VM has the data.
1016 			 */
1017 			bp->b_flags |= B_RELBUF;
1018 			brelse(bp);
1019 		} else {
1020 			/*
1021 			 * Otherwise let whoever
1022 			 * made the request take care of
1023 			 * freeing it. We just queue
1024 			 * it onto another list.
1025 			 */
1026 			bqrelse(bp);
1027 		}
1028 	}
1029 
1030 	/*
1031 	 * This can only happen in the case of an error
1032 	 * because the loop above resets bp to NULL on each iteration
1033 	 * and on normal completion has not set a new value into it.
1034 	 * so it must have come from a 'break' statement
1035 	 */
1036 	if (bp != NULL) {
1037 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1038 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1039 			bp->b_flags |= B_RELBUF;
1040 			brelse(bp);
1041 		} else {
1042 			bqrelse(bp);
1043 		}
1044 	}
1045 
1046 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1047 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1048 		ip->i_flag |= IN_ACCESS;
1049 	return (error);
1050 }
1051 
1052 /*
1053  * Extended attribute area writing.
1054  */
1055 static int
1056 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1057 {
1058 	struct inode *ip;
1059 	struct ufs2_dinode *dp;
1060 	struct fs *fs;
1061 	struct buf *bp;
1062 	ufs_lbn_t lbn;
1063 	off_t osize;
1064 	int blkoffset, error, flags, resid, size, xfersize;
1065 
1066 	GIANT_REQUIRED;
1067 
1068 	ip = VTOI(vp);
1069 	fs = ip->i_fs;
1070 	dp = ip->i_din2;
1071 
1072 #ifdef DIAGNOSTIC
1073 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1074 		panic("ffs_extwrite: mode");
1075 #endif
1076 
1077 	if (ioflag & IO_APPEND)
1078 		uio->uio_offset = dp->di_extsize;
1079 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1080 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1081 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1082 		return (EFBIG);
1083 
1084 	resid = uio->uio_resid;
1085 	osize = dp->di_extsize;
1086 	flags = IO_EXT;
1087 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1088 		flags |= IO_SYNC;
1089 
1090 	for (error = 0; uio->uio_resid > 0;) {
1091 		lbn = lblkno(fs, uio->uio_offset);
1092 		blkoffset = blkoff(fs, uio->uio_offset);
1093 		xfersize = fs->fs_bsize - blkoffset;
1094 		if (uio->uio_resid < xfersize)
1095 			xfersize = uio->uio_resid;
1096 
1097                 /*
1098 		 * We must perform a read-before-write if the transfer size
1099 		 * does not cover the entire buffer.
1100                  */
1101 		if (fs->fs_bsize > xfersize)
1102 			flags |= BA_CLRBUF;
1103 		else
1104 			flags &= ~BA_CLRBUF;
1105 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1106 		    ucred, flags, &bp);
1107 		if (error != 0)
1108 			break;
1109 		/*
1110 		 * If the buffer is not valid we have to clear out any
1111 		 * garbage data from the pages instantiated for the buffer.
1112 		 * If we do not, a failed uiomove() during a write can leave
1113 		 * the prior contents of the pages exposed to a userland
1114 		 * mmap().  XXX deal with uiomove() errors a better way.
1115 		 */
1116 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1117 			vfs_bio_clrbuf(bp);
1118 		if (ioflag & IO_DIRECT)
1119 			bp->b_flags |= B_DIRECT;
1120 
1121 		if (uio->uio_offset + xfersize > dp->di_extsize)
1122 			dp->di_extsize = uio->uio_offset + xfersize;
1123 
1124 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1125 		if (size < xfersize)
1126 			xfersize = size;
1127 
1128 		error =
1129 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1130 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1131 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1132 			bp->b_flags |= B_RELBUF;
1133 		}
1134 
1135 		/*
1136 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1137 		 * if we have a severe page deficiency write the buffer
1138 		 * asynchronously.  Otherwise try to cluster, and if that
1139 		 * doesn't do it then either do an async write (if O_DIRECT),
1140 		 * or a delayed write (if not).
1141 		 */
1142 		if (ioflag & IO_SYNC) {
1143 			(void)bwrite(bp);
1144 		} else if (vm_page_count_severe() ||
1145 			    buf_dirty_count_severe() ||
1146 			    xfersize + blkoffset == fs->fs_bsize ||
1147 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1148 			bawrite(bp);
1149 		else
1150 			bdwrite(bp);
1151 		if (error || xfersize == 0)
1152 			break;
1153 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1154 	}
1155 	/*
1156 	 * If we successfully wrote any data, and we are not the superuser
1157 	 * we clear the setuid and setgid bits as a precaution against
1158 	 * tampering.
1159 	 */
1160 	if (resid > uio->uio_resid && ucred &&
1161 	    suser_cred(ucred, SUSER_ALLOWJAIL)) {
1162 		ip->i_mode &= ~(ISUID | ISGID);
1163 		dp->di_mode = ip->i_mode;
1164 	}
1165 	if (error) {
1166 		if (ioflag & IO_UNIT) {
1167 			(void)UFS_TRUNCATE(vp, osize,
1168 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1169 			uio->uio_offset -= resid - uio->uio_resid;
1170 			uio->uio_resid = resid;
1171 		}
1172 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1173 		error = UFS_UPDATE(vp, 1);
1174 	return (error);
1175 }
1176 
1177 
1178 /*
1179  * Vnode operating to retrieve a named extended attribute.
1180  *
1181  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1182  * the length of the EA, and possibly the pointer to the entry and to the data.
1183  */
1184 static int
1185 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1186 {
1187 	u_char *p, *pe, *pn, *p0;
1188 	int eapad1, eapad2, ealength, ealen, nlen;
1189 	uint32_t ul;
1190 
1191 	pe = ptr + length;
1192 	nlen = strlen(name);
1193 
1194 	for (p = ptr; p < pe; p = pn) {
1195 		p0 = p;
1196 		bcopy(p, &ul, sizeof(ul));
1197 		pn = p + ul;
1198 		/* make sure this entry is complete */
1199 		if (pn > pe)
1200 			break;
1201 		p += sizeof(uint32_t);
1202 		if (*p != nspace)
1203 			continue;
1204 		p++;
1205 		eapad2 = *p++;
1206 		if (*p != nlen)
1207 			continue;
1208 		p++;
1209 		if (bcmp(p, name, nlen))
1210 			continue;
1211 		ealength = sizeof(uint32_t) + 3 + nlen;
1212 		eapad1 = 8 - (ealength % 8);
1213 		if (eapad1 == 8)
1214 			eapad1 = 0;
1215 		ealength += eapad1;
1216 		ealen = ul - ealength - eapad2;
1217 		p += nlen + eapad1;
1218 		if (eap != NULL)
1219 			*eap = p0;
1220 		if (eac != NULL)
1221 			*eac = p;
1222 		return (ealen);
1223 	}
1224 	return(-1);
1225 }
1226 
1227 static int
1228 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1229 {
1230 	struct inode *ip;
1231 	struct ufs2_dinode *dp;
1232 	struct uio luio;
1233 	struct iovec liovec;
1234 	int easize, error;
1235 	u_char *eae;
1236 
1237 	ip = VTOI(vp);
1238 	dp = ip->i_din2;
1239 	easize = dp->di_extsize;
1240 
1241 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1242 
1243 	liovec.iov_base = eae;
1244 	liovec.iov_len = easize;
1245 	luio.uio_iov = &liovec;
1246 	luio.uio_iovcnt = 1;
1247 	luio.uio_offset = 0;
1248 	luio.uio_resid = easize;
1249 	luio.uio_segflg = UIO_SYSSPACE;
1250 	luio.uio_rw = UIO_READ;
1251 	luio.uio_td = td;
1252 
1253 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1254 	if (error) {
1255 		free(eae, M_TEMP);
1256 		return(error);
1257 	}
1258 	*p = eae;
1259 	return (0);
1260 }
1261 
1262 static int
1263 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1264 {
1265 	struct inode *ip;
1266 	struct ufs2_dinode *dp;
1267 	int error;
1268 
1269 	ip = VTOI(vp);
1270 
1271 	if (ip->i_ea_area != NULL)
1272 		return (EBUSY);
1273 	dp = ip->i_din2;
1274 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1275 	if (error)
1276 		return (error);
1277 	ip->i_ea_len = dp->di_extsize;
1278 	ip->i_ea_error = 0;
1279 	return (0);
1280 }
1281 
1282 /*
1283  * Vnode extattr transaction commit/abort
1284  */
1285 static int
1286 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1287 {
1288 	struct inode *ip;
1289 	struct uio luio;
1290 	struct iovec liovec;
1291 	int error;
1292 	struct ufs2_dinode *dp;
1293 
1294 	ip = VTOI(vp);
1295 	if (ip->i_ea_area == NULL)
1296 		return (EINVAL);
1297 	dp = ip->i_din2;
1298 	error = ip->i_ea_error;
1299 	if (commit && error == 0) {
1300 		if (cred == NOCRED)
1301 			cred =  vp->v_mount->mnt_cred;
1302 		liovec.iov_base = ip->i_ea_area;
1303 		liovec.iov_len = ip->i_ea_len;
1304 		luio.uio_iov = &liovec;
1305 		luio.uio_iovcnt = 1;
1306 		luio.uio_offset = 0;
1307 		luio.uio_resid = ip->i_ea_len;
1308 		luio.uio_segflg = UIO_SYSSPACE;
1309 		luio.uio_rw = UIO_WRITE;
1310 		luio.uio_td = td;
1311 		/* XXX: I'm not happy about truncating to zero size */
1312 		if (ip->i_ea_len < dp->di_extsize)
1313 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1314 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1315 	}
1316 	free(ip->i_ea_area, M_TEMP);
1317 	ip->i_ea_area = NULL;
1318 	ip->i_ea_len = 0;
1319 	ip->i_ea_error = 0;
1320 	return (error);
1321 }
1322 
1323 /*
1324  * Vnode extattr strategy routine for special devices and fifos.
1325  *
1326  * We need to check for a read or write of the external attributes.
1327  * Otherwise we just fall through and do the usual thing.
1328  */
1329 static int
1330 ffsext_strategy(struct vop_strategy_args *ap)
1331 /*
1332 struct vop_strategy_args {
1333 	struct vnodeop_desc *a_desc;
1334 	struct vnode *a_vp;
1335 	struct buf *a_bp;
1336 };
1337 */
1338 {
1339 	struct vnode *vp;
1340 	daddr_t lbn;
1341 
1342 	KASSERT(ap->a_vp == ap->a_bp->b_vp, ("%s(%p != %p)",
1343 	    __func__, ap->a_vp, ap->a_bp->b_vp));
1344 	vp = ap->a_vp;
1345 	lbn = ap->a_bp->b_lblkno;
1346 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1347 	    lbn < 0 && lbn >= -NXADDR)
1348 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1349 	if (vp->v_type == VFIFO)
1350 		return (ufs_vnoperatefifo((struct vop_generic_args *)ap));
1351 	return (ufs_vnoperatespec((struct vop_generic_args *)ap));
1352 }
1353 
1354 /*
1355  * Vnode extattr transaction commit/abort
1356  */
1357 static int
1358 ffs_openextattr(struct vop_openextattr_args *ap)
1359 /*
1360 struct vop_openextattr_args {
1361 	struct vnodeop_desc *a_desc;
1362 	struct vnode *a_vp;
1363 	IN struct ucred *a_cred;
1364 	IN struct thread *a_td;
1365 };
1366 */
1367 {
1368 	struct inode *ip;
1369 	struct fs *fs;
1370 
1371 	ip = VTOI(ap->a_vp);
1372 	fs = ip->i_fs;
1373 	if (fs->fs_magic == FS_UFS1_MAGIC)
1374 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1375 
1376 	if (ap->a_vp->v_type == VCHR)
1377 		return (EOPNOTSUPP);
1378 
1379 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1380 }
1381 
1382 
1383 /*
1384  * Vnode extattr transaction commit/abort
1385  */
1386 static int
1387 ffs_closeextattr(struct vop_closeextattr_args *ap)
1388 /*
1389 struct vop_closeextattr_args {
1390 	struct vnodeop_desc *a_desc;
1391 	struct vnode *a_vp;
1392 	int a_commit;
1393 	IN struct ucred *a_cred;
1394 	IN struct thread *a_td;
1395 };
1396 */
1397 {
1398 	struct inode *ip;
1399 	struct fs *fs;
1400 
1401 	ip = VTOI(ap->a_vp);
1402 	fs = ip->i_fs;
1403 	if (fs->fs_magic == FS_UFS1_MAGIC)
1404 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1405 
1406 	if (ap->a_vp->v_type == VCHR)
1407 		return (EOPNOTSUPP);
1408 
1409 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1410 }
1411 
1412 /*
1413  * Vnode operation to remove a named attribute.
1414  */
1415 static int
1416 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1417 /*
1418 vop_deleteextattr {
1419 	IN struct vnode *a_vp;
1420 	IN int a_attrnamespace;
1421 	IN const char *a_name;
1422 	IN struct ucred *a_cred;
1423 	IN struct thread *a_td;
1424 };
1425 */
1426 {
1427 	struct inode *ip;
1428 	struct fs *fs;
1429 	uint32_t ealength, ul;
1430 	int ealen, olen, eapad1, eapad2, error, i, easize;
1431 	u_char *eae, *p;
1432 	int stand_alone;
1433 
1434 	ip = VTOI(ap->a_vp);
1435 	fs = ip->i_fs;
1436 
1437 	if (fs->fs_magic == FS_UFS1_MAGIC)
1438 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1439 
1440 	if (ap->a_vp->v_type == VCHR)
1441 		return (EOPNOTSUPP);
1442 
1443 	if (strlen(ap->a_name) == 0)
1444 		return (EINVAL);
1445 
1446 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1447 	    ap->a_cred, ap->a_td, IWRITE);
1448 	if (error) {
1449 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1450 			ip->i_ea_error = error;
1451 		return (error);
1452 	}
1453 
1454 	if (ip->i_ea_area == NULL) {
1455 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1456 		if (error)
1457 			return (error);
1458 		stand_alone = 1;
1459 	} else {
1460 		stand_alone = 0;
1461 	}
1462 
1463 	ealength = eapad1 = ealen = eapad2 = 0;
1464 
1465 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1466 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1467 	easize = ip->i_ea_len;
1468 
1469 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1470 	    &p, NULL);
1471 	if (olen == -1) {
1472 		/* delete but nonexistent */
1473 		free(eae, M_TEMP);
1474 		if (stand_alone)
1475 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1476 		return(ENOATTR);
1477 	}
1478 	bcopy(p, &ul, sizeof ul);
1479 	i = p - eae + ul;
1480 	if (ul != ealength) {
1481 		bcopy(p + ul, p + ealength, easize - i);
1482 		easize += (ealength - ul);
1483 	}
1484 	if (easize > NXADDR * fs->fs_bsize) {
1485 		free(eae, M_TEMP);
1486 		if (stand_alone)
1487 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1488 		else if (ip->i_ea_error == 0)
1489 			ip->i_ea_error = ENOSPC;
1490 		return(ENOSPC);
1491 	}
1492 	p = ip->i_ea_area;
1493 	ip->i_ea_area = eae;
1494 	ip->i_ea_len = easize;
1495 	free(p, M_TEMP);
1496 	if (stand_alone)
1497 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1498 	return(error);
1499 }
1500 
1501 /*
1502  * Vnode operation to retrieve a named extended attribute.
1503  */
1504 static int
1505 ffs_getextattr(struct vop_getextattr_args *ap)
1506 /*
1507 vop_getextattr {
1508 	IN struct vnode *a_vp;
1509 	IN int a_attrnamespace;
1510 	IN const char *a_name;
1511 	INOUT struct uio *a_uio;
1512 	OUT size_t *a_size;
1513 	IN struct ucred *a_cred;
1514 	IN struct thread *a_td;
1515 };
1516 */
1517 {
1518 	struct inode *ip;
1519 	struct fs *fs;
1520 	u_char *eae, *p;
1521 	unsigned easize;
1522 	int error, ealen, stand_alone;
1523 
1524 	ip = VTOI(ap->a_vp);
1525 	fs = ip->i_fs;
1526 
1527 	if (fs->fs_magic == FS_UFS1_MAGIC)
1528 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1529 
1530 	if (ap->a_vp->v_type == VCHR)
1531 		return (EOPNOTSUPP);
1532 
1533 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1534 	    ap->a_cred, ap->a_td, IREAD);
1535 	if (error)
1536 		return (error);
1537 
1538 	if (ip->i_ea_area == NULL) {
1539 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1540 		if (error)
1541 			return (error);
1542 		stand_alone = 1;
1543 	} else {
1544 		stand_alone = 0;
1545 	}
1546 	eae = ip->i_ea_area;
1547 	easize = ip->i_ea_len;
1548 
1549 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1550 	    NULL, &p);
1551 	if (ealen >= 0) {
1552 		error = 0;
1553 		if (ap->a_size != NULL)
1554 			*ap->a_size = ealen;
1555 		else if (ap->a_uio != NULL)
1556 			error = uiomove(p, ealen, ap->a_uio);
1557 	} else
1558 		error = ENOATTR;
1559 	if (stand_alone)
1560 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1561 	return(error);
1562 }
1563 
1564 /*
1565  * Vnode operation to retrieve extended attributes on a vnode.
1566  */
1567 static int
1568 ffs_listextattr(struct vop_listextattr_args *ap)
1569 /*
1570 vop_listextattr {
1571 	IN struct vnode *a_vp;
1572 	IN int a_attrnamespace;
1573 	INOUT struct uio *a_uio;
1574 	OUT size_t *a_size;
1575 	IN struct ucred *a_cred;
1576 	IN struct thread *a_td;
1577 };
1578 */
1579 {
1580 	struct inode *ip;
1581 	struct fs *fs;
1582 	u_char *eae, *p, *pe, *pn;
1583 	unsigned easize;
1584 	uint32_t ul;
1585 	int error, ealen, stand_alone;
1586 
1587 	ip = VTOI(ap->a_vp);
1588 	fs = ip->i_fs;
1589 
1590 	if (fs->fs_magic == FS_UFS1_MAGIC)
1591 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1592 
1593 	if (ap->a_vp->v_type == VCHR)
1594 		return (EOPNOTSUPP);
1595 
1596 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1597 	    ap->a_cred, ap->a_td, IREAD);
1598 	if (error)
1599 		return (error);
1600 
1601 	if (ip->i_ea_area == NULL) {
1602 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1603 		if (error)
1604 			return (error);
1605 		stand_alone = 1;
1606 	} else {
1607 		stand_alone = 0;
1608 	}
1609 	eae = ip->i_ea_area;
1610 	easize = ip->i_ea_len;
1611 
1612 	error = 0;
1613 	if (ap->a_size != NULL)
1614 		*ap->a_size = 0;
1615 	pe = eae + easize;
1616 	for(p = eae; error == 0 && p < pe; p = pn) {
1617 		bcopy(p, &ul, sizeof(ul));
1618 		pn = p + ul;
1619 		if (pn > pe)
1620 			break;
1621 		p += sizeof(ul);
1622 		if (*p++ != ap->a_attrnamespace)
1623 			continue;
1624 		p++;	/* pad2 */
1625 		ealen = *p;
1626 		if (ap->a_size != NULL) {
1627 			*ap->a_size += ealen + 1;
1628 		} else if (ap->a_uio != NULL) {
1629 			error = uiomove(p, ealen + 1, ap->a_uio);
1630 		}
1631 	}
1632 	if (stand_alone)
1633 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1634 	return(error);
1635 }
1636 
1637 /*
1638  * Vnode operation to set a named attribute.
1639  */
1640 static int
1641 ffs_setextattr(struct vop_setextattr_args *ap)
1642 /*
1643 vop_setextattr {
1644 	IN struct vnode *a_vp;
1645 	IN int a_attrnamespace;
1646 	IN const char *a_name;
1647 	INOUT struct uio *a_uio;
1648 	IN struct ucred *a_cred;
1649 	IN struct thread *a_td;
1650 };
1651 */
1652 {
1653 	struct inode *ip;
1654 	struct fs *fs;
1655 	uint32_t ealength, ul;
1656 	int ealen, olen, eapad1, eapad2, error, i, easize;
1657 	u_char *eae, *p;
1658 	int stand_alone;
1659 
1660 	ip = VTOI(ap->a_vp);
1661 	fs = ip->i_fs;
1662 
1663 	if (fs->fs_magic == FS_UFS1_MAGIC)
1664 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1665 
1666 	if (ap->a_vp->v_type == VCHR)
1667 		return (EOPNOTSUPP);
1668 
1669 	if (strlen(ap->a_name) == 0)
1670 		return (EINVAL);
1671 
1672 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1673 	if (ap->a_uio == NULL)
1674 		return (EOPNOTSUPP);
1675 
1676 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1677 	    ap->a_cred, ap->a_td, IWRITE);
1678 	if (error) {
1679 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1680 			ip->i_ea_error = error;
1681 		return (error);
1682 	}
1683 
1684 	if (ip->i_ea_area == NULL) {
1685 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1686 		if (error)
1687 			return (error);
1688 		stand_alone = 1;
1689 	} else {
1690 		stand_alone = 0;
1691 	}
1692 
1693 	ealen = ap->a_uio->uio_resid;
1694 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1695 	eapad1 = 8 - (ealength % 8);
1696 	if (eapad1 == 8)
1697 		eapad1 = 0;
1698 	eapad2 = 8 - (ealen % 8);
1699 	if (eapad2 == 8)
1700 		eapad2 = 0;
1701 	ealength += eapad1 + ealen + eapad2;
1702 
1703 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1704 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1705 	easize = ip->i_ea_len;
1706 
1707 	olen = ffs_findextattr(eae, easize,
1708 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1709         if (olen == -1) {
1710 		/* new, append at end */
1711 		p = eae + easize;
1712 		easize += ealength;
1713 	} else {
1714 		bcopy(p, &ul, sizeof ul);
1715 		i = p - eae + ul;
1716 		if (ul != ealength) {
1717 			bcopy(p + ul, p + ealength, easize - i);
1718 			easize += (ealength - ul);
1719 		}
1720 	}
1721 	if (easize > NXADDR * fs->fs_bsize) {
1722 		free(eae, M_TEMP);
1723 		if (stand_alone)
1724 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1725 		else if (ip->i_ea_error == 0)
1726 			ip->i_ea_error = ENOSPC;
1727 		return(ENOSPC);
1728 	}
1729 	bcopy(&ealength, p, sizeof(ealength));
1730 	p += sizeof(ealength);
1731 	*p++ = ap->a_attrnamespace;
1732 	*p++ = eapad2;
1733 	*p++ = strlen(ap->a_name);
1734 	strcpy(p, ap->a_name);
1735 	p += strlen(ap->a_name);
1736 	bzero(p, eapad1);
1737 	p += eapad1;
1738 	error = uiomove(p, ealen, ap->a_uio);
1739 	if (error) {
1740 		free(eae, M_TEMP);
1741 		if (stand_alone)
1742 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1743 		else if (ip->i_ea_error == 0)
1744 			ip->i_ea_error = error;
1745 		return(error);
1746 	}
1747 	p += ealen;
1748 	bzero(p, eapad2);
1749 
1750 	p = ip->i_ea_area;
1751 	ip->i_ea_area = eae;
1752 	ip->i_ea_len = easize;
1753 	free(p, M_TEMP);
1754 	if (stand_alone)
1755 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1756 	return(error);
1757 }
1758