xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 69718b786d3943ea9a99eeeb5f5f6162f11c78b7)
1 /*-
2  * Copyright (c) 2002, 2003 Networks Associates Technology, Inc.
3  * All rights reserved.
4  *
5  * This software was developed for the FreeBSD Project by Marshall
6  * Kirk McKusick and Network Associates Laboratories, the Security
7  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
8  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
9  * research program
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * Copyright (c) 1982, 1986, 1989, 1993
33  *	The Regents of the University of California.  All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 4. Neither the name of the University nor the names of its contributors
44  *    may be used to endorse or promote products derived from this software
45  *    without specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
57  * SUCH DAMAGE.
58  *
59  *	from: @(#)ufs_readwrite.c	8.11 (Berkeley) 5/8/95
60  * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
61  *	@(#)ffs_vnops.c	8.15 (Berkeley) 5/14/95
62  */
63 
64 #include <sys/cdefs.h>
65 __FBSDID("$FreeBSD$");
66 
67 #include <sys/param.h>
68 #include <sys/bio.h>
69 #include <sys/systm.h>
70 #include <sys/buf.h>
71 #include <sys/conf.h>
72 #include <sys/extattr.h>
73 #include <sys/kernel.h>
74 #include <sys/limits.h>
75 #include <sys/malloc.h>
76 #include <sys/mount.h>
77 #include <sys/priv.h>
78 #include <sys/rwlock.h>
79 #include <sys/stat.h>
80 #include <sys/sysctl.h>
81 #include <sys/vmmeter.h>
82 #include <sys/vnode.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_param.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pager.h>
90 #include <vm/vnode_pager.h>
91 
92 #include <ufs/ufs/extattr.h>
93 #include <ufs/ufs/quota.h>
94 #include <ufs/ufs/inode.h>
95 #include <ufs/ufs/ufs_extern.h>
96 #include <ufs/ufs/ufsmount.h>
97 
98 #include <ufs/ffs/fs.h>
99 #include <ufs/ffs/ffs_extern.h>
100 #include "opt_directio.h"
101 #include "opt_ffs.h"
102 
103 #ifdef DIRECTIO
104 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
105 #endif
106 static vop_fdatasync_t	ffs_fdatasync;
107 static vop_fsync_t	ffs_fsync;
108 static vop_getpages_t	ffs_getpages;
109 static vop_lock1_t	ffs_lock;
110 static vop_read_t	ffs_read;
111 static vop_write_t	ffs_write;
112 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
113 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
114 		    struct ucred *cred);
115 static vop_strategy_t	ffsext_strategy;
116 static vop_closeextattr_t	ffs_closeextattr;
117 static vop_deleteextattr_t	ffs_deleteextattr;
118 static vop_getextattr_t	ffs_getextattr;
119 static vop_listextattr_t	ffs_listextattr;
120 static vop_openextattr_t	ffs_openextattr;
121 static vop_setextattr_t	ffs_setextattr;
122 static vop_vptofh_t	ffs_vptofh;
123 
124 /* Global vfs data structures for ufs. */
125 struct vop_vector ffs_vnodeops1 = {
126 	.vop_default =		&ufs_vnodeops,
127 	.vop_fsync =		ffs_fsync,
128 	.vop_fdatasync =	ffs_fdatasync,
129 	.vop_getpages =		ffs_getpages,
130 	.vop_getpages_async =	vnode_pager_local_getpages_async,
131 	.vop_lock1 =		ffs_lock,
132 	.vop_read =		ffs_read,
133 	.vop_reallocblks =	ffs_reallocblks,
134 	.vop_write =		ffs_write,
135 	.vop_vptofh =		ffs_vptofh,
136 };
137 
138 struct vop_vector ffs_fifoops1 = {
139 	.vop_default =		&ufs_fifoops,
140 	.vop_fsync =		ffs_fsync,
141 	.vop_fdatasync =	ffs_fdatasync,
142 	.vop_reallocblks =	ffs_reallocblks, /* XXX: really ??? */
143 	.vop_vptofh =		ffs_vptofh,
144 };
145 
146 /* Global vfs data structures for ufs. */
147 struct vop_vector ffs_vnodeops2 = {
148 	.vop_default =		&ufs_vnodeops,
149 	.vop_fsync =		ffs_fsync,
150 	.vop_fdatasync =	ffs_fdatasync,
151 	.vop_getpages =		ffs_getpages,
152 	.vop_getpages_async =	vnode_pager_local_getpages_async,
153 	.vop_lock1 =		ffs_lock,
154 	.vop_read =		ffs_read,
155 	.vop_reallocblks =	ffs_reallocblks,
156 	.vop_write =		ffs_write,
157 	.vop_closeextattr =	ffs_closeextattr,
158 	.vop_deleteextattr =	ffs_deleteextattr,
159 	.vop_getextattr =	ffs_getextattr,
160 	.vop_listextattr =	ffs_listextattr,
161 	.vop_openextattr =	ffs_openextattr,
162 	.vop_setextattr =	ffs_setextattr,
163 	.vop_vptofh =		ffs_vptofh,
164 };
165 
166 struct vop_vector ffs_fifoops2 = {
167 	.vop_default =		&ufs_fifoops,
168 	.vop_fsync =		ffs_fsync,
169 	.vop_fdatasync =	ffs_fdatasync,
170 	.vop_lock1 =		ffs_lock,
171 	.vop_reallocblks =	ffs_reallocblks,
172 	.vop_strategy =		ffsext_strategy,
173 	.vop_closeextattr =	ffs_closeextattr,
174 	.vop_deleteextattr =	ffs_deleteextattr,
175 	.vop_getextattr =	ffs_getextattr,
176 	.vop_listextattr =	ffs_listextattr,
177 	.vop_openextattr =	ffs_openextattr,
178 	.vop_setextattr =	ffs_setextattr,
179 	.vop_vptofh =		ffs_vptofh,
180 };
181 
182 /*
183  * Synch an open file.
184  */
185 /* ARGSUSED */
186 static int
187 ffs_fsync(struct vop_fsync_args *ap)
188 {
189 	struct vnode *vp;
190 	struct bufobj *bo;
191 	int error;
192 
193 	vp = ap->a_vp;
194 	bo = &vp->v_bufobj;
195 retry:
196 	error = ffs_syncvnode(vp, ap->a_waitfor, 0);
197 	if (error)
198 		return (error);
199 	if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
200 		error = softdep_fsync(vp);
201 		if (error)
202 			return (error);
203 
204 		/*
205 		 * The softdep_fsync() function may drop vp lock,
206 		 * allowing for dirty buffers to reappear on the
207 		 * bo_dirty list. Recheck and resync as needed.
208 		 */
209 		BO_LOCK(bo);
210 		if ((vp->v_type == VREG || vp->v_type == VDIR) &&
211 		    (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
212 			BO_UNLOCK(bo);
213 			goto retry;
214 		}
215 		BO_UNLOCK(bo);
216 	}
217 	return (0);
218 }
219 
220 int
221 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
222 {
223 	struct inode *ip;
224 	struct bufobj *bo;
225 	struct buf *bp, *nbp;
226 	ufs_lbn_t lbn;
227 	int error, passes;
228 	bool still_dirty, wait;
229 
230 	ip = VTOI(vp);
231 	ip->i_flag &= ~IN_NEEDSYNC;
232 	bo = &vp->v_bufobj;
233 
234 	/*
235 	 * When doing MNT_WAIT we must first flush all dependencies
236 	 * on the inode.
237 	 */
238 	if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
239 	    (error = softdep_sync_metadata(vp)) != 0)
240 		return (error);
241 
242 	/*
243 	 * Flush all dirty buffers associated with a vnode.
244 	 */
245 	error = 0;
246 	passes = 0;
247 	wait = false;	/* Always do an async pass first. */
248 	lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1));
249 	BO_LOCK(bo);
250 loop:
251 	TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
252 		bp->b_vflags &= ~BV_SCANNED;
253 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
254 		/*
255 		 * Reasons to skip this buffer: it has already been considered
256 		 * on this pass, the buffer has dependencies that will cause
257 		 * it to be redirtied and it has not already been deferred,
258 		 * or it is already being written.
259 		 */
260 		if ((bp->b_vflags & BV_SCANNED) != 0)
261 			continue;
262 		bp->b_vflags |= BV_SCANNED;
263 		/*
264 		 * Flush indirects in order, if requested.
265 		 *
266 		 * Note that if only datasync is requested, we can
267 		 * skip indirect blocks when softupdates are not
268 		 * active.  Otherwise we must flush them with data,
269 		 * since dependencies prevent data block writes.
270 		 */
271 		if (waitfor == MNT_WAIT && bp->b_lblkno <= -NDADDR &&
272 		    (lbn_level(bp->b_lblkno) >= passes ||
273 		    ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp))))
274 			continue;
275 		if (bp->b_lblkno > lbn)
276 			panic("ffs_syncvnode: syncing truncated data.");
277 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) {
278 			BO_UNLOCK(bo);
279 		} else if (wait) {
280 			if (BUF_LOCK(bp,
281 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
282 			    BO_LOCKPTR(bo)) != 0) {
283 				bp->b_vflags &= ~BV_SCANNED;
284 				goto next;
285 			}
286 		} else
287 			continue;
288 		if ((bp->b_flags & B_DELWRI) == 0)
289 			panic("ffs_fsync: not dirty");
290 		/*
291 		 * Check for dependencies and potentially complete them.
292 		 */
293 		if (!LIST_EMPTY(&bp->b_dep) &&
294 		    (error = softdep_sync_buf(vp, bp,
295 		    wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
296 			/* I/O error. */
297 			if (error != EBUSY) {
298 				BUF_UNLOCK(bp);
299 				return (error);
300 			}
301 			/* If we deferred once, don't defer again. */
302 		    	if ((bp->b_flags & B_DEFERRED) == 0) {
303 				bp->b_flags |= B_DEFERRED;
304 				BUF_UNLOCK(bp);
305 				goto next;
306 			}
307 		}
308 		if (wait) {
309 			bremfree(bp);
310 			if ((error = bwrite(bp)) != 0)
311 				return (error);
312 		} else if ((bp->b_flags & B_CLUSTEROK)) {
313 			(void) vfs_bio_awrite(bp);
314 		} else {
315 			bremfree(bp);
316 			(void) bawrite(bp);
317 		}
318 next:
319 		/*
320 		 * Since we may have slept during the I/O, we need
321 		 * to start from a known point.
322 		 */
323 		BO_LOCK(bo);
324 		nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
325 	}
326 	if (waitfor != MNT_WAIT) {
327 		BO_UNLOCK(bo);
328 		if ((flags & NO_INO_UPDT) != 0)
329 			return (0);
330 		else
331 			return (ffs_update(vp, 0));
332 	}
333 	/* Drain IO to see if we're done. */
334 	bufobj_wwait(bo, 0, 0);
335 	/*
336 	 * Block devices associated with filesystems may have new I/O
337 	 * requests posted for them even if the vnode is locked, so no
338 	 * amount of trying will get them clean.  We make several passes
339 	 * as a best effort.
340 	 *
341 	 * Regular files may need multiple passes to flush all dependency
342 	 * work as it is possible that we must write once per indirect
343 	 * level, once for the leaf, and once for the inode and each of
344 	 * these will be done with one sync and one async pass.
345 	 */
346 	if (bo->bo_dirty.bv_cnt > 0) {
347 		if ((flags & DATA_ONLY) == 0) {
348 			still_dirty = true;
349 		} else {
350 			/*
351 			 * For data-only sync, dirty indirect buffers
352 			 * are ignored.
353 			 */
354 			still_dirty = false;
355 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
356 				if (bp->b_lblkno > -NDADDR) {
357 					still_dirty = true;
358 					break;
359 				}
360 			}
361 		}
362 
363 		if (still_dirty) {
364 			/* Write the inode after sync passes to flush deps. */
365 			if (wait && DOINGSOFTDEP(vp) &&
366 			    (flags & NO_INO_UPDT) == 0) {
367 				BO_UNLOCK(bo);
368 				ffs_update(vp, 1);
369 				BO_LOCK(bo);
370 			}
371 			/* switch between sync/async. */
372 			wait = !wait;
373 			if (wait || ++passes < NIADDR + 2)
374 				goto loop;
375 #ifdef INVARIANTS
376 			if (!vn_isdisk(vp, NULL))
377 				vn_printf(vp, "ffs_fsync: dirty ");
378 #endif
379 		}
380 	}
381 	BO_UNLOCK(bo);
382 	error = 0;
383 	if ((flags & DATA_ONLY) == 0) {
384 		if ((flags & NO_INO_UPDT) == 0)
385 			error = ffs_update(vp, 1);
386 		if (DOINGSUJ(vp))
387 			softdep_journal_fsync(VTOI(vp));
388 	}
389 	return (error);
390 }
391 
392 static int
393 ffs_fdatasync(struct vop_fdatasync_args *ap)
394 {
395 
396 	return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY));
397 }
398 
399 static int
400 ffs_lock(ap)
401 	struct vop_lock1_args /* {
402 		struct vnode *a_vp;
403 		int a_flags;
404 		struct thread *a_td;
405 		char *file;
406 		int line;
407 	} */ *ap;
408 {
409 #ifndef NO_FFS_SNAPSHOT
410 	struct vnode *vp;
411 	int flags;
412 	struct lock *lkp;
413 	int result;
414 
415 	switch (ap->a_flags & LK_TYPE_MASK) {
416 	case LK_SHARED:
417 	case LK_UPGRADE:
418 	case LK_EXCLUSIVE:
419 		vp = ap->a_vp;
420 		flags = ap->a_flags;
421 		for (;;) {
422 #ifdef DEBUG_VFS_LOCKS
423 			KASSERT(vp->v_holdcnt != 0,
424 			    ("ffs_lock %p: zero hold count", vp));
425 #endif
426 			lkp = vp->v_vnlock;
427 			result = _lockmgr_args(lkp, flags, VI_MTX(vp),
428 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
429 			    ap->a_file, ap->a_line);
430 			if (lkp == vp->v_vnlock || result != 0)
431 				break;
432 			/*
433 			 * Apparent success, except that the vnode
434 			 * mutated between snapshot file vnode and
435 			 * regular file vnode while this process
436 			 * slept.  The lock currently held is not the
437 			 * right lock.  Release it, and try to get the
438 			 * new lock.
439 			 */
440 			(void) _lockmgr_args(lkp, LK_RELEASE, NULL,
441 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
442 			    ap->a_file, ap->a_line);
443 			if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
444 			    (LK_INTERLOCK | LK_NOWAIT))
445 				return (EBUSY);
446 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
447 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
448 			flags &= ~LK_INTERLOCK;
449 		}
450 		break;
451 	default:
452 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
453 	}
454 	return (result);
455 #else
456 	return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
457 #endif
458 }
459 
460 /*
461  * Vnode op for reading.
462  */
463 static int
464 ffs_read(ap)
465 	struct vop_read_args /* {
466 		struct vnode *a_vp;
467 		struct uio *a_uio;
468 		int a_ioflag;
469 		struct ucred *a_cred;
470 	} */ *ap;
471 {
472 	struct vnode *vp;
473 	struct inode *ip;
474 	struct uio *uio;
475 	struct fs *fs;
476 	struct buf *bp;
477 	ufs_lbn_t lbn, nextlbn;
478 	off_t bytesinfile;
479 	long size, xfersize, blkoffset;
480 	ssize_t orig_resid;
481 	int error;
482 	int seqcount;
483 	int ioflag;
484 
485 	vp = ap->a_vp;
486 	uio = ap->a_uio;
487 	ioflag = ap->a_ioflag;
488 	if (ap->a_ioflag & IO_EXT)
489 #ifdef notyet
490 		return (ffs_extread(vp, uio, ioflag));
491 #else
492 		panic("ffs_read+IO_EXT");
493 #endif
494 #ifdef DIRECTIO
495 	if ((ioflag & IO_DIRECT) != 0) {
496 		int workdone;
497 
498 		error = ffs_rawread(vp, uio, &workdone);
499 		if (error != 0 || workdone != 0)
500 			return error;
501 	}
502 #endif
503 
504 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
505 	ip = VTOI(vp);
506 
507 #ifdef INVARIANTS
508 	if (uio->uio_rw != UIO_READ)
509 		panic("ffs_read: mode");
510 
511 	if (vp->v_type == VLNK) {
512 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
513 			panic("ffs_read: short symlink");
514 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
515 		panic("ffs_read: type %d",  vp->v_type);
516 #endif
517 	orig_resid = uio->uio_resid;
518 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
519 	if (orig_resid == 0)
520 		return (0);
521 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
522 	fs = ITOFS(ip);
523 	if (uio->uio_offset < ip->i_size &&
524 	    uio->uio_offset >= fs->fs_maxfilesize)
525 		return (EOVERFLOW);
526 
527 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
528 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
529 			break;
530 		lbn = lblkno(fs, uio->uio_offset);
531 		nextlbn = lbn + 1;
532 
533 		/*
534 		 * size of buffer.  The buffer representing the
535 		 * end of the file is rounded up to the size of
536 		 * the block type ( fragment or full block,
537 		 * depending ).
538 		 */
539 		size = blksize(fs, ip, lbn);
540 		blkoffset = blkoff(fs, uio->uio_offset);
541 
542 		/*
543 		 * The amount we want to transfer in this iteration is
544 		 * one FS block less the amount of the data before
545 		 * our startpoint (duh!)
546 		 */
547 		xfersize = fs->fs_bsize - blkoffset;
548 
549 		/*
550 		 * But if we actually want less than the block,
551 		 * or the file doesn't have a whole block more of data,
552 		 * then use the lesser number.
553 		 */
554 		if (uio->uio_resid < xfersize)
555 			xfersize = uio->uio_resid;
556 		if (bytesinfile < xfersize)
557 			xfersize = bytesinfile;
558 
559 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
560 			/*
561 			 * Don't do readahead if this is the end of the file.
562 			 */
563 			error = bread_gb(vp, lbn, size, NOCRED,
564 			    GB_UNMAPPED, &bp);
565 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
566 			/*
567 			 * Otherwise if we are allowed to cluster,
568 			 * grab as much as we can.
569 			 *
570 			 * XXX  This may not be a win if we are not
571 			 * doing sequential access.
572 			 */
573 			error = cluster_read(vp, ip->i_size, lbn,
574 			    size, NOCRED, blkoffset + uio->uio_resid,
575 			    seqcount, GB_UNMAPPED, &bp);
576 		} else if (seqcount > 1) {
577 			/*
578 			 * If we are NOT allowed to cluster, then
579 			 * if we appear to be acting sequentially,
580 			 * fire off a request for a readahead
581 			 * as well as a read. Note that the 4th and 5th
582 			 * arguments point to arrays of the size specified in
583 			 * the 6th argument.
584 			 */
585 			u_int nextsize = blksize(fs, ip, nextlbn);
586 			error = breadn_flags(vp, lbn, size, &nextlbn,
587 			    &nextsize, 1, NOCRED, GB_UNMAPPED, &bp);
588 		} else {
589 			/*
590 			 * Failing all of the above, just read what the
591 			 * user asked for. Interestingly, the same as
592 			 * the first option above.
593 			 */
594 			error = bread_gb(vp, lbn, size, NOCRED,
595 			    GB_UNMAPPED, &bp);
596 		}
597 		if (error) {
598 			brelse(bp);
599 			bp = NULL;
600 			break;
601 		}
602 
603 		/*
604 		 * We should only get non-zero b_resid when an I/O error
605 		 * has occurred, which should cause us to break above.
606 		 * However, if the short read did not cause an error,
607 		 * then we want to ensure that we do not uiomove bad
608 		 * or uninitialized data.
609 		 */
610 		size -= bp->b_resid;
611 		if (size < xfersize) {
612 			if (size == 0)
613 				break;
614 			xfersize = size;
615 		}
616 
617 		if (buf_mapped(bp)) {
618 			error = vn_io_fault_uiomove((char *)bp->b_data +
619 			    blkoffset, (int)xfersize, uio);
620 		} else {
621 			error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
622 			    (int)xfersize, uio);
623 		}
624 		if (error)
625 			break;
626 
627 		vfs_bio_brelse(bp, ioflag);
628 	}
629 
630 	/*
631 	 * This can only happen in the case of an error
632 	 * because the loop above resets bp to NULL on each iteration
633 	 * and on normal completion has not set a new value into it.
634 	 * so it must have come from a 'break' statement
635 	 */
636 	if (bp != NULL)
637 		vfs_bio_brelse(bp, ioflag);
638 
639 	if ((error == 0 || uio->uio_resid != orig_resid) &&
640 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0 &&
641 	    (ip->i_flag & IN_ACCESS) == 0) {
642 		VI_LOCK(vp);
643 		ip->i_flag |= IN_ACCESS;
644 		VI_UNLOCK(vp);
645 	}
646 	return (error);
647 }
648 
649 /*
650  * Vnode op for writing.
651  */
652 static int
653 ffs_write(ap)
654 	struct vop_write_args /* {
655 		struct vnode *a_vp;
656 		struct uio *a_uio;
657 		int a_ioflag;
658 		struct ucred *a_cred;
659 	} */ *ap;
660 {
661 	struct vnode *vp;
662 	struct uio *uio;
663 	struct inode *ip;
664 	struct fs *fs;
665 	struct buf *bp;
666 	ufs_lbn_t lbn;
667 	off_t osize;
668 	ssize_t resid;
669 	int seqcount;
670 	int blkoffset, error, flags, ioflag, size, xfersize;
671 
672 	vp = ap->a_vp;
673 	uio = ap->a_uio;
674 	ioflag = ap->a_ioflag;
675 	if (ap->a_ioflag & IO_EXT)
676 #ifdef notyet
677 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
678 #else
679 		panic("ffs_write+IO_EXT");
680 #endif
681 
682 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
683 	ip = VTOI(vp);
684 
685 #ifdef INVARIANTS
686 	if (uio->uio_rw != UIO_WRITE)
687 		panic("ffs_write: mode");
688 #endif
689 
690 	switch (vp->v_type) {
691 	case VREG:
692 		if (ioflag & IO_APPEND)
693 			uio->uio_offset = ip->i_size;
694 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
695 			return (EPERM);
696 		/* FALLTHROUGH */
697 	case VLNK:
698 		break;
699 	case VDIR:
700 		panic("ffs_write: dir write");
701 		break;
702 	default:
703 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
704 			(int)uio->uio_offset,
705 			(int)uio->uio_resid
706 		);
707 	}
708 
709 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
710 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
711 	fs = ITOFS(ip);
712 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
713 		return (EFBIG);
714 	/*
715 	 * Maybe this should be above the vnode op call, but so long as
716 	 * file servers have no limits, I don't think it matters.
717 	 */
718 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
719 		return (EFBIG);
720 
721 	resid = uio->uio_resid;
722 	osize = ip->i_size;
723 	if (seqcount > BA_SEQMAX)
724 		flags = BA_SEQMAX << BA_SEQSHIFT;
725 	else
726 		flags = seqcount << BA_SEQSHIFT;
727 	if (ioflag & IO_SYNC)
728 		flags |= IO_SYNC;
729 	flags |= BA_UNMAPPED;
730 
731 	for (error = 0; uio->uio_resid > 0;) {
732 		lbn = lblkno(fs, uio->uio_offset);
733 		blkoffset = blkoff(fs, uio->uio_offset);
734 		xfersize = fs->fs_bsize - blkoffset;
735 		if (uio->uio_resid < xfersize)
736 			xfersize = uio->uio_resid;
737 		if (uio->uio_offset + xfersize > ip->i_size)
738 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
739 
740 		/*
741 		 * We must perform a read-before-write if the transfer size
742 		 * does not cover the entire buffer.
743 		 */
744 		if (fs->fs_bsize > xfersize)
745 			flags |= BA_CLRBUF;
746 		else
747 			flags &= ~BA_CLRBUF;
748 /* XXX is uio->uio_offset the right thing here? */
749 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
750 		    ap->a_cred, flags, &bp);
751 		if (error != 0) {
752 			vnode_pager_setsize(vp, ip->i_size);
753 			break;
754 		}
755 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
756 			bp->b_flags |= B_NOCACHE;
757 
758 		if (uio->uio_offset + xfersize > ip->i_size) {
759 			ip->i_size = uio->uio_offset + xfersize;
760 			DIP_SET(ip, i_size, ip->i_size);
761 		}
762 
763 		size = blksize(fs, ip, lbn) - bp->b_resid;
764 		if (size < xfersize)
765 			xfersize = size;
766 
767 		if (buf_mapped(bp)) {
768 			error = vn_io_fault_uiomove((char *)bp->b_data +
769 			    blkoffset, (int)xfersize, uio);
770 		} else {
771 			error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
772 			    (int)xfersize, uio);
773 		}
774 		/*
775 		 * If the buffer is not already filled and we encounter an
776 		 * error while trying to fill it, we have to clear out any
777 		 * garbage data from the pages instantiated for the buffer.
778 		 * If we do not, a failed uiomove() during a write can leave
779 		 * the prior contents of the pages exposed to a userland mmap.
780 		 *
781 		 * Note that we need only clear buffers with a transfer size
782 		 * equal to the block size because buffers with a shorter
783 		 * transfer size were cleared above by the call to UFS_BALLOC()
784 		 * with the BA_CLRBUF flag set.
785 		 *
786 		 * If the source region for uiomove identically mmaps the
787 		 * buffer, uiomove() performed the NOP copy, and the buffer
788 		 * content remains valid because the page fault handler
789 		 * validated the pages.
790 		 */
791 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
792 		    fs->fs_bsize == xfersize)
793 			vfs_bio_clrbuf(bp);
794 
795 		vfs_bio_set_flags(bp, ioflag);
796 
797 		/*
798 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
799 		 * if we have a severe page deficiency write the buffer
800 		 * asynchronously.  Otherwise try to cluster, and if that
801 		 * doesn't do it then either do an async write (if O_DIRECT),
802 		 * or a delayed write (if not).
803 		 */
804 		if (ioflag & IO_SYNC) {
805 			(void)bwrite(bp);
806 		} else if (vm_page_count_severe() ||
807 			    buf_dirty_count_severe() ||
808 			    (ioflag & IO_ASYNC)) {
809 			bp->b_flags |= B_CLUSTEROK;
810 			bawrite(bp);
811 		} else if (xfersize + blkoffset == fs->fs_bsize) {
812 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
813 				bp->b_flags |= B_CLUSTEROK;
814 				cluster_write(vp, bp, ip->i_size, seqcount,
815 				    GB_UNMAPPED);
816 			} else {
817 				bawrite(bp);
818 			}
819 		} else if (ioflag & IO_DIRECT) {
820 			bp->b_flags |= B_CLUSTEROK;
821 			bawrite(bp);
822 		} else {
823 			bp->b_flags |= B_CLUSTEROK;
824 			bdwrite(bp);
825 		}
826 		if (error || xfersize == 0)
827 			break;
828 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
829 	}
830 	/*
831 	 * If we successfully wrote any data, and we are not the superuser
832 	 * we clear the setuid and setgid bits as a precaution against
833 	 * tampering.
834 	 */
835 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
836 	    ap->a_cred) {
837 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
838 			ip->i_mode &= ~(ISUID | ISGID);
839 			DIP_SET(ip, i_mode, ip->i_mode);
840 		}
841 	}
842 	if (error) {
843 		if (ioflag & IO_UNIT) {
844 			(void)ffs_truncate(vp, osize,
845 			    IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
846 			uio->uio_offset -= resid - uio->uio_resid;
847 			uio->uio_resid = resid;
848 		}
849 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
850 		error = ffs_update(vp, 1);
851 	return (error);
852 }
853 
854 /*
855  * Extended attribute area reading.
856  */
857 static int
858 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
859 {
860 	struct inode *ip;
861 	struct ufs2_dinode *dp;
862 	struct fs *fs;
863 	struct buf *bp;
864 	ufs_lbn_t lbn, nextlbn;
865 	off_t bytesinfile;
866 	long size, xfersize, blkoffset;
867 	ssize_t orig_resid;
868 	int error;
869 
870 	ip = VTOI(vp);
871 	fs = ITOFS(ip);
872 	dp = ip->i_din2;
873 
874 #ifdef INVARIANTS
875 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
876 		panic("ffs_extread: mode");
877 
878 #endif
879 	orig_resid = uio->uio_resid;
880 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
881 	if (orig_resid == 0)
882 		return (0);
883 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
884 
885 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
886 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
887 			break;
888 		lbn = lblkno(fs, uio->uio_offset);
889 		nextlbn = lbn + 1;
890 
891 		/*
892 		 * size of buffer.  The buffer representing the
893 		 * end of the file is rounded up to the size of
894 		 * the block type ( fragment or full block,
895 		 * depending ).
896 		 */
897 		size = sblksize(fs, dp->di_extsize, lbn);
898 		blkoffset = blkoff(fs, uio->uio_offset);
899 
900 		/*
901 		 * The amount we want to transfer in this iteration is
902 		 * one FS block less the amount of the data before
903 		 * our startpoint (duh!)
904 		 */
905 		xfersize = fs->fs_bsize - blkoffset;
906 
907 		/*
908 		 * But if we actually want less than the block,
909 		 * or the file doesn't have a whole block more of data,
910 		 * then use the lesser number.
911 		 */
912 		if (uio->uio_resid < xfersize)
913 			xfersize = uio->uio_resid;
914 		if (bytesinfile < xfersize)
915 			xfersize = bytesinfile;
916 
917 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
918 			/*
919 			 * Don't do readahead if this is the end of the info.
920 			 */
921 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
922 		} else {
923 			/*
924 			 * If we have a second block, then
925 			 * fire off a request for a readahead
926 			 * as well as a read. Note that the 4th and 5th
927 			 * arguments point to arrays of the size specified in
928 			 * the 6th argument.
929 			 */
930 			u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
931 
932 			nextlbn = -1 - nextlbn;
933 			error = breadn(vp, -1 - lbn,
934 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
935 		}
936 		if (error) {
937 			brelse(bp);
938 			bp = NULL;
939 			break;
940 		}
941 
942 		/*
943 		 * We should only get non-zero b_resid when an I/O error
944 		 * has occurred, which should cause us to break above.
945 		 * However, if the short read did not cause an error,
946 		 * then we want to ensure that we do not uiomove bad
947 		 * or uninitialized data.
948 		 */
949 		size -= bp->b_resid;
950 		if (size < xfersize) {
951 			if (size == 0)
952 				break;
953 			xfersize = size;
954 		}
955 
956 		error = uiomove((char *)bp->b_data + blkoffset,
957 					(int)xfersize, uio);
958 		if (error)
959 			break;
960 		vfs_bio_brelse(bp, ioflag);
961 	}
962 
963 	/*
964 	 * This can only happen in the case of an error
965 	 * because the loop above resets bp to NULL on each iteration
966 	 * and on normal completion has not set a new value into it.
967 	 * so it must have come from a 'break' statement
968 	 */
969 	if (bp != NULL)
970 		vfs_bio_brelse(bp, ioflag);
971 	return (error);
972 }
973 
974 /*
975  * Extended attribute area writing.
976  */
977 static int
978 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
979 {
980 	struct inode *ip;
981 	struct ufs2_dinode *dp;
982 	struct fs *fs;
983 	struct buf *bp;
984 	ufs_lbn_t lbn;
985 	off_t osize;
986 	ssize_t resid;
987 	int blkoffset, error, flags, size, xfersize;
988 
989 	ip = VTOI(vp);
990 	fs = ITOFS(ip);
991 	dp = ip->i_din2;
992 
993 #ifdef INVARIANTS
994 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
995 		panic("ffs_extwrite: mode");
996 #endif
997 
998 	if (ioflag & IO_APPEND)
999 		uio->uio_offset = dp->di_extsize;
1000 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1001 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1002 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1003 		return (EFBIG);
1004 
1005 	resid = uio->uio_resid;
1006 	osize = dp->di_extsize;
1007 	flags = IO_EXT;
1008 	if (ioflag & IO_SYNC)
1009 		flags |= IO_SYNC;
1010 
1011 	for (error = 0; uio->uio_resid > 0;) {
1012 		lbn = lblkno(fs, uio->uio_offset);
1013 		blkoffset = blkoff(fs, uio->uio_offset);
1014 		xfersize = fs->fs_bsize - blkoffset;
1015 		if (uio->uio_resid < xfersize)
1016 			xfersize = uio->uio_resid;
1017 
1018 		/*
1019 		 * We must perform a read-before-write if the transfer size
1020 		 * does not cover the entire buffer.
1021 		 */
1022 		if (fs->fs_bsize > xfersize)
1023 			flags |= BA_CLRBUF;
1024 		else
1025 			flags &= ~BA_CLRBUF;
1026 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1027 		    ucred, flags, &bp);
1028 		if (error != 0)
1029 			break;
1030 		/*
1031 		 * If the buffer is not valid we have to clear out any
1032 		 * garbage data from the pages instantiated for the buffer.
1033 		 * If we do not, a failed uiomove() during a write can leave
1034 		 * the prior contents of the pages exposed to a userland
1035 		 * mmap().  XXX deal with uiomove() errors a better way.
1036 		 */
1037 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1038 			vfs_bio_clrbuf(bp);
1039 
1040 		if (uio->uio_offset + xfersize > dp->di_extsize)
1041 			dp->di_extsize = uio->uio_offset + xfersize;
1042 
1043 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1044 		if (size < xfersize)
1045 			xfersize = size;
1046 
1047 		error =
1048 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1049 
1050 		vfs_bio_set_flags(bp, ioflag);
1051 
1052 		/*
1053 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1054 		 * if we have a severe page deficiency write the buffer
1055 		 * asynchronously.  Otherwise try to cluster, and if that
1056 		 * doesn't do it then either do an async write (if O_DIRECT),
1057 		 * or a delayed write (if not).
1058 		 */
1059 		if (ioflag & IO_SYNC) {
1060 			(void)bwrite(bp);
1061 		} else if (vm_page_count_severe() ||
1062 			    buf_dirty_count_severe() ||
1063 			    xfersize + blkoffset == fs->fs_bsize ||
1064 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1065 			bawrite(bp);
1066 		else
1067 			bdwrite(bp);
1068 		if (error || xfersize == 0)
1069 			break;
1070 		ip->i_flag |= IN_CHANGE;
1071 	}
1072 	/*
1073 	 * If we successfully wrote any data, and we are not the superuser
1074 	 * we clear the setuid and setgid bits as a precaution against
1075 	 * tampering.
1076 	 */
1077 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1078 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1079 			ip->i_mode &= ~(ISUID | ISGID);
1080 			dp->di_mode = ip->i_mode;
1081 		}
1082 	}
1083 	if (error) {
1084 		if (ioflag & IO_UNIT) {
1085 			(void)ffs_truncate(vp, osize,
1086 			    IO_EXT | (ioflag&IO_SYNC), ucred);
1087 			uio->uio_offset -= resid - uio->uio_resid;
1088 			uio->uio_resid = resid;
1089 		}
1090 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1091 		error = ffs_update(vp, 1);
1092 	return (error);
1093 }
1094 
1095 
1096 /*
1097  * Vnode operating to retrieve a named extended attribute.
1098  *
1099  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1100  * the length of the EA, and possibly the pointer to the entry and to the data.
1101  */
1102 static int
1103 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1104 {
1105 	u_char *p, *pe, *pn, *p0;
1106 	int eapad1, eapad2, ealength, ealen, nlen;
1107 	uint32_t ul;
1108 
1109 	pe = ptr + length;
1110 	nlen = strlen(name);
1111 
1112 	for (p = ptr; p < pe; p = pn) {
1113 		p0 = p;
1114 		bcopy(p, &ul, sizeof(ul));
1115 		pn = p + ul;
1116 		/* make sure this entry is complete */
1117 		if (pn > pe)
1118 			break;
1119 		p += sizeof(uint32_t);
1120 		if (*p != nspace)
1121 			continue;
1122 		p++;
1123 		eapad2 = *p++;
1124 		if (*p != nlen)
1125 			continue;
1126 		p++;
1127 		if (bcmp(p, name, nlen))
1128 			continue;
1129 		ealength = sizeof(uint32_t) + 3 + nlen;
1130 		eapad1 = 8 - (ealength % 8);
1131 		if (eapad1 == 8)
1132 			eapad1 = 0;
1133 		ealength += eapad1;
1134 		ealen = ul - ealength - eapad2;
1135 		p += nlen + eapad1;
1136 		if (eap != NULL)
1137 			*eap = p0;
1138 		if (eac != NULL)
1139 			*eac = p;
1140 		return (ealen);
1141 	}
1142 	return(-1);
1143 }
1144 
1145 static int
1146 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1147 {
1148 	struct inode *ip;
1149 	struct ufs2_dinode *dp;
1150 	struct fs *fs;
1151 	struct uio luio;
1152 	struct iovec liovec;
1153 	u_int easize;
1154 	int error;
1155 	u_char *eae;
1156 
1157 	ip = VTOI(vp);
1158 	fs = ITOFS(ip);
1159 	dp = ip->i_din2;
1160 	easize = dp->di_extsize;
1161 	if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1162 		return (EFBIG);
1163 
1164 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1165 
1166 	liovec.iov_base = eae;
1167 	liovec.iov_len = easize;
1168 	luio.uio_iov = &liovec;
1169 	luio.uio_iovcnt = 1;
1170 	luio.uio_offset = 0;
1171 	luio.uio_resid = easize;
1172 	luio.uio_segflg = UIO_SYSSPACE;
1173 	luio.uio_rw = UIO_READ;
1174 	luio.uio_td = td;
1175 
1176 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1177 	if (error) {
1178 		free(eae, M_TEMP);
1179 		return(error);
1180 	}
1181 	*p = eae;
1182 	return (0);
1183 }
1184 
1185 static void
1186 ffs_lock_ea(struct vnode *vp)
1187 {
1188 	struct inode *ip;
1189 
1190 	ip = VTOI(vp);
1191 	VI_LOCK(vp);
1192 	while (ip->i_flag & IN_EA_LOCKED) {
1193 		ip->i_flag |= IN_EA_LOCKWAIT;
1194 		msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1195 		    0);
1196 	}
1197 	ip->i_flag |= IN_EA_LOCKED;
1198 	VI_UNLOCK(vp);
1199 }
1200 
1201 static void
1202 ffs_unlock_ea(struct vnode *vp)
1203 {
1204 	struct inode *ip;
1205 
1206 	ip = VTOI(vp);
1207 	VI_LOCK(vp);
1208 	if (ip->i_flag & IN_EA_LOCKWAIT)
1209 		wakeup(&ip->i_ea_refs);
1210 	ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1211 	VI_UNLOCK(vp);
1212 }
1213 
1214 static int
1215 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1216 {
1217 	struct inode *ip;
1218 	struct ufs2_dinode *dp;
1219 	int error;
1220 
1221 	ip = VTOI(vp);
1222 
1223 	ffs_lock_ea(vp);
1224 	if (ip->i_ea_area != NULL) {
1225 		ip->i_ea_refs++;
1226 		ffs_unlock_ea(vp);
1227 		return (0);
1228 	}
1229 	dp = ip->i_din2;
1230 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1231 	if (error) {
1232 		ffs_unlock_ea(vp);
1233 		return (error);
1234 	}
1235 	ip->i_ea_len = dp->di_extsize;
1236 	ip->i_ea_error = 0;
1237 	ip->i_ea_refs++;
1238 	ffs_unlock_ea(vp);
1239 	return (0);
1240 }
1241 
1242 /*
1243  * Vnode extattr transaction commit/abort
1244  */
1245 static int
1246 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1247 {
1248 	struct inode *ip;
1249 	struct uio luio;
1250 	struct iovec liovec;
1251 	int error;
1252 	struct ufs2_dinode *dp;
1253 
1254 	ip = VTOI(vp);
1255 
1256 	ffs_lock_ea(vp);
1257 	if (ip->i_ea_area == NULL) {
1258 		ffs_unlock_ea(vp);
1259 		return (EINVAL);
1260 	}
1261 	dp = ip->i_din2;
1262 	error = ip->i_ea_error;
1263 	if (commit && error == 0) {
1264 		ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1265 		if (cred == NOCRED)
1266 			cred =  vp->v_mount->mnt_cred;
1267 		liovec.iov_base = ip->i_ea_area;
1268 		liovec.iov_len = ip->i_ea_len;
1269 		luio.uio_iov = &liovec;
1270 		luio.uio_iovcnt = 1;
1271 		luio.uio_offset = 0;
1272 		luio.uio_resid = ip->i_ea_len;
1273 		luio.uio_segflg = UIO_SYSSPACE;
1274 		luio.uio_rw = UIO_WRITE;
1275 		luio.uio_td = td;
1276 		/* XXX: I'm not happy about truncating to zero size */
1277 		if (ip->i_ea_len < dp->di_extsize)
1278 			error = ffs_truncate(vp, 0, IO_EXT, cred);
1279 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1280 	}
1281 	if (--ip->i_ea_refs == 0) {
1282 		free(ip->i_ea_area, M_TEMP);
1283 		ip->i_ea_area = NULL;
1284 		ip->i_ea_len = 0;
1285 		ip->i_ea_error = 0;
1286 	}
1287 	ffs_unlock_ea(vp);
1288 	return (error);
1289 }
1290 
1291 /*
1292  * Vnode extattr strategy routine for fifos.
1293  *
1294  * We need to check for a read or write of the external attributes.
1295  * Otherwise we just fall through and do the usual thing.
1296  */
1297 static int
1298 ffsext_strategy(struct vop_strategy_args *ap)
1299 /*
1300 struct vop_strategy_args {
1301 	struct vnodeop_desc *a_desc;
1302 	struct vnode *a_vp;
1303 	struct buf *a_bp;
1304 };
1305 */
1306 {
1307 	struct vnode *vp;
1308 	daddr_t lbn;
1309 
1310 	vp = ap->a_vp;
1311 	lbn = ap->a_bp->b_lblkno;
1312 	if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -NXADDR)
1313 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1314 	if (vp->v_type == VFIFO)
1315 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1316 	panic("spec nodes went here");
1317 }
1318 
1319 /*
1320  * Vnode extattr transaction commit/abort
1321  */
1322 static int
1323 ffs_openextattr(struct vop_openextattr_args *ap)
1324 /*
1325 struct vop_openextattr_args {
1326 	struct vnodeop_desc *a_desc;
1327 	struct vnode *a_vp;
1328 	IN struct ucred *a_cred;
1329 	IN struct thread *a_td;
1330 };
1331 */
1332 {
1333 
1334 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1335 		return (EOPNOTSUPP);
1336 
1337 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1338 }
1339 
1340 
1341 /*
1342  * Vnode extattr transaction commit/abort
1343  */
1344 static int
1345 ffs_closeextattr(struct vop_closeextattr_args *ap)
1346 /*
1347 struct vop_closeextattr_args {
1348 	struct vnodeop_desc *a_desc;
1349 	struct vnode *a_vp;
1350 	int a_commit;
1351 	IN struct ucred *a_cred;
1352 	IN struct thread *a_td;
1353 };
1354 */
1355 {
1356 
1357 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1358 		return (EOPNOTSUPP);
1359 
1360 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1361 		return (EROFS);
1362 
1363 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1364 }
1365 
1366 /*
1367  * Vnode operation to remove a named attribute.
1368  */
1369 static int
1370 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1371 /*
1372 vop_deleteextattr {
1373 	IN struct vnode *a_vp;
1374 	IN int a_attrnamespace;
1375 	IN const char *a_name;
1376 	IN struct ucred *a_cred;
1377 	IN struct thread *a_td;
1378 };
1379 */
1380 {
1381 	struct inode *ip;
1382 	struct fs *fs;
1383 	uint32_t ealength, ul;
1384 	int ealen, olen, eapad1, eapad2, error, i, easize;
1385 	u_char *eae, *p;
1386 
1387 	ip = VTOI(ap->a_vp);
1388 	fs = ITOFS(ip);
1389 
1390 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1391 		return (EOPNOTSUPP);
1392 
1393 	if (strlen(ap->a_name) == 0)
1394 		return (EINVAL);
1395 
1396 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1397 		return (EROFS);
1398 
1399 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1400 	    ap->a_cred, ap->a_td, VWRITE);
1401 	if (error) {
1402 
1403 		/*
1404 		 * ffs_lock_ea is not needed there, because the vnode
1405 		 * must be exclusively locked.
1406 		 */
1407 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1408 			ip->i_ea_error = error;
1409 		return (error);
1410 	}
1411 
1412 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1413 	if (error)
1414 		return (error);
1415 
1416 	ealength = eapad1 = ealen = eapad2 = 0;
1417 
1418 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1419 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1420 	easize = ip->i_ea_len;
1421 
1422 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1423 	    &p, NULL);
1424 	if (olen == -1) {
1425 		/* delete but nonexistent */
1426 		free(eae, M_TEMP);
1427 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1428 		return(ENOATTR);
1429 	}
1430 	bcopy(p, &ul, sizeof ul);
1431 	i = p - eae + ul;
1432 	if (ul != ealength) {
1433 		bcopy(p + ul, p + ealength, easize - i);
1434 		easize += (ealength - ul);
1435 	}
1436 	if (easize > NXADDR * fs->fs_bsize) {
1437 		free(eae, M_TEMP);
1438 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1439 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1440 			ip->i_ea_error = ENOSPC;
1441 		return(ENOSPC);
1442 	}
1443 	p = ip->i_ea_area;
1444 	ip->i_ea_area = eae;
1445 	ip->i_ea_len = easize;
1446 	free(p, M_TEMP);
1447 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1448 	return(error);
1449 }
1450 
1451 /*
1452  * Vnode operation to retrieve a named extended attribute.
1453  */
1454 static int
1455 ffs_getextattr(struct vop_getextattr_args *ap)
1456 /*
1457 vop_getextattr {
1458 	IN struct vnode *a_vp;
1459 	IN int a_attrnamespace;
1460 	IN const char *a_name;
1461 	INOUT struct uio *a_uio;
1462 	OUT size_t *a_size;
1463 	IN struct ucred *a_cred;
1464 	IN struct thread *a_td;
1465 };
1466 */
1467 {
1468 	struct inode *ip;
1469 	u_char *eae, *p;
1470 	unsigned easize;
1471 	int error, ealen;
1472 
1473 	ip = VTOI(ap->a_vp);
1474 
1475 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1476 		return (EOPNOTSUPP);
1477 
1478 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1479 	    ap->a_cred, ap->a_td, VREAD);
1480 	if (error)
1481 		return (error);
1482 
1483 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1484 	if (error)
1485 		return (error);
1486 
1487 	eae = ip->i_ea_area;
1488 	easize = ip->i_ea_len;
1489 
1490 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1491 	    NULL, &p);
1492 	if (ealen >= 0) {
1493 		error = 0;
1494 		if (ap->a_size != NULL)
1495 			*ap->a_size = ealen;
1496 		else if (ap->a_uio != NULL)
1497 			error = uiomove(p, ealen, ap->a_uio);
1498 	} else
1499 		error = ENOATTR;
1500 
1501 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1502 	return(error);
1503 }
1504 
1505 /*
1506  * Vnode operation to retrieve extended attributes on a vnode.
1507  */
1508 static int
1509 ffs_listextattr(struct vop_listextattr_args *ap)
1510 /*
1511 vop_listextattr {
1512 	IN struct vnode *a_vp;
1513 	IN int a_attrnamespace;
1514 	INOUT struct uio *a_uio;
1515 	OUT size_t *a_size;
1516 	IN struct ucred *a_cred;
1517 	IN struct thread *a_td;
1518 };
1519 */
1520 {
1521 	struct inode *ip;
1522 	u_char *eae, *p, *pe, *pn;
1523 	unsigned easize;
1524 	uint32_t ul;
1525 	int error, ealen;
1526 
1527 	ip = VTOI(ap->a_vp);
1528 
1529 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1530 		return (EOPNOTSUPP);
1531 
1532 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1533 	    ap->a_cred, ap->a_td, VREAD);
1534 	if (error)
1535 		return (error);
1536 
1537 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1538 	if (error)
1539 		return (error);
1540 	eae = ip->i_ea_area;
1541 	easize = ip->i_ea_len;
1542 
1543 	error = 0;
1544 	if (ap->a_size != NULL)
1545 		*ap->a_size = 0;
1546 	pe = eae + easize;
1547 	for(p = eae; error == 0 && p < pe; p = pn) {
1548 		bcopy(p, &ul, sizeof(ul));
1549 		pn = p + ul;
1550 		if (pn > pe)
1551 			break;
1552 		p += sizeof(ul);
1553 		if (*p++ != ap->a_attrnamespace)
1554 			continue;
1555 		p++;	/* pad2 */
1556 		ealen = *p;
1557 		if (ap->a_size != NULL) {
1558 			*ap->a_size += ealen + 1;
1559 		} else if (ap->a_uio != NULL) {
1560 			error = uiomove(p, ealen + 1, ap->a_uio);
1561 		}
1562 	}
1563 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1564 	return(error);
1565 }
1566 
1567 /*
1568  * Vnode operation to set a named attribute.
1569  */
1570 static int
1571 ffs_setextattr(struct vop_setextattr_args *ap)
1572 /*
1573 vop_setextattr {
1574 	IN struct vnode *a_vp;
1575 	IN int a_attrnamespace;
1576 	IN const char *a_name;
1577 	INOUT struct uio *a_uio;
1578 	IN struct ucred *a_cred;
1579 	IN struct thread *a_td;
1580 };
1581 */
1582 {
1583 	struct inode *ip;
1584 	struct fs *fs;
1585 	uint32_t ealength, ul;
1586 	ssize_t ealen;
1587 	int olen, eapad1, eapad2, error, i, easize;
1588 	u_char *eae, *p;
1589 
1590 	ip = VTOI(ap->a_vp);
1591 	fs = ITOFS(ip);
1592 
1593 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1594 		return (EOPNOTSUPP);
1595 
1596 	if (strlen(ap->a_name) == 0)
1597 		return (EINVAL);
1598 
1599 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1600 	if (ap->a_uio == NULL)
1601 		return (EOPNOTSUPP);
1602 
1603 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1604 		return (EROFS);
1605 
1606 	ealen = ap->a_uio->uio_resid;
1607 	if (ealen < 0 || ealen > lblktosize(fs, NXADDR))
1608 		return (EINVAL);
1609 
1610 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1611 	    ap->a_cred, ap->a_td, VWRITE);
1612 	if (error) {
1613 
1614 		/*
1615 		 * ffs_lock_ea is not needed there, because the vnode
1616 		 * must be exclusively locked.
1617 		 */
1618 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1619 			ip->i_ea_error = error;
1620 		return (error);
1621 	}
1622 
1623 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1624 	if (error)
1625 		return (error);
1626 
1627 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1628 	eapad1 = 8 - (ealength % 8);
1629 	if (eapad1 == 8)
1630 		eapad1 = 0;
1631 	eapad2 = 8 - (ealen % 8);
1632 	if (eapad2 == 8)
1633 		eapad2 = 0;
1634 	ealength += eapad1 + ealen + eapad2;
1635 
1636 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1637 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1638 	easize = ip->i_ea_len;
1639 
1640 	olen = ffs_findextattr(eae, easize,
1641 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1642         if (olen == -1) {
1643 		/* new, append at end */
1644 		p = eae + easize;
1645 		easize += ealength;
1646 	} else {
1647 		bcopy(p, &ul, sizeof ul);
1648 		i = p - eae + ul;
1649 		if (ul != ealength) {
1650 			bcopy(p + ul, p + ealength, easize - i);
1651 			easize += (ealength - ul);
1652 		}
1653 	}
1654 	if (easize > lblktosize(fs, NXADDR)) {
1655 		free(eae, M_TEMP);
1656 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1657 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1658 			ip->i_ea_error = ENOSPC;
1659 		return(ENOSPC);
1660 	}
1661 	bcopy(&ealength, p, sizeof(ealength));
1662 	p += sizeof(ealength);
1663 	*p++ = ap->a_attrnamespace;
1664 	*p++ = eapad2;
1665 	*p++ = strlen(ap->a_name);
1666 	strcpy(p, ap->a_name);
1667 	p += strlen(ap->a_name);
1668 	bzero(p, eapad1);
1669 	p += eapad1;
1670 	error = uiomove(p, ealen, ap->a_uio);
1671 	if (error) {
1672 		free(eae, M_TEMP);
1673 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1674 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1675 			ip->i_ea_error = error;
1676 		return(error);
1677 	}
1678 	p += ealen;
1679 	bzero(p, eapad2);
1680 
1681 	p = ip->i_ea_area;
1682 	ip->i_ea_area = eae;
1683 	ip->i_ea_len = easize;
1684 	free(p, M_TEMP);
1685 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1686 	return(error);
1687 }
1688 
1689 /*
1690  * Vnode pointer to File handle
1691  */
1692 static int
1693 ffs_vptofh(struct vop_vptofh_args *ap)
1694 /*
1695 vop_vptofh {
1696 	IN struct vnode *a_vp;
1697 	IN struct fid *a_fhp;
1698 };
1699 */
1700 {
1701 	struct inode *ip;
1702 	struct ufid *ufhp;
1703 
1704 	ip = VTOI(ap->a_vp);
1705 	ufhp = (struct ufid *)ap->a_fhp;
1706 	ufhp->ufid_len = sizeof(struct ufid);
1707 	ufhp->ufid_ino = ip->i_number;
1708 	ufhp->ufid_gen = ip->i_gen;
1709 	return (0);
1710 }
1711 
1712 SYSCTL_DECL(_vfs_ffs);
1713 static int use_buf_pager = 1;
1714 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0,
1715     "Always use buffer pager instead of bmap");
1716 
1717 static daddr_t
1718 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1719 {
1720 
1721 	return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off));
1722 }
1723 
1724 static int
1725 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1726 {
1727 
1728 	return (blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn));
1729 }
1730 
1731 static int
1732 ffs_getpages(struct vop_getpages_args *ap)
1733 {
1734 	struct vnode *vp;
1735 	struct ufsmount *um;
1736 
1737 	vp = ap->a_vp;
1738 	um = VFSTOUFS(vp->v_mount);
1739 
1740 	if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE)
1741 		return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1742 		    ap->a_rbehind, ap->a_rahead, NULL, NULL));
1743 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1744 	    ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz));
1745 }
1746