xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision d0ba1baed3f6e4936a0c1b89c25f6c59168ef6de)
1 /*-
2  * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND BSD-3-Clause)
3  *
4  * Copyright (c) 2002, 2003 Networks Associates Technology, Inc.
5  * All rights reserved.
6  *
7  * This software was developed for the FreeBSD Project by Marshall
8  * Kirk McKusick and Network Associates Laboratories, the Security
9  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
10  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
11  * research program
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * Copyright (c) 1982, 1986, 1989, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	from: @(#)ufs_readwrite.c	8.11 (Berkeley) 5/8/95
62  * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
63  *	@(#)ffs_vnops.c	8.15 (Berkeley) 5/14/95
64  */
65 
66 #include <sys/cdefs.h>
67 __FBSDID("$FreeBSD$");
68 
69 #include <sys/param.h>
70 #include <sys/bio.h>
71 #include <sys/systm.h>
72 #include <sys/buf.h>
73 #include <sys/conf.h>
74 #include <sys/extattr.h>
75 #include <sys/kernel.h>
76 #include <sys/limits.h>
77 #include <sys/malloc.h>
78 #include <sys/mount.h>
79 #include <sys/priv.h>
80 #include <sys/rwlock.h>
81 #include <sys/stat.h>
82 #include <sys/sysctl.h>
83 #include <sys/vmmeter.h>
84 #include <sys/vnode.h>
85 
86 #include <vm/vm.h>
87 #include <vm/vm_param.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vnode_pager.h>
93 
94 #include <ufs/ufs/extattr.h>
95 #include <ufs/ufs/quota.h>
96 #include <ufs/ufs/inode.h>
97 #include <ufs/ufs/ufs_extern.h>
98 #include <ufs/ufs/ufsmount.h>
99 
100 #include <ufs/ffs/fs.h>
101 #include <ufs/ffs/ffs_extern.h>
102 #include "opt_directio.h"
103 #include "opt_ffs.h"
104 
105 #define	ALIGNED_TO(ptr, s)	\
106 	(((uintptr_t)(ptr) & (_Alignof(s) - 1)) == 0)
107 
108 #ifdef DIRECTIO
109 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
110 #endif
111 static vop_fdatasync_t	ffs_fdatasync;
112 static vop_fsync_t	ffs_fsync;
113 static vop_getpages_t	ffs_getpages;
114 static vop_lock1_t	ffs_lock;
115 static vop_read_t	ffs_read;
116 static vop_write_t	ffs_write;
117 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
118 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
119 		    struct ucred *cred);
120 static vop_strategy_t	ffsext_strategy;
121 static vop_closeextattr_t	ffs_closeextattr;
122 static vop_deleteextattr_t	ffs_deleteextattr;
123 static vop_getextattr_t	ffs_getextattr;
124 static vop_listextattr_t	ffs_listextattr;
125 static vop_openextattr_t	ffs_openextattr;
126 static vop_setextattr_t	ffs_setextattr;
127 static vop_vptofh_t	ffs_vptofh;
128 
129 /* Global vfs data structures for ufs. */
130 struct vop_vector ffs_vnodeops1 = {
131 	.vop_default =		&ufs_vnodeops,
132 	.vop_fsync =		ffs_fsync,
133 	.vop_fdatasync =	ffs_fdatasync,
134 	.vop_getpages =		ffs_getpages,
135 	.vop_getpages_async =	vnode_pager_local_getpages_async,
136 	.vop_lock1 =		ffs_lock,
137 	.vop_read =		ffs_read,
138 	.vop_reallocblks =	ffs_reallocblks,
139 	.vop_write =		ffs_write,
140 	.vop_vptofh =		ffs_vptofh,
141 };
142 
143 struct vop_vector ffs_fifoops1 = {
144 	.vop_default =		&ufs_fifoops,
145 	.vop_fsync =		ffs_fsync,
146 	.vop_fdatasync =	ffs_fdatasync,
147 	.vop_reallocblks =	ffs_reallocblks, /* XXX: really ??? */
148 	.vop_vptofh =		ffs_vptofh,
149 };
150 
151 /* Global vfs data structures for ufs. */
152 struct vop_vector ffs_vnodeops2 = {
153 	.vop_default =		&ufs_vnodeops,
154 	.vop_fsync =		ffs_fsync,
155 	.vop_fdatasync =	ffs_fdatasync,
156 	.vop_getpages =		ffs_getpages,
157 	.vop_getpages_async =	vnode_pager_local_getpages_async,
158 	.vop_lock1 =		ffs_lock,
159 	.vop_read =		ffs_read,
160 	.vop_reallocblks =	ffs_reallocblks,
161 	.vop_write =		ffs_write,
162 	.vop_closeextattr =	ffs_closeextattr,
163 	.vop_deleteextattr =	ffs_deleteextattr,
164 	.vop_getextattr =	ffs_getextattr,
165 	.vop_listextattr =	ffs_listextattr,
166 	.vop_openextattr =	ffs_openextattr,
167 	.vop_setextattr =	ffs_setextattr,
168 	.vop_vptofh =		ffs_vptofh,
169 };
170 
171 struct vop_vector ffs_fifoops2 = {
172 	.vop_default =		&ufs_fifoops,
173 	.vop_fsync =		ffs_fsync,
174 	.vop_fdatasync =	ffs_fdatasync,
175 	.vop_lock1 =		ffs_lock,
176 	.vop_reallocblks =	ffs_reallocblks,
177 	.vop_strategy =		ffsext_strategy,
178 	.vop_closeextattr =	ffs_closeextattr,
179 	.vop_deleteextattr =	ffs_deleteextattr,
180 	.vop_getextattr =	ffs_getextattr,
181 	.vop_listextattr =	ffs_listextattr,
182 	.vop_openextattr =	ffs_openextattr,
183 	.vop_setextattr =	ffs_setextattr,
184 	.vop_vptofh =		ffs_vptofh,
185 };
186 
187 /*
188  * Synch an open file.
189  */
190 /* ARGSUSED */
191 static int
192 ffs_fsync(struct vop_fsync_args *ap)
193 {
194 	struct vnode *vp;
195 	struct bufobj *bo;
196 	int error;
197 
198 	vp = ap->a_vp;
199 	bo = &vp->v_bufobj;
200 retry:
201 	error = ffs_syncvnode(vp, ap->a_waitfor, 0);
202 	if (error)
203 		return (error);
204 	if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
205 		error = softdep_fsync(vp);
206 		if (error)
207 			return (error);
208 
209 		/*
210 		 * The softdep_fsync() function may drop vp lock,
211 		 * allowing for dirty buffers to reappear on the
212 		 * bo_dirty list. Recheck and resync as needed.
213 		 */
214 		BO_LOCK(bo);
215 		if ((vp->v_type == VREG || vp->v_type == VDIR) &&
216 		    (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
217 			BO_UNLOCK(bo);
218 			goto retry;
219 		}
220 		BO_UNLOCK(bo);
221 	}
222 	return (0);
223 }
224 
225 int
226 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
227 {
228 	struct inode *ip;
229 	struct bufobj *bo;
230 	struct buf *bp, *nbp;
231 	ufs_lbn_t lbn;
232 	int error, passes;
233 	bool still_dirty, wait;
234 
235 	ip = VTOI(vp);
236 	ip->i_flag &= ~IN_NEEDSYNC;
237 	bo = &vp->v_bufobj;
238 
239 	/*
240 	 * When doing MNT_WAIT we must first flush all dependencies
241 	 * on the inode.
242 	 */
243 	if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
244 	    (error = softdep_sync_metadata(vp)) != 0)
245 		return (error);
246 
247 	/*
248 	 * Flush all dirty buffers associated with a vnode.
249 	 */
250 	error = 0;
251 	passes = 0;
252 	wait = false;	/* Always do an async pass first. */
253 	lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1));
254 	BO_LOCK(bo);
255 loop:
256 	TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
257 		bp->b_vflags &= ~BV_SCANNED;
258 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
259 		/*
260 		 * Reasons to skip this buffer: it has already been considered
261 		 * on this pass, the buffer has dependencies that will cause
262 		 * it to be redirtied and it has not already been deferred,
263 		 * or it is already being written.
264 		 */
265 		if ((bp->b_vflags & BV_SCANNED) != 0)
266 			continue;
267 		bp->b_vflags |= BV_SCANNED;
268 		/*
269 		 * Flush indirects in order, if requested.
270 		 *
271 		 * Note that if only datasync is requested, we can
272 		 * skip indirect blocks when softupdates are not
273 		 * active.  Otherwise we must flush them with data,
274 		 * since dependencies prevent data block writes.
275 		 */
276 		if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR &&
277 		    (lbn_level(bp->b_lblkno) >= passes ||
278 		    ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp))))
279 			continue;
280 		if (bp->b_lblkno > lbn)
281 			panic("ffs_syncvnode: syncing truncated data.");
282 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) {
283 			BO_UNLOCK(bo);
284 		} else if (wait) {
285 			if (BUF_LOCK(bp,
286 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
287 			    BO_LOCKPTR(bo)) != 0) {
288 				bp->b_vflags &= ~BV_SCANNED;
289 				goto next;
290 			}
291 		} else
292 			continue;
293 		if ((bp->b_flags & B_DELWRI) == 0)
294 			panic("ffs_fsync: not dirty");
295 		/*
296 		 * Check for dependencies and potentially complete them.
297 		 */
298 		if (!LIST_EMPTY(&bp->b_dep) &&
299 		    (error = softdep_sync_buf(vp, bp,
300 		    wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
301 			/* I/O error. */
302 			if (error != EBUSY) {
303 				BUF_UNLOCK(bp);
304 				return (error);
305 			}
306 			/* If we deferred once, don't defer again. */
307 		    	if ((bp->b_flags & B_DEFERRED) == 0) {
308 				bp->b_flags |= B_DEFERRED;
309 				BUF_UNLOCK(bp);
310 				goto next;
311 			}
312 		}
313 		if (wait) {
314 			bremfree(bp);
315 			if ((error = bwrite(bp)) != 0)
316 				return (error);
317 		} else if ((bp->b_flags & B_CLUSTEROK)) {
318 			(void) vfs_bio_awrite(bp);
319 		} else {
320 			bremfree(bp);
321 			(void) bawrite(bp);
322 		}
323 next:
324 		/*
325 		 * Since we may have slept during the I/O, we need
326 		 * to start from a known point.
327 		 */
328 		BO_LOCK(bo);
329 		nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
330 	}
331 	if (waitfor != MNT_WAIT) {
332 		BO_UNLOCK(bo);
333 		if ((flags & NO_INO_UPDT) != 0)
334 			return (0);
335 		else
336 			return (ffs_update(vp, 0));
337 	}
338 	/* Drain IO to see if we're done. */
339 	bufobj_wwait(bo, 0, 0);
340 	/*
341 	 * Block devices associated with filesystems may have new I/O
342 	 * requests posted for them even if the vnode is locked, so no
343 	 * amount of trying will get them clean.  We make several passes
344 	 * as a best effort.
345 	 *
346 	 * Regular files may need multiple passes to flush all dependency
347 	 * work as it is possible that we must write once per indirect
348 	 * level, once for the leaf, and once for the inode and each of
349 	 * these will be done with one sync and one async pass.
350 	 */
351 	if (bo->bo_dirty.bv_cnt > 0) {
352 		if ((flags & DATA_ONLY) == 0) {
353 			still_dirty = true;
354 		} else {
355 			/*
356 			 * For data-only sync, dirty indirect buffers
357 			 * are ignored.
358 			 */
359 			still_dirty = false;
360 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
361 				if (bp->b_lblkno > -UFS_NDADDR) {
362 					still_dirty = true;
363 					break;
364 				}
365 			}
366 		}
367 
368 		if (still_dirty) {
369 			/* Write the inode after sync passes to flush deps. */
370 			if (wait && DOINGSOFTDEP(vp) &&
371 			    (flags & NO_INO_UPDT) == 0) {
372 				BO_UNLOCK(bo);
373 				ffs_update(vp, 1);
374 				BO_LOCK(bo);
375 			}
376 			/* switch between sync/async. */
377 			wait = !wait;
378 			if (wait || ++passes < UFS_NIADDR + 2)
379 				goto loop;
380 #ifdef INVARIANTS
381 			if (!vn_isdisk(vp, NULL))
382 				vn_printf(vp, "ffs_fsync: dirty ");
383 #endif
384 		}
385 	}
386 	BO_UNLOCK(bo);
387 	error = 0;
388 	if ((flags & DATA_ONLY) == 0) {
389 		if ((flags & NO_INO_UPDT) == 0)
390 			error = ffs_update(vp, 1);
391 		if (DOINGSUJ(vp))
392 			softdep_journal_fsync(VTOI(vp));
393 	}
394 	return (error);
395 }
396 
397 static int
398 ffs_fdatasync(struct vop_fdatasync_args *ap)
399 {
400 
401 	return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY));
402 }
403 
404 static int
405 ffs_lock(ap)
406 	struct vop_lock1_args /* {
407 		struct vnode *a_vp;
408 		int a_flags;
409 		struct thread *a_td;
410 		char *file;
411 		int line;
412 	} */ *ap;
413 {
414 #ifndef NO_FFS_SNAPSHOT
415 	struct vnode *vp;
416 	int flags;
417 	struct lock *lkp;
418 	int result;
419 
420 	switch (ap->a_flags & LK_TYPE_MASK) {
421 	case LK_SHARED:
422 	case LK_UPGRADE:
423 	case LK_EXCLUSIVE:
424 		vp = ap->a_vp;
425 		flags = ap->a_flags;
426 		for (;;) {
427 #ifdef DEBUG_VFS_LOCKS
428 			KASSERT(vp->v_holdcnt != 0,
429 			    ("ffs_lock %p: zero hold count", vp));
430 #endif
431 			lkp = vp->v_vnlock;
432 			result = _lockmgr_args(lkp, flags, VI_MTX(vp),
433 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
434 			    ap->a_file, ap->a_line);
435 			if (lkp == vp->v_vnlock || result != 0)
436 				break;
437 			/*
438 			 * Apparent success, except that the vnode
439 			 * mutated between snapshot file vnode and
440 			 * regular file vnode while this process
441 			 * slept.  The lock currently held is not the
442 			 * right lock.  Release it, and try to get the
443 			 * new lock.
444 			 */
445 			(void) _lockmgr_args(lkp, LK_RELEASE, NULL,
446 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
447 			    ap->a_file, ap->a_line);
448 			if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
449 			    (LK_INTERLOCK | LK_NOWAIT))
450 				return (EBUSY);
451 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
452 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
453 			flags &= ~LK_INTERLOCK;
454 		}
455 		break;
456 	default:
457 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
458 	}
459 	return (result);
460 #else
461 	return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
462 #endif
463 }
464 
465 static int
466 ffs_read_hole(struct uio *uio, long xfersize, long *size)
467 {
468 	ssize_t saved_resid, tlen;
469 	int error;
470 
471 	while (xfersize > 0) {
472 		tlen = min(xfersize, ZERO_REGION_SIZE);
473 		saved_resid = uio->uio_resid;
474 		error = vn_io_fault_uiomove(__DECONST(void *, zero_region),
475 		    tlen, uio);
476 		if (error != 0)
477 			return (error);
478 		tlen = saved_resid - uio->uio_resid;
479 		xfersize -= tlen;
480 		*size -= tlen;
481 	}
482 	return (0);
483 }
484 
485 /*
486  * Vnode op for reading.
487  */
488 static int
489 ffs_read(ap)
490 	struct vop_read_args /* {
491 		struct vnode *a_vp;
492 		struct uio *a_uio;
493 		int a_ioflag;
494 		struct ucred *a_cred;
495 	} */ *ap;
496 {
497 	struct vnode *vp;
498 	struct inode *ip;
499 	struct uio *uio;
500 	struct fs *fs;
501 	struct buf *bp;
502 	ufs_lbn_t lbn, nextlbn;
503 	off_t bytesinfile;
504 	long size, xfersize, blkoffset;
505 	ssize_t orig_resid;
506 	int bflag, error, ioflag, seqcount;
507 
508 	vp = ap->a_vp;
509 	uio = ap->a_uio;
510 	ioflag = ap->a_ioflag;
511 	if (ap->a_ioflag & IO_EXT)
512 #ifdef notyet
513 		return (ffs_extread(vp, uio, ioflag));
514 #else
515 		panic("ffs_read+IO_EXT");
516 #endif
517 #ifdef DIRECTIO
518 	if ((ioflag & IO_DIRECT) != 0) {
519 		int workdone;
520 
521 		error = ffs_rawread(vp, uio, &workdone);
522 		if (error != 0 || workdone != 0)
523 			return error;
524 	}
525 #endif
526 
527 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
528 	ip = VTOI(vp);
529 
530 #ifdef INVARIANTS
531 	if (uio->uio_rw != UIO_READ)
532 		panic("ffs_read: mode");
533 
534 	if (vp->v_type == VLNK) {
535 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
536 			panic("ffs_read: short symlink");
537 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
538 		panic("ffs_read: type %d",  vp->v_type);
539 #endif
540 	orig_resid = uio->uio_resid;
541 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
542 	if (orig_resid == 0)
543 		return (0);
544 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
545 	fs = ITOFS(ip);
546 	if (uio->uio_offset < ip->i_size &&
547 	    uio->uio_offset >= fs->fs_maxfilesize)
548 		return (EOVERFLOW);
549 
550 	bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE);
551 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
552 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
553 			break;
554 		lbn = lblkno(fs, uio->uio_offset);
555 		nextlbn = lbn + 1;
556 
557 		/*
558 		 * size of buffer.  The buffer representing the
559 		 * end of the file is rounded up to the size of
560 		 * the block type ( fragment or full block,
561 		 * depending ).
562 		 */
563 		size = blksize(fs, ip, lbn);
564 		blkoffset = blkoff(fs, uio->uio_offset);
565 
566 		/*
567 		 * The amount we want to transfer in this iteration is
568 		 * one FS block less the amount of the data before
569 		 * our startpoint (duh!)
570 		 */
571 		xfersize = fs->fs_bsize - blkoffset;
572 
573 		/*
574 		 * But if we actually want less than the block,
575 		 * or the file doesn't have a whole block more of data,
576 		 * then use the lesser number.
577 		 */
578 		if (uio->uio_resid < xfersize)
579 			xfersize = uio->uio_resid;
580 		if (bytesinfile < xfersize)
581 			xfersize = bytesinfile;
582 
583 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
584 			/*
585 			 * Don't do readahead if this is the end of the file.
586 			 */
587 			error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
588 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
589 			/*
590 			 * Otherwise if we are allowed to cluster,
591 			 * grab as much as we can.
592 			 *
593 			 * XXX  This may not be a win if we are not
594 			 * doing sequential access.
595 			 */
596 			error = cluster_read(vp, ip->i_size, lbn,
597 			    size, NOCRED, blkoffset + uio->uio_resid,
598 			    seqcount, bflag, &bp);
599 		} else if (seqcount > 1) {
600 			/*
601 			 * If we are NOT allowed to cluster, then
602 			 * if we appear to be acting sequentially,
603 			 * fire off a request for a readahead
604 			 * as well as a read. Note that the 4th and 5th
605 			 * arguments point to arrays of the size specified in
606 			 * the 6th argument.
607 			 */
608 			u_int nextsize = blksize(fs, ip, nextlbn);
609 			error = breadn_flags(vp, lbn, size, &nextlbn,
610 			    &nextsize, 1, NOCRED, bflag, NULL, &bp);
611 		} else {
612 			/*
613 			 * Failing all of the above, just read what the
614 			 * user asked for. Interestingly, the same as
615 			 * the first option above.
616 			 */
617 			error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
618 		}
619 		if (error == EJUSTRETURN) {
620 			error = ffs_read_hole(uio, xfersize, &size);
621 			if (error == 0)
622 				continue;
623 		}
624 		if (error != 0) {
625 			brelse(bp);
626 			bp = NULL;
627 			break;
628 		}
629 
630 		/*
631 		 * We should only get non-zero b_resid when an I/O error
632 		 * has occurred, which should cause us to break above.
633 		 * However, if the short read did not cause an error,
634 		 * then we want to ensure that we do not uiomove bad
635 		 * or uninitialized data.
636 		 */
637 		size -= bp->b_resid;
638 		if (size < xfersize) {
639 			if (size == 0)
640 				break;
641 			xfersize = size;
642 		}
643 
644 		if (buf_mapped(bp)) {
645 			error = vn_io_fault_uiomove((char *)bp->b_data +
646 			    blkoffset, (int)xfersize, uio);
647 		} else {
648 			error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
649 			    (int)xfersize, uio);
650 		}
651 		if (error)
652 			break;
653 
654 		vfs_bio_brelse(bp, ioflag);
655 	}
656 
657 	/*
658 	 * This can only happen in the case of an error
659 	 * because the loop above resets bp to NULL on each iteration
660 	 * and on normal completion has not set a new value into it.
661 	 * so it must have come from a 'break' statement
662 	 */
663 	if (bp != NULL)
664 		vfs_bio_brelse(bp, ioflag);
665 
666 	if ((error == 0 || uio->uio_resid != orig_resid) &&
667 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0 &&
668 	    (ip->i_flag & IN_ACCESS) == 0) {
669 		VI_LOCK(vp);
670 		ip->i_flag |= IN_ACCESS;
671 		VI_UNLOCK(vp);
672 	}
673 	return (error);
674 }
675 
676 /*
677  * Vnode op for writing.
678  */
679 static int
680 ffs_write(ap)
681 	struct vop_write_args /* {
682 		struct vnode *a_vp;
683 		struct uio *a_uio;
684 		int a_ioflag;
685 		struct ucred *a_cred;
686 	} */ *ap;
687 {
688 	struct vnode *vp;
689 	struct uio *uio;
690 	struct inode *ip;
691 	struct fs *fs;
692 	struct buf *bp;
693 	ufs_lbn_t lbn;
694 	off_t osize;
695 	ssize_t resid;
696 	int seqcount;
697 	int blkoffset, error, flags, ioflag, size, xfersize;
698 
699 	vp = ap->a_vp;
700 	uio = ap->a_uio;
701 	ioflag = ap->a_ioflag;
702 	if (ap->a_ioflag & IO_EXT)
703 #ifdef notyet
704 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
705 #else
706 		panic("ffs_write+IO_EXT");
707 #endif
708 
709 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
710 	ip = VTOI(vp);
711 
712 #ifdef INVARIANTS
713 	if (uio->uio_rw != UIO_WRITE)
714 		panic("ffs_write: mode");
715 #endif
716 
717 	switch (vp->v_type) {
718 	case VREG:
719 		if (ioflag & IO_APPEND)
720 			uio->uio_offset = ip->i_size;
721 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
722 			return (EPERM);
723 		/* FALLTHROUGH */
724 	case VLNK:
725 		break;
726 	case VDIR:
727 		panic("ffs_write: dir write");
728 		break;
729 	default:
730 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
731 			(int)uio->uio_offset,
732 			(int)uio->uio_resid
733 		);
734 	}
735 
736 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
737 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
738 	fs = ITOFS(ip);
739 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
740 		return (EFBIG);
741 	/*
742 	 * Maybe this should be above the vnode op call, but so long as
743 	 * file servers have no limits, I don't think it matters.
744 	 */
745 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
746 		return (EFBIG);
747 
748 	resid = uio->uio_resid;
749 	osize = ip->i_size;
750 	if (seqcount > BA_SEQMAX)
751 		flags = BA_SEQMAX << BA_SEQSHIFT;
752 	else
753 		flags = seqcount << BA_SEQSHIFT;
754 	if (ioflag & IO_SYNC)
755 		flags |= IO_SYNC;
756 	flags |= BA_UNMAPPED;
757 
758 	for (error = 0; uio->uio_resid > 0;) {
759 		lbn = lblkno(fs, uio->uio_offset);
760 		blkoffset = blkoff(fs, uio->uio_offset);
761 		xfersize = fs->fs_bsize - blkoffset;
762 		if (uio->uio_resid < xfersize)
763 			xfersize = uio->uio_resid;
764 		if (uio->uio_offset + xfersize > ip->i_size)
765 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
766 
767 		/*
768 		 * We must perform a read-before-write if the transfer size
769 		 * does not cover the entire buffer.
770 		 */
771 		if (fs->fs_bsize > xfersize)
772 			flags |= BA_CLRBUF;
773 		else
774 			flags &= ~BA_CLRBUF;
775 /* XXX is uio->uio_offset the right thing here? */
776 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
777 		    ap->a_cred, flags, &bp);
778 		if (error != 0) {
779 			vnode_pager_setsize(vp, ip->i_size);
780 			break;
781 		}
782 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
783 			bp->b_flags |= B_NOCACHE;
784 
785 		if (uio->uio_offset + xfersize > ip->i_size) {
786 			ip->i_size = uio->uio_offset + xfersize;
787 			DIP_SET(ip, i_size, ip->i_size);
788 		}
789 
790 		size = blksize(fs, ip, lbn) - bp->b_resid;
791 		if (size < xfersize)
792 			xfersize = size;
793 
794 		if (buf_mapped(bp)) {
795 			error = vn_io_fault_uiomove((char *)bp->b_data +
796 			    blkoffset, (int)xfersize, uio);
797 		} else {
798 			error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
799 			    (int)xfersize, uio);
800 		}
801 		/*
802 		 * If the buffer is not already filled and we encounter an
803 		 * error while trying to fill it, we have to clear out any
804 		 * garbage data from the pages instantiated for the buffer.
805 		 * If we do not, a failed uiomove() during a write can leave
806 		 * the prior contents of the pages exposed to a userland mmap.
807 		 *
808 		 * Note that we need only clear buffers with a transfer size
809 		 * equal to the block size because buffers with a shorter
810 		 * transfer size were cleared above by the call to UFS_BALLOC()
811 		 * with the BA_CLRBUF flag set.
812 		 *
813 		 * If the source region for uiomove identically mmaps the
814 		 * buffer, uiomove() performed the NOP copy, and the buffer
815 		 * content remains valid because the page fault handler
816 		 * validated the pages.
817 		 */
818 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
819 		    fs->fs_bsize == xfersize)
820 			vfs_bio_clrbuf(bp);
821 
822 		vfs_bio_set_flags(bp, ioflag);
823 
824 		/*
825 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
826 		 * if we have a severe page deficiency write the buffer
827 		 * asynchronously.  Otherwise try to cluster, and if that
828 		 * doesn't do it then either do an async write (if O_DIRECT),
829 		 * or a delayed write (if not).
830 		 */
831 		if (ioflag & IO_SYNC) {
832 			(void)bwrite(bp);
833 		} else if (vm_page_count_severe() ||
834 			    buf_dirty_count_severe() ||
835 			    (ioflag & IO_ASYNC)) {
836 			bp->b_flags |= B_CLUSTEROK;
837 			bawrite(bp);
838 		} else if (xfersize + blkoffset == fs->fs_bsize) {
839 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
840 				bp->b_flags |= B_CLUSTEROK;
841 				cluster_write(vp, bp, ip->i_size, seqcount,
842 				    GB_UNMAPPED);
843 			} else {
844 				bawrite(bp);
845 			}
846 		} else if (ioflag & IO_DIRECT) {
847 			bp->b_flags |= B_CLUSTEROK;
848 			bawrite(bp);
849 		} else {
850 			bp->b_flags |= B_CLUSTEROK;
851 			bdwrite(bp);
852 		}
853 		if (error || xfersize == 0)
854 			break;
855 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
856 	}
857 	/*
858 	 * If we successfully wrote any data, and we are not the superuser
859 	 * we clear the setuid and setgid bits as a precaution against
860 	 * tampering.
861 	 */
862 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
863 	    ap->a_cred) {
864 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
865 			ip->i_mode &= ~(ISUID | ISGID);
866 			DIP_SET(ip, i_mode, ip->i_mode);
867 		}
868 	}
869 	if (error) {
870 		if (ioflag & IO_UNIT) {
871 			(void)ffs_truncate(vp, osize,
872 			    IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
873 			uio->uio_offset -= resid - uio->uio_resid;
874 			uio->uio_resid = resid;
875 		}
876 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
877 		error = ffs_update(vp, 1);
878 	return (error);
879 }
880 
881 /*
882  * Extended attribute area reading.
883  */
884 static int
885 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
886 {
887 	struct inode *ip;
888 	struct ufs2_dinode *dp;
889 	struct fs *fs;
890 	struct buf *bp;
891 	ufs_lbn_t lbn, nextlbn;
892 	off_t bytesinfile;
893 	long size, xfersize, blkoffset;
894 	ssize_t orig_resid;
895 	int error;
896 
897 	ip = VTOI(vp);
898 	fs = ITOFS(ip);
899 	dp = ip->i_din2;
900 
901 #ifdef INVARIANTS
902 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
903 		panic("ffs_extread: mode");
904 
905 #endif
906 	orig_resid = uio->uio_resid;
907 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
908 	if (orig_resid == 0)
909 		return (0);
910 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
911 
912 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
913 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
914 			break;
915 		lbn = lblkno(fs, uio->uio_offset);
916 		nextlbn = lbn + 1;
917 
918 		/*
919 		 * size of buffer.  The buffer representing the
920 		 * end of the file is rounded up to the size of
921 		 * the block type ( fragment or full block,
922 		 * depending ).
923 		 */
924 		size = sblksize(fs, dp->di_extsize, lbn);
925 		blkoffset = blkoff(fs, uio->uio_offset);
926 
927 		/*
928 		 * The amount we want to transfer in this iteration is
929 		 * one FS block less the amount of the data before
930 		 * our startpoint (duh!)
931 		 */
932 		xfersize = fs->fs_bsize - blkoffset;
933 
934 		/*
935 		 * But if we actually want less than the block,
936 		 * or the file doesn't have a whole block more of data,
937 		 * then use the lesser number.
938 		 */
939 		if (uio->uio_resid < xfersize)
940 			xfersize = uio->uio_resid;
941 		if (bytesinfile < xfersize)
942 			xfersize = bytesinfile;
943 
944 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
945 			/*
946 			 * Don't do readahead if this is the end of the info.
947 			 */
948 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
949 		} else {
950 			/*
951 			 * If we have a second block, then
952 			 * fire off a request for a readahead
953 			 * as well as a read. Note that the 4th and 5th
954 			 * arguments point to arrays of the size specified in
955 			 * the 6th argument.
956 			 */
957 			u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
958 
959 			nextlbn = -1 - nextlbn;
960 			error = breadn(vp, -1 - lbn,
961 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
962 		}
963 		if (error) {
964 			brelse(bp);
965 			bp = NULL;
966 			break;
967 		}
968 
969 		/*
970 		 * We should only get non-zero b_resid when an I/O error
971 		 * has occurred, which should cause us to break above.
972 		 * However, if the short read did not cause an error,
973 		 * then we want to ensure that we do not uiomove bad
974 		 * or uninitialized data.
975 		 */
976 		size -= bp->b_resid;
977 		if (size < xfersize) {
978 			if (size == 0)
979 				break;
980 			xfersize = size;
981 		}
982 
983 		error = uiomove((char *)bp->b_data + blkoffset,
984 					(int)xfersize, uio);
985 		if (error)
986 			break;
987 		vfs_bio_brelse(bp, ioflag);
988 	}
989 
990 	/*
991 	 * This can only happen in the case of an error
992 	 * because the loop above resets bp to NULL on each iteration
993 	 * and on normal completion has not set a new value into it.
994 	 * so it must have come from a 'break' statement
995 	 */
996 	if (bp != NULL)
997 		vfs_bio_brelse(bp, ioflag);
998 	return (error);
999 }
1000 
1001 /*
1002  * Extended attribute area writing.
1003  */
1004 static int
1005 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1006 {
1007 	struct inode *ip;
1008 	struct ufs2_dinode *dp;
1009 	struct fs *fs;
1010 	struct buf *bp;
1011 	ufs_lbn_t lbn;
1012 	off_t osize;
1013 	ssize_t resid;
1014 	int blkoffset, error, flags, size, xfersize;
1015 
1016 	ip = VTOI(vp);
1017 	fs = ITOFS(ip);
1018 	dp = ip->i_din2;
1019 
1020 #ifdef INVARIANTS
1021 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1022 		panic("ffs_extwrite: mode");
1023 #endif
1024 
1025 	if (ioflag & IO_APPEND)
1026 		uio->uio_offset = dp->di_extsize;
1027 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1028 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1029 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
1030 	    UFS_NXADDR * fs->fs_bsize)
1031 		return (EFBIG);
1032 
1033 	resid = uio->uio_resid;
1034 	osize = dp->di_extsize;
1035 	flags = IO_EXT;
1036 	if (ioflag & IO_SYNC)
1037 		flags |= IO_SYNC;
1038 
1039 	for (error = 0; uio->uio_resid > 0;) {
1040 		lbn = lblkno(fs, uio->uio_offset);
1041 		blkoffset = blkoff(fs, uio->uio_offset);
1042 		xfersize = fs->fs_bsize - blkoffset;
1043 		if (uio->uio_resid < xfersize)
1044 			xfersize = uio->uio_resid;
1045 
1046 		/*
1047 		 * We must perform a read-before-write if the transfer size
1048 		 * does not cover the entire buffer.
1049 		 */
1050 		if (fs->fs_bsize > xfersize)
1051 			flags |= BA_CLRBUF;
1052 		else
1053 			flags &= ~BA_CLRBUF;
1054 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1055 		    ucred, flags, &bp);
1056 		if (error != 0)
1057 			break;
1058 		/*
1059 		 * If the buffer is not valid we have to clear out any
1060 		 * garbage data from the pages instantiated for the buffer.
1061 		 * If we do not, a failed uiomove() during a write can leave
1062 		 * the prior contents of the pages exposed to a userland
1063 		 * mmap().  XXX deal with uiomove() errors a better way.
1064 		 */
1065 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1066 			vfs_bio_clrbuf(bp);
1067 
1068 		if (uio->uio_offset + xfersize > dp->di_extsize)
1069 			dp->di_extsize = uio->uio_offset + xfersize;
1070 
1071 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1072 		if (size < xfersize)
1073 			xfersize = size;
1074 
1075 		error =
1076 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1077 
1078 		vfs_bio_set_flags(bp, ioflag);
1079 
1080 		/*
1081 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1082 		 * if we have a severe page deficiency write the buffer
1083 		 * asynchronously.  Otherwise try to cluster, and if that
1084 		 * doesn't do it then either do an async write (if O_DIRECT),
1085 		 * or a delayed write (if not).
1086 		 */
1087 		if (ioflag & IO_SYNC) {
1088 			(void)bwrite(bp);
1089 		} else if (vm_page_count_severe() ||
1090 			    buf_dirty_count_severe() ||
1091 			    xfersize + blkoffset == fs->fs_bsize ||
1092 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1093 			bawrite(bp);
1094 		else
1095 			bdwrite(bp);
1096 		if (error || xfersize == 0)
1097 			break;
1098 		ip->i_flag |= IN_CHANGE;
1099 	}
1100 	/*
1101 	 * If we successfully wrote any data, and we are not the superuser
1102 	 * we clear the setuid and setgid bits as a precaution against
1103 	 * tampering.
1104 	 */
1105 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1106 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1107 			ip->i_mode &= ~(ISUID | ISGID);
1108 			dp->di_mode = ip->i_mode;
1109 		}
1110 	}
1111 	if (error) {
1112 		if (ioflag & IO_UNIT) {
1113 			(void)ffs_truncate(vp, osize,
1114 			    IO_EXT | (ioflag&IO_SYNC), ucred);
1115 			uio->uio_offset -= resid - uio->uio_resid;
1116 			uio->uio_resid = resid;
1117 		}
1118 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1119 		error = ffs_update(vp, 1);
1120 	return (error);
1121 }
1122 
1123 
1124 /*
1125  * Vnode operating to retrieve a named extended attribute.
1126  *
1127  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1128  * the length of the EA, and possibly the pointer to the entry and to the data.
1129  */
1130 static int
1131 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name,
1132     struct extattr **eapp, u_char **eac)
1133 {
1134 	struct extattr *eap, *eaend;
1135 	size_t nlen;
1136 
1137 	nlen = strlen(name);
1138 	KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned"));
1139 	eap = (struct extattr *)ptr;
1140 	eaend = (struct extattr *)(ptr + length);
1141 	for (; eap < eaend; eap = EXTATTR_NEXT(eap)) {
1142 		/* make sure this entry is complete */
1143 		if (EXTATTR_NEXT(eap) > eaend)
1144 			break;
1145 		if (eap->ea_namespace != nspace || eap->ea_namelength != nlen
1146 		    || memcmp(eap->ea_name, name, nlen) != 0)
1147 			continue;
1148 		if (eapp != NULL)
1149 			*eapp = eap;
1150 		if (eac != NULL)
1151 			*eac = EXTATTR_CONTENT(eap);
1152 		return (EXTATTR_CONTENT_SIZE(eap));
1153 	}
1154 	return (-1);
1155 }
1156 
1157 static int
1158 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1159 {
1160 	struct inode *ip;
1161 	struct ufs2_dinode *dp;
1162 	struct fs *fs;
1163 	struct uio luio;
1164 	struct iovec liovec;
1165 	u_int easize;
1166 	int error;
1167 	u_char *eae;
1168 
1169 	ip = VTOI(vp);
1170 	fs = ITOFS(ip);
1171 	dp = ip->i_din2;
1172 	easize = dp->di_extsize;
1173 	if ((uoff_t)easize + extra > UFS_NXADDR * fs->fs_bsize)
1174 		return (EFBIG);
1175 
1176 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1177 
1178 	liovec.iov_base = eae;
1179 	liovec.iov_len = easize;
1180 	luio.uio_iov = &liovec;
1181 	luio.uio_iovcnt = 1;
1182 	luio.uio_offset = 0;
1183 	luio.uio_resid = easize;
1184 	luio.uio_segflg = UIO_SYSSPACE;
1185 	luio.uio_rw = UIO_READ;
1186 	luio.uio_td = td;
1187 
1188 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1189 	if (error) {
1190 		free(eae, M_TEMP);
1191 		return(error);
1192 	}
1193 	*p = eae;
1194 	return (0);
1195 }
1196 
1197 static void
1198 ffs_lock_ea(struct vnode *vp)
1199 {
1200 	struct inode *ip;
1201 
1202 	ip = VTOI(vp);
1203 	VI_LOCK(vp);
1204 	while (ip->i_flag & IN_EA_LOCKED) {
1205 		ip->i_flag |= IN_EA_LOCKWAIT;
1206 		msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1207 		    0);
1208 	}
1209 	ip->i_flag |= IN_EA_LOCKED;
1210 	VI_UNLOCK(vp);
1211 }
1212 
1213 static void
1214 ffs_unlock_ea(struct vnode *vp)
1215 {
1216 	struct inode *ip;
1217 
1218 	ip = VTOI(vp);
1219 	VI_LOCK(vp);
1220 	if (ip->i_flag & IN_EA_LOCKWAIT)
1221 		wakeup(&ip->i_ea_refs);
1222 	ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1223 	VI_UNLOCK(vp);
1224 }
1225 
1226 static int
1227 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1228 {
1229 	struct inode *ip;
1230 	struct ufs2_dinode *dp;
1231 	int error;
1232 
1233 	ip = VTOI(vp);
1234 
1235 	ffs_lock_ea(vp);
1236 	if (ip->i_ea_area != NULL) {
1237 		ip->i_ea_refs++;
1238 		ffs_unlock_ea(vp);
1239 		return (0);
1240 	}
1241 	dp = ip->i_din2;
1242 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1243 	if (error) {
1244 		ffs_unlock_ea(vp);
1245 		return (error);
1246 	}
1247 	ip->i_ea_len = dp->di_extsize;
1248 	ip->i_ea_error = 0;
1249 	ip->i_ea_refs++;
1250 	ffs_unlock_ea(vp);
1251 	return (0);
1252 }
1253 
1254 /*
1255  * Vnode extattr transaction commit/abort
1256  */
1257 static int
1258 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1259 {
1260 	struct inode *ip;
1261 	struct uio luio;
1262 	struct iovec liovec;
1263 	int error;
1264 	struct ufs2_dinode *dp;
1265 
1266 	ip = VTOI(vp);
1267 
1268 	ffs_lock_ea(vp);
1269 	if (ip->i_ea_area == NULL) {
1270 		ffs_unlock_ea(vp);
1271 		return (EINVAL);
1272 	}
1273 	dp = ip->i_din2;
1274 	error = ip->i_ea_error;
1275 	if (commit && error == 0) {
1276 		ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1277 		if (cred == NOCRED)
1278 			cred =  vp->v_mount->mnt_cred;
1279 		liovec.iov_base = ip->i_ea_area;
1280 		liovec.iov_len = ip->i_ea_len;
1281 		luio.uio_iov = &liovec;
1282 		luio.uio_iovcnt = 1;
1283 		luio.uio_offset = 0;
1284 		luio.uio_resid = ip->i_ea_len;
1285 		luio.uio_segflg = UIO_SYSSPACE;
1286 		luio.uio_rw = UIO_WRITE;
1287 		luio.uio_td = td;
1288 		/* XXX: I'm not happy about truncating to zero size */
1289 		if (ip->i_ea_len < dp->di_extsize)
1290 			error = ffs_truncate(vp, 0, IO_EXT, cred);
1291 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1292 	}
1293 	if (--ip->i_ea_refs == 0) {
1294 		free(ip->i_ea_area, M_TEMP);
1295 		ip->i_ea_area = NULL;
1296 		ip->i_ea_len = 0;
1297 		ip->i_ea_error = 0;
1298 	}
1299 	ffs_unlock_ea(vp);
1300 	return (error);
1301 }
1302 
1303 /*
1304  * Vnode extattr strategy routine for fifos.
1305  *
1306  * We need to check for a read or write of the external attributes.
1307  * Otherwise we just fall through and do the usual thing.
1308  */
1309 static int
1310 ffsext_strategy(struct vop_strategy_args *ap)
1311 /*
1312 struct vop_strategy_args {
1313 	struct vnodeop_desc *a_desc;
1314 	struct vnode *a_vp;
1315 	struct buf *a_bp;
1316 };
1317 */
1318 {
1319 	struct vnode *vp;
1320 	daddr_t lbn;
1321 
1322 	vp = ap->a_vp;
1323 	lbn = ap->a_bp->b_lblkno;
1324 	if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR)
1325 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1326 	if (vp->v_type == VFIFO)
1327 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1328 	panic("spec nodes went here");
1329 }
1330 
1331 /*
1332  * Vnode extattr transaction commit/abort
1333  */
1334 static int
1335 ffs_openextattr(struct vop_openextattr_args *ap)
1336 /*
1337 struct vop_openextattr_args {
1338 	struct vnodeop_desc *a_desc;
1339 	struct vnode *a_vp;
1340 	IN struct ucred *a_cred;
1341 	IN struct thread *a_td;
1342 };
1343 */
1344 {
1345 
1346 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1347 		return (EOPNOTSUPP);
1348 
1349 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1350 }
1351 
1352 
1353 /*
1354  * Vnode extattr transaction commit/abort
1355  */
1356 static int
1357 ffs_closeextattr(struct vop_closeextattr_args *ap)
1358 /*
1359 struct vop_closeextattr_args {
1360 	struct vnodeop_desc *a_desc;
1361 	struct vnode *a_vp;
1362 	int a_commit;
1363 	IN struct ucred *a_cred;
1364 	IN struct thread *a_td;
1365 };
1366 */
1367 {
1368 
1369 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1370 		return (EOPNOTSUPP);
1371 
1372 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1373 		return (EROFS);
1374 
1375 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1376 }
1377 
1378 /*
1379  * Vnode operation to remove a named attribute.
1380  */
1381 static int
1382 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1383 /*
1384 vop_deleteextattr {
1385 	IN struct vnode *a_vp;
1386 	IN int a_attrnamespace;
1387 	IN const char *a_name;
1388 	IN struct ucred *a_cred;
1389 	IN struct thread *a_td;
1390 };
1391 */
1392 {
1393 	struct inode *ip;
1394 	struct extattr *eap;
1395 	uint32_t ul;
1396 	int olen, error, i, easize;
1397 	u_char *eae;
1398 	void *tmp;
1399 
1400 	ip = VTOI(ap->a_vp);
1401 
1402 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1403 		return (EOPNOTSUPP);
1404 
1405 	if (strlen(ap->a_name) == 0)
1406 		return (EINVAL);
1407 
1408 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1409 		return (EROFS);
1410 
1411 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1412 	    ap->a_cred, ap->a_td, VWRITE);
1413 	if (error) {
1414 
1415 		/*
1416 		 * ffs_lock_ea is not needed there, because the vnode
1417 		 * must be exclusively locked.
1418 		 */
1419 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1420 			ip->i_ea_error = error;
1421 		return (error);
1422 	}
1423 
1424 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1425 	if (error)
1426 		return (error);
1427 
1428 	/* CEM: delete could be done in-place instead */
1429 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1430 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1431 	easize = ip->i_ea_len;
1432 
1433 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1434 	    &eap, NULL);
1435 	if (olen == -1) {
1436 		/* delete but nonexistent */
1437 		free(eae, M_TEMP);
1438 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1439 		return (ENOATTR);
1440 	}
1441 	ul = eap->ea_length;
1442 	i = (u_char *)EXTATTR_NEXT(eap) - eae;
1443 	bcopy(EXTATTR_NEXT(eap), eap, easize - i);
1444 	easize -= ul;
1445 
1446 	tmp = ip->i_ea_area;
1447 	ip->i_ea_area = eae;
1448 	ip->i_ea_len = easize;
1449 	free(tmp, M_TEMP);
1450 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1451 	return (error);
1452 }
1453 
1454 /*
1455  * Vnode operation to retrieve a named extended attribute.
1456  */
1457 static int
1458 ffs_getextattr(struct vop_getextattr_args *ap)
1459 /*
1460 vop_getextattr {
1461 	IN struct vnode *a_vp;
1462 	IN int a_attrnamespace;
1463 	IN const char *a_name;
1464 	INOUT struct uio *a_uio;
1465 	OUT size_t *a_size;
1466 	IN struct ucred *a_cred;
1467 	IN struct thread *a_td;
1468 };
1469 */
1470 {
1471 	struct inode *ip;
1472 	u_char *eae, *p;
1473 	unsigned easize;
1474 	int error, ealen;
1475 
1476 	ip = VTOI(ap->a_vp);
1477 
1478 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1479 		return (EOPNOTSUPP);
1480 
1481 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1482 	    ap->a_cred, ap->a_td, VREAD);
1483 	if (error)
1484 		return (error);
1485 
1486 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1487 	if (error)
1488 		return (error);
1489 
1490 	eae = ip->i_ea_area;
1491 	easize = ip->i_ea_len;
1492 
1493 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1494 	    NULL, &p);
1495 	if (ealen >= 0) {
1496 		error = 0;
1497 		if (ap->a_size != NULL)
1498 			*ap->a_size = ealen;
1499 		else if (ap->a_uio != NULL)
1500 			error = uiomove(p, ealen, ap->a_uio);
1501 	} else
1502 		error = ENOATTR;
1503 
1504 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1505 	return (error);
1506 }
1507 
1508 /*
1509  * Vnode operation to retrieve extended attributes on a vnode.
1510  */
1511 static int
1512 ffs_listextattr(struct vop_listextattr_args *ap)
1513 /*
1514 vop_listextattr {
1515 	IN struct vnode *a_vp;
1516 	IN int a_attrnamespace;
1517 	INOUT struct uio *a_uio;
1518 	OUT size_t *a_size;
1519 	IN struct ucred *a_cred;
1520 	IN struct thread *a_td;
1521 };
1522 */
1523 {
1524 	struct inode *ip;
1525 	struct extattr *eap, *eaend;
1526 	int error, ealen;
1527 
1528 	ip = VTOI(ap->a_vp);
1529 
1530 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1531 		return (EOPNOTSUPP);
1532 
1533 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1534 	    ap->a_cred, ap->a_td, VREAD);
1535 	if (error)
1536 		return (error);
1537 
1538 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1539 	if (error)
1540 		return (error);
1541 
1542 	error = 0;
1543 	if (ap->a_size != NULL)
1544 		*ap->a_size = 0;
1545 
1546 	KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned"));
1547 	eap = (struct extattr *)ip->i_ea_area;
1548 	eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len);
1549 	for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) {
1550 		/* make sure this entry is complete */
1551 		if (EXTATTR_NEXT(eap) > eaend)
1552 			break;
1553 		if (eap->ea_namespace != ap->a_attrnamespace)
1554 			continue;
1555 
1556 		ealen = eap->ea_namelength;
1557 		if (ap->a_size != NULL)
1558 			*ap->a_size += ealen + 1;
1559 		else if (ap->a_uio != NULL)
1560 			error = uiomove(&eap->ea_namelength, ealen + 1,
1561 			    ap->a_uio);
1562 	}
1563 
1564 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1565 	return (error);
1566 }
1567 
1568 /*
1569  * Vnode operation to set a named attribute.
1570  */
1571 static int
1572 ffs_setextattr(struct vop_setextattr_args *ap)
1573 /*
1574 vop_setextattr {
1575 	IN struct vnode *a_vp;
1576 	IN int a_attrnamespace;
1577 	IN const char *a_name;
1578 	INOUT struct uio *a_uio;
1579 	IN struct ucred *a_cred;
1580 	IN struct thread *a_td;
1581 };
1582 */
1583 {
1584 	struct inode *ip;
1585 	struct fs *fs;
1586 	struct extattr *eap;
1587 	uint32_t ealength, ul;
1588 	ssize_t ealen;
1589 	int olen, eapad1, eapad2, error, i, easize;
1590 	u_char *eae;
1591 	void *tmp;
1592 
1593 	ip = VTOI(ap->a_vp);
1594 	fs = ITOFS(ip);
1595 
1596 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1597 		return (EOPNOTSUPP);
1598 
1599 	if (strlen(ap->a_name) == 0)
1600 		return (EINVAL);
1601 
1602 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1603 	if (ap->a_uio == NULL)
1604 		return (EOPNOTSUPP);
1605 
1606 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1607 		return (EROFS);
1608 
1609 	ealen = ap->a_uio->uio_resid;
1610 	if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR))
1611 		return (EINVAL);
1612 
1613 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1614 	    ap->a_cred, ap->a_td, VWRITE);
1615 	if (error) {
1616 
1617 		/*
1618 		 * ffs_lock_ea is not needed there, because the vnode
1619 		 * must be exclusively locked.
1620 		 */
1621 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1622 			ip->i_ea_error = error;
1623 		return (error);
1624 	}
1625 
1626 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1627 	if (error)
1628 		return (error);
1629 
1630 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1631 	eapad1 = roundup2(ealength, 8) - ealength;
1632 	eapad2 = roundup2(ealen, 8) - ealen;
1633 	ealength += eapad1 + ealen + eapad2;
1634 
1635 	/*
1636 	 * CEM: rewrites of the same size or smaller could be done in-place
1637 	 * instead.  (We don't acquire any fine-grained locks in here either,
1638 	 * so we could also do bigger writes in-place.)
1639 	 */
1640 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1641 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1642 	easize = ip->i_ea_len;
1643 
1644 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1645 	    &eap, NULL);
1646         if (olen == -1) {
1647 		/* new, append at end */
1648 		KASSERT(ALIGNED_TO(eae + easize, struct extattr),
1649 		    ("unaligned"));
1650 		eap = (struct extattr *)(eae + easize);
1651 		easize += ealength;
1652 	} else {
1653 		ul = eap->ea_length;
1654 		i = (u_char *)EXTATTR_NEXT(eap) - eae;
1655 		if (ul != ealength) {
1656 			bcopy(EXTATTR_NEXT(eap), (u_char *)eap + ealength,
1657 			    easize - i);
1658 			easize += (ealength - ul);
1659 		}
1660 	}
1661 	if (easize > lblktosize(fs, UFS_NXADDR)) {
1662 		free(eae, M_TEMP);
1663 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1664 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1665 			ip->i_ea_error = ENOSPC;
1666 		return (ENOSPC);
1667 	}
1668 	eap->ea_length = ealength;
1669 	eap->ea_namespace = ap->a_attrnamespace;
1670 	eap->ea_contentpadlen = eapad2;
1671 	eap->ea_namelength = strlen(ap->a_name);
1672 	memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name));
1673 	bzero(&eap->ea_name[strlen(ap->a_name)], eapad1);
1674 	error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio);
1675 	if (error) {
1676 		free(eae, M_TEMP);
1677 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1678 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1679 			ip->i_ea_error = error;
1680 		return (error);
1681 	}
1682 	bzero((u_char *)EXTATTR_CONTENT(eap) + ealen, eapad2);
1683 
1684 	tmp = ip->i_ea_area;
1685 	ip->i_ea_area = eae;
1686 	ip->i_ea_len = easize;
1687 	free(tmp, M_TEMP);
1688 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1689 	return (error);
1690 }
1691 
1692 /*
1693  * Vnode pointer to File handle
1694  */
1695 static int
1696 ffs_vptofh(struct vop_vptofh_args *ap)
1697 /*
1698 vop_vptofh {
1699 	IN struct vnode *a_vp;
1700 	IN struct fid *a_fhp;
1701 };
1702 */
1703 {
1704 	struct inode *ip;
1705 	struct ufid *ufhp;
1706 
1707 	ip = VTOI(ap->a_vp);
1708 	ufhp = (struct ufid *)ap->a_fhp;
1709 	ufhp->ufid_len = sizeof(struct ufid);
1710 	ufhp->ufid_ino = ip->i_number;
1711 	ufhp->ufid_gen = ip->i_gen;
1712 	return (0);
1713 }
1714 
1715 SYSCTL_DECL(_vfs_ffs);
1716 static int use_buf_pager = 1;
1717 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0,
1718     "Always use buffer pager instead of bmap");
1719 
1720 static daddr_t
1721 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1722 {
1723 
1724 	return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off));
1725 }
1726 
1727 static int
1728 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1729 {
1730 
1731 	return (blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn));
1732 }
1733 
1734 static int
1735 ffs_getpages(struct vop_getpages_args *ap)
1736 {
1737 	struct vnode *vp;
1738 	struct ufsmount *um;
1739 
1740 	vp = ap->a_vp;
1741 	um = VFSTOUFS(vp->v_mount);
1742 
1743 	if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE)
1744 		return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1745 		    ap->a_rbehind, ap->a_rahead, NULL, NULL));
1746 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1747 	    ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz));
1748 }
1749