xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision bc7512cc58af2e8bbe5bbf5ca0059b1daa1da897)
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 "opt_directio.h"
70 #include "opt_ffs.h"
71 #include "opt_ufs.h"
72 
73 #include <sys/param.h>
74 #include <sys/bio.h>
75 #include <sys/systm.h>
76 #include <sys/buf.h>
77 #include <sys/conf.h>
78 #include <sys/extattr.h>
79 #include <sys/kernel.h>
80 #include <sys/limits.h>
81 #include <sys/malloc.h>
82 #include <sys/mount.h>
83 #include <sys/priv.h>
84 #include <sys/rwlock.h>
85 #include <sys/stat.h>
86 #include <sys/sysctl.h>
87 #include <sys/vmmeter.h>
88 #include <sys/vnode.h>
89 
90 #include <vm/vm.h>
91 #include <vm/vm_param.h>
92 #include <vm/vm_extern.h>
93 #include <vm/vm_object.h>
94 #include <vm/vm_page.h>
95 #include <vm/vm_pager.h>
96 #include <vm/vnode_pager.h>
97 
98 #include <ufs/ufs/extattr.h>
99 #include <ufs/ufs/quota.h>
100 #include <ufs/ufs/inode.h>
101 #include <ufs/ufs/ufs_extern.h>
102 #include <ufs/ufs/ufsmount.h>
103 #include <ufs/ufs/dir.h>
104 #ifdef UFS_DIRHASH
105 #include <ufs/ufs/dirhash.h>
106 #endif
107 
108 #include <ufs/ffs/fs.h>
109 #include <ufs/ffs/ffs_extern.h>
110 
111 #define	ALIGNED_TO(ptr, s)	\
112 	(((uintptr_t)(ptr) & (_Alignof(s) - 1)) == 0)
113 
114 #ifdef DIRECTIO
115 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
116 #endif
117 static vop_fdatasync_t	ffs_fdatasync;
118 static vop_fsync_t	ffs_fsync;
119 static vop_getpages_t	ffs_getpages;
120 static vop_getpages_async_t	ffs_getpages_async;
121 static vop_lock1_t	ffs_lock;
122 #ifdef INVARIANTS
123 static vop_unlock_t	ffs_unlock_debug;
124 #endif
125 static vop_read_t	ffs_read;
126 static vop_write_t	ffs_write;
127 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
128 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
129 		    struct ucred *cred);
130 static vop_strategy_t	ffsext_strategy;
131 static vop_closeextattr_t	ffs_closeextattr;
132 static vop_deleteextattr_t	ffs_deleteextattr;
133 static vop_getextattr_t	ffs_getextattr;
134 static vop_listextattr_t	ffs_listextattr;
135 static vop_openextattr_t	ffs_openextattr;
136 static vop_setextattr_t	ffs_setextattr;
137 static vop_vptofh_t	ffs_vptofh;
138 static vop_vput_pair_t	ffs_vput_pair;
139 
140 /* Global vfs data structures for ufs. */
141 struct vop_vector ffs_vnodeops1 = {
142 	.vop_default =		&ufs_vnodeops,
143 	.vop_fsync =		ffs_fsync,
144 	.vop_fdatasync =	ffs_fdatasync,
145 	.vop_getpages =		ffs_getpages,
146 	.vop_getpages_async =	ffs_getpages_async,
147 	.vop_lock1 =		ffs_lock,
148 #ifdef INVARIANTS
149 	.vop_unlock =		ffs_unlock_debug,
150 #endif
151 	.vop_read =		ffs_read,
152 	.vop_reallocblks =	ffs_reallocblks,
153 	.vop_write =		ffs_write,
154 	.vop_vptofh =		ffs_vptofh,
155 	.vop_vput_pair =	ffs_vput_pair,
156 };
157 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops1);
158 
159 struct vop_vector ffs_fifoops1 = {
160 	.vop_default =		&ufs_fifoops,
161 	.vop_fsync =		ffs_fsync,
162 	.vop_fdatasync =	ffs_fdatasync,
163 	.vop_lock1 =		ffs_lock,
164 #ifdef INVARIANTS
165 	.vop_unlock =		ffs_unlock_debug,
166 #endif
167 	.vop_vptofh =		ffs_vptofh,
168 };
169 VFS_VOP_VECTOR_REGISTER(ffs_fifoops1);
170 
171 /* Global vfs data structures for ufs. */
172 struct vop_vector ffs_vnodeops2 = {
173 	.vop_default =		&ufs_vnodeops,
174 	.vop_fsync =		ffs_fsync,
175 	.vop_fdatasync =	ffs_fdatasync,
176 	.vop_getpages =		ffs_getpages,
177 	.vop_getpages_async =	ffs_getpages_async,
178 	.vop_lock1 =		ffs_lock,
179 #ifdef INVARIANTS
180 	.vop_unlock =		ffs_unlock_debug,
181 #endif
182 	.vop_read =		ffs_read,
183 	.vop_reallocblks =	ffs_reallocblks,
184 	.vop_write =		ffs_write,
185 	.vop_closeextattr =	ffs_closeextattr,
186 	.vop_deleteextattr =	ffs_deleteextattr,
187 	.vop_getextattr =	ffs_getextattr,
188 	.vop_listextattr =	ffs_listextattr,
189 	.vop_openextattr =	ffs_openextattr,
190 	.vop_setextattr =	ffs_setextattr,
191 	.vop_vptofh =		ffs_vptofh,
192 	.vop_vput_pair =	ffs_vput_pair,
193 };
194 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops2);
195 
196 struct vop_vector ffs_fifoops2 = {
197 	.vop_default =		&ufs_fifoops,
198 	.vop_fsync =		ffs_fsync,
199 	.vop_fdatasync =	ffs_fdatasync,
200 	.vop_lock1 =		ffs_lock,
201 #ifdef INVARIANTS
202 	.vop_unlock =		ffs_unlock_debug,
203 #endif
204 	.vop_reallocblks =	ffs_reallocblks,
205 	.vop_strategy =		ffsext_strategy,
206 	.vop_closeextattr =	ffs_closeextattr,
207 	.vop_deleteextattr =	ffs_deleteextattr,
208 	.vop_getextattr =	ffs_getextattr,
209 	.vop_listextattr =	ffs_listextattr,
210 	.vop_openextattr =	ffs_openextattr,
211 	.vop_setextattr =	ffs_setextattr,
212 	.vop_vptofh =		ffs_vptofh,
213 };
214 VFS_VOP_VECTOR_REGISTER(ffs_fifoops2);
215 
216 /*
217  * Synch an open file.
218  */
219 /* ARGSUSED */
220 static int
221 ffs_fsync(struct vop_fsync_args *ap)
222 {
223 	struct vnode *vp;
224 	struct bufobj *bo;
225 	int error;
226 
227 	vp = ap->a_vp;
228 	bo = &vp->v_bufobj;
229 retry:
230 	error = ffs_syncvnode(vp, ap->a_waitfor, 0);
231 	if (error)
232 		return (error);
233 	if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
234 		error = softdep_fsync(vp);
235 		if (error)
236 			return (error);
237 
238 		/*
239 		 * The softdep_fsync() function may drop vp lock,
240 		 * allowing for dirty buffers to reappear on the
241 		 * bo_dirty list. Recheck and resync as needed.
242 		 */
243 		BO_LOCK(bo);
244 		if ((vp->v_type == VREG || vp->v_type == VDIR) &&
245 		    (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
246 			BO_UNLOCK(bo);
247 			goto retry;
248 		}
249 		BO_UNLOCK(bo);
250 	}
251 	if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), 0))
252 		return (ENXIO);
253 	return (0);
254 }
255 
256 int
257 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
258 {
259 	struct inode *ip;
260 	struct bufobj *bo;
261 	struct ufsmount *ump;
262 	struct buf *bp, *nbp;
263 	ufs_lbn_t lbn;
264 	int error, passes, wflag;
265 	bool still_dirty, unlocked, wait;
266 
267 	ip = VTOI(vp);
268 	bo = &vp->v_bufobj;
269 	ump = VFSTOUFS(vp->v_mount);
270 #ifdef WITNESS
271 	wflag = IS_SNAPSHOT(ip) ? LK_NOWITNESS : 0;
272 #else
273 	wflag = 0;
274 #endif
275 
276 	/*
277 	 * When doing MNT_WAIT we must first flush all dependencies
278 	 * on the inode.
279 	 */
280 	if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
281 	    (error = softdep_sync_metadata(vp)) != 0) {
282 		if (ffs_fsfail_cleanup(ump, error))
283 			error = 0;
284 		return (error);
285 	}
286 
287 	/*
288 	 * Flush all dirty buffers associated with a vnode.
289 	 */
290 	error = 0;
291 	passes = 0;
292 	wait = false;	/* Always do an async pass first. */
293 	unlocked = false;
294 	lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1));
295 	BO_LOCK(bo);
296 loop:
297 	TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
298 		bp->b_vflags &= ~BV_SCANNED;
299 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
300 		/*
301 		 * Reasons to skip this buffer: it has already been considered
302 		 * on this pass, the buffer has dependencies that will cause
303 		 * it to be redirtied and it has not already been deferred,
304 		 * or it is already being written.
305 		 */
306 		if ((bp->b_vflags & BV_SCANNED) != 0)
307 			continue;
308 		bp->b_vflags |= BV_SCANNED;
309 		/*
310 		 * Flush indirects in order, if requested.
311 		 *
312 		 * Note that if only datasync is requested, we can
313 		 * skip indirect blocks when softupdates are not
314 		 * active.  Otherwise we must flush them with data,
315 		 * since dependencies prevent data block writes.
316 		 */
317 		if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR &&
318 		    (lbn_level(bp->b_lblkno) >= passes ||
319 		    ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp))))
320 			continue;
321 		if (bp->b_lblkno > lbn)
322 			panic("ffs_syncvnode: syncing truncated data.");
323 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) {
324 			BO_UNLOCK(bo);
325 		} else if (wait) {
326 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
327 			    LK_INTERLOCK | wflag, BO_LOCKPTR(bo)) != 0) {
328 				BO_LOCK(bo);
329 				bp->b_vflags &= ~BV_SCANNED;
330 				goto next_locked;
331 			}
332 		} else
333 			continue;
334 		if ((bp->b_flags & B_DELWRI) == 0)
335 			panic("ffs_fsync: not dirty");
336 		/*
337 		 * Check for dependencies and potentially complete them.
338 		 */
339 		if (!LIST_EMPTY(&bp->b_dep) &&
340 		    (error = softdep_sync_buf(vp, bp,
341 		    wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
342 			/*
343 			 * Lock order conflict, buffer was already unlocked,
344 			 * and vnode possibly unlocked.
345 			 */
346 			if (error == ERELOOKUP) {
347 				if (vp->v_data == NULL)
348 					return (EBADF);
349 				unlocked = true;
350 				if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
351 				    (error = softdep_sync_metadata(vp)) != 0) {
352 					if (ffs_fsfail_cleanup(ump, error))
353 						error = 0;
354 					return (unlocked && error == 0 ?
355 					    ERELOOKUP : error);
356 				}
357 				/* Re-evaluate inode size */
358 				lbn = lblkno(ITOFS(ip), (ip->i_size +
359 				    ITOFS(ip)->fs_bsize - 1));
360 				goto next;
361 			}
362 			/* I/O error. */
363 			if (error != EBUSY) {
364 				BUF_UNLOCK(bp);
365 				return (error);
366 			}
367 			/* If we deferred once, don't defer again. */
368 		    	if ((bp->b_flags & B_DEFERRED) == 0) {
369 				bp->b_flags |= B_DEFERRED;
370 				BUF_UNLOCK(bp);
371 				goto next;
372 			}
373 		}
374 		if (wait) {
375 			bremfree(bp);
376 			error = bwrite(bp);
377 			if (ffs_fsfail_cleanup(ump, error))
378 				error = 0;
379 			if (error != 0)
380 				return (error);
381 		} else if ((bp->b_flags & B_CLUSTEROK)) {
382 			(void) vfs_bio_awrite(bp);
383 		} else {
384 			bremfree(bp);
385 			(void) bawrite(bp);
386 		}
387 next:
388 		/*
389 		 * Since we may have slept during the I/O, we need
390 		 * to start from a known point.
391 		 */
392 		BO_LOCK(bo);
393 next_locked:
394 		nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
395 	}
396 	if (waitfor != MNT_WAIT) {
397 		BO_UNLOCK(bo);
398 		if ((flags & NO_INO_UPDT) != 0)
399 			return (unlocked ? ERELOOKUP : 0);
400 		error = ffs_update(vp, 0);
401 		if (error == 0 && unlocked)
402 			error = ERELOOKUP;
403 		return (error);
404 	}
405 	/* Drain IO to see if we're done. */
406 	bufobj_wwait(bo, 0, 0);
407 	/*
408 	 * Block devices associated with filesystems may have new I/O
409 	 * requests posted for them even if the vnode is locked, so no
410 	 * amount of trying will get them clean.  We make several passes
411 	 * as a best effort.
412 	 *
413 	 * Regular files may need multiple passes to flush all dependency
414 	 * work as it is possible that we must write once per indirect
415 	 * level, once for the leaf, and once for the inode and each of
416 	 * these will be done with one sync and one async pass.
417 	 */
418 	if (bo->bo_dirty.bv_cnt > 0) {
419 		if ((flags & DATA_ONLY) == 0) {
420 			still_dirty = true;
421 		} else {
422 			/*
423 			 * For data-only sync, dirty indirect buffers
424 			 * are ignored.
425 			 */
426 			still_dirty = false;
427 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
428 				if (bp->b_lblkno > -UFS_NDADDR) {
429 					still_dirty = true;
430 					break;
431 				}
432 			}
433 		}
434 
435 		if (still_dirty) {
436 			/* Write the inode after sync passes to flush deps. */
437 			if (wait && DOINGSOFTDEP(vp) &&
438 			    (flags & NO_INO_UPDT) == 0) {
439 				BO_UNLOCK(bo);
440 				ffs_update(vp, 1);
441 				BO_LOCK(bo);
442 			}
443 			/* switch between sync/async. */
444 			wait = !wait;
445 			if (wait || ++passes < UFS_NIADDR + 2)
446 				goto loop;
447 		}
448 	}
449 	BO_UNLOCK(bo);
450 	error = 0;
451 	if ((flags & DATA_ONLY) == 0) {
452 		if ((flags & NO_INO_UPDT) == 0)
453 			error = ffs_update(vp, 1);
454 		if (DOINGSUJ(vp))
455 			softdep_journal_fsync(VTOI(vp));
456 	} else if ((ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)) != 0) {
457 		error = ffs_update(vp, 1);
458 	}
459 	if (error == 0 && unlocked)
460 		error = ERELOOKUP;
461 	if (error == 0)
462 		ip->i_flag &= ~IN_NEEDSYNC;
463 	return (error);
464 }
465 
466 static int
467 ffs_fdatasync(struct vop_fdatasync_args *ap)
468 {
469 
470 	return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY));
471 }
472 
473 static int
474 ffs_lock(ap)
475 	struct vop_lock1_args /* {
476 		struct vnode *a_vp;
477 		int a_flags;
478 		char *file;
479 		int line;
480 	} */ *ap;
481 {
482 #if !defined(NO_FFS_SNAPSHOT) || defined(DIAGNOSTIC)
483 	struct vnode *vp = ap->a_vp;
484 #endif	/* !NO_FFS_SNAPSHOT || DIAGNOSTIC */
485 #ifdef DIAGNOSTIC
486 	struct inode *ip;
487 #endif	/* DIAGNOSTIC */
488 	int result;
489 #ifndef NO_FFS_SNAPSHOT
490 	int flags;
491 	struct lock *lkp;
492 
493 	/*
494 	 * Adaptive spinning mixed with SU leads to trouble. use a giant hammer
495 	 * and only use it when LK_NODDLKTREAT is set. Currently this means it
496 	 * is only used during path lookup.
497 	 */
498 	if ((ap->a_flags & LK_NODDLKTREAT) != 0)
499 		ap->a_flags |= LK_ADAPTIVE;
500 	switch (ap->a_flags & LK_TYPE_MASK) {
501 	case LK_SHARED:
502 	case LK_UPGRADE:
503 	case LK_EXCLUSIVE:
504 		flags = ap->a_flags;
505 		for (;;) {
506 #ifdef DEBUG_VFS_LOCKS
507 			VNPASS(vp->v_holdcnt != 0, vp);
508 #endif	/* DEBUG_VFS_LOCKS */
509 			lkp = vp->v_vnlock;
510 			result = lockmgr_lock_flags(lkp, flags,
511 			    &VI_MTX(vp)->lock_object, ap->a_file, ap->a_line);
512 			if (lkp == vp->v_vnlock || result != 0)
513 				break;
514 			/*
515 			 * Apparent success, except that the vnode
516 			 * mutated between snapshot file vnode and
517 			 * regular file vnode while this process
518 			 * slept.  The lock currently held is not the
519 			 * right lock.  Release it, and try to get the
520 			 * new lock.
521 			 */
522 			lockmgr_unlock(lkp);
523 			if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
524 			    (LK_INTERLOCK | LK_NOWAIT))
525 				return (EBUSY);
526 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
527 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
528 			flags &= ~LK_INTERLOCK;
529 		}
530 #ifdef DIAGNOSTIC
531 		switch (ap->a_flags & LK_TYPE_MASK) {
532 		case LK_UPGRADE:
533 		case LK_EXCLUSIVE:
534 			if (result == 0 && vp->v_vnlock->lk_recurse == 0) {
535 				ip = VTOI(vp);
536 				if (ip != NULL)
537 					ip->i_lock_gen++;
538 			}
539 		}
540 #endif	/* DIAGNOSTIC */
541 		break;
542 	default:
543 #ifdef DIAGNOSTIC
544 		if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) {
545 			ip = VTOI(vp);
546 			if (ip != NULL)
547 				ufs_unlock_tracker(ip);
548 		}
549 #endif	/* DIAGNOSTIC */
550 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
551 		break;
552 	}
553 #else	/* NO_FFS_SNAPSHOT */
554 	/*
555 	 * See above for an explanation.
556 	 */
557 	if ((ap->a_flags & LK_NODDLKTREAT) != 0)
558 		ap->a_flags |= LK_ADAPTIVE;
559 #ifdef DIAGNOSTIC
560 	if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) {
561 		ip = VTOI(vp);
562 		if (ip != NULL)
563 			ufs_unlock_tracker(ip);
564 	}
565 #endif	/* DIAGNOSTIC */
566 	result =  VOP_LOCK1_APV(&ufs_vnodeops, ap);
567 #endif	/* NO_FFS_SNAPSHOT */
568 #ifdef DIAGNOSTIC
569 	switch (ap->a_flags & LK_TYPE_MASK) {
570 	case LK_UPGRADE:
571 	case LK_EXCLUSIVE:
572 		if (result == 0 && vp->v_vnlock->lk_recurse == 0) {
573 			ip = VTOI(vp);
574 			if (ip != NULL)
575 				ip->i_lock_gen++;
576 		}
577 	}
578 #endif	/* DIAGNOSTIC */
579 	return (result);
580 }
581 
582 #ifdef INVARIANTS
583 static int
584 ffs_unlock_debug(struct vop_unlock_args *ap)
585 {
586 	struct vnode *vp;
587 	struct inode *ip;
588 
589 	vp = ap->a_vp;
590 	ip = VTOI(vp);
591 	if (ip->i_flag & UFS_INODE_FLAG_LAZY_MASK_ASSERTABLE) {
592 		if ((vp->v_mflag & VMP_LAZYLIST) == 0) {
593 			VI_LOCK(vp);
594 			VNASSERT((vp->v_mflag & VMP_LAZYLIST), vp,
595 			    ("%s: modified vnode (%x) not on lazy list",
596 			    __func__, ip->i_flag));
597 			VI_UNLOCK(vp);
598 		}
599 	}
600 	KASSERT(vp->v_type != VDIR || vp->v_vnlock->lk_recurse != 0 ||
601 	    (ip->i_flag & IN_ENDOFF) == 0,
602 	    ("ufs dir vp %p ip %p flags %#x", vp, ip, ip->i_flag));
603 #ifdef DIAGNOSTIC
604 	if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && ip != NULL &&
605 	    vp->v_vnlock->lk_recurse == 0)
606 		ufs_unlock_tracker(ip);
607 #endif
608 	return (VOP_UNLOCK_APV(&ufs_vnodeops, ap));
609 }
610 #endif
611 
612 static int
613 ffs_read_hole(struct uio *uio, long xfersize, long *size)
614 {
615 	ssize_t saved_resid, tlen;
616 	int error;
617 
618 	while (xfersize > 0) {
619 		tlen = min(xfersize, ZERO_REGION_SIZE);
620 		saved_resid = uio->uio_resid;
621 		error = vn_io_fault_uiomove(__DECONST(void *, zero_region),
622 		    tlen, uio);
623 		if (error != 0)
624 			return (error);
625 		tlen = saved_resid - uio->uio_resid;
626 		xfersize -= tlen;
627 		*size -= tlen;
628 	}
629 	return (0);
630 }
631 
632 /*
633  * Vnode op for reading.
634  */
635 static int
636 ffs_read(ap)
637 	struct vop_read_args /* {
638 		struct vnode *a_vp;
639 		struct uio *a_uio;
640 		int a_ioflag;
641 		struct ucred *a_cred;
642 	} */ *ap;
643 {
644 	struct vnode *vp;
645 	struct inode *ip;
646 	struct uio *uio;
647 	struct fs *fs;
648 	struct buf *bp;
649 	ufs_lbn_t lbn, nextlbn;
650 	off_t bytesinfile;
651 	long size, xfersize, blkoffset;
652 	ssize_t orig_resid;
653 	int bflag, error, ioflag, seqcount;
654 
655 	vp = ap->a_vp;
656 	uio = ap->a_uio;
657 	ioflag = ap->a_ioflag;
658 	if (ap->a_ioflag & IO_EXT)
659 #ifdef notyet
660 		return (ffs_extread(vp, uio, ioflag));
661 #else
662 		panic("ffs_read+IO_EXT");
663 #endif
664 #ifdef DIRECTIO
665 	if ((ioflag & IO_DIRECT) != 0) {
666 		int workdone;
667 
668 		error = ffs_rawread(vp, uio, &workdone);
669 		if (error != 0 || workdone != 0)
670 			return error;
671 	}
672 #endif
673 
674 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
675 	ip = VTOI(vp);
676 
677 #ifdef INVARIANTS
678 	if (uio->uio_rw != UIO_READ)
679 		panic("ffs_read: mode");
680 
681 	if (vp->v_type == VLNK) {
682 		if ((int)ip->i_size < VFSTOUFS(vp->v_mount)->um_maxsymlinklen)
683 			panic("ffs_read: short symlink");
684 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
685 		panic("ffs_read: type %d",  vp->v_type);
686 #endif
687 	orig_resid = uio->uio_resid;
688 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
689 	if (orig_resid == 0)
690 		return (0);
691 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
692 	fs = ITOFS(ip);
693 	if (uio->uio_offset < ip->i_size &&
694 	    uio->uio_offset >= fs->fs_maxfilesize)
695 		return (EOVERFLOW);
696 
697 	bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE);
698 #ifdef WITNESS
699 	bflag |= IS_SNAPSHOT(ip) ? GB_NOWITNESS : 0;
700 #endif
701 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
702 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
703 			break;
704 		lbn = lblkno(fs, uio->uio_offset);
705 		nextlbn = lbn + 1;
706 
707 		/*
708 		 * size of buffer.  The buffer representing the
709 		 * end of the file is rounded up to the size of
710 		 * the block type ( fragment or full block,
711 		 * depending ).
712 		 */
713 		size = blksize(fs, ip, lbn);
714 		blkoffset = blkoff(fs, uio->uio_offset);
715 
716 		/*
717 		 * The amount we want to transfer in this iteration is
718 		 * one FS block less the amount of the data before
719 		 * our startpoint (duh!)
720 		 */
721 		xfersize = fs->fs_bsize - blkoffset;
722 
723 		/*
724 		 * But if we actually want less than the block,
725 		 * or the file doesn't have a whole block more of data,
726 		 * then use the lesser number.
727 		 */
728 		if (uio->uio_resid < xfersize)
729 			xfersize = uio->uio_resid;
730 		if (bytesinfile < xfersize)
731 			xfersize = bytesinfile;
732 
733 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
734 			/*
735 			 * Don't do readahead if this is the end of the file.
736 			 */
737 			error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
738 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
739 			/*
740 			 * Otherwise if we are allowed to cluster,
741 			 * grab as much as we can.
742 			 *
743 			 * XXX  This may not be a win if we are not
744 			 * doing sequential access.
745 			 */
746 			error = cluster_read(vp, ip->i_size, lbn,
747 			    size, NOCRED, blkoffset + uio->uio_resid,
748 			    seqcount, bflag, &bp);
749 		} else if (seqcount > 1) {
750 			/*
751 			 * If we are NOT allowed to cluster, then
752 			 * if we appear to be acting sequentially,
753 			 * fire off a request for a readahead
754 			 * as well as a read. Note that the 4th and 5th
755 			 * arguments point to arrays of the size specified in
756 			 * the 6th argument.
757 			 */
758 			u_int nextsize = blksize(fs, ip, nextlbn);
759 			error = breadn_flags(vp, lbn, lbn, size, &nextlbn,
760 			    &nextsize, 1, NOCRED, bflag, NULL, &bp);
761 		} else {
762 			/*
763 			 * Failing all of the above, just read what the
764 			 * user asked for. Interestingly, the same as
765 			 * the first option above.
766 			 */
767 			error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
768 		}
769 		if (error == EJUSTRETURN) {
770 			error = ffs_read_hole(uio, xfersize, &size);
771 			if (error == 0)
772 				continue;
773 		}
774 		if (error != 0) {
775 			brelse(bp);
776 			bp = NULL;
777 			break;
778 		}
779 
780 		/*
781 		 * We should only get non-zero b_resid when an I/O error
782 		 * has occurred, which should cause us to break above.
783 		 * However, if the short read did not cause an error,
784 		 * then we want to ensure that we do not uiomove bad
785 		 * or uninitialized data.
786 		 */
787 		size -= bp->b_resid;
788 		if (size < xfersize) {
789 			if (size == 0)
790 				break;
791 			xfersize = size;
792 		}
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 		if (error)
802 			break;
803 
804 		vfs_bio_brelse(bp, ioflag);
805 	}
806 
807 	/*
808 	 * This can only happen in the case of an error
809 	 * because the loop above resets bp to NULL on each iteration
810 	 * and on normal completion has not set a new value into it.
811 	 * so it must have come from a 'break' statement
812 	 */
813 	if (bp != NULL)
814 		vfs_bio_brelse(bp, ioflag);
815 
816 	if ((error == 0 || uio->uio_resid != orig_resid) &&
817 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
818 		UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS);
819 	return (error);
820 }
821 
822 /*
823  * Vnode op for writing.
824  */
825 static int
826 ffs_write(ap)
827 	struct vop_write_args /* {
828 		struct vnode *a_vp;
829 		struct uio *a_uio;
830 		int a_ioflag;
831 		struct ucred *a_cred;
832 	} */ *ap;
833 {
834 	struct vnode *vp;
835 	struct uio *uio;
836 	struct inode *ip;
837 	struct fs *fs;
838 	struct buf *bp;
839 	ufs_lbn_t lbn;
840 	off_t osize;
841 	ssize_t resid;
842 	int seqcount;
843 	int blkoffset, error, flags, ioflag, size, xfersize;
844 
845 	vp = ap->a_vp;
846 	if (DOINGSUJ(vp))
847 		softdep_prealloc(vp, MNT_WAIT);
848 	if (vp->v_data == NULL)
849 		return (EBADF);
850 
851 	uio = ap->a_uio;
852 	ioflag = ap->a_ioflag;
853 	if (ap->a_ioflag & IO_EXT)
854 #ifdef notyet
855 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
856 #else
857 		panic("ffs_write+IO_EXT");
858 #endif
859 
860 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
861 	ip = VTOI(vp);
862 
863 #ifdef INVARIANTS
864 	if (uio->uio_rw != UIO_WRITE)
865 		panic("ffs_write: mode");
866 #endif
867 
868 	switch (vp->v_type) {
869 	case VREG:
870 		if (ioflag & IO_APPEND)
871 			uio->uio_offset = ip->i_size;
872 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
873 			return (EPERM);
874 		/* FALLTHROUGH */
875 	case VLNK:
876 		break;
877 	case VDIR:
878 		panic("ffs_write: dir write");
879 		break;
880 	default:
881 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
882 			(int)uio->uio_offset,
883 			(int)uio->uio_resid
884 		);
885 	}
886 
887 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
888 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
889 	fs = ITOFS(ip);
890 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
891 		return (EFBIG);
892 	/*
893 	 * Maybe this should be above the vnode op call, but so long as
894 	 * file servers have no limits, I don't think it matters.
895 	 */
896 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
897 		return (EFBIG);
898 
899 	resid = uio->uio_resid;
900 	osize = ip->i_size;
901 	if (seqcount > BA_SEQMAX)
902 		flags = BA_SEQMAX << BA_SEQSHIFT;
903 	else
904 		flags = seqcount << BA_SEQSHIFT;
905 	if (ioflag & IO_SYNC)
906 		flags |= IO_SYNC;
907 	flags |= BA_UNMAPPED;
908 
909 	for (error = 0; uio->uio_resid > 0;) {
910 		lbn = lblkno(fs, uio->uio_offset);
911 		blkoffset = blkoff(fs, uio->uio_offset);
912 		xfersize = fs->fs_bsize - blkoffset;
913 		if (uio->uio_resid < xfersize)
914 			xfersize = uio->uio_resid;
915 		if (uio->uio_offset + xfersize > ip->i_size)
916 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
917 
918 		/*
919 		 * We must perform a read-before-write if the transfer size
920 		 * does not cover the entire buffer.
921 		 */
922 		if (fs->fs_bsize > xfersize)
923 			flags |= BA_CLRBUF;
924 		else
925 			flags &= ~BA_CLRBUF;
926 /* XXX is uio->uio_offset the right thing here? */
927 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
928 		    ap->a_cred, flags, &bp);
929 		if (error != 0) {
930 			vnode_pager_setsize(vp, ip->i_size);
931 			break;
932 		}
933 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
934 			bp->b_flags |= B_NOCACHE;
935 
936 		if (uio->uio_offset + xfersize > ip->i_size) {
937 			ip->i_size = uio->uio_offset + xfersize;
938 			DIP_SET(ip, i_size, ip->i_size);
939 			UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
940 		}
941 
942 		size = blksize(fs, ip, lbn) - bp->b_resid;
943 		if (size < xfersize)
944 			xfersize = size;
945 
946 		if (buf_mapped(bp)) {
947 			error = vn_io_fault_uiomove((char *)bp->b_data +
948 			    blkoffset, (int)xfersize, uio);
949 		} else {
950 			error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
951 			    (int)xfersize, uio);
952 		}
953 		/*
954 		 * If the buffer is not already filled and we encounter an
955 		 * error while trying to fill it, we have to clear out any
956 		 * garbage data from the pages instantiated for the buffer.
957 		 * If we do not, a failed uiomove() during a write can leave
958 		 * the prior contents of the pages exposed to a userland mmap.
959 		 *
960 		 * Note that we need only clear buffers with a transfer size
961 		 * equal to the block size because buffers with a shorter
962 		 * transfer size were cleared above by the call to UFS_BALLOC()
963 		 * with the BA_CLRBUF flag set.
964 		 *
965 		 * If the source region for uiomove identically mmaps the
966 		 * buffer, uiomove() performed the NOP copy, and the buffer
967 		 * content remains valid because the page fault handler
968 		 * validated the pages.
969 		 */
970 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
971 		    fs->fs_bsize == xfersize)
972 			vfs_bio_clrbuf(bp);
973 
974 		vfs_bio_set_flags(bp, ioflag);
975 
976 		/*
977 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
978 		 * if we have a severe page deficiency write the buffer
979 		 * asynchronously.  Otherwise try to cluster, and if that
980 		 * doesn't do it then either do an async write (if O_DIRECT),
981 		 * or a delayed write (if not).
982 		 */
983 		if (ioflag & IO_SYNC) {
984 			(void)bwrite(bp);
985 		} else if (vm_page_count_severe() ||
986 			    buf_dirty_count_severe() ||
987 			    (ioflag & IO_ASYNC)) {
988 			bp->b_flags |= B_CLUSTEROK;
989 			bawrite(bp);
990 		} else if (xfersize + blkoffset == fs->fs_bsize) {
991 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
992 				bp->b_flags |= B_CLUSTEROK;
993 				cluster_write(vp, &ip->i_clusterw, bp,
994 				    ip->i_size, seqcount, GB_UNMAPPED);
995 			} else {
996 				bawrite(bp);
997 			}
998 		} else if (ioflag & IO_DIRECT) {
999 			bp->b_flags |= B_CLUSTEROK;
1000 			bawrite(bp);
1001 		} else {
1002 			bp->b_flags |= B_CLUSTEROK;
1003 			bdwrite(bp);
1004 		}
1005 		if (error || xfersize == 0)
1006 			break;
1007 		UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE);
1008 	}
1009 	/*
1010 	 * If we successfully wrote any data, and we are not the superuser
1011 	 * we clear the setuid and setgid bits as a precaution against
1012 	 * tampering.
1013 	 */
1014 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
1015 	    ap->a_cred) {
1016 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) {
1017 			vn_seqc_write_begin(vp);
1018 			UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1019 			DIP_SET(ip, i_mode, ip->i_mode);
1020 			vn_seqc_write_end(vp);
1021 		}
1022 	}
1023 	if (error) {
1024 		if (ioflag & IO_UNIT) {
1025 			(void)ffs_truncate(vp, osize,
1026 			    IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
1027 			uio->uio_offset -= resid - uio->uio_resid;
1028 			uio->uio_resid = resid;
1029 		}
1030 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
1031 		if (!(ioflag & IO_DATASYNC) ||
1032 		    (ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)))
1033 			error = ffs_update(vp, 1);
1034 		if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error))
1035 			error = ENXIO;
1036 	}
1037 	return (error);
1038 }
1039 
1040 /*
1041  * Extended attribute area reading.
1042  */
1043 static int
1044 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
1045 {
1046 	struct inode *ip;
1047 	struct ufs2_dinode *dp;
1048 	struct fs *fs;
1049 	struct buf *bp;
1050 	ufs_lbn_t lbn, nextlbn;
1051 	off_t bytesinfile;
1052 	long size, xfersize, blkoffset;
1053 	ssize_t orig_resid;
1054 	int error;
1055 
1056 	ip = VTOI(vp);
1057 	fs = ITOFS(ip);
1058 	dp = ip->i_din2;
1059 
1060 #ifdef INVARIANTS
1061 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
1062 		panic("ffs_extread: mode");
1063 
1064 #endif
1065 	orig_resid = uio->uio_resid;
1066 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
1067 	if (orig_resid == 0)
1068 		return (0);
1069 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
1070 
1071 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1072 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
1073 			break;
1074 		lbn = lblkno(fs, uio->uio_offset);
1075 		nextlbn = lbn + 1;
1076 
1077 		/*
1078 		 * size of buffer.  The buffer representing the
1079 		 * end of the file is rounded up to the size of
1080 		 * the block type ( fragment or full block,
1081 		 * depending ).
1082 		 */
1083 		size = sblksize(fs, dp->di_extsize, lbn);
1084 		blkoffset = blkoff(fs, uio->uio_offset);
1085 
1086 		/*
1087 		 * The amount we want to transfer in this iteration is
1088 		 * one FS block less the amount of the data before
1089 		 * our startpoint (duh!)
1090 		 */
1091 		xfersize = fs->fs_bsize - blkoffset;
1092 
1093 		/*
1094 		 * But if we actually want less than the block,
1095 		 * or the file doesn't have a whole block more of data,
1096 		 * then use the lesser number.
1097 		 */
1098 		if (uio->uio_resid < xfersize)
1099 			xfersize = uio->uio_resid;
1100 		if (bytesinfile < xfersize)
1101 			xfersize = bytesinfile;
1102 
1103 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
1104 			/*
1105 			 * Don't do readahead if this is the end of the info.
1106 			 */
1107 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
1108 		} else {
1109 			/*
1110 			 * If we have a second block, then
1111 			 * fire off a request for a readahead
1112 			 * as well as a read. Note that the 4th and 5th
1113 			 * arguments point to arrays of the size specified in
1114 			 * the 6th argument.
1115 			 */
1116 			u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
1117 
1118 			nextlbn = -1 - nextlbn;
1119 			error = breadn(vp, -1 - lbn,
1120 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1121 		}
1122 		if (error) {
1123 			brelse(bp);
1124 			bp = NULL;
1125 			break;
1126 		}
1127 
1128 		/*
1129 		 * We should only get non-zero b_resid when an I/O error
1130 		 * has occurred, which should cause us to break above.
1131 		 * However, if the short read did not cause an error,
1132 		 * then we want to ensure that we do not uiomove bad
1133 		 * or uninitialized data.
1134 		 */
1135 		size -= bp->b_resid;
1136 		if (size < xfersize) {
1137 			if (size == 0)
1138 				break;
1139 			xfersize = size;
1140 		}
1141 
1142 		error = uiomove((char *)bp->b_data + blkoffset,
1143 					(int)xfersize, uio);
1144 		if (error)
1145 			break;
1146 		vfs_bio_brelse(bp, ioflag);
1147 	}
1148 
1149 	/*
1150 	 * This can only happen in the case of an error
1151 	 * because the loop above resets bp to NULL on each iteration
1152 	 * and on normal completion has not set a new value into it.
1153 	 * so it must have come from a 'break' statement
1154 	 */
1155 	if (bp != NULL)
1156 		vfs_bio_brelse(bp, ioflag);
1157 	return (error);
1158 }
1159 
1160 /*
1161  * Extended attribute area writing.
1162  */
1163 static int
1164 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1165 {
1166 	struct inode *ip;
1167 	struct ufs2_dinode *dp;
1168 	struct fs *fs;
1169 	struct buf *bp;
1170 	ufs_lbn_t lbn;
1171 	off_t osize;
1172 	ssize_t resid;
1173 	int blkoffset, error, flags, size, xfersize;
1174 
1175 	ip = VTOI(vp);
1176 	fs = ITOFS(ip);
1177 	dp = ip->i_din2;
1178 
1179 #ifdef INVARIANTS
1180 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1181 		panic("ffs_extwrite: mode");
1182 #endif
1183 
1184 	if (ioflag & IO_APPEND)
1185 		uio->uio_offset = dp->di_extsize;
1186 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1187 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1188 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
1189 	    UFS_NXADDR * fs->fs_bsize)
1190 		return (EFBIG);
1191 
1192 	resid = uio->uio_resid;
1193 	osize = dp->di_extsize;
1194 	flags = IO_EXT;
1195 	if (ioflag & IO_SYNC)
1196 		flags |= IO_SYNC;
1197 
1198 	for (error = 0; uio->uio_resid > 0;) {
1199 		lbn = lblkno(fs, uio->uio_offset);
1200 		blkoffset = blkoff(fs, uio->uio_offset);
1201 		xfersize = fs->fs_bsize - blkoffset;
1202 		if (uio->uio_resid < xfersize)
1203 			xfersize = uio->uio_resid;
1204 
1205 		/*
1206 		 * We must perform a read-before-write if the transfer size
1207 		 * does not cover the entire buffer.
1208 		 */
1209 		if (fs->fs_bsize > xfersize)
1210 			flags |= BA_CLRBUF;
1211 		else
1212 			flags &= ~BA_CLRBUF;
1213 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1214 		    ucred, flags, &bp);
1215 		if (error != 0)
1216 			break;
1217 		/*
1218 		 * If the buffer is not valid we have to clear out any
1219 		 * garbage data from the pages instantiated for the buffer.
1220 		 * If we do not, a failed uiomove() during a write can leave
1221 		 * the prior contents of the pages exposed to a userland
1222 		 * mmap().  XXX deal with uiomove() errors a better way.
1223 		 */
1224 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1225 			vfs_bio_clrbuf(bp);
1226 
1227 		if (uio->uio_offset + xfersize > dp->di_extsize) {
1228 			dp->di_extsize = uio->uio_offset + xfersize;
1229 			UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
1230 		}
1231 
1232 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1233 		if (size < xfersize)
1234 			xfersize = size;
1235 
1236 		error =
1237 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1238 
1239 		vfs_bio_set_flags(bp, ioflag);
1240 
1241 		/*
1242 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1243 		 * if we have a severe page deficiency write the buffer
1244 		 * asynchronously.  Otherwise try to cluster, and if that
1245 		 * doesn't do it then either do an async write (if O_DIRECT),
1246 		 * or a delayed write (if not).
1247 		 */
1248 		if (ioflag & IO_SYNC) {
1249 			(void)bwrite(bp);
1250 		} else if (vm_page_count_severe() ||
1251 			    buf_dirty_count_severe() ||
1252 			    xfersize + blkoffset == fs->fs_bsize ||
1253 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1254 			bawrite(bp);
1255 		else
1256 			bdwrite(bp);
1257 		if (error || xfersize == 0)
1258 			break;
1259 		UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1260 	}
1261 	/*
1262 	 * If we successfully wrote any data, and we are not the superuser
1263 	 * we clear the setuid and setgid bits as a precaution against
1264 	 * tampering.
1265 	 */
1266 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1267 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID)) {
1268 			vn_seqc_write_begin(vp);
1269 			UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1270 			dp->di_mode = ip->i_mode;
1271 			vn_seqc_write_end(vp);
1272 		}
1273 	}
1274 	if (error) {
1275 		if (ioflag & IO_UNIT) {
1276 			(void)ffs_truncate(vp, osize,
1277 			    IO_EXT | (ioflag&IO_SYNC), ucred);
1278 			uio->uio_offset -= resid - uio->uio_resid;
1279 			uio->uio_resid = resid;
1280 		}
1281 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1282 		error = ffs_update(vp, 1);
1283 	return (error);
1284 }
1285 
1286 /*
1287  * Vnode operating to retrieve a named extended attribute.
1288  *
1289  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1290  * the length of the EA, and possibly the pointer to the entry and to the data.
1291  */
1292 static int
1293 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name,
1294     struct extattr **eapp, u_char **eac)
1295 {
1296 	struct extattr *eap, *eaend;
1297 	size_t nlen;
1298 
1299 	nlen = strlen(name);
1300 	KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned"));
1301 	eap = (struct extattr *)ptr;
1302 	eaend = (struct extattr *)(ptr + length);
1303 	for (; eap < eaend; eap = EXTATTR_NEXT(eap)) {
1304 		KASSERT(EXTATTR_NEXT(eap) <= eaend,
1305 		    ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend));
1306 		if (eap->ea_namespace != nspace || eap->ea_namelength != nlen
1307 		    || memcmp(eap->ea_name, name, nlen) != 0)
1308 			continue;
1309 		if (eapp != NULL)
1310 			*eapp = eap;
1311 		if (eac != NULL)
1312 			*eac = EXTATTR_CONTENT(eap);
1313 		return (EXTATTR_CONTENT_SIZE(eap));
1314 	}
1315 	return (-1);
1316 }
1317 
1318 static int
1319 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td)
1320 {
1321 	const struct extattr *eap, *eaend, *eapnext;
1322 	struct inode *ip;
1323 	struct ufs2_dinode *dp;
1324 	struct fs *fs;
1325 	struct uio luio;
1326 	struct iovec liovec;
1327 	u_int easize;
1328 	int error;
1329 	u_char *eae;
1330 
1331 	ip = VTOI(vp);
1332 	fs = ITOFS(ip);
1333 	dp = ip->i_din2;
1334 	easize = dp->di_extsize;
1335 	if ((uoff_t)easize > UFS_NXADDR * fs->fs_bsize)
1336 		return (EFBIG);
1337 
1338 	eae = malloc(easize, M_TEMP, M_WAITOK);
1339 
1340 	liovec.iov_base = eae;
1341 	liovec.iov_len = easize;
1342 	luio.uio_iov = &liovec;
1343 	luio.uio_iovcnt = 1;
1344 	luio.uio_offset = 0;
1345 	luio.uio_resid = easize;
1346 	luio.uio_segflg = UIO_SYSSPACE;
1347 	luio.uio_rw = UIO_READ;
1348 	luio.uio_td = td;
1349 
1350 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1351 	if (error) {
1352 		free(eae, M_TEMP);
1353 		return (error);
1354 	}
1355 	/* Validate disk xattrfile contents. */
1356 	for (eap = (void *)eae, eaend = (void *)(eae + easize); eap < eaend;
1357 	    eap = eapnext) {
1358 		/* Detect zeroed out tail */
1359 		if (eap->ea_length < sizeof(*eap) || eap->ea_length == 0) {
1360 			easize = (const u_char *)eap - eae;
1361 			break;
1362 		}
1363 
1364 		eapnext = EXTATTR_NEXT(eap);
1365 		/* Bogusly long entry. */
1366 		if (eapnext > eaend) {
1367 			free(eae, M_TEMP);
1368 			return (EINTEGRITY);
1369 		}
1370 	}
1371 	ip->i_ea_len = easize;
1372 	*p = eae;
1373 	return (0);
1374 }
1375 
1376 static void
1377 ffs_lock_ea(struct vnode *vp)
1378 {
1379 	struct inode *ip;
1380 
1381 	ip = VTOI(vp);
1382 	VI_LOCK(vp);
1383 	while (ip->i_flag & IN_EA_LOCKED) {
1384 		UFS_INODE_SET_FLAG(ip, IN_EA_LOCKWAIT);
1385 		msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1386 		    0);
1387 	}
1388 	UFS_INODE_SET_FLAG(ip, IN_EA_LOCKED);
1389 	VI_UNLOCK(vp);
1390 }
1391 
1392 static void
1393 ffs_unlock_ea(struct vnode *vp)
1394 {
1395 	struct inode *ip;
1396 
1397 	ip = VTOI(vp);
1398 	VI_LOCK(vp);
1399 	if (ip->i_flag & IN_EA_LOCKWAIT)
1400 		wakeup(&ip->i_ea_refs);
1401 	ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1402 	VI_UNLOCK(vp);
1403 }
1404 
1405 static int
1406 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1407 {
1408 	struct inode *ip;
1409 	int error;
1410 
1411 	ip = VTOI(vp);
1412 
1413 	ffs_lock_ea(vp);
1414 	if (ip->i_ea_area != NULL) {
1415 		ip->i_ea_refs++;
1416 		ffs_unlock_ea(vp);
1417 		return (0);
1418 	}
1419 	error = ffs_rdextattr(&ip->i_ea_area, vp, td);
1420 	if (error) {
1421 		ffs_unlock_ea(vp);
1422 		return (error);
1423 	}
1424 	ip->i_ea_error = 0;
1425 	ip->i_ea_refs++;
1426 	ffs_unlock_ea(vp);
1427 	return (0);
1428 }
1429 
1430 /*
1431  * Vnode extattr transaction commit/abort
1432  */
1433 static int
1434 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1435 {
1436 	struct inode *ip;
1437 	struct uio luio;
1438 	struct iovec *liovec;
1439 	struct ufs2_dinode *dp;
1440 	size_t ea_len, tlen;
1441 	int error, i, lcnt;
1442 	bool truncate;
1443 
1444 	ip = VTOI(vp);
1445 
1446 	ffs_lock_ea(vp);
1447 	if (ip->i_ea_area == NULL) {
1448 		ffs_unlock_ea(vp);
1449 		return (EINVAL);
1450 	}
1451 	dp = ip->i_din2;
1452 	error = ip->i_ea_error;
1453 	truncate = false;
1454 	if (commit && error == 0) {
1455 		ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1456 		if (cred == NOCRED)
1457 			cred =  vp->v_mount->mnt_cred;
1458 
1459 		ea_len = MAX(ip->i_ea_len, dp->di_extsize);
1460 		for (lcnt = 1, tlen = ea_len - ip->i_ea_len; tlen > 0;) {
1461 			tlen -= MIN(ZERO_REGION_SIZE, tlen);
1462 			lcnt++;
1463 		}
1464 
1465 		liovec = __builtin_alloca(lcnt * sizeof(struct iovec));
1466 		luio.uio_iovcnt = lcnt;
1467 
1468 		liovec[0].iov_base = ip->i_ea_area;
1469 		liovec[0].iov_len = ip->i_ea_len;
1470 		for (i = 1, tlen = ea_len - ip->i_ea_len; i < lcnt; i++) {
1471 			liovec[i].iov_base = __DECONST(void *, zero_region);
1472 			liovec[i].iov_len = MIN(ZERO_REGION_SIZE, tlen);
1473 			tlen -= liovec[i].iov_len;
1474 		}
1475 		MPASS(tlen == 0);
1476 
1477 		luio.uio_iov = liovec;
1478 		luio.uio_offset = 0;
1479 		luio.uio_resid = ea_len;
1480 		luio.uio_segflg = UIO_SYSSPACE;
1481 		luio.uio_rw = UIO_WRITE;
1482 		luio.uio_td = td;
1483 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1484 		if (error == 0 && ip->i_ea_len == 0)
1485 			truncate = true;
1486 	}
1487 	if (--ip->i_ea_refs == 0) {
1488 		free(ip->i_ea_area, M_TEMP);
1489 		ip->i_ea_area = NULL;
1490 		ip->i_ea_len = 0;
1491 		ip->i_ea_error = 0;
1492 	}
1493 	ffs_unlock_ea(vp);
1494 
1495 	if (truncate)
1496 		ffs_truncate(vp, 0, IO_EXT, cred);
1497 	return (error);
1498 }
1499 
1500 /*
1501  * Vnode extattr strategy routine for fifos.
1502  *
1503  * We need to check for a read or write of the external attributes.
1504  * Otherwise we just fall through and do the usual thing.
1505  */
1506 static int
1507 ffsext_strategy(struct vop_strategy_args *ap)
1508 /*
1509 struct vop_strategy_args {
1510 	struct vnodeop_desc *a_desc;
1511 	struct vnode *a_vp;
1512 	struct buf *a_bp;
1513 };
1514 */
1515 {
1516 	struct vnode *vp;
1517 	daddr_t lbn;
1518 
1519 	vp = ap->a_vp;
1520 	lbn = ap->a_bp->b_lblkno;
1521 	if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR)
1522 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1523 	if (vp->v_type == VFIFO)
1524 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1525 	panic("spec nodes went here");
1526 }
1527 
1528 /*
1529  * Vnode extattr transaction commit/abort
1530  */
1531 static int
1532 ffs_openextattr(struct vop_openextattr_args *ap)
1533 /*
1534 struct vop_openextattr_args {
1535 	struct vnodeop_desc *a_desc;
1536 	struct vnode *a_vp;
1537 	IN struct ucred *a_cred;
1538 	IN struct thread *a_td;
1539 };
1540 */
1541 {
1542 
1543 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1544 		return (EOPNOTSUPP);
1545 
1546 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1547 }
1548 
1549 /*
1550  * Vnode extattr transaction commit/abort
1551  */
1552 static int
1553 ffs_closeextattr(struct vop_closeextattr_args *ap)
1554 /*
1555 struct vop_closeextattr_args {
1556 	struct vnodeop_desc *a_desc;
1557 	struct vnode *a_vp;
1558 	int a_commit;
1559 	IN struct ucred *a_cred;
1560 	IN struct thread *a_td;
1561 };
1562 */
1563 {
1564 	struct vnode *vp;
1565 
1566 	vp = ap->a_vp;
1567 	if (vp->v_type == VCHR || vp->v_type == VBLK)
1568 		return (EOPNOTSUPP);
1569 	if (ap->a_commit && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0)
1570 		return (EROFS);
1571 
1572 	if (ap->a_commit && DOINGSUJ(vp)) {
1573 		ASSERT_VOP_ELOCKED(vp, "ffs_closeextattr commit");
1574 		softdep_prealloc(vp, MNT_WAIT);
1575 		if (vp->v_data == NULL)
1576 			return (EBADF);
1577 	}
1578 	return (ffs_close_ea(vp, ap->a_commit, ap->a_cred, ap->a_td));
1579 }
1580 
1581 /*
1582  * Vnode operation to remove a named attribute.
1583  */
1584 static int
1585 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1586 /*
1587 vop_deleteextattr {
1588 	IN struct vnode *a_vp;
1589 	IN int a_attrnamespace;
1590 	IN const char *a_name;
1591 	IN struct ucred *a_cred;
1592 	IN struct thread *a_td;
1593 };
1594 */
1595 {
1596 	struct vnode *vp;
1597 	struct inode *ip;
1598 	struct extattr *eap;
1599 	uint32_t ul;
1600 	int olen, error, i, easize;
1601 	u_char *eae;
1602 	void *tmp;
1603 
1604 	vp = ap->a_vp;
1605 	ip = VTOI(vp);
1606 
1607 	if (vp->v_type == VCHR || vp->v_type == VBLK)
1608 		return (EOPNOTSUPP);
1609 	if (strlen(ap->a_name) == 0)
1610 		return (EINVAL);
1611 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1612 		return (EROFS);
1613 
1614 	error = extattr_check_cred(vp, ap->a_attrnamespace,
1615 	    ap->a_cred, ap->a_td, VWRITE);
1616 	if (error) {
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 	if (DOINGSUJ(vp)) {
1627 		ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr");
1628 		softdep_prealloc(vp, MNT_WAIT);
1629 		if (vp->v_data == NULL)
1630 			return (EBADF);
1631 	}
1632 
1633 	error = ffs_open_ea(vp, ap->a_cred, ap->a_td);
1634 	if (error)
1635 		return (error);
1636 
1637 	/* CEM: delete could be done in-place instead */
1638 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1639 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1640 	easize = ip->i_ea_len;
1641 
1642 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1643 	    &eap, NULL);
1644 	if (olen == -1) {
1645 		/* delete but nonexistent */
1646 		free(eae, M_TEMP);
1647 		ffs_close_ea(vp, 0, ap->a_cred, ap->a_td);
1648 		return (ENOATTR);
1649 	}
1650 	ul = eap->ea_length;
1651 	i = (u_char *)EXTATTR_NEXT(eap) - eae;
1652 	bcopy(EXTATTR_NEXT(eap), eap, easize - i);
1653 	easize -= ul;
1654 
1655 	tmp = ip->i_ea_area;
1656 	ip->i_ea_area = eae;
1657 	ip->i_ea_len = easize;
1658 	free(tmp, M_TEMP);
1659 	error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td);
1660 	return (error);
1661 }
1662 
1663 /*
1664  * Vnode operation to retrieve a named extended attribute.
1665  */
1666 static int
1667 ffs_getextattr(struct vop_getextattr_args *ap)
1668 /*
1669 vop_getextattr {
1670 	IN struct vnode *a_vp;
1671 	IN int a_attrnamespace;
1672 	IN const char *a_name;
1673 	INOUT struct uio *a_uio;
1674 	OUT size_t *a_size;
1675 	IN struct ucred *a_cred;
1676 	IN struct thread *a_td;
1677 };
1678 */
1679 {
1680 	struct inode *ip;
1681 	u_char *eae, *p;
1682 	unsigned easize;
1683 	int error, ealen;
1684 
1685 	ip = VTOI(ap->a_vp);
1686 
1687 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1688 		return (EOPNOTSUPP);
1689 
1690 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1691 	    ap->a_cred, ap->a_td, VREAD);
1692 	if (error)
1693 		return (error);
1694 
1695 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1696 	if (error)
1697 		return (error);
1698 
1699 	eae = ip->i_ea_area;
1700 	easize = ip->i_ea_len;
1701 
1702 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1703 	    NULL, &p);
1704 	if (ealen >= 0) {
1705 		error = 0;
1706 		if (ap->a_size != NULL)
1707 			*ap->a_size = ealen;
1708 		else if (ap->a_uio != NULL)
1709 			error = uiomove(p, ealen, ap->a_uio);
1710 	} else
1711 		error = ENOATTR;
1712 
1713 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1714 	return (error);
1715 }
1716 
1717 /*
1718  * Vnode operation to retrieve extended attributes on a vnode.
1719  */
1720 static int
1721 ffs_listextattr(struct vop_listextattr_args *ap)
1722 /*
1723 vop_listextattr {
1724 	IN struct vnode *a_vp;
1725 	IN int a_attrnamespace;
1726 	INOUT struct uio *a_uio;
1727 	OUT size_t *a_size;
1728 	IN struct ucred *a_cred;
1729 	IN struct thread *a_td;
1730 };
1731 */
1732 {
1733 	struct inode *ip;
1734 	struct extattr *eap, *eaend;
1735 	int error, ealen;
1736 
1737 	ip = VTOI(ap->a_vp);
1738 
1739 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1740 		return (EOPNOTSUPP);
1741 
1742 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1743 	    ap->a_cred, ap->a_td, VREAD);
1744 	if (error)
1745 		return (error);
1746 
1747 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1748 	if (error)
1749 		return (error);
1750 
1751 	error = 0;
1752 	if (ap->a_size != NULL)
1753 		*ap->a_size = 0;
1754 
1755 	KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned"));
1756 	eap = (struct extattr *)ip->i_ea_area;
1757 	eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len);
1758 	for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) {
1759 		KASSERT(EXTATTR_NEXT(eap) <= eaend,
1760 		    ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend));
1761 		if (eap->ea_namespace != ap->a_attrnamespace)
1762 			continue;
1763 
1764 		ealen = eap->ea_namelength;
1765 		if (ap->a_size != NULL)
1766 			*ap->a_size += ealen + 1;
1767 		else if (ap->a_uio != NULL)
1768 			error = uiomove(&eap->ea_namelength, ealen + 1,
1769 			    ap->a_uio);
1770 	}
1771 
1772 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1773 	return (error);
1774 }
1775 
1776 /*
1777  * Vnode operation to set a named attribute.
1778  */
1779 static int
1780 ffs_setextattr(struct vop_setextattr_args *ap)
1781 /*
1782 vop_setextattr {
1783 	IN struct vnode *a_vp;
1784 	IN int a_attrnamespace;
1785 	IN const char *a_name;
1786 	INOUT struct uio *a_uio;
1787 	IN struct ucred *a_cred;
1788 	IN struct thread *a_td;
1789 };
1790 */
1791 {
1792 	struct vnode *vp;
1793 	struct inode *ip;
1794 	struct fs *fs;
1795 	struct extattr *eap;
1796 	uint32_t ealength, ul;
1797 	ssize_t ealen;
1798 	int olen, eapad1, eapad2, error, i, easize;
1799 	u_char *eae;
1800 	void *tmp;
1801 
1802 	vp = ap->a_vp;
1803 	ip = VTOI(vp);
1804 	fs = ITOFS(ip);
1805 
1806 	if (vp->v_type == VCHR || vp->v_type == VBLK)
1807 		return (EOPNOTSUPP);
1808 	if (strlen(ap->a_name) == 0)
1809 		return (EINVAL);
1810 
1811 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1812 	if (ap->a_uio == NULL)
1813 		return (EOPNOTSUPP);
1814 
1815 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1816 		return (EROFS);
1817 
1818 	ealen = ap->a_uio->uio_resid;
1819 	if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR))
1820 		return (EINVAL);
1821 
1822 	error = extattr_check_cred(vp, ap->a_attrnamespace,
1823 	    ap->a_cred, ap->a_td, VWRITE);
1824 	if (error) {
1825 		/*
1826 		 * ffs_lock_ea is not needed there, because the vnode
1827 		 * must be exclusively locked.
1828 		 */
1829 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1830 			ip->i_ea_error = error;
1831 		return (error);
1832 	}
1833 
1834 	if (DOINGSUJ(vp)) {
1835 		ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr");
1836 		softdep_prealloc(vp, MNT_WAIT);
1837 		if (vp->v_data == NULL)
1838 			return (EBADF);
1839 	}
1840 
1841 	error = ffs_open_ea(vp, ap->a_cred, ap->a_td);
1842 	if (error)
1843 		return (error);
1844 
1845 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1846 	eapad1 = roundup2(ealength, 8) - ealength;
1847 	eapad2 = roundup2(ealen, 8) - ealen;
1848 	ealength += eapad1 + ealen + eapad2;
1849 
1850 	/*
1851 	 * CEM: rewrites of the same size or smaller could be done in-place
1852 	 * instead.  (We don't acquire any fine-grained locks in here either,
1853 	 * so we could also do bigger writes in-place.)
1854 	 */
1855 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1856 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1857 	easize = ip->i_ea_len;
1858 
1859 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1860 	    &eap, NULL);
1861         if (olen == -1) {
1862 		/* new, append at end */
1863 		KASSERT(ALIGNED_TO(eae + easize, struct extattr),
1864 		    ("unaligned"));
1865 		eap = (struct extattr *)(eae + easize);
1866 		easize += ealength;
1867 	} else {
1868 		ul = eap->ea_length;
1869 		i = (u_char *)EXTATTR_NEXT(eap) - eae;
1870 		if (ul != ealength) {
1871 			bcopy(EXTATTR_NEXT(eap), (u_char *)eap + ealength,
1872 			    easize - i);
1873 			easize += (ealength - ul);
1874 		}
1875 	}
1876 	if (easize > lblktosize(fs, UFS_NXADDR)) {
1877 		free(eae, M_TEMP);
1878 		ffs_close_ea(vp, 0, ap->a_cred, ap->a_td);
1879 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1880 			ip->i_ea_error = ENOSPC;
1881 		return (ENOSPC);
1882 	}
1883 	eap->ea_length = ealength;
1884 	eap->ea_namespace = ap->a_attrnamespace;
1885 	eap->ea_contentpadlen = eapad2;
1886 	eap->ea_namelength = strlen(ap->a_name);
1887 	memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name));
1888 	bzero(&eap->ea_name[strlen(ap->a_name)], eapad1);
1889 	error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio);
1890 	if (error) {
1891 		free(eae, M_TEMP);
1892 		ffs_close_ea(vp, 0, ap->a_cred, ap->a_td);
1893 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1894 			ip->i_ea_error = error;
1895 		return (error);
1896 	}
1897 	bzero((u_char *)EXTATTR_CONTENT(eap) + ealen, eapad2);
1898 
1899 	tmp = ip->i_ea_area;
1900 	ip->i_ea_area = eae;
1901 	ip->i_ea_len = easize;
1902 	free(tmp, M_TEMP);
1903 	error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td);
1904 	return (error);
1905 }
1906 
1907 /*
1908  * Vnode pointer to File handle
1909  */
1910 static int
1911 ffs_vptofh(struct vop_vptofh_args *ap)
1912 /*
1913 vop_vptofh {
1914 	IN struct vnode *a_vp;
1915 	IN struct fid *a_fhp;
1916 };
1917 */
1918 {
1919 	struct inode *ip;
1920 	struct ufid *ufhp;
1921 
1922 	ip = VTOI(ap->a_vp);
1923 	ufhp = (struct ufid *)ap->a_fhp;
1924 	ufhp->ufid_len = sizeof(struct ufid);
1925 	ufhp->ufid_ino = ip->i_number;
1926 	ufhp->ufid_gen = ip->i_gen;
1927 	return (0);
1928 }
1929 
1930 SYSCTL_DECL(_vfs_ffs);
1931 static int use_buf_pager = 1;
1932 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0,
1933     "Always use buffer pager instead of bmap");
1934 
1935 static daddr_t
1936 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1937 {
1938 
1939 	return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off));
1940 }
1941 
1942 static int
1943 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *sz)
1944 {
1945 
1946 	*sz = blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn);
1947 	return (0);
1948 }
1949 
1950 static int
1951 ffs_getpages(struct vop_getpages_args *ap)
1952 {
1953 	struct vnode *vp;
1954 	struct ufsmount *um;
1955 
1956 	vp = ap->a_vp;
1957 	um = VFSTOUFS(vp->v_mount);
1958 
1959 	if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE)
1960 		return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1961 		    ap->a_rbehind, ap->a_rahead, NULL, NULL));
1962 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1963 	    ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz));
1964 }
1965 
1966 static int
1967 ffs_getpages_async(struct vop_getpages_async_args *ap)
1968 {
1969 	struct vnode *vp;
1970 	struct ufsmount *um;
1971 	bool do_iodone;
1972 	int error;
1973 
1974 	vp = ap->a_vp;
1975 	um = VFSTOUFS(vp->v_mount);
1976 	do_iodone = true;
1977 
1978 	if (um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) {
1979 		error = vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1980 		    ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg);
1981 		if (error == 0)
1982 			do_iodone = false;
1983 	} else {
1984 		error = vfs_bio_getpages(vp, ap->a_m, ap->a_count,
1985 		    ap->a_rbehind, ap->a_rahead, ffs_gbp_getblkno,
1986 		    ffs_gbp_getblksz);
1987 	}
1988 	if (do_iodone && ap->a_iodone != NULL)
1989 		ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);
1990 
1991 	return (error);
1992 }
1993 
1994 static int
1995 ffs_vput_pair(struct vop_vput_pair_args *ap)
1996 {
1997 	struct mount *mp;
1998 	struct vnode *dvp, *vp, *vp1, **vpp;
1999 	struct inode *dp, *ip;
2000 	ino_t ip_ino;
2001 	u_int64_t ip_gen;
2002 	int error, vp_locked;
2003 
2004 	dvp = ap->a_dvp;
2005 	dp = VTOI(dvp);
2006 	vpp = ap->a_vpp;
2007 	vp = vpp != NULL ? *vpp : NULL;
2008 
2009 	if ((dp->i_flag & (IN_NEEDSYNC | IN_ENDOFF)) == 0) {
2010 		vput(dvp);
2011 		if (vp != NULL && ap->a_unlock_vp)
2012 			vput(vp);
2013 		return (0);
2014 	}
2015 
2016 	mp = dvp->v_mount;
2017 	if (vp != NULL) {
2018 		if (ap->a_unlock_vp) {
2019 			vput(vp);
2020 		} else {
2021 			MPASS(vp->v_type != VNON);
2022 			vp_locked = VOP_ISLOCKED(vp);
2023 			ip = VTOI(vp);
2024 			ip_ino = ip->i_number;
2025 			ip_gen = ip->i_gen;
2026 			VOP_UNLOCK(vp);
2027 		}
2028 	}
2029 
2030 	/*
2031 	 * If compaction or fsync was requested do it in ffs_vput_pair()
2032 	 * now that other locks are no longer held.
2033          */
2034 	if ((dp->i_flag & IN_ENDOFF) != 0) {
2035 		VNASSERT(I_ENDOFF(dp) != 0 && I_ENDOFF(dp) < dp->i_size, dvp,
2036 		    ("IN_ENDOFF set but I_ENDOFF() is not"));
2037 		dp->i_flag &= ~IN_ENDOFF;
2038 		error = UFS_TRUNCATE(dvp, (off_t)I_ENDOFF(dp), IO_NORMAL |
2039 		    (DOINGASYNC(dvp) ? 0 : IO_SYNC), curthread->td_ucred);
2040 		if (error != 0 && error != ERELOOKUP) {
2041 			if (!ffs_fsfail_cleanup(VFSTOUFS(mp), error)) {
2042 				vn_printf(dvp,
2043 				    "IN_ENDOFF: failed to truncate, "
2044 				    "error %d\n", error);
2045 			}
2046 #ifdef UFS_DIRHASH
2047 			ufsdirhash_free(dp);
2048 #endif
2049 		}
2050 		SET_I_ENDOFF(dp, 0);
2051 	}
2052 	if ((dp->i_flag & IN_NEEDSYNC) != 0) {
2053 		do {
2054 			error = ffs_syncvnode(dvp, MNT_WAIT, 0);
2055 		} while (error == ERELOOKUP);
2056 	}
2057 
2058 	vput(dvp);
2059 
2060 	if (vp == NULL || ap->a_unlock_vp)
2061 		return (0);
2062 	MPASS(mp != NULL);
2063 
2064 	/*
2065 	 * It is possible that vp is reclaimed at this point. Only
2066 	 * routines that call us with a_unlock_vp == false can find
2067 	 * that their vp has been reclaimed. There are three areas
2068 	 * that are affected:
2069 	 * 1) vn_open_cred() - later VOPs could fail, but
2070 	 *    dead_open() returns 0 to simulate successful open.
2071 	 * 2) ffs_snapshot() - creation of snapshot fails with EBADF.
2072 	 * 3) NFS server (several places) - code is prepared to detect
2073 	 *    and respond to dead vnodes by returning ESTALE.
2074 	 */
2075 	VOP_LOCK(vp, vp_locked | LK_RETRY);
2076 	if (IS_UFS(vp))
2077 		return (0);
2078 
2079 	/*
2080 	 * Try harder to recover from reclaimed vp if reclaim was not
2081 	 * because underlying inode was cleared.  We saved inode
2082 	 * number and inode generation, so we can try to reinstantiate
2083 	 * exactly same version of inode.  If this fails, return
2084 	 * original doomed vnode and let caller to handle
2085 	 * consequences.
2086 	 *
2087 	 * Note that callers must keep write started around
2088 	 * VOP_VPUT_PAIR() calls, so it is safe to use mp without
2089 	 * busying it.
2090 	 */
2091 	VOP_UNLOCK(vp);
2092 	error = ffs_inotovp(mp, ip_ino, ip_gen, LK_EXCLUSIVE, &vp1,
2093 	    FFSV_REPLACE_DOOMED);
2094 	if (error != 0) {
2095 		VOP_LOCK(vp, vp_locked | LK_RETRY);
2096 	} else {
2097 		vrele(vp);
2098 		*vpp = vp1;
2099 	}
2100 	return (error);
2101 }
2102