xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 51a9219f5780e61e1437d25220bf8750d9df7f8b)
1 /*
2  * Copyright (c) 2002 Networks Associates Technology, Inc.
3  * All rights reserved.
4  *
5  * This software was developed for the FreeBSD Project by Marshall
6  * Kirk McKusick and Network Associates Laboratories, the Security
7  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
8  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
9  * research program
10  *
11  * Copyright (c) 1982, 1986, 1989, 1993
12  *	The Regents of the University of California.  All rights reserved.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. All advertising materials mentioning features or use of this software
23  *    must display the following acknowledgement:
24  *	This product includes software developed by the University of
25  *	California, Berkeley and its contributors.
26  * 4. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)ffs_vnops.c	8.15 (Berkeley) 5/14/95
43  * $FreeBSD$
44  */
45 
46 #include <sys/param.h>
47 #include <sys/bio.h>
48 #include <sys/systm.h>
49 #include <sys/buf.h>
50 #include <sys/conf.h>
51 #include <sys/extattr.h>
52 #include <sys/kernel.h>
53 #include <sys/malloc.h>
54 #include <sys/mount.h>
55 #include <sys/proc.h>
56 #include <sys/resourcevar.h>
57 #include <sys/signalvar.h>
58 #include <sys/stat.h>
59 #include <sys/vmmeter.h>
60 #include <sys/vnode.h>
61 
62 #include <machine/limits.h>
63 
64 #include <vm/vm.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_page.h>
68 #include <vm/vm_pager.h>
69 #include <vm/vnode_pager.h>
70 
71 #include <ufs/ufs/extattr.h>
72 #include <ufs/ufs/quota.h>
73 #include <ufs/ufs/inode.h>
74 #include <ufs/ufs/ufs_extern.h>
75 #include <ufs/ufs/ufsmount.h>
76 
77 #include <ufs/ffs/fs.h>
78 #include <ufs/ffs/ffs_extern.h>
79 
80 static int	ffs_fsync(struct vop_fsync_args *);
81 static int	ffs_getpages(struct vop_getpages_args *);
82 static int	ffs_read(struct vop_read_args *);
83 static int	ffs_write(struct vop_write_args *);
84 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
85 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
86 		    struct ucred *cred);
87 static int	ffsext_strategy(struct vop_strategy_args *);
88 static int	ffs_closeextattr(struct vop_closeextattr_args *);
89 static int	ffs_getextattr(struct vop_getextattr_args *);
90 static int	ffs_openextattr(struct vop_openextattr_args *);
91 static int	ffs_setextattr(struct vop_setextattr_args *);
92 
93 
94 /* Global vfs data structures for ufs. */
95 vop_t **ffs_vnodeop_p;
96 static struct vnodeopv_entry_desc ffs_vnodeop_entries[] = {
97 	{ &vop_default_desc,		(vop_t *) ufs_vnoperate },
98 	{ &vop_fsync_desc,		(vop_t *) ffs_fsync },
99 	{ &vop_getpages_desc,		(vop_t *) ffs_getpages },
100 	{ &vop_read_desc,		(vop_t *) ffs_read },
101 	{ &vop_reallocblks_desc,	(vop_t *) ffs_reallocblks },
102 	{ &vop_write_desc,		(vop_t *) ffs_write },
103 	{ &vop_closeextattr_desc,	(vop_t *) ffs_closeextattr },
104 	{ &vop_getextattr_desc,		(vop_t *) ffs_getextattr },
105 	{ &vop_openextattr_desc,	(vop_t *) ffs_openextattr },
106 	{ &vop_setextattr_desc,		(vop_t *) ffs_setextattr },
107 	{ NULL, NULL }
108 };
109 static struct vnodeopv_desc ffs_vnodeop_opv_desc =
110 	{ &ffs_vnodeop_p, ffs_vnodeop_entries };
111 
112 vop_t **ffs_specop_p;
113 static struct vnodeopv_entry_desc ffs_specop_entries[] = {
114 	{ &vop_default_desc,		(vop_t *) ufs_vnoperatespec },
115 	{ &vop_fsync_desc,		(vop_t *) ffs_fsync },
116 	{ &vop_reallocblks_desc,	(vop_t *) ffs_reallocblks },
117 	{ &vop_strategy_desc,		(vop_t *) ffsext_strategy },
118 	{ &vop_closeextattr_desc,	(vop_t *) ffs_closeextattr },
119 	{ &vop_getextattr_desc,		(vop_t *) ffs_getextattr },
120 	{ &vop_openextattr_desc,	(vop_t *) ffs_openextattr },
121 	{ &vop_setextattr_desc,		(vop_t *) ffs_setextattr },
122 	{ NULL, NULL }
123 };
124 static struct vnodeopv_desc ffs_specop_opv_desc =
125 	{ &ffs_specop_p, ffs_specop_entries };
126 
127 vop_t **ffs_fifoop_p;
128 static struct vnodeopv_entry_desc ffs_fifoop_entries[] = {
129 	{ &vop_default_desc,		(vop_t *) ufs_vnoperatefifo },
130 	{ &vop_fsync_desc,		(vop_t *) ffs_fsync },
131 	{ &vop_reallocblks_desc,	(vop_t *) ffs_reallocblks },
132 	{ &vop_strategy_desc,		(vop_t *) ffsext_strategy },
133 	{ &vop_closeextattr_desc,	(vop_t *) ffs_closeextattr },
134 	{ &vop_getextattr_desc,		(vop_t *) ffs_getextattr },
135 	{ &vop_openextattr_desc,	(vop_t *) ffs_openextattr },
136 	{ &vop_setextattr_desc,		(vop_t *) ffs_setextattr },
137 	{ NULL, NULL }
138 };
139 static struct vnodeopv_desc ffs_fifoop_opv_desc =
140 	{ &ffs_fifoop_p, ffs_fifoop_entries };
141 
142 VNODEOP_SET(ffs_vnodeop_opv_desc);
143 VNODEOP_SET(ffs_specop_opv_desc);
144 VNODEOP_SET(ffs_fifoop_opv_desc);
145 
146 /*
147  * Synch an open file.
148  */
149 /* ARGSUSED */
150 static int
151 ffs_fsync(ap)
152 	struct vop_fsync_args /* {
153 		struct vnode *a_vp;
154 		struct ucred *a_cred;
155 		int a_waitfor;
156 		struct thread *a_td;
157 	} */ *ap;
158 {
159 	struct vnode *vp = ap->a_vp;
160 	struct inode *ip = VTOI(vp);
161 	struct buf *bp;
162 	struct buf *nbp;
163 	int s, error, wait, passes, skipmeta;
164 	ufs_lbn_t lbn;
165 
166 	wait = (ap->a_waitfor == MNT_WAIT);
167 	if (vn_isdisk(vp, NULL)) {
168 		lbn = INT_MAX;
169 		if (vp->v_rdev->si_mountpoint != NULL &&
170 		    (vp->v_rdev->si_mountpoint->mnt_flag & MNT_SOFTDEP))
171 			softdep_fsync_mountdev(vp);
172 	} else {
173 		lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
174 	}
175 
176 	/*
177 	 * Flush all dirty buffers associated with a vnode.
178 	 */
179 	passes = NIADDR + 1;
180 	skipmeta = 0;
181 	if (wait)
182 		skipmeta = 1;
183 	s = splbio();
184 	VI_LOCK(vp);
185 loop:
186 	TAILQ_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs)
187 		bp->b_flags &= ~B_SCANNED;
188 	for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
189 		nbp = TAILQ_NEXT(bp, b_vnbufs);
190 		/*
191 		 * Reasons to skip this buffer: it has already been considered
192 		 * on this pass, this pass is the first time through on a
193 		 * synchronous flush request and the buffer being considered
194 		 * is metadata, the buffer has dependencies that will cause
195 		 * it to be redirtied and it has not already been deferred,
196 		 * or it is already being written.
197 		 */
198 		if ((bp->b_flags & B_SCANNED) != 0)
199 			continue;
200 		bp->b_flags |= B_SCANNED;
201 		if ((skipmeta == 1 && bp->b_lblkno < 0))
202 			continue;
203 		if (!wait && LIST_FIRST(&bp->b_dep) != NULL &&
204 		    (bp->b_flags & B_DEFERRED) == 0 &&
205 		    buf_countdeps(bp, 0)) {
206 			bp->b_flags |= B_DEFERRED;
207 			continue;
208 		}
209 		VI_UNLOCK(vp);
210 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
211 			VI_LOCK(vp);
212 			continue;
213 		}
214 		if ((bp->b_flags & B_DELWRI) == 0)
215 			panic("ffs_fsync: not dirty");
216 		if (vp != bp->b_vp)
217 			panic("ffs_fsync: vp != vp->b_vp");
218 		/*
219 		 * If this is a synchronous flush request, or it is not a
220 		 * file or device, start the write on this buffer immediatly.
221 		 */
222 		if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
223 
224 			/*
225 			 * On our final pass through, do all I/O synchronously
226 			 * so that we can find out if our flush is failing
227 			 * because of write errors.
228 			 */
229 			if (passes > 0 || !wait) {
230 				if ((bp->b_flags & B_CLUSTEROK) && !wait) {
231 					BUF_UNLOCK(bp);
232 					(void) vfs_bio_awrite(bp);
233 				} else {
234 					bremfree(bp);
235 					splx(s);
236 					(void) bawrite(bp);
237 					s = splbio();
238 				}
239 			} else {
240 				bremfree(bp);
241 				splx(s);
242 				if ((error = bwrite(bp)) != 0)
243 					return (error);
244 				s = splbio();
245 			}
246 		} else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
247 			/*
248 			 * If the buffer is for data that has been truncated
249 			 * off the file, then throw it away.
250 			 */
251 			bremfree(bp);
252 			bp->b_flags |= B_INVAL | B_NOCACHE;
253 			splx(s);
254 			brelse(bp);
255 			s = splbio();
256 		} else {
257 			BUF_UNLOCK(bp);
258 			vfs_bio_awrite(bp);
259 		}
260 		/*
261 		 * Since we may have slept during the I/O, we need
262 		 * to start from a known point.
263 		 */
264 		VI_LOCK(vp);
265 		nbp = TAILQ_FIRST(&vp->v_dirtyblkhd);
266 	}
267 	/*
268 	 * If we were asked to do this synchronously, then go back for
269 	 * another pass, this time doing the metadata.
270 	 */
271 	if (skipmeta) {
272 		skipmeta = 0;
273 		goto loop;
274 	}
275 
276 	if (wait) {
277 		while (vp->v_numoutput) {
278 			vp->v_iflag |= VI_BWAIT;
279 			msleep((caddr_t)&vp->v_numoutput, VI_MTX(vp),
280 			    PRIBIO + 4, "ffsfsn", 0);
281   		}
282 		VI_UNLOCK(vp);
283 
284 		/*
285 		 * Ensure that any filesystem metatdata associated
286 		 * with the vnode has been written.
287 		 */
288 		splx(s);
289 		if ((error = softdep_sync_metadata(ap)) != 0)
290 			return (error);
291 		s = splbio();
292 
293 		VI_LOCK(vp);
294 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
295 			/*
296 			 * Block devices associated with filesystems may
297 			 * have new I/O requests posted for them even if
298 			 * the vnode is locked, so no amount of trying will
299 			 * get them clean. Thus we give block devices a
300 			 * good effort, then just give up. For all other file
301 			 * types, go around and try again until it is clean.
302 			 */
303 			if (passes > 0) {
304 				passes -= 1;
305 				goto loop;
306 			}
307 #ifdef DIAGNOSTIC
308 			if (!vn_isdisk(vp, NULL))
309 				vprint("ffs_fsync: dirty", vp);
310 #endif
311 		}
312 	}
313 	VI_UNLOCK(vp);
314 	splx(s);
315 	return (UFS_UPDATE(vp, wait));
316 }
317 
318 
319 /*
320  * Vnode op for reading.
321  */
322 /* ARGSUSED */
323 static int
324 ffs_read(ap)
325 	struct vop_read_args /* {
326 		struct vnode *a_vp;
327 		struct uio *a_uio;
328 		int a_ioflag;
329 		struct ucred *a_cred;
330 	} */ *ap;
331 {
332 	struct vnode *vp;
333 	struct inode *ip;
334 	struct uio *uio;
335 	struct fs *fs;
336 	struct buf *bp;
337 	ufs_lbn_t lbn, nextlbn;
338 	off_t bytesinfile;
339 	long size, xfersize, blkoffset;
340 	int error, orig_resid;
341 	mode_t mode;
342 	int seqcount;
343 	int ioflag;
344 	vm_object_t object;
345 
346 	vp = ap->a_vp;
347 	uio = ap->a_uio;
348 	ioflag = ap->a_ioflag;
349 	if (ap->a_ioflag & IO_EXT)
350 #ifdef notyet
351 		return (ffs_extread(vp, uio, ioflag));
352 #else
353 		panic("ffs_read+IO_EXT");
354 #endif
355 
356 	GIANT_REQUIRED;
357 
358 	seqcount = ap->a_ioflag >> 16;
359 	ip = VTOI(vp);
360 	mode = ip->i_mode;
361 
362 #ifdef DIAGNOSTIC
363 	if (uio->uio_rw != UIO_READ)
364 		panic("ffs_read: mode");
365 
366 	if (vp->v_type == VLNK) {
367 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
368 			panic("ffs_read: short symlink");
369 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
370 		panic("ffs_read: type %d",  vp->v_type);
371 #endif
372 	fs = ip->i_fs;
373 	if ((u_int64_t)uio->uio_offset > fs->fs_maxfilesize)
374 		return (EFBIG);
375 
376 	orig_resid = uio->uio_resid;
377 	if (orig_resid <= 0)
378 		return (0);
379 
380 	object = vp->v_object;
381 
382 	bytesinfile = ip->i_size - uio->uio_offset;
383 	if (bytesinfile <= 0) {
384 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
385 			ip->i_flag |= IN_ACCESS;
386 		return 0;
387 	}
388 
389 	if (object) {
390 		vm_object_reference(object);
391 	}
392 
393 #ifdef ENABLE_VFS_IOOPT
394 	/*
395 	 * If IO optimisation is turned on,
396 	 * and we are NOT a VM based IO request,
397 	 * (i.e. not headed for the buffer cache)
398 	 * but there IS a vm object associated with it.
399 	 */
400 	if ((ioflag & IO_VMIO) == 0 && (vfs_ioopt > 1) && object) {
401 		int nread, toread;
402 
403 		toread = uio->uio_resid;
404 		if (toread > bytesinfile)
405 			toread = bytesinfile;
406 		if (toread >= PAGE_SIZE) {
407 			/*
408 			 * Then if it's at least a page in size, try
409 			 * get the data from the object using vm tricks
410 			 */
411 			error = uioread(toread, uio, object, &nread);
412 			if ((uio->uio_resid == 0) || (error != 0)) {
413 				/*
414 				 * If we finished or there was an error
415 				 * then finish up (the reference previously
416 				 * obtained on object must be released).
417 				 */
418 				if ((error == 0 ||
419 				    uio->uio_resid != orig_resid) &&
420 				    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
421 					ip->i_flag |= IN_ACCESS;
422 
423 				if (object) {
424 					vm_object_vndeallocate(object);
425 				}
426 				return error;
427 			}
428 		}
429 	}
430 #endif
431 
432 	/*
433 	 * Ok so we couldn't do it all in one vm trick...
434 	 * so cycle around trying smaller bites..
435 	 */
436 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
437 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
438 			break;
439 #ifdef ENABLE_VFS_IOOPT
440 		if ((ioflag & IO_VMIO) == 0 && (vfs_ioopt > 1) && object) {
441 			/*
442 			 * Obviously we didn't finish above, but we
443 			 * didn't get an error either. Try the same trick again.
444 			 * but this time we are looping.
445 			 */
446 			int nread, toread;
447 			toread = uio->uio_resid;
448 			if (toread > bytesinfile)
449 				toread = bytesinfile;
450 
451 			/*
452 			 * Once again, if there isn't enough for a
453 			 * whole page, don't try optimising.
454 			 */
455 			if (toread >= PAGE_SIZE) {
456 				error = uioread(toread, uio, object, &nread);
457 				if ((uio->uio_resid == 0) || (error != 0)) {
458 					/*
459 					 * If we finished or there was an
460 					 * error then finish up (the reference
461 					 * previously obtained on object must
462 					 * be released).
463 					 */
464 					if ((error == 0 ||
465 					    uio->uio_resid != orig_resid) &&
466 					    (vp->v_mount->mnt_flag &
467 					    MNT_NOATIME) == 0)
468 						ip->i_flag |= IN_ACCESS;
469 					if (object) {
470 						vm_object_vndeallocate(object);
471 					}
472 					return error;
473 				}
474 				/*
475 				 * To get here we didnt't finish or err.
476 				 * If we did get some data,
477 				 * loop to try another bite.
478 				 */
479 				if (nread > 0) {
480 					continue;
481 				}
482 			}
483 		}
484 #endif
485 
486 		lbn = lblkno(fs, uio->uio_offset);
487 		nextlbn = lbn + 1;
488 
489 		/*
490 		 * size of buffer.  The buffer representing the
491 		 * end of the file is rounded up to the size of
492 		 * the block type ( fragment or full block,
493 		 * depending ).
494 		 */
495 		size = blksize(fs, ip, lbn);
496 		blkoffset = blkoff(fs, uio->uio_offset);
497 
498 		/*
499 		 * The amount we want to transfer in this iteration is
500 		 * one FS block less the amount of the data before
501 		 * our startpoint (duh!)
502 		 */
503 		xfersize = fs->fs_bsize - blkoffset;
504 
505 		/*
506 		 * But if we actually want less than the block,
507 		 * or the file doesn't have a whole block more of data,
508 		 * then use the lesser number.
509 		 */
510 		if (uio->uio_resid < xfersize)
511 			xfersize = uio->uio_resid;
512 		if (bytesinfile < xfersize)
513 			xfersize = bytesinfile;
514 
515 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
516 			/*
517 			 * Don't do readahead if this is the end of the file.
518 			 */
519 			error = bread(vp, lbn, size, NOCRED, &bp);
520 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
521 			/*
522 			 * Otherwise if we are allowed to cluster,
523 			 * grab as much as we can.
524 			 *
525 			 * XXX  This may not be a win if we are not
526 			 * doing sequential access.
527 			 */
528 			error = cluster_read(vp, ip->i_size, lbn,
529 				size, NOCRED, uio->uio_resid, seqcount, &bp);
530 		} else if (seqcount > 1) {
531 			/*
532 			 * If we are NOT allowed to cluster, then
533 			 * if we appear to be acting sequentially,
534 			 * fire off a request for a readahead
535 			 * as well as a read. Note that the 4th and 5th
536 			 * arguments point to arrays of the size specified in
537 			 * the 6th argument.
538 			 */
539 			int nextsize = blksize(fs, ip, nextlbn);
540 			error = breadn(vp, lbn,
541 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
542 		} else {
543 			/*
544 			 * Failing all of the above, just read what the
545 			 * user asked for. Interestingly, the same as
546 			 * the first option above.
547 			 */
548 			error = bread(vp, lbn, size, NOCRED, &bp);
549 		}
550 		if (error) {
551 			brelse(bp);
552 			bp = NULL;
553 			break;
554 		}
555 
556 		/*
557 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
558 		 * will cause us to attempt to release the buffer later on
559 		 * and will cause the buffer cache to attempt to free the
560 		 * underlying pages.
561 		 */
562 		if (ioflag & IO_DIRECT)
563 			bp->b_flags |= B_DIRECT;
564 
565 		/*
566 		 * We should only get non-zero b_resid when an I/O error
567 		 * has occurred, which should cause us to break above.
568 		 * However, if the short read did not cause an error,
569 		 * then we want to ensure that we do not uiomove bad
570 		 * or uninitialized data.
571 		 */
572 		size -= bp->b_resid;
573 		if (size < xfersize) {
574 			if (size == 0)
575 				break;
576 			xfersize = size;
577 		}
578 
579 #ifdef ENABLE_VFS_IOOPT
580 		if (vfs_ioopt && object &&
581 		    (bp->b_flags & B_VMIO) &&
582 		    ((blkoffset & PAGE_MASK) == 0) &&
583 		    ((xfersize & PAGE_MASK) == 0)) {
584 			/*
585 			 * If VFS IO  optimisation is turned on,
586 			 * and it's an exact page multiple
587 			 * And a normal VM based op,
588 			 * then use uiomiveco()
589 			 */
590 			error =
591 				uiomoveco((char *)bp->b_data + blkoffset,
592 					(int)xfersize, uio, object, 0);
593 		} else
594 #endif
595 		{
596 			/*
597 			 * otherwise use the general form
598 			 */
599 			error =
600 				uiomove((char *)bp->b_data + blkoffset,
601 					(int)xfersize, uio);
602 		}
603 
604 		if (error)
605 			break;
606 
607 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
608 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
609 			/*
610 			 * If there are no dependencies, and it's VMIO,
611 			 * then we don't need the buf, mark it available
612 			 * for freeing. The VM has the data.
613 			 */
614 			bp->b_flags |= B_RELBUF;
615 			brelse(bp);
616 		} else {
617 			/*
618 			 * Otherwise let whoever
619 			 * made the request take care of
620 			 * freeing it. We just queue
621 			 * it onto another list.
622 			 */
623 			bqrelse(bp);
624 		}
625 	}
626 
627 	/*
628 	 * This can only happen in the case of an error
629 	 * because the loop above resets bp to NULL on each iteration
630 	 * and on normal completion has not set a new value into it.
631 	 * so it must have come from a 'break' statement
632 	 */
633 	if (bp != NULL) {
634 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
635 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
636 			bp->b_flags |= B_RELBUF;
637 			brelse(bp);
638 		} else {
639 			bqrelse(bp);
640 		}
641 	}
642 
643 	if (object) {
644 		vm_object_vndeallocate(object);
645 	}
646 	if ((error == 0 || uio->uio_resid != orig_resid) &&
647 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
648 		ip->i_flag |= IN_ACCESS;
649 	return (error);
650 }
651 
652 /*
653  * Vnode op for writing.
654  */
655 static int
656 ffs_write(ap)
657 	struct vop_write_args /* {
658 		struct vnode *a_vp;
659 		struct uio *a_uio;
660 		int a_ioflag;
661 		struct ucred *a_cred;
662 	} */ *ap;
663 {
664 	struct vnode *vp;
665 	struct uio *uio;
666 	struct inode *ip;
667 	struct fs *fs;
668 	struct buf *bp;
669 	struct thread *td;
670 	ufs_lbn_t lbn;
671 	off_t osize;
672 	int seqcount;
673 	int blkoffset, error, extended, flags, ioflag, resid, size, xfersize;
674 	vm_object_t object;
675 
676 	vp = ap->a_vp;
677 	uio = ap->a_uio;
678 	ioflag = ap->a_ioflag;
679 	if (ap->a_ioflag & IO_EXT)
680 #ifdef notyet
681 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
682 #else
683 		panic("ffs_read+IO_EXT");
684 #endif
685 
686 	GIANT_REQUIRED;
687 
688 	extended = 0;
689 	seqcount = ap->a_ioflag >> 16;
690 	ip = VTOI(vp);
691 
692 	object = vp->v_object;
693 	if (object) {
694 		vm_object_reference(object);
695 	}
696 
697 #ifdef DIAGNOSTIC
698 	if (uio->uio_rw != UIO_WRITE)
699 		panic("ffswrite: mode");
700 #endif
701 
702 	switch (vp->v_type) {
703 	case VREG:
704 		if (ioflag & IO_APPEND)
705 			uio->uio_offset = ip->i_size;
706 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) {
707 			if (object) {
708 				vm_object_vndeallocate(object);
709 			}
710 			return (EPERM);
711 		}
712 		/* FALLTHROUGH */
713 	case VLNK:
714 		break;
715 	case VDIR:
716 		panic("ffswrite: dir write");
717 		break;
718 	default:
719 		panic("ffswrite: type %p %d (%d,%d)", vp, (int)vp->v_type,
720 			(int)uio->uio_offset,
721 			(int)uio->uio_resid
722 		);
723 	}
724 
725 	fs = ip->i_fs;
726 	if (uio->uio_offset < 0 ||
727 	    (u_int64_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) {
728 		if (object) {
729 			vm_object_vndeallocate(object);
730 		}
731 		return (EFBIG);
732 	}
733 	/*
734 	 * Maybe this should be above the vnode op call, but so long as
735 	 * file servers have no limits, I don't think it matters.
736 	 */
737 	td = uio->uio_td;
738 	if (vp->v_type == VREG && td &&
739 	    uio->uio_offset + uio->uio_resid >
740 	    td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
741 		PROC_LOCK(td->td_proc);
742 		psignal(td->td_proc, SIGXFSZ);
743 		PROC_UNLOCK(td->td_proc);
744 		if (object) {
745 			vm_object_vndeallocate(object);
746 		}
747 		return (EFBIG);
748 	}
749 
750 	resid = uio->uio_resid;
751 	osize = ip->i_size;
752 	if (seqcount > BA_SEQMAX)
753 		flags = BA_SEQMAX << BA_SEQSHIFT;
754 	else
755 		flags = seqcount << BA_SEQSHIFT;
756 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
757 		flags |= IO_SYNC;
758 
759 #ifdef ENABLE_VFS_IOOPT
760 	if (object && (object->flags & OBJ_OPT)) {
761 		vm_freeze_copyopts(object,
762 			OFF_TO_IDX(uio->uio_offset),
763 			OFF_TO_IDX(uio->uio_offset + uio->uio_resid + PAGE_MASK));
764 	}
765 #endif
766 	for (error = 0; uio->uio_resid > 0;) {
767 		lbn = lblkno(fs, uio->uio_offset);
768 		blkoffset = blkoff(fs, uio->uio_offset);
769 		xfersize = fs->fs_bsize - blkoffset;
770 		if (uio->uio_resid < xfersize)
771 			xfersize = uio->uio_resid;
772 
773 		if (uio->uio_offset + xfersize > ip->i_size)
774 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
775 
776                 /*
777 		 * We must perform a read-before-write if the transfer size
778 		 * does not cover the entire buffer.
779                  */
780 		if (fs->fs_bsize > xfersize)
781 			flags |= BA_CLRBUF;
782 		else
783 			flags &= ~BA_CLRBUF;
784 /* XXX is uio->uio_offset the right thing here? */
785 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
786 		    ap->a_cred, flags, &bp);
787 		if (error != 0)
788 			break;
789 		/*
790 		 * If the buffer is not valid we have to clear out any
791 		 * garbage data from the pages instantiated for the buffer.
792 		 * If we do not, a failed uiomove() during a write can leave
793 		 * the prior contents of the pages exposed to a userland
794 		 * mmap().  XXX deal with uiomove() errors a better way.
795 		 */
796 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
797 			vfs_bio_clrbuf(bp);
798 		if (ioflag & IO_DIRECT)
799 			bp->b_flags |= B_DIRECT;
800 		if (ioflag & IO_NOWDRAIN)
801 			bp->b_flags |= B_NOWDRAIN;
802 
803 		if (uio->uio_offset + xfersize > ip->i_size) {
804 			ip->i_size = uio->uio_offset + xfersize;
805 			DIP(ip, i_size) = ip->i_size;
806 			extended = 1;
807 		}
808 
809 		size = blksize(fs, ip, lbn) - bp->b_resid;
810 		if (size < xfersize)
811 			xfersize = size;
812 
813 		error =
814 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
815 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
816 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
817 			bp->b_flags |= B_RELBUF;
818 		}
819 
820 		/*
821 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
822 		 * if we have a severe page deficiency write the buffer
823 		 * asynchronously.  Otherwise try to cluster, and if that
824 		 * doesn't do it then either do an async write (if O_DIRECT),
825 		 * or a delayed write (if not).
826 		 */
827 		if (ioflag & IO_SYNC) {
828 			(void)bwrite(bp);
829 		} else if (vm_page_count_severe() ||
830 			    buf_dirty_count_severe() ||
831 			    (ioflag & IO_ASYNC)) {
832 			bp->b_flags |= B_CLUSTEROK;
833 			bawrite(bp);
834 		} else if (xfersize + blkoffset == fs->fs_bsize) {
835 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
836 				bp->b_flags |= B_CLUSTEROK;
837 				cluster_write(bp, ip->i_size, seqcount);
838 			} else {
839 				bawrite(bp);
840 			}
841 		} else if (ioflag & IO_DIRECT) {
842 			bp->b_flags |= B_CLUSTEROK;
843 			bawrite(bp);
844 		} else {
845 			bp->b_flags |= B_CLUSTEROK;
846 			bdwrite(bp);
847 		}
848 		if (error || xfersize == 0)
849 			break;
850 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
851 	}
852 	/*
853 	 * If we successfully wrote any data, and we are not the superuser
854 	 * we clear the setuid and setgid bits as a precaution against
855 	 * tampering.
856 	 */
857 	if (resid > uio->uio_resid && ap->a_cred &&
858 	    suser_cred(ap->a_cred, PRISON_ROOT)) {
859 		ip->i_mode &= ~(ISUID | ISGID);
860 		DIP(ip, i_mode) = ip->i_mode;
861 	}
862 	if (resid > uio->uio_resid)
863 		VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
864 	if (error) {
865 		if (ioflag & IO_UNIT) {
866 			(void)UFS_TRUNCATE(vp, osize,
867 			    IO_NORMAL | (ioflag & IO_SYNC),
868 			    ap->a_cred, uio->uio_td);
869 			uio->uio_offset -= resid - uio->uio_resid;
870 			uio->uio_resid = resid;
871 		}
872 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
873 		error = UFS_UPDATE(vp, 1);
874 
875 	if (object) {
876 		vm_object_vndeallocate(object);
877 	}
878 
879 	return (error);
880 }
881 
882 /*
883  * get page routine
884  */
885 static int
886 ffs_getpages(ap)
887 	struct vop_getpages_args *ap;
888 {
889 	off_t foff, physoffset;
890 	int i, size, bsize;
891 	struct vnode *dp, *vp;
892 	vm_object_t obj;
893 	vm_pindex_t pindex, firstindex;
894 	vm_page_t mreq;
895 	int bbackwards, bforwards;
896 	int pbackwards, pforwards;
897 	int firstpage;
898 	ufs2_daddr_t reqblkno, reqlblkno;
899 	int poff;
900 	int pcount;
901 	int rtval;
902 	int pagesperblock;
903 
904 	GIANT_REQUIRED;
905 
906 	pcount = round_page(ap->a_count) / PAGE_SIZE;
907 	mreq = ap->a_m[ap->a_reqpage];
908 	firstindex = ap->a_m[0]->pindex;
909 
910 	/*
911 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
912 	 * then the entire page is valid.  Since the page may be mapped,
913 	 * user programs might reference data beyond the actual end of file
914 	 * occuring within the page.  We have to zero that data.
915 	 */
916 	if (mreq->valid) {
917 		if (mreq->valid != VM_PAGE_BITS_ALL)
918 			vm_page_zero_invalid(mreq, TRUE);
919 		vm_page_lock_queues();
920 		for (i = 0; i < pcount; i++) {
921 			if (i != ap->a_reqpage) {
922 				vm_page_free(ap->a_m[i]);
923 			}
924 		}
925 		vm_page_unlock_queues();
926 		return VM_PAGER_OK;
927 	}
928 
929 	vp = ap->a_vp;
930 	obj = vp->v_object;
931 	bsize = vp->v_mount->mnt_stat.f_iosize;
932 	pindex = mreq->pindex;
933 	foff = IDX_TO_OFF(pindex) /* + ap->a_offset should be zero */;
934 
935 	if (bsize < PAGE_SIZE)
936 		return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
937 						    ap->a_count,
938 						    ap->a_reqpage);
939 
940 	/*
941 	 * foff is the file offset of the required page
942 	 * reqlblkno is the logical block that contains the page
943 	 * poff is the index of the page into the logical block
944 	 */
945 	reqlblkno = foff / bsize;
946 	poff = (foff % bsize) / PAGE_SIZE;
947 
948 	dp = VTOI(vp)->i_devvp;
949 	if (ufs_bmaparray(vp, reqlblkno, &reqblkno, 0, &bforwards, &bbackwards)
950 	    || (reqblkno == -1)) {
951 		vm_page_lock_queues();
952 		for(i = 0; i < pcount; i++) {
953 			if (i != ap->a_reqpage)
954 				vm_page_free(ap->a_m[i]);
955 		}
956 		vm_page_unlock_queues();
957 		if (reqblkno == -1) {
958 			if ((mreq->flags & PG_ZERO) == 0)
959 				pmap_zero_page(mreq);
960 			vm_page_undirty(mreq);
961 			mreq->valid = VM_PAGE_BITS_ALL;
962 			return VM_PAGER_OK;
963 		} else {
964 			return VM_PAGER_ERROR;
965 		}
966 	}
967 
968 	physoffset = (off_t)reqblkno * DEV_BSIZE + poff * PAGE_SIZE;
969 	pagesperblock = bsize / PAGE_SIZE;
970 	/*
971 	 * find the first page that is contiguous...
972 	 * note that pbackwards is the number of pages that are contiguous
973 	 * backwards.
974 	 */
975 	firstpage = 0;
976 	if (ap->a_count) {
977 		pbackwards = poff + bbackwards * pagesperblock;
978 		if (ap->a_reqpage > pbackwards) {
979 			firstpage = ap->a_reqpage - pbackwards;
980 			vm_page_lock_queues();
981 			for(i=0;i<firstpage;i++)
982 				vm_page_free(ap->a_m[i]);
983 			vm_page_unlock_queues();
984 		}
985 
986 	/*
987 	 * pforwards is the number of pages that are contiguous
988 	 * after the current page.
989 	 */
990 		pforwards = (pagesperblock - (poff + 1)) +
991 			bforwards * pagesperblock;
992 		if (pforwards < (pcount - (ap->a_reqpage + 1))) {
993 			vm_page_lock_queues();
994 			for( i = ap->a_reqpage + pforwards + 1; i < pcount; i++)
995 				vm_page_free(ap->a_m[i]);
996 			vm_page_unlock_queues();
997 			pcount = ap->a_reqpage + pforwards + 1;
998 		}
999 
1000 	/*
1001 	 * number of pages for I/O corrected for the non-contig pages at
1002 	 * the beginning of the array.
1003 	 */
1004 		pcount -= firstpage;
1005 	}
1006 
1007 	/*
1008 	 * calculate the size of the transfer
1009 	 */
1010 
1011 	size = pcount * PAGE_SIZE;
1012 
1013 	if ((IDX_TO_OFF(ap->a_m[firstpage]->pindex) + size) >
1014 		obj->un_pager.vnp.vnp_size)
1015 		size = obj->un_pager.vnp.vnp_size -
1016 			IDX_TO_OFF(ap->a_m[firstpage]->pindex);
1017 
1018 	physoffset -= foff;
1019 	rtval = VOP_GETPAGES(dp, &ap->a_m[firstpage], size,
1020 		(ap->a_reqpage - firstpage), physoffset);
1021 
1022 	return (rtval);
1023 }
1024 
1025 /*
1026  * Extended attribute area reading.
1027  */
1028 static int
1029 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
1030 {
1031 	struct inode *ip;
1032 	struct ufs2_dinode *dp;
1033 	struct fs *fs;
1034 	struct buf *bp;
1035 	ufs_lbn_t lbn, nextlbn;
1036 	off_t bytesinfile;
1037 	long size, xfersize, blkoffset;
1038 	int error, orig_resid;
1039 	mode_t mode;
1040 
1041 	GIANT_REQUIRED;
1042 
1043 	ip = VTOI(vp);
1044 	fs = ip->i_fs;
1045 	dp = ip->i_din2;
1046 	mode = ip->i_mode;
1047 
1048 #ifdef DIAGNOSTIC
1049 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
1050 		panic("ffs_extread: mode");
1051 
1052 #endif
1053 	orig_resid = uio->uio_resid;
1054 	if (orig_resid <= 0)
1055 		return (0);
1056 
1057 	bytesinfile = dp->di_extsize - uio->uio_offset;
1058 	if (bytesinfile <= 0) {
1059 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1060 			ip->i_flag |= IN_ACCESS;
1061 		return 0;
1062 	}
1063 
1064 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1065 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
1066 			break;
1067 
1068 		lbn = lblkno(fs, uio->uio_offset);
1069 		nextlbn = lbn + 1;
1070 
1071 		/*
1072 		 * size of buffer.  The buffer representing the
1073 		 * end of the file is rounded up to the size of
1074 		 * the block type ( fragment or full block,
1075 		 * depending ).
1076 		 */
1077 		size = sblksize(fs, dp->di_extsize, lbn);
1078 		blkoffset = blkoff(fs, uio->uio_offset);
1079 
1080 		/*
1081 		 * The amount we want to transfer in this iteration is
1082 		 * one FS block less the amount of the data before
1083 		 * our startpoint (duh!)
1084 		 */
1085 		xfersize = fs->fs_bsize - blkoffset;
1086 
1087 		/*
1088 		 * But if we actually want less than the block,
1089 		 * or the file doesn't have a whole block more of data,
1090 		 * then use the lesser number.
1091 		 */
1092 		if (uio->uio_resid < xfersize)
1093 			xfersize = uio->uio_resid;
1094 		if (bytesinfile < xfersize)
1095 			xfersize = bytesinfile;
1096 
1097 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
1098 			/*
1099 			 * Don't do readahead if this is the end of the info.
1100 			 */
1101 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
1102 		} else {
1103 			/*
1104 			 * If we have a second block, then
1105 			 * fire off a request for a readahead
1106 			 * as well as a read. Note that the 4th and 5th
1107 			 * arguments point to arrays of the size specified in
1108 			 * the 6th argument.
1109 			 */
1110 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
1111 
1112 			nextlbn = -1 - nextlbn;
1113 			error = breadn(vp, -1 - lbn,
1114 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1115 		}
1116 		if (error) {
1117 			brelse(bp);
1118 			bp = NULL;
1119 			break;
1120 		}
1121 
1122 		/*
1123 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
1124 		 * will cause us to attempt to release the buffer later on
1125 		 * and will cause the buffer cache to attempt to free the
1126 		 * underlying pages.
1127 		 */
1128 		if (ioflag & IO_DIRECT)
1129 			bp->b_flags |= B_DIRECT;
1130 
1131 		/*
1132 		 * We should only get non-zero b_resid when an I/O error
1133 		 * has occurred, which should cause us to break above.
1134 		 * However, if the short read did not cause an error,
1135 		 * then we want to ensure that we do not uiomove bad
1136 		 * or uninitialized data.
1137 		 */
1138 		size -= bp->b_resid;
1139 		if (size < xfersize) {
1140 			if (size == 0)
1141 				break;
1142 			xfersize = size;
1143 		}
1144 
1145 		error = uiomove((char *)bp->b_data + blkoffset,
1146 					(int)xfersize, uio);
1147 		if (error)
1148 			break;
1149 
1150 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1151 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1152 			/*
1153 			 * If there are no dependencies, and it's VMIO,
1154 			 * then we don't need the buf, mark it available
1155 			 * for freeing. The VM has the data.
1156 			 */
1157 			bp->b_flags |= B_RELBUF;
1158 			brelse(bp);
1159 		} else {
1160 			/*
1161 			 * Otherwise let whoever
1162 			 * made the request take care of
1163 			 * freeing it. We just queue
1164 			 * it onto another list.
1165 			 */
1166 			bqrelse(bp);
1167 		}
1168 	}
1169 
1170 	/*
1171 	 * This can only happen in the case of an error
1172 	 * because the loop above resets bp to NULL on each iteration
1173 	 * and on normal completion has not set a new value into it.
1174 	 * so it must have come from a 'break' statement
1175 	 */
1176 	if (bp != NULL) {
1177 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1178 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1179 			bp->b_flags |= B_RELBUF;
1180 			brelse(bp);
1181 		} else {
1182 			bqrelse(bp);
1183 		}
1184 	}
1185 
1186 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1187 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1188 		ip->i_flag |= IN_ACCESS;
1189 	return (error);
1190 }
1191 
1192 /*
1193  * Extended attribute area writing.
1194  */
1195 static int
1196 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1197 {
1198 	struct inode *ip;
1199 	struct ufs2_dinode *dp;
1200 	struct fs *fs;
1201 	struct buf *bp;
1202 	ufs_lbn_t lbn;
1203 	off_t osize;
1204 	int blkoffset, error, flags, resid, size, xfersize;
1205 
1206 	GIANT_REQUIRED;
1207 
1208 	ip = VTOI(vp);
1209 	fs = ip->i_fs;
1210 	dp = ip->i_din2;
1211 
1212 #ifdef DIAGNOSTIC
1213 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1214 		panic("ext_write: mode");
1215 #endif
1216 
1217 	if (ioflag & IO_APPEND)
1218 		uio->uio_offset = dp->di_extsize;
1219 
1220 	if (uio->uio_offset < 0 ||
1221 	    (u_int64_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1222 		return (EFBIG);
1223 
1224 	resid = uio->uio_resid;
1225 	osize = dp->di_extsize;
1226 	flags = IO_EXT;
1227 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1228 		flags |= IO_SYNC;
1229 
1230 	for (error = 0; uio->uio_resid > 0;) {
1231 		lbn = lblkno(fs, uio->uio_offset);
1232 		blkoffset = blkoff(fs, uio->uio_offset);
1233 		xfersize = fs->fs_bsize - blkoffset;
1234 		if (uio->uio_resid < xfersize)
1235 			xfersize = uio->uio_resid;
1236 
1237                 /*
1238 		 * We must perform a read-before-write if the transfer size
1239 		 * does not cover the entire buffer.
1240                  */
1241 		if (fs->fs_bsize > xfersize)
1242 			flags |= BA_CLRBUF;
1243 		else
1244 			flags &= ~BA_CLRBUF;
1245 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1246 		    ucred, flags, &bp);
1247 		if (error != 0)
1248 			break;
1249 		/*
1250 		 * If the buffer is not valid we have to clear out any
1251 		 * garbage data from the pages instantiated for the buffer.
1252 		 * If we do not, a failed uiomove() during a write can leave
1253 		 * the prior contents of the pages exposed to a userland
1254 		 * mmap().  XXX deal with uiomove() errors a better way.
1255 		 */
1256 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1257 			vfs_bio_clrbuf(bp);
1258 		if (ioflag & IO_DIRECT)
1259 			bp->b_flags |= B_DIRECT;
1260 		if (ioflag & IO_NOWDRAIN)
1261 			bp->b_flags |= B_NOWDRAIN;
1262 
1263 		if (uio->uio_offset + xfersize > dp->di_extsize)
1264 			dp->di_extsize = uio->uio_offset + xfersize;
1265 
1266 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1267 		if (size < xfersize)
1268 			xfersize = size;
1269 
1270 		error =
1271 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1272 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1273 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1274 			bp->b_flags |= B_RELBUF;
1275 		}
1276 
1277 		/*
1278 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1279 		 * if we have a severe page deficiency write the buffer
1280 		 * asynchronously.  Otherwise try to cluster, and if that
1281 		 * doesn't do it then either do an async write (if O_DIRECT),
1282 		 * or a delayed write (if not).
1283 		 */
1284 		if (ioflag & IO_SYNC) {
1285 			(void)bwrite(bp);
1286 		} else if (vm_page_count_severe() ||
1287 			    buf_dirty_count_severe() ||
1288 			    xfersize + blkoffset == fs->fs_bsize ||
1289 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1290 			bawrite(bp);
1291 		else
1292 			bdwrite(bp);
1293 		if (error || xfersize == 0)
1294 			break;
1295 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1296 	}
1297 	/*
1298 	 * If we successfully wrote any data, and we are not the superuser
1299 	 * we clear the setuid and setgid bits as a precaution against
1300 	 * tampering.
1301 	 */
1302 	if (resid > uio->uio_resid && ucred &&
1303 	    suser_cred(ucred, PRISON_ROOT)) {
1304 		ip->i_mode &= ~(ISUID | ISGID);
1305 		dp->di_mode = ip->i_mode;
1306 	}
1307 	if (error) {
1308 		if (ioflag & IO_UNIT) {
1309 			(void)UFS_TRUNCATE(vp, osize,
1310 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1311 			uio->uio_offset -= resid - uio->uio_resid;
1312 			uio->uio_resid = resid;
1313 		}
1314 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1315 		error = UFS_UPDATE(vp, 1);
1316 	return (error);
1317 }
1318 
1319 
1320 /*
1321  * Vnode operating to retrieve a named extended attribute.
1322  *
1323  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1324  * the length of the EA, and possibly the pointer to the entry and to the data.
1325  */
1326 static int
1327 ffs_findextattr(u_char *ptr, uint length, int nspace, const char *name, u_char **eap, u_char **eac)
1328 {
1329 	u_char *p, *pe, *pn, *p0;
1330 	int eapad1, eapad2, ealength, ealen, nlen;
1331 	uint32_t ul;
1332 
1333 	pe = ptr + length;
1334 	nlen = strlen(name);
1335 
1336 	for (p = ptr; p < pe; p = pn) {
1337 		p0 = p;
1338 		bcopy(p, &ul, sizeof(ul));
1339 		pn = p + ul;
1340 		/* make sure this entry is complete */
1341 		if (pn > pe)
1342 			break;
1343 		p += sizeof(uint32_t);
1344 		if (*p != nspace)
1345 			continue;
1346 		p++;
1347 		eapad2 = *p++;
1348 		if (*p != nlen)
1349 			continue;
1350 		p++;
1351 		if (bcmp(p, name, nlen))
1352 			continue;
1353 		ealength = sizeof(uint32_t) + 3 + nlen;
1354 		eapad1 = 8 - (ealength % 8);
1355 		if (eapad1 == 8)
1356 			eapad1 = 0;
1357 		ealength += eapad1;
1358 		ealen = ul - ealength - eapad2;
1359 		p += nlen + eapad1;
1360 		if (eap != NULL)
1361 			*eap = p0;
1362 		if (eac != NULL)
1363 			*eac = p;
1364 		return (ealen);
1365 	}
1366 	return(-1);
1367 }
1368 
1369 static int
1370 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1371 {
1372 	struct inode *ip;
1373 	struct fs *fs;
1374 	struct ufs2_dinode *dp;
1375 	struct uio luio;
1376 	struct iovec liovec;
1377 	int easize, error;
1378 	u_char *eae;
1379 
1380 	ip = VTOI(vp);
1381 	fs = ip->i_fs;
1382 	dp = ip->i_din2;
1383 	easize = dp->di_extsize;
1384 
1385 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1386 
1387 	liovec.iov_base = eae;
1388 	liovec.iov_len = easize;
1389 	luio.uio_iov = &liovec;
1390 	luio.uio_iovcnt = 1;
1391 	luio.uio_offset = 0;
1392 	luio.uio_resid = easize;
1393 	luio.uio_segflg = UIO_SYSSPACE;
1394 	luio.uio_rw = UIO_READ;
1395 	luio.uio_td = td;
1396 
1397 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1398 	if (error) {
1399 		free(eae, M_TEMP);
1400 		return(error);
1401 	}
1402 	*p = eae;
1403 	return (0);
1404 }
1405 
1406 static int
1407 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1408 {
1409 	struct inode *ip;
1410 	struct fs *fs;
1411 	struct ufs2_dinode *dp;
1412 	int error;
1413 
1414 	ip = VTOI(vp);
1415 	fs = ip->i_fs;
1416 
1417 	if (ip->i_ea_area != NULL)
1418 		return (EBUSY);
1419 	dp = ip->i_din2;
1420 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1421 	if (error)
1422 		return (error);
1423 	ip->i_ea_len = dp->di_extsize;
1424 	ip->i_ea_error = 0;
1425 	return (0);
1426 }
1427 
1428 /*
1429  * Vnode extattr transaction commit/abort
1430  */
1431 static int
1432 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1433 {
1434 	struct inode *ip;
1435 	struct fs *fs;
1436 	struct uio luio;
1437 	struct iovec liovec;
1438 	int error;
1439 	struct ufs2_dinode *dp;
1440 
1441 	ip = VTOI(vp);
1442 	fs = ip->i_fs;
1443 	if (ip->i_ea_area == NULL)
1444 		return (EINVAL);
1445 	dp = ip->i_din2;
1446 	error = ip->i_ea_error;
1447 	if (commit && error == 0) {
1448 		if (cred == NOCRED)
1449 			cred =  vp->v_mount->mnt_cred;
1450 		liovec.iov_base = ip->i_ea_area;
1451 		liovec.iov_len = ip->i_ea_len;
1452 		luio.uio_iov = &liovec;
1453 		luio.uio_iovcnt = 1;
1454 		luio.uio_offset = 0;
1455 		luio.uio_resid = ip->i_ea_len;
1456 		luio.uio_segflg = UIO_SYSSPACE;
1457 		luio.uio_rw = UIO_WRITE;
1458 		luio.uio_td = td;
1459 		/* XXX: I'm not happy about truncating to zero size */
1460 		if (ip->i_ea_len < dp->di_extsize)
1461 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1462 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1463 	}
1464 	free(ip->i_ea_area, M_TEMP);
1465 	ip->i_ea_area = NULL;
1466 	ip->i_ea_len = 0;
1467 	ip->i_ea_error = 0;
1468 	return (error);
1469 }
1470 
1471 /*
1472  * Vnode extattr strategy routine for special devices and fifos.
1473  *
1474  * We need to check for a read or write of the external attributes.
1475  * Otherwise we just fall through and do the usual thing.
1476  */
1477 static int
1478 ffsext_strategy(struct vop_strategy_args *ap)
1479 /*
1480 struct vop_strategy_args {
1481 	struct vnodeop_desc *a_desc;
1482 	struct vnode *a_vp;
1483 	struct buf *a_bp;
1484 };
1485 */
1486 {
1487 	struct vnode *vp;
1488 	daddr_t lbn;
1489 
1490 	vp = ap->a_vp;
1491 	lbn = ap->a_bp->b_lblkno;
1492 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1493 	    lbn < 0 && lbn >= -NXADDR)
1494 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1495 	if (vp->v_type == VFIFO)
1496 		return (ufs_vnoperatefifo((struct vop_generic_args *)ap));
1497 	return (ufs_vnoperatespec((struct vop_generic_args *)ap));
1498 }
1499 
1500 /*
1501  * Vnode extattr transaction commit/abort
1502  */
1503 static int
1504 ffs_openextattr(struct vop_openextattr_args *ap)
1505 /*
1506 struct vop_openextattr_args {
1507 	struct vnodeop_desc *a_desc;
1508 	struct vnode *a_vp;
1509 	IN struct ucred *a_cred;
1510 	IN struct thread *a_td;
1511 };
1512 */
1513 {
1514 	struct inode *ip;
1515 	struct fs *fs;
1516 
1517 	ip = VTOI(ap->a_vp);
1518 	fs = ip->i_fs;
1519 	if (fs->fs_magic == FS_UFS1_MAGIC)
1520 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1521 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1522 }
1523 
1524 
1525 /*
1526  * Vnode extattr transaction commit/abort
1527  */
1528 static int
1529 ffs_closeextattr(struct vop_closeextattr_args *ap)
1530 /*
1531 struct vop_closeextattr_args {
1532 	struct vnodeop_desc *a_desc;
1533 	struct vnode *a_vp;
1534 	int a_commit;
1535 	IN struct ucred *a_cred;
1536 	IN struct thread *a_td;
1537 };
1538 */
1539 {
1540 	struct inode *ip;
1541 	struct fs *fs;
1542 
1543 	ip = VTOI(ap->a_vp);
1544 	fs = ip->i_fs;
1545 	if (fs->fs_magic == FS_UFS1_MAGIC)
1546 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1547 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1548 }
1549 
1550 
1551 
1552 /*
1553  * Vnode operation to retrieve a named extended attribute.
1554  */
1555 static int
1556 ffs_getextattr(struct vop_getextattr_args *ap)
1557 /*
1558 vop_getextattr {
1559 	IN struct vnode *a_vp;
1560 	IN int a_attrnamespace;
1561 	IN const char *a_name;
1562 	INOUT struct uio *a_uio;
1563 	OUT size_t *a_size;
1564 	IN struct ucred *a_cred;
1565 	IN struct thread *a_td;
1566 };
1567 */
1568 {
1569 	struct inode *ip;
1570 	struct fs *fs;
1571 	u_char *eae, *p, *pe, *pn;
1572 	struct ufs2_dinode *dp;
1573 	unsigned easize;
1574 	uint32_t ul;
1575 	int error, ealen, stand_alone;
1576 
1577 	ip = VTOI(ap->a_vp);
1578 	fs = ip->i_fs;
1579 
1580 	if (fs->fs_magic == FS_UFS1_MAGIC)
1581 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1582 
1583 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1584 	    ap->a_cred, ap->a_td, IREAD);
1585 	if (error)
1586 		return (error);
1587 
1588 	if (ip->i_ea_area == NULL) {
1589 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1590 		if (error)
1591 			return (error);
1592 		stand_alone = 1;
1593 	} else {
1594 		stand_alone = 0;
1595 	}
1596 	dp = ip->i_din2;
1597 	eae = ip->i_ea_area;
1598 	easize = ip->i_ea_len;
1599 	if (strlen(ap->a_name) > 0) {
1600 		ealen = ffs_findextattr(eae, easize,
1601 		    ap->a_attrnamespace, ap->a_name, NULL, &p);
1602 		if (ealen >= 0) {
1603 			error = 0;
1604 			if (ap->a_size != NULL)
1605 				*ap->a_size = ealen;
1606 			else if (ap->a_uio != NULL)
1607 				error = uiomove(p, ealen, ap->a_uio);
1608 		} else {
1609 			error = ENOATTR;
1610 		}
1611 	} else {
1612 		error = 0;
1613 		if (ap->a_size != NULL)
1614 			*ap->a_size = 0;
1615 		pe = eae + easize;
1616 		for(p = eae; error == 0 && p < pe; p = pn) {
1617 			bcopy(p, &ul, sizeof(ul));
1618 			pn = p + ul;
1619 			if (pn > pe)
1620 				break;
1621 			p += sizeof(ul);
1622 			if (*p++ != ap->a_attrnamespace)
1623 				continue;
1624 			p++;	/* pad2 */
1625 			ealen = *p;
1626 			if (ap->a_size != NULL) {
1627 				*ap->a_size += ealen + 1;
1628 			} else if (ap->a_uio != NULL) {
1629 				error = uiomove(p, ealen + 1, ap->a_uio);
1630 			}
1631 		}
1632 	}
1633 	if (stand_alone)
1634 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1635 	return(error);
1636 }
1637 
1638 /*
1639  * Vnode operation to set a named attribute.
1640  */
1641 static int
1642 ffs_setextattr(struct vop_setextattr_args *ap)
1643 /*
1644 vop_setextattr {
1645 	IN struct vnode *a_vp;
1646 	IN int a_attrnamespace;
1647 	IN const char *a_name;
1648 	INOUT struct uio *a_uio;
1649 	IN struct ucred *a_cred;
1650 	IN struct thread *a_td;
1651 };
1652 */
1653 {
1654 	struct inode *ip;
1655 	struct fs *fs;
1656 	uint32_t ealength, ul;
1657 	int ealen, olen, eacont, eapad1, eapad2, error, i, easize;
1658 	u_char *eae, *p;
1659 	struct ufs2_dinode *dp;
1660 	struct ucred *cred;
1661 	int stand_alone;
1662 
1663 	ip = VTOI(ap->a_vp);
1664 	fs = ip->i_fs;
1665 
1666 	if (fs->fs_magic == FS_UFS1_MAGIC)
1667 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1668 
1669 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1670 	    ap->a_cred, ap->a_td, IWRITE);
1671 	if (error) {
1672 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1673 			ip->i_ea_error = error;
1674 		return (error);
1675 	}
1676 
1677 	if (ap->a_cred != NOCRED)
1678 		cred = ap->a_cred;
1679 	else
1680 		cred = ap->a_vp->v_mount->mnt_cred;
1681 
1682 	dp = ip->i_din2;
1683 
1684 	if (ip->i_ea_area == NULL) {
1685 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1686 		if (error)
1687 			return (error);
1688 		stand_alone = 1;
1689 	} else {
1690 		stand_alone = 0;
1691 	}
1692 
1693 	/* Calculate the length of the EA entry */
1694 	if (ap->a_uio == NULL) {
1695 		/* delete */
1696 		ealength = eapad1 = ealen = eapad2 = eacont = 0;
1697 	} else {
1698 		ealen = ap->a_uio->uio_resid;
1699 		ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1700 		eapad1 = 8 - (ealength % 8);
1701 		if (eapad1 == 8)
1702 			eapad1 = 0;
1703 		eacont = ealength + eapad1;
1704 		eapad2 = 8 - (ealen % 8);
1705 		if (eapad2 == 8)
1706 			eapad2 = 0;
1707 		ealength += eapad1 + ealen + eapad2;
1708 	}
1709 
1710 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1711 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1712 	easize = ip->i_ea_len;
1713 
1714 	olen = ffs_findextattr(eae, easize,
1715 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1716 	if (olen == -1 && ealength == 0) {
1717 		/* delete but nonexistent */
1718 		free(eae, M_TEMP);
1719 		if (stand_alone)
1720 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1721 		return(ENOATTR);
1722 	}
1723         if (olen == -1) {
1724 		/* new, append at end */
1725 		p = eae + easize;
1726 		easize += ealength;
1727 	} else {
1728 		bcopy(p, &ul, sizeof ul);
1729 		i = p - eae + ul;
1730 		if (ul != ealength) {
1731 			bcopy(p + ul, p + ealength, easize - i);
1732 			easize += (ealength - ul);
1733 		}
1734 	}
1735 	if (easize > NXADDR * fs->fs_bsize) {
1736 		free(eae, M_TEMP);
1737 		if (stand_alone)
1738 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1739 		else if (ip->i_ea_error == 0)
1740 			ip->i_ea_error = ENOSPC;
1741 		return(ENOSPC);
1742 	}
1743 	if (ealength != 0) {
1744 		bcopy(&ealength, p, sizeof(ealength));
1745 		p += sizeof(ealength);
1746 		*p++ = ap->a_attrnamespace;
1747 		*p++ = eapad2;
1748 		*p++ = strlen(ap->a_name);
1749 		strcpy(p, ap->a_name);
1750 		p += strlen(ap->a_name);
1751 		bzero(p, eapad1);
1752 		p += eapad1;
1753 		error = uiomove(p, ealen, ap->a_uio);
1754 		if (error) {
1755 			free(eae, M_TEMP);
1756 			if (stand_alone)
1757 				ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1758 			else if (ip->i_ea_error == 0)
1759 				ip->i_ea_error = error;
1760 			return(error);
1761 		}
1762 		p += ealen;
1763 		bzero(p, eapad2);
1764 	}
1765 	p = ip->i_ea_area;
1766 	ip->i_ea_area = eae;
1767 	ip->i_ea_len = easize;
1768 	free(p, M_TEMP);
1769 	if (stand_alone)
1770 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1771 	return(error);
1772 }
1773