xref: /freebsd/sys/ufs/ffs/ffs_balloc.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
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  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * Copyright (c) 1982, 1986, 1989, 1993
33  *	The Regents of the University of California.  All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 4. Neither the name of the University nor the names of its contributors
44  *    may be used to endorse or promote products derived from this software
45  *    without specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
57  * SUCH DAMAGE.
58  *
59  *	@(#)ffs_balloc.c	8.8 (Berkeley) 6/16/95
60  */
61 
62 #include <sys/cdefs.h>
63 __FBSDID("$FreeBSD$");
64 
65 #include <sys/param.h>
66 #include <sys/systm.h>
67 #include <sys/bio.h>
68 #include <sys/buf.h>
69 #include <sys/lock.h>
70 #include <sys/mount.h>
71 #include <sys/vnode.h>
72 
73 #include <ufs/ufs/quota.h>
74 #include <ufs/ufs/inode.h>
75 #include <ufs/ufs/ufs_extern.h>
76 #include <ufs/ufs/extattr.h>
77 #include <ufs/ufs/ufsmount.h>
78 
79 #include <ufs/ffs/fs.h>
80 #include <ufs/ffs/ffs_extern.h>
81 
82 /*
83  * Balloc defines the structure of filesystem storage
84  * by allocating the physical blocks on a device given
85  * the inode and the logical block number in a file.
86  * This is the allocation strategy for UFS1. Below is
87  * the allocation strategy for UFS2.
88  */
89 int
90 ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
91     struct ucred *cred, int flags, struct buf **bpp)
92 {
93 	struct inode *ip;
94 	struct ufs1_dinode *dp;
95 	ufs_lbn_t lbn, lastlbn;
96 	struct fs *fs;
97 	ufs1_daddr_t nb;
98 	struct buf *bp, *nbp;
99 	struct ufsmount *ump;
100 	struct indir indirs[NIADDR + 2];
101 	int deallocated, osize, nsize, num, i, error;
102 	ufs2_daddr_t newb;
103 	ufs1_daddr_t *bap, pref;
104 	ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
105 	ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
106 	int unwindidx = -1;
107 	int saved_inbdflush;
108 	static struct timeval lastfail;
109 	static int curfail;
110 	int gbflags, reclaimed;
111 
112 	ip = VTOI(vp);
113 	dp = ip->i_din1;
114 	fs = ip->i_fs;
115 	ump = ip->i_ump;
116 	lbn = lblkno(fs, startoffset);
117 	size = blkoff(fs, startoffset) + size;
118 	reclaimed = 0;
119 	if (size > fs->fs_bsize)
120 		panic("ffs_balloc_ufs1: blk too big");
121 	*bpp = NULL;
122 	if (flags & IO_EXT)
123 		return (EOPNOTSUPP);
124 	if (lbn < 0)
125 		return (EFBIG);
126 	gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
127 
128 	if (DOINGSOFTDEP(vp))
129 		softdep_prealloc(vp, MNT_WAIT);
130 	/*
131 	 * If the next write will extend the file into a new block,
132 	 * and the file is currently composed of a fragment
133 	 * this fragment has to be extended to be a full block.
134 	 */
135 	lastlbn = lblkno(fs, ip->i_size);
136 	if (lastlbn < NDADDR && lastlbn < lbn) {
137 		nb = lastlbn;
138 		osize = blksize(fs, ip, nb);
139 		if (osize < fs->fs_bsize && osize > 0) {
140 			UFS_LOCK(ump);
141 			error = ffs_realloccg(ip, nb, dp->di_db[nb],
142 			   ffs_blkpref_ufs1(ip, lastlbn, (int)nb,
143 			   &dp->di_db[0]), osize, (int)fs->fs_bsize, flags,
144 			   cred, &bp);
145 			if (error)
146 				return (error);
147 			if (DOINGSOFTDEP(vp))
148 				softdep_setup_allocdirect(ip, nb,
149 				    dbtofsb(fs, bp->b_blkno), dp->di_db[nb],
150 				    fs->fs_bsize, osize, bp);
151 			ip->i_size = smalllblktosize(fs, nb + 1);
152 			dp->di_size = ip->i_size;
153 			dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
154 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
155 			if (flags & IO_SYNC)
156 				bwrite(bp);
157 			else
158 				bawrite(bp);
159 		}
160 	}
161 	/*
162 	 * The first NDADDR blocks are direct blocks
163 	 */
164 	if (lbn < NDADDR) {
165 		if (flags & BA_METAONLY)
166 			panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
167 		nb = dp->di_db[lbn];
168 		if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
169 			error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
170 			if (error) {
171 				brelse(bp);
172 				return (error);
173 			}
174 			bp->b_blkno = fsbtodb(fs, nb);
175 			*bpp = bp;
176 			return (0);
177 		}
178 		if (nb != 0) {
179 			/*
180 			 * Consider need to reallocate a fragment.
181 			 */
182 			osize = fragroundup(fs, blkoff(fs, ip->i_size));
183 			nsize = fragroundup(fs, size);
184 			if (nsize <= osize) {
185 				error = bread(vp, lbn, osize, NOCRED, &bp);
186 				if (error) {
187 					brelse(bp);
188 					return (error);
189 				}
190 				bp->b_blkno = fsbtodb(fs, nb);
191 			} else {
192 				UFS_LOCK(ump);
193 				error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
194 				    ffs_blkpref_ufs1(ip, lbn, (int)lbn,
195 				    &dp->di_db[0]), osize, nsize, flags,
196 				    cred, &bp);
197 				if (error)
198 					return (error);
199 				if (DOINGSOFTDEP(vp))
200 					softdep_setup_allocdirect(ip, lbn,
201 					    dbtofsb(fs, bp->b_blkno), nb,
202 					    nsize, osize, bp);
203 			}
204 		} else {
205 			if (ip->i_size < smalllblktosize(fs, lbn + 1))
206 				nsize = fragroundup(fs, size);
207 			else
208 				nsize = fs->fs_bsize;
209 			UFS_LOCK(ump);
210 			error = ffs_alloc(ip, lbn,
211 			    ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
212 			    nsize, flags, cred, &newb);
213 			if (error)
214 				return (error);
215 			bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
216 			bp->b_blkno = fsbtodb(fs, newb);
217 			if (flags & BA_CLRBUF)
218 				vfs_bio_clrbuf(bp);
219 			if (DOINGSOFTDEP(vp))
220 				softdep_setup_allocdirect(ip, lbn, newb, 0,
221 				    nsize, 0, bp);
222 		}
223 		dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
224 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
225 		*bpp = bp;
226 		return (0);
227 	}
228 	/*
229 	 * Determine the number of levels of indirection.
230 	 */
231 	pref = 0;
232 	if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
233 		return(error);
234 #ifdef INVARIANTS
235 	if (num < 1)
236 		panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block");
237 #endif
238 	saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
239 	/*
240 	 * Fetch the first indirect block allocating if necessary.
241 	 */
242 	--num;
243 	nb = dp->di_ib[indirs[0].in_off];
244 	allocib = NULL;
245 	allocblk = allociblk;
246 	lbns_remfree = lbns;
247 	if (nb == 0) {
248 		UFS_LOCK(ump);
249 		pref = ffs_blkpref_ufs1(ip, lbn, -indirs[0].in_off - 1,
250 		    (ufs1_daddr_t *)0);
251 		if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
252 		    flags, cred, &newb)) != 0) {
253 			curthread_pflags_restore(saved_inbdflush);
254 			return (error);
255 		}
256 		pref = newb + fs->fs_frag;
257 		nb = newb;
258 		*allocblk++ = nb;
259 		*lbns_remfree++ = indirs[1].in_lbn;
260 		bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, gbflags);
261 		bp->b_blkno = fsbtodb(fs, nb);
262 		vfs_bio_clrbuf(bp);
263 		if (DOINGSOFTDEP(vp)) {
264 			softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
265 			    newb, 0, fs->fs_bsize, 0, bp);
266 			bdwrite(bp);
267 		} else {
268 			/*
269 			 * Write synchronously so that indirect blocks
270 			 * never point at garbage.
271 			 */
272 			if (DOINGASYNC(vp))
273 				bdwrite(bp);
274 			else if ((error = bwrite(bp)) != 0)
275 				goto fail;
276 		}
277 		allocib = &dp->di_ib[indirs[0].in_off];
278 		*allocib = nb;
279 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
280 	}
281 	/*
282 	 * Fetch through the indirect blocks, allocating as necessary.
283 	 */
284 retry:
285 	for (i = 1;;) {
286 		error = bread(vp,
287 		    indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
288 		if (error) {
289 			brelse(bp);
290 			goto fail;
291 		}
292 		bap = (ufs1_daddr_t *)bp->b_data;
293 		nb = bap[indirs[i].in_off];
294 		if (i == num)
295 			break;
296 		i += 1;
297 		if (nb != 0) {
298 			bqrelse(bp);
299 			continue;
300 		}
301 		UFS_LOCK(ump);
302 		/*
303 		 * If parent indirect has just been allocated, try to cluster
304 		 * immediately following it.
305 		 */
306 		if (pref == 0)
307 			pref = ffs_blkpref_ufs1(ip, lbn, i - num - 1,
308 			    (ufs1_daddr_t *)0);
309 		if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
310 		    flags | IO_BUFLOCKED, cred, &newb)) != 0) {
311 			brelse(bp);
312 			if (++reclaimed == 1) {
313 				UFS_LOCK(ump);
314 				softdep_request_cleanup(fs, vp, cred,
315 				    FLUSH_BLOCKS_WAIT);
316 				UFS_UNLOCK(ump);
317 				goto retry;
318 			}
319 			if (ppsratecheck(&lastfail, &curfail, 1)) {
320 				ffs_fserr(fs, ip->i_number, "filesystem full");
321 				uprintf("\n%s: write failed, filesystem "
322 				    "is full\n", fs->fs_fsmnt);
323 			}
324 			goto fail;
325 		}
326 		pref = newb + fs->fs_frag;
327 		nb = newb;
328 		*allocblk++ = nb;
329 		*lbns_remfree++ = indirs[i].in_lbn;
330 		nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
331 		nbp->b_blkno = fsbtodb(fs, nb);
332 		vfs_bio_clrbuf(nbp);
333 		if (DOINGSOFTDEP(vp)) {
334 			softdep_setup_allocindir_meta(nbp, ip, bp,
335 			    indirs[i - 1].in_off, nb);
336 			bdwrite(nbp);
337 		} else {
338 			/*
339 			 * Write synchronously so that indirect blocks
340 			 * never point at garbage.
341 			 */
342 			if ((error = bwrite(nbp)) != 0) {
343 				brelse(bp);
344 				goto fail;
345 			}
346 		}
347 		bap[indirs[i - 1].in_off] = nb;
348 		if (allocib == NULL && unwindidx < 0)
349 			unwindidx = i - 1;
350 		/*
351 		 * If required, write synchronously, otherwise use
352 		 * delayed write.
353 		 */
354 		if (flags & IO_SYNC) {
355 			bwrite(bp);
356 		} else {
357 			if (bp->b_bufsize == fs->fs_bsize)
358 				bp->b_flags |= B_CLUSTEROK;
359 			bdwrite(bp);
360 		}
361 	}
362 	/*
363 	 * If asked only for the indirect block, then return it.
364 	 */
365 	if (flags & BA_METAONLY) {
366 		curthread_pflags_restore(saved_inbdflush);
367 		*bpp = bp;
368 		return (0);
369 	}
370 	/*
371 	 * Get the data block, allocating if necessary.
372 	 */
373 	if (nb == 0) {
374 		UFS_LOCK(ump);
375 		/*
376 		 * If allocating metadata at the front of the cylinder
377 		 * group and parent indirect block has just been allocated,
378 		 * then cluster next to it if it is the first indirect in
379 		 * the file. Otherwise it has been allocated in the metadata
380 		 * area, so we want to find our own place out in the data area.
381 		 */
382 		if (pref == 0 || (lbn > NDADDR && fs->fs_metaspace != 0))
383 			pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off,
384 			    &bap[0]);
385 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
386 		    flags | IO_BUFLOCKED, cred, &newb);
387 		if (error) {
388 			brelse(bp);
389 			if (++reclaimed == 1) {
390 				UFS_LOCK(ump);
391 				softdep_request_cleanup(fs, vp, cred,
392 				    FLUSH_BLOCKS_WAIT);
393 				UFS_UNLOCK(ump);
394 				goto retry;
395 			}
396 			if (ppsratecheck(&lastfail, &curfail, 1)) {
397 				ffs_fserr(fs, ip->i_number, "filesystem full");
398 				uprintf("\n%s: write failed, filesystem "
399 				    "is full\n", fs->fs_fsmnt);
400 			}
401 			goto fail;
402 		}
403 		nb = newb;
404 		*allocblk++ = nb;
405 		*lbns_remfree++ = lbn;
406 		nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
407 		nbp->b_blkno = fsbtodb(fs, nb);
408 		if (flags & BA_CLRBUF)
409 			vfs_bio_clrbuf(nbp);
410 		if (DOINGSOFTDEP(vp))
411 			softdep_setup_allocindir_page(ip, lbn, bp,
412 			    indirs[i].in_off, nb, 0, nbp);
413 		bap[indirs[i].in_off] = nb;
414 		/*
415 		 * If required, write synchronously, otherwise use
416 		 * delayed write.
417 		 */
418 		if (flags & IO_SYNC) {
419 			bwrite(bp);
420 		} else {
421 			if (bp->b_bufsize == fs->fs_bsize)
422 				bp->b_flags |= B_CLUSTEROK;
423 			bdwrite(bp);
424 		}
425 		curthread_pflags_restore(saved_inbdflush);
426 		*bpp = nbp;
427 		return (0);
428 	}
429 	brelse(bp);
430 	if (flags & BA_CLRBUF) {
431 		int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
432 		if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
433 			error = cluster_read(vp, ip->i_size, lbn,
434 			    (int)fs->fs_bsize, NOCRED,
435 			    MAXBSIZE, seqcount, gbflags, &nbp);
436 		} else {
437 			error = bread_gb(vp, lbn, (int)fs->fs_bsize, NOCRED,
438 			    gbflags, &nbp);
439 		}
440 		if (error) {
441 			brelse(nbp);
442 			goto fail;
443 		}
444 	} else {
445 		nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
446 		nbp->b_blkno = fsbtodb(fs, nb);
447 	}
448 	curthread_pflags_restore(saved_inbdflush);
449 	*bpp = nbp;
450 	return (0);
451 fail:
452 	curthread_pflags_restore(saved_inbdflush);
453 	/*
454 	 * If we have failed to allocate any blocks, simply return the error.
455 	 * This is the usual case and avoids the need to fsync the file.
456 	 */
457 	if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
458 		return (error);
459 	/*
460 	 * If we have failed part way through block allocation, we
461 	 * have to deallocate any indirect blocks that we have allocated.
462 	 * We have to fsync the file before we start to get rid of all
463 	 * of its dependencies so that we do not leave them dangling.
464 	 * We have to sync it at the end so that the soft updates code
465 	 * does not find any untracked changes. Although this is really
466 	 * slow, running out of disk space is not expected to be a common
467 	 * occurrence. The error return from fsync is ignored as we already
468 	 * have an error to return to the user.
469 	 *
470 	 * XXX Still have to journal the free below
471 	 */
472 	(void) ffs_syncvnode(vp, MNT_WAIT, 0);
473 	for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
474 	     blkp < allocblk; blkp++, lbns_remfree++) {
475 		/*
476 		 * We shall not leave the freed blocks on the vnode
477 		 * buffer object lists.
478 		 */
479 		bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0, GB_NOCREAT);
480 		if (bp != NULL) {
481 			bp->b_flags |= (B_INVAL | B_RELBUF);
482 			bp->b_flags &= ~B_ASYNC;
483 			brelse(bp);
484 		}
485 		deallocated += fs->fs_bsize;
486 	}
487 	if (allocib != NULL) {
488 		*allocib = 0;
489 	} else if (unwindidx >= 0) {
490 		int r;
491 
492 		r = bread(vp, indirs[unwindidx].in_lbn,
493 		    (int)fs->fs_bsize, NOCRED, &bp);
494 		if (r) {
495 			panic("Could not unwind indirect block, error %d", r);
496 			brelse(bp);
497 		} else {
498 			bap = (ufs1_daddr_t *)bp->b_data;
499 			bap[indirs[unwindidx].in_off] = 0;
500 			if (flags & IO_SYNC) {
501 				bwrite(bp);
502 			} else {
503 				if (bp->b_bufsize == fs->fs_bsize)
504 					bp->b_flags |= B_CLUSTEROK;
505 				bdwrite(bp);
506 			}
507 		}
508 	}
509 	if (deallocated) {
510 #ifdef QUOTA
511 		/*
512 		 * Restore user's disk quota because allocation failed.
513 		 */
514 		(void) chkdq(ip, -btodb(deallocated), cred, FORCE);
515 #endif
516 		dp->di_blocks -= btodb(deallocated);
517 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
518 	}
519 	(void) ffs_syncvnode(vp, MNT_WAIT, 0);
520 	/*
521 	 * After the buffers are invalidated and on-disk pointers are
522 	 * cleared, free the blocks.
523 	 */
524 	for (blkp = allociblk; blkp < allocblk; blkp++) {
525 		ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize,
526 		    ip->i_number, vp->v_type, NULL);
527 	}
528 	return (error);
529 }
530 
531 /*
532  * Balloc defines the structure of file system storage
533  * by allocating the physical blocks on a device given
534  * the inode and the logical block number in a file.
535  * This is the allocation strategy for UFS2. Above is
536  * the allocation strategy for UFS1.
537  */
538 int
539 ffs_balloc_ufs2(struct vnode *vp, off_t startoffset, int size,
540     struct ucred *cred, int flags, struct buf **bpp)
541 {
542 	struct inode *ip;
543 	struct ufs2_dinode *dp;
544 	ufs_lbn_t lbn, lastlbn;
545 	struct fs *fs;
546 	struct buf *bp, *nbp;
547 	struct ufsmount *ump;
548 	struct indir indirs[NIADDR + 2];
549 	ufs2_daddr_t nb, newb, *bap, pref;
550 	ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
551 	ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
552 	int deallocated, osize, nsize, num, i, error;
553 	int unwindidx = -1;
554 	int saved_inbdflush;
555 	static struct timeval lastfail;
556 	static int curfail;
557 	int gbflags, reclaimed;
558 
559 	ip = VTOI(vp);
560 	dp = ip->i_din2;
561 	fs = ip->i_fs;
562 	ump = ip->i_ump;
563 	lbn = lblkno(fs, startoffset);
564 	size = blkoff(fs, startoffset) + size;
565 	reclaimed = 0;
566 	if (size > fs->fs_bsize)
567 		panic("ffs_balloc_ufs2: blk too big");
568 	*bpp = NULL;
569 	if (lbn < 0)
570 		return (EFBIG);
571 	gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
572 
573 	if (DOINGSOFTDEP(vp))
574 		softdep_prealloc(vp, MNT_WAIT);
575 
576 	/*
577 	 * Check for allocating external data.
578 	 */
579 	if (flags & IO_EXT) {
580 		if (lbn >= NXADDR)
581 			return (EFBIG);
582 		/*
583 		 * If the next write will extend the data into a new block,
584 		 * and the data is currently composed of a fragment
585 		 * this fragment has to be extended to be a full block.
586 		 */
587 		lastlbn = lblkno(fs, dp->di_extsize);
588 		if (lastlbn < lbn) {
589 			nb = lastlbn;
590 			osize = sblksize(fs, dp->di_extsize, nb);
591 			if (osize < fs->fs_bsize && osize > 0) {
592 				UFS_LOCK(ump);
593 				error = ffs_realloccg(ip, -1 - nb,
594 				    dp->di_extb[nb],
595 				    ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
596 				    &dp->di_extb[0]), osize,
597 				    (int)fs->fs_bsize, flags, cred, &bp);
598 				if (error)
599 					return (error);
600 				if (DOINGSOFTDEP(vp))
601 					softdep_setup_allocext(ip, nb,
602 					    dbtofsb(fs, bp->b_blkno),
603 					    dp->di_extb[nb],
604 					    fs->fs_bsize, osize, bp);
605 				dp->di_extsize = smalllblktosize(fs, nb + 1);
606 				dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno);
607 				bp->b_xflags |= BX_ALTDATA;
608 				ip->i_flag |= IN_CHANGE;
609 				if (flags & IO_SYNC)
610 					bwrite(bp);
611 				else
612 					bawrite(bp);
613 			}
614 		}
615 		/*
616 		 * All blocks are direct blocks
617 		 */
618 		if (flags & BA_METAONLY)
619 			panic("ffs_balloc_ufs2: BA_METAONLY for ext block");
620 		nb = dp->di_extb[lbn];
621 		if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) {
622 			error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED,
623 			    gbflags, &bp);
624 			if (error) {
625 				brelse(bp);
626 				return (error);
627 			}
628 			bp->b_blkno = fsbtodb(fs, nb);
629 			bp->b_xflags |= BX_ALTDATA;
630 			*bpp = bp;
631 			return (0);
632 		}
633 		if (nb != 0) {
634 			/*
635 			 * Consider need to reallocate a fragment.
636 			 */
637 			osize = fragroundup(fs, blkoff(fs, dp->di_extsize));
638 			nsize = fragroundup(fs, size);
639 			if (nsize <= osize) {
640 				error = bread_gb(vp, -1 - lbn, osize, NOCRED,
641 				    gbflags, &bp);
642 				if (error) {
643 					brelse(bp);
644 					return (error);
645 				}
646 				bp->b_blkno = fsbtodb(fs, nb);
647 				bp->b_xflags |= BX_ALTDATA;
648 			} else {
649 				UFS_LOCK(ump);
650 				error = ffs_realloccg(ip, -1 - lbn,
651 				    dp->di_extb[lbn],
652 				    ffs_blkpref_ufs2(ip, lbn, (int)lbn,
653 				    &dp->di_extb[0]), osize, nsize, flags,
654 				    cred, &bp);
655 				if (error)
656 					return (error);
657 				bp->b_xflags |= BX_ALTDATA;
658 				if (DOINGSOFTDEP(vp))
659 					softdep_setup_allocext(ip, lbn,
660 					    dbtofsb(fs, bp->b_blkno), nb,
661 					    nsize, osize, bp);
662 			}
663 		} else {
664 			if (dp->di_extsize < smalllblktosize(fs, lbn + 1))
665 				nsize = fragroundup(fs, size);
666 			else
667 				nsize = fs->fs_bsize;
668 			UFS_LOCK(ump);
669 			error = ffs_alloc(ip, lbn,
670 			   ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]),
671 			   nsize, flags, cred, &newb);
672 			if (error)
673 				return (error);
674 			bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags);
675 			bp->b_blkno = fsbtodb(fs, newb);
676 			bp->b_xflags |= BX_ALTDATA;
677 			if (flags & BA_CLRBUF)
678 				vfs_bio_clrbuf(bp);
679 			if (DOINGSOFTDEP(vp))
680 				softdep_setup_allocext(ip, lbn, newb, 0,
681 				    nsize, 0, bp);
682 		}
683 		dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno);
684 		ip->i_flag |= IN_CHANGE;
685 		*bpp = bp;
686 		return (0);
687 	}
688 	/*
689 	 * If the next write will extend the file into a new block,
690 	 * and the file is currently composed of a fragment
691 	 * this fragment has to be extended to be a full block.
692 	 */
693 	lastlbn = lblkno(fs, ip->i_size);
694 	if (lastlbn < NDADDR && lastlbn < lbn) {
695 		nb = lastlbn;
696 		osize = blksize(fs, ip, nb);
697 		if (osize < fs->fs_bsize && osize > 0) {
698 			UFS_LOCK(ump);
699 			error = ffs_realloccg(ip, nb, dp->di_db[nb],
700 			    ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
701 			    &dp->di_db[0]), osize, (int)fs->fs_bsize,
702 			    flags, cred, &bp);
703 			if (error)
704 				return (error);
705 			if (DOINGSOFTDEP(vp))
706 				softdep_setup_allocdirect(ip, nb,
707 				    dbtofsb(fs, bp->b_blkno),
708 				    dp->di_db[nb],
709 				    fs->fs_bsize, osize, bp);
710 			ip->i_size = smalllblktosize(fs, nb + 1);
711 			dp->di_size = ip->i_size;
712 			dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
713 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
714 			if (flags & IO_SYNC)
715 				bwrite(bp);
716 			else
717 				bawrite(bp);
718 		}
719 	}
720 	/*
721 	 * The first NDADDR blocks are direct blocks
722 	 */
723 	if (lbn < NDADDR) {
724 		if (flags & BA_METAONLY)
725 			panic("ffs_balloc_ufs2: BA_METAONLY for direct block");
726 		nb = dp->di_db[lbn];
727 		if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
728 			error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED,
729 			    gbflags, &bp);
730 			if (error) {
731 				brelse(bp);
732 				return (error);
733 			}
734 			bp->b_blkno = fsbtodb(fs, nb);
735 			*bpp = bp;
736 			return (0);
737 		}
738 		if (nb != 0) {
739 			/*
740 			 * Consider need to reallocate a fragment.
741 			 */
742 			osize = fragroundup(fs, blkoff(fs, ip->i_size));
743 			nsize = fragroundup(fs, size);
744 			if (nsize <= osize) {
745 				error = bread_gb(vp, lbn, osize, NOCRED,
746 				    gbflags, &bp);
747 				if (error) {
748 					brelse(bp);
749 					return (error);
750 				}
751 				bp->b_blkno = fsbtodb(fs, nb);
752 			} else {
753 				UFS_LOCK(ump);
754 				error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
755 				    ffs_blkpref_ufs2(ip, lbn, (int)lbn,
756 				    &dp->di_db[0]), osize, nsize, flags,
757 				    cred, &bp);
758 				if (error)
759 					return (error);
760 				if (DOINGSOFTDEP(vp))
761 					softdep_setup_allocdirect(ip, lbn,
762 					    dbtofsb(fs, bp->b_blkno), nb,
763 					    nsize, osize, bp);
764 			}
765 		} else {
766 			if (ip->i_size < smalllblktosize(fs, lbn + 1))
767 				nsize = fragroundup(fs, size);
768 			else
769 				nsize = fs->fs_bsize;
770 			UFS_LOCK(ump);
771 			error = ffs_alloc(ip, lbn,
772 			    ffs_blkpref_ufs2(ip, lbn, (int)lbn,
773 				&dp->di_db[0]), nsize, flags, cred, &newb);
774 			if (error)
775 				return (error);
776 			bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
777 			bp->b_blkno = fsbtodb(fs, newb);
778 			if (flags & BA_CLRBUF)
779 				vfs_bio_clrbuf(bp);
780 			if (DOINGSOFTDEP(vp))
781 				softdep_setup_allocdirect(ip, lbn, newb, 0,
782 				    nsize, 0, bp);
783 		}
784 		dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
785 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
786 		*bpp = bp;
787 		return (0);
788 	}
789 	/*
790 	 * Determine the number of levels of indirection.
791 	 */
792 	pref = 0;
793 	if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
794 		return(error);
795 #ifdef INVARIANTS
796 	if (num < 1)
797 		panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block");
798 #endif
799 	saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
800 	/*
801 	 * Fetch the first indirect block allocating if necessary.
802 	 */
803 	--num;
804 	nb = dp->di_ib[indirs[0].in_off];
805 	allocib = NULL;
806 	allocblk = allociblk;
807 	lbns_remfree = lbns;
808 	if (nb == 0) {
809 		UFS_LOCK(ump);
810 		pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1,
811 		    (ufs2_daddr_t *)0);
812 		if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
813 		    flags, cred, &newb)) != 0) {
814 			curthread_pflags_restore(saved_inbdflush);
815 			return (error);
816 		}
817 		pref = newb + fs->fs_frag;
818 		nb = newb;
819 		*allocblk++ = nb;
820 		*lbns_remfree++ = indirs[1].in_lbn;
821 		bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0,
822 		    GB_UNMAPPED);
823 		bp->b_blkno = fsbtodb(fs, nb);
824 		vfs_bio_clrbuf(bp);
825 		if (DOINGSOFTDEP(vp)) {
826 			softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
827 			    newb, 0, fs->fs_bsize, 0, bp);
828 			bdwrite(bp);
829 		} else {
830 			/*
831 			 * Write synchronously so that indirect blocks
832 			 * never point at garbage.
833 			 */
834 			if (DOINGASYNC(vp))
835 				bdwrite(bp);
836 			else if ((error = bwrite(bp)) != 0)
837 				goto fail;
838 		}
839 		allocib = &dp->di_ib[indirs[0].in_off];
840 		*allocib = nb;
841 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
842 	}
843 	/*
844 	 * Fetch through the indirect blocks, allocating as necessary.
845 	 */
846 retry:
847 	for (i = 1;;) {
848 		error = bread(vp,
849 		    indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
850 		if (error) {
851 			brelse(bp);
852 			goto fail;
853 		}
854 		bap = (ufs2_daddr_t *)bp->b_data;
855 		nb = bap[indirs[i].in_off];
856 		if (i == num)
857 			break;
858 		i += 1;
859 		if (nb != 0) {
860 			bqrelse(bp);
861 			continue;
862 		}
863 		UFS_LOCK(ump);
864 		/*
865 		 * If parent indirect has just been allocated, try to cluster
866 		 * immediately following it.
867 		 */
868 		if (pref == 0)
869 			pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1,
870 			    (ufs2_daddr_t *)0);
871 		if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
872 		    flags | IO_BUFLOCKED, cred, &newb)) != 0) {
873 			brelse(bp);
874 			if (++reclaimed == 1) {
875 				UFS_LOCK(ump);
876 				softdep_request_cleanup(fs, vp, cred,
877 				    FLUSH_BLOCKS_WAIT);
878 				UFS_UNLOCK(ump);
879 				goto retry;
880 			}
881 			if (ppsratecheck(&lastfail, &curfail, 1)) {
882 				ffs_fserr(fs, ip->i_number, "filesystem full");
883 				uprintf("\n%s: write failed, filesystem "
884 				    "is full\n", fs->fs_fsmnt);
885 			}
886 			goto fail;
887 		}
888 		pref = newb + fs->fs_frag;
889 		nb = newb;
890 		*allocblk++ = nb;
891 		*lbns_remfree++ = indirs[i].in_lbn;
892 		nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0,
893 		    GB_UNMAPPED);
894 		nbp->b_blkno = fsbtodb(fs, nb);
895 		vfs_bio_clrbuf(nbp);
896 		if (DOINGSOFTDEP(vp)) {
897 			softdep_setup_allocindir_meta(nbp, ip, bp,
898 			    indirs[i - 1].in_off, nb);
899 			bdwrite(nbp);
900 		} else {
901 			/*
902 			 * Write synchronously so that indirect blocks
903 			 * never point at garbage.
904 			 */
905 			if ((error = bwrite(nbp)) != 0) {
906 				brelse(bp);
907 				goto fail;
908 			}
909 		}
910 		bap[indirs[i - 1].in_off] = nb;
911 		if (allocib == NULL && unwindidx < 0)
912 			unwindidx = i - 1;
913 		/*
914 		 * If required, write synchronously, otherwise use
915 		 * delayed write.
916 		 */
917 		if (flags & IO_SYNC) {
918 			bwrite(bp);
919 		} else {
920 			if (bp->b_bufsize == fs->fs_bsize)
921 				bp->b_flags |= B_CLUSTEROK;
922 			bdwrite(bp);
923 		}
924 	}
925 	/*
926 	 * If asked only for the indirect block, then return it.
927 	 */
928 	if (flags & BA_METAONLY) {
929 		curthread_pflags_restore(saved_inbdflush);
930 		*bpp = bp;
931 		return (0);
932 	}
933 	/*
934 	 * Get the data block, allocating if necessary.
935 	 */
936 	if (nb == 0) {
937 		UFS_LOCK(ump);
938 		/*
939 		 * If allocating metadata at the front of the cylinder
940 		 * group and parent indirect block has just been allocated,
941 		 * then cluster next to it if it is the first indirect in
942 		 * the file. Otherwise it has been allocated in the metadata
943 		 * area, so we want to find our own place out in the data area.
944 		 */
945 		if (pref == 0 || (lbn > NDADDR && fs->fs_metaspace != 0))
946 			pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off,
947 			    &bap[0]);
948 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
949 		    flags | IO_BUFLOCKED, cred, &newb);
950 		if (error) {
951 			brelse(bp);
952 			if (++reclaimed == 1) {
953 				UFS_LOCK(ump);
954 				softdep_request_cleanup(fs, vp, cred,
955 				    FLUSH_BLOCKS_WAIT);
956 				UFS_UNLOCK(ump);
957 				goto retry;
958 			}
959 			if (ppsratecheck(&lastfail, &curfail, 1)) {
960 				ffs_fserr(fs, ip->i_number, "filesystem full");
961 				uprintf("\n%s: write failed, filesystem "
962 				    "is full\n", fs->fs_fsmnt);
963 			}
964 			goto fail;
965 		}
966 		nb = newb;
967 		*allocblk++ = nb;
968 		*lbns_remfree++ = lbn;
969 		nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
970 		nbp->b_blkno = fsbtodb(fs, nb);
971 		if (flags & BA_CLRBUF)
972 			vfs_bio_clrbuf(nbp);
973 		if (DOINGSOFTDEP(vp))
974 			softdep_setup_allocindir_page(ip, lbn, bp,
975 			    indirs[i].in_off, nb, 0, nbp);
976 		bap[indirs[i].in_off] = nb;
977 		/*
978 		 * If required, write synchronously, otherwise use
979 		 * delayed write.
980 		 */
981 		if (flags & IO_SYNC) {
982 			bwrite(bp);
983 		} else {
984 			if (bp->b_bufsize == fs->fs_bsize)
985 				bp->b_flags |= B_CLUSTEROK;
986 			bdwrite(bp);
987 		}
988 		curthread_pflags_restore(saved_inbdflush);
989 		*bpp = nbp;
990 		return (0);
991 	}
992 	brelse(bp);
993 	/*
994 	 * If requested clear invalid portions of the buffer.  If we
995 	 * have to do a read-before-write (typical if BA_CLRBUF is set),
996 	 * try to do some read-ahead in the sequential case to reduce
997 	 * the number of I/O transactions.
998 	 */
999 	if (flags & BA_CLRBUF) {
1000 		int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
1001 		if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
1002 			error = cluster_read(vp, ip->i_size, lbn,
1003 			    (int)fs->fs_bsize, NOCRED,
1004 			    MAXBSIZE, seqcount, gbflags, &nbp);
1005 		} else {
1006 			error = bread_gb(vp, lbn, (int)fs->fs_bsize,
1007 			    NOCRED, gbflags, &nbp);
1008 		}
1009 		if (error) {
1010 			brelse(nbp);
1011 			goto fail;
1012 		}
1013 	} else {
1014 		nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
1015 		nbp->b_blkno = fsbtodb(fs, nb);
1016 	}
1017 	curthread_pflags_restore(saved_inbdflush);
1018 	*bpp = nbp;
1019 	return (0);
1020 fail:
1021 	curthread_pflags_restore(saved_inbdflush);
1022 	/*
1023 	 * If we have failed to allocate any blocks, simply return the error.
1024 	 * This is the usual case and avoids the need to fsync the file.
1025 	 */
1026 	if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
1027 		return (error);
1028 	/*
1029 	 * If we have failed part way through block allocation, we
1030 	 * have to deallocate any indirect blocks that we have allocated.
1031 	 * We have to fsync the file before we start to get rid of all
1032 	 * of its dependencies so that we do not leave them dangling.
1033 	 * We have to sync it at the end so that the soft updates code
1034 	 * does not find any untracked changes. Although this is really
1035 	 * slow, running out of disk space is not expected to be a common
1036 	 * occurrence. The error return from fsync is ignored as we already
1037 	 * have an error to return to the user.
1038 	 *
1039 	 * XXX Still have to journal the free below
1040 	 */
1041 	(void) ffs_syncvnode(vp, MNT_WAIT, 0);
1042 	for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
1043 	     blkp < allocblk; blkp++, lbns_remfree++) {
1044 		/*
1045 		 * We shall not leave the freed blocks on the vnode
1046 		 * buffer object lists.
1047 		 */
1048 		bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0, GB_NOCREAT);
1049 		if (bp != NULL) {
1050 			bp->b_flags |= (B_INVAL | B_RELBUF);
1051 			bp->b_flags &= ~B_ASYNC;
1052 			brelse(bp);
1053 		}
1054 		deallocated += fs->fs_bsize;
1055 	}
1056 	if (allocib != NULL) {
1057 		*allocib = 0;
1058 	} else if (unwindidx >= 0) {
1059 		int r;
1060 
1061 		r = bread(vp, indirs[unwindidx].in_lbn,
1062 		    (int)fs->fs_bsize, NOCRED, &bp);
1063 		if (r) {
1064 			panic("Could not unwind indirect block, error %d", r);
1065 			brelse(bp);
1066 		} else {
1067 			bap = (ufs2_daddr_t *)bp->b_data;
1068 			bap[indirs[unwindidx].in_off] = 0;
1069 			if (flags & IO_SYNC) {
1070 				bwrite(bp);
1071 			} else {
1072 				if (bp->b_bufsize == fs->fs_bsize)
1073 					bp->b_flags |= B_CLUSTEROK;
1074 				bdwrite(bp);
1075 			}
1076 		}
1077 	}
1078 	if (deallocated) {
1079 #ifdef QUOTA
1080 		/*
1081 		 * Restore user's disk quota because allocation failed.
1082 		 */
1083 		(void) chkdq(ip, -btodb(deallocated), cred, FORCE);
1084 #endif
1085 		dp->di_blocks -= btodb(deallocated);
1086 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1087 	}
1088 	(void) ffs_syncvnode(vp, MNT_WAIT, 0);
1089 	/*
1090 	 * After the buffers are invalidated and on-disk pointers are
1091 	 * cleared, free the blocks.
1092 	 */
1093 	for (blkp = allociblk; blkp < allocblk; blkp++) {
1094 		ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize,
1095 		    ip->i_number, vp->v_type, NULL);
1096 	}
1097 	return (error);
1098 }
1099