xref: /freebsd/sys/ufs/ufs/ufs_bmap.c (revision 8ad8643a66735d28dac53a772856c94ca65b2bf3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/systm.h>
38 #include <sys/bio.h>
39 #include <sys/buf.h>
40 #include <sys/proc.h>
41 #include <sys/rwlock.h>
42 #include <sys/vnode.h>
43 #include <sys/mount.h>
44 #include <sys/racct.h>
45 #include <sys/resourcevar.h>
46 #include <sys/sched.h>
47 #include <sys/sf_buf.h>
48 #include <sys/stat.h>
49 
50 #include <vm/vm.h>
51 #include <vm/vm_object.h>
52 #include <vm/vm_page.h>
53 #include <vm/vnode_pager.h>
54 
55 #include <ufs/ufs/extattr.h>
56 #include <ufs/ufs/quota.h>
57 #include <ufs/ufs/inode.h>
58 #include <ufs/ufs/ufsmount.h>
59 #include <ufs/ufs/ufs_extern.h>
60 
61 static ufs_lbn_t lbn_count(struct ufsmount *, int);
62 static int readindir(struct vnode *, ufs_lbn_t, ufs2_daddr_t, bool,
63     struct buf **);
64 
65 static int ufs_bmap_use_unmapped = 1;
66 
67 SYSCTL_INT(_vfs_ufs, OID_AUTO, bmap_use_unmapped, CTLFLAG_RWTUN,
68     &ufs_bmap_use_unmapped, 0, "UFS bmap uses unmapped bufs");
69 
70 /*
71  * Bmap converts the logical block number of a file to its physical block
72  * number on the disk. The conversion is done by using the logical block
73  * number to index into the array of block pointers described by the dinode.
74  */
75 int
ufs_bmap(struct vop_bmap_args * ap)76 ufs_bmap(
77 	struct vop_bmap_args /* {
78 		struct vnode *a_vp;
79 		daddr_t a_bn;
80 		struct bufobj **a_bop;
81 		daddr_t *a_bnp;
82 		int *a_runp;
83 		int *a_runb;
84 	} */ *ap)
85 {
86 	ufs2_daddr_t blkno;
87 	int error;
88 
89 	/*
90 	 * Check for underlying vnode requests and ensure that logical
91 	 * to physical mapping is requested.
92 	 */
93 	if (ap->a_bop != NULL)
94 		*ap->a_bop = &VFSTOUFS(ap->a_vp->v_mount)->um_devvp->v_bufobj;
95 	if (ap->a_bnp == NULL)
96 		return (0);
97 
98 	error = ufs_bmaparray(ap->a_vp, ap->a_bn, &blkno, NULL,
99 	    ap->a_runp, ap->a_runb);
100 	*ap->a_bnp = blkno;
101 	return (error);
102 }
103 
104 static int
readindir(struct vnode * vp,ufs_lbn_t lbn,ufs2_daddr_t daddr,bool allow_unmapped,struct buf ** bpp)105 readindir(struct vnode *vp,
106 	ufs_lbn_t lbn,
107 	ufs2_daddr_t daddr,
108 	bool allow_unmapped,
109 	struct buf **bpp)
110 {
111 	struct buf *bp;
112 	struct mount *mp;
113 	struct ufsmount *ump;
114 	struct inode *ip;
115 	int error, gbflags;
116 
117 	mp = vp->v_mount;
118 	ump = VFSTOUFS(mp);
119 	ip = VTOI(vp);
120 
121 	gbflags = allow_unmapped && !I_IS_UFS1(ip) &&
122 	    ufs_bmap_use_unmapped ? GB_UNMAPPED : 0;
123 	bp = getblk(vp, lbn, mp->mnt_stat.f_iosize, 0, 0, gbflags);
124 	if ((bp->b_flags & B_CACHE) == 0) {
125 		KASSERT(daddr != 0,
126 		    ("readindir: indirect block not in cache"));
127 
128 		bp->b_blkno = blkptrtodb(ump, daddr);
129 		bp->b_iocmd = BIO_READ;
130 		bp->b_flags &= ~B_INVAL;
131 		bp->b_ioflags &= ~BIO_ERROR;
132 		vfs_busy_pages(bp, 0);
133 		bp->b_iooffset = dbtob(bp->b_blkno);
134 		bstrategy(bp);
135 #ifdef RACCT
136 		if (racct_enable) {
137 			PROC_LOCK(curproc);
138 			racct_add_buf(curproc, bp, 0);
139 			PROC_UNLOCK(curproc);
140 		}
141 #endif
142 		curthread->td_ru.ru_inblock++;
143 		error = bufwait(bp);
144 		if (error != 0) {
145 			brelse(bp);
146 			return (error);
147 		}
148 	}
149 	*bpp = bp;
150 	return (0);
151 }
152 
153 /*
154  * Indirect blocks are now on the vnode for the file.  They are given negative
155  * logical block numbers.  Indirect blocks are addressed by the negative
156  * address of the first data block to which they point.  Double indirect blocks
157  * are addressed by one less than the address of the first indirect block to
158  * which they point.  Triple indirect blocks are addressed by one less than
159  * the address of the first double indirect block to which they point.
160  *
161  * ufs_bmaparray does the bmap conversion, and if requested returns the
162  * array of logical blocks which must be traversed to get to a block.
163  * Each entry contains the offset into that block that gets you to the
164  * next block and the disk address of the block (if it is assigned).
165  */
166 
167 static void *
ufs_bm_sf_get(struct buf * bp,int32_t pgidx,struct sf_buf ** sfp)168 ufs_bm_sf_get(struct buf *bp, int32_t pgidx, struct sf_buf **sfp)
169 {
170 	struct sf_buf *sf;
171 
172 	sched_pin();
173 	sf = sf_buf_alloc(bp->b_pages[pgidx], SFB_CPUPRIVATE);
174 	*sfp = sf;
175 	return (sf_buf_kva(sf));
176 }
177 
178 static void
ufs_bm_sf_put(struct sf_buf * sf)179 ufs_bm_sf_put(struct sf_buf *sf)
180 {
181 	sf_buf_free(sf);
182 	sched_unpin();
183 }
184 
185 int
ufs_bmaparray(struct vnode * vp,ufs2_daddr_t bn,ufs2_daddr_t * bnp,struct buf * nbp,int * runp,int * runb)186 ufs_bmaparray(struct vnode *vp,
187 	ufs2_daddr_t bn,
188 	ufs2_daddr_t *bnp,
189 	struct buf *nbp,
190 	int *runp,
191 	int *runb)
192 {
193 	struct inode *ip;
194 	struct buf *bp;
195 	struct ufsmount *ump;
196 	struct mount *mp;
197 	struct indir a[UFS_NIADDR+1], *ap;
198 	struct sf_buf *sf;
199 	ufs2_daddr_t daddr;
200 	ufs_lbn_t metalbn;
201 	int error, num, maxrun = 0;
202 	int *nump;
203 	ufs1_daddr_t *daddr1p;
204 	ufs2_daddr_t pgbn, daddrppg, prevdaddr, *daddr2p;
205 	int32_t daddrsz, boff, pgidx, pgoff;
206 	void *pgaddr;
207 	bool isseq;
208 
209 	ap = NULL;
210 	ip = VTOI(vp);
211 	mp = vp->v_mount;
212 	ump = VFSTOUFS(mp);
213 
214 	if (runp) {
215 		maxrun = mp->mnt_iosize_max / mp->mnt_stat.f_iosize - 1;
216 		*runp = 0;
217 	}
218 
219 	if (runb) {
220 		*runb = 0;
221 	}
222 
223 	ap = a;
224 	nump = &num;
225 	error = ufs_getlbns(vp, bn, ap, nump);
226 	if (error)
227 		return (error);
228 
229 	num = *nump;
230 	if (num == 0) {
231 		if (bn >= 0 && bn < UFS_NDADDR) {
232 			*bnp = blkptrtodb(ump, DIP(ip, i_db[bn]));
233 		} else if (bn < 0 && bn >= -UFS_NXADDR) {
234 			*bnp = blkptrtodb(ump, ip->i_din2->di_extb[-1 - bn]);
235 			if (*bnp == 0)
236 				*bnp = -1;
237 			if (nbp == NULL) {
238 				/* indirect block not found */
239 				return (EINVAL);
240 			}
241 			nbp->b_xflags |= BX_ALTDATA;
242 			return (0);
243 		} else {
244 			/* blkno out of range */
245 			return (EINVAL);
246 		}
247 		/*
248 		 * Since this is FFS independent code, we are out of
249 		 * scope for the definitions of BLK_NOCOPY and
250 		 * BLK_SNAP, but we do know that they will fall in
251 		 * the range 1..um_seqinc, so we use that test and
252 		 * return a request for a zeroed out buffer if attempts
253 		 * are made to read a BLK_NOCOPY or BLK_SNAP block.
254 		 */
255 		if (IS_SNAPSHOT(ip) && DIP(ip, i_db[bn]) > 0 &&
256 		    DIP(ip, i_db[bn]) < ump->um_seqinc) {
257 			*bnp = -1;
258 		} else if (*bnp == 0) {
259 			*bnp = IS_SNAPSHOT(ip) ? blkptrtodb(ump,
260 			    bn * ump->um_seqinc) : -1;
261 		} else if (runp) {
262 			ufs2_daddr_t bnb = bn;
263 			for (++bn; bn < UFS_NDADDR && *runp < maxrun &&
264 			    is_sequential(ump, DIP(ip, i_db[bn - 1]),
265 			    DIP(ip, i_db[bn]));
266 			    ++bn, ++*runp);
267 			bn = bnb;
268 			if (runb && (bn > 0)) {
269 				for (--bn; (bn >= 0) && (*runb < maxrun) &&
270 					is_sequential(ump, DIP(ip, i_db[bn]),
271 						DIP(ip, i_db[bn+1]));
272 						--bn, ++*runb);
273 			}
274 		}
275 		return (0);
276 	}
277 
278 	/* Get disk address out of indirect block array */
279 	daddr = DIP(ip, i_ib[ap->in_off]);
280 
281 	for (bp = NULL, ++ap; --num; ++ap) {
282 		/*
283 		 * Exit the loop if there is no disk address assigned yet and
284 		 * the indirect block isn't in the cache, or if we were
285 		 * looking for an indirect block and we've found it.
286 		 */
287 
288 		metalbn = ap->in_lbn;
289 		if ((daddr == 0 && !incore(&vp->v_bufobj, metalbn)) || metalbn == bn)
290 			break;
291 		/*
292 		 * If we get here, we've either got the block in the cache
293 		 * or we have a disk address for it, go fetch it.
294 		 */
295 		if (bp)
296 			bqrelse(bp);
297 		error = readindir(vp, metalbn, daddr, true, &bp);
298 		if (error != 0)
299 			return (error);
300 
301 		daddrsz = I_IS_UFS1(ip) ? sizeof(ufs1_daddr_t) : sizeof(ufs2_daddr_t);
302 		if (!buf_mapped(bp)) {
303 			boff = ap->in_off * daddrsz;
304 			pgidx = boff / PAGE_SIZE;
305 			pgoff = (boff & PAGE_MASK) / daddrsz;
306 			pgaddr = ufs_bm_sf_get(bp, pgidx, &sf);
307 			if (I_IS_UFS1(ip))
308 				daddr = ((ufs1_daddr_t *)pgaddr)[pgoff];
309 			else
310 				daddr = ((ufs2_daddr_t *)pgaddr)[pgoff];
311 			ufs_bm_sf_put(sf);
312 		} else {
313 			if (I_IS_UFS1(ip))
314 				daddr = ((ufs1_daddr_t *)bp->b_data)[ap->in_off];
315 			else
316 				daddr = ((ufs2_daddr_t *)bp->b_data)[ap->in_off];
317 		}
318 
319 		if ((error = UFS_CHECK_BLKNO(mp, ip->i_number, daddr,
320 		     mp->mnt_stat.f_iosize)) != 0) {
321 			bqrelse(bp);
322 			return (error);
323 		}
324 		if (num > 1 || daddr == 0 || runp == NULL)
325 			continue;
326 
327 		daddrppg = PAGE_SIZE / daddrsz;
328 		if (I_IS_UFS1(ip)) {
329 			if (!buf_mapped(bp)) {
330 				prevdaddr = daddr;
331 				isseq = true;
332 				for (bn = ap->in_off + 1;
333 				    bn < MNINDIR(ump) && *runp < maxrun && isseq; ) {
334 					boff = bn * daddrsz;
335 					pgidx = boff / PAGE_SIZE;
336 					pgoff = (boff & PAGE_MASK) / daddrsz;
337 					KASSERT(pgidx >= 0 && pgidx < bp->b_npages,
338 						("pgidx %d vs b_npages %d", pgidx, bp->b_npages));
339 					pgaddr = ufs_bm_sf_get(bp, pgidx, &sf);
340 					daddr1p = (ufs1_daddr_t *)pgaddr;
341 					for (pgbn = pgoff;
342 					     pgbn < daddrppg && *runp < maxrun &&
343 					     (isseq = is_sequential(ump, prevdaddr, daddr1p[pgbn]));
344 					     prevdaddr = daddr1p[pgbn], ++pgbn, ++bn, ++*runp);
345 					ufs_bm_sf_put(sf);
346 				}
347 				prevdaddr = daddr;
348 				bn = ap->in_off;
349 				if (runb && bn) {
350 					isseq = true;
351 					for (--bn; bn >= 0 && *runb < maxrun && isseq; ) {
352 						boff = bn * daddrsz;
353 						pgidx = boff / PAGE_SIZE;
354 						pgoff = (boff & PAGE_MASK) / daddrsz;
355 						KASSERT(pgidx >= 0 && pgidx < bp->b_npages,
356 							("pgidx %d vs b_npages %d", pgidx, bp->b_npages));
357 						pgaddr = ufs_bm_sf_get(bp, pgidx, &sf);
358 						daddr1p = (ufs1_daddr_t *)pgaddr;
359 						for (pgbn = pgoff; pgbn >= 0 && *runb < maxrun &&
360 						     (isseq = is_sequential(ump, daddr1p[pgbn], prevdaddr));
361 						     prevdaddr = daddr1p[pgbn], --pgbn, --bn, ++*runb);
362 						ufs_bm_sf_put(sf);
363 					}
364 				}
365 			} else {
366 				for (bn = ap->in_off + 1;
367 				    bn < MNINDIR(ump) && *runp < maxrun &&
368 				    is_sequential(ump,
369 				    ((ufs1_daddr_t *)bp->b_data)[bn - 1],
370 				    ((ufs1_daddr_t *)bp->b_data)[bn]);
371 				    ++bn, ++*runp);
372 				bn = ap->in_off;
373 				if (runb && bn) {
374 					for (--bn; bn >= 0 && *runb < maxrun &&
375 					    is_sequential(ump,
376 					    ((ufs1_daddr_t *)bp->b_data)[bn],
377 					    ((ufs1_daddr_t *)bp->b_data)[bn+1]);
378 					    --bn, ++*runb);
379 				}
380 			}
381 			continue;
382 		}
383 
384 		if (!buf_mapped(bp)) {
385 			prevdaddr = daddr;
386 			isseq = true;
387 			for (bn = ap->in_off + 1;
388 			    bn < MNINDIR(ump) && *runp < maxrun && isseq; ) {
389 				boff = bn * daddrsz;
390 				pgidx = boff / PAGE_SIZE;
391 				pgoff = (boff & PAGE_MASK) / daddrsz;
392 				KASSERT(pgidx >= 0 && pgidx < bp->b_npages,
393 					("pgidx %d vs b_npages %d", pgidx, bp->b_npages));
394 				pgaddr = ufs_bm_sf_get(bp, pgidx, &sf);
395 				daddr2p = (ufs2_daddr_t *)pgaddr;
396 				for (pgbn = pgoff;
397 				     pgbn < daddrppg && *runp < maxrun &&
398 				     (isseq = is_sequential(ump, prevdaddr, daddr2p[pgbn]));
399 				     prevdaddr = daddr2p[pgbn], ++pgbn, ++bn, ++*runp);
400 				ufs_bm_sf_put(sf);
401 			}
402 			prevdaddr = daddr;
403 			bn = ap->in_off;
404 			if (runb && bn) {
405 				isseq = true;
406 				for (--bn; bn >= 0 && *runb < maxrun && isseq; ) {
407 					boff = bn * daddrsz;
408 					pgidx = boff / PAGE_SIZE;
409 					pgoff = (boff & PAGE_MASK) / daddrsz;
410 					KASSERT(pgidx >= 0 && pgidx < bp->b_npages,
411 						("pgidx %d vs b_npages %d", pgidx, bp->b_npages));
412 					pgaddr = ufs_bm_sf_get(bp, pgidx, &sf);
413 					daddr2p = (ufs2_daddr_t *)pgaddr;
414 					for (pgbn = pgoff; pgbn >= 0 && *runb < maxrun &&
415 					     (isseq = is_sequential(ump, daddr2p[pgbn], prevdaddr));
416 					     prevdaddr = daddr2p[pgbn], --pgbn, --bn, ++*runb);
417 					ufs_bm_sf_put(sf);
418 				}
419 			}
420 		} else {
421 			for (bn = ap->in_off + 1;
422 			    bn < MNINDIR(ump) && *runp < maxrun &&
423 			    is_sequential(ump,
424 			    ((ufs2_daddr_t *)bp->b_data)[bn - 1],
425 			    ((ufs2_daddr_t *)bp->b_data)[bn]);
426 			    ++bn, ++*runp);
427 			bn = ap->in_off;
428 			if (runb && bn) {
429 				for (--bn; bn >= 0 && *runb < maxrun &&
430 				    is_sequential(ump,
431 				    ((ufs2_daddr_t *)bp->b_data)[bn],
432 				    ((ufs2_daddr_t *)bp->b_data)[bn + 1]);
433 				    --bn, ++*runb);
434 			}
435 		}
436 	}
437 	if (bp)
438 		bqrelse(bp);
439 
440 	/*
441 	 * Since this is FFS independent code, we are out of scope for the
442 	 * definitions of BLK_NOCOPY and BLK_SNAP, but we do know that they
443 	 * will fall in the range 1..um_seqinc, so we use that test and
444 	 * return a request for a zeroed out buffer if attempts are made
445 	 * to read a BLK_NOCOPY or BLK_SNAP block.
446 	 */
447 	if (IS_SNAPSHOT(ip) && daddr > 0 && daddr < ump->um_seqinc){
448 		*bnp = -1;
449 		return (0);
450 	}
451 	*bnp = blkptrtodb(ump, daddr);
452 	if (*bnp == 0) {
453 		if (IS_SNAPSHOT(ip))
454 			*bnp = blkptrtodb(ump, bn * ump->um_seqinc);
455 		else
456 			*bnp = -1;
457 	}
458 	return (0);
459 }
460 
461 static ufs_lbn_t
lbn_count(struct ufsmount * ump,int level)462 lbn_count(struct ufsmount *ump, int level)
463 {
464 	ufs_lbn_t blockcnt;
465 
466 	for (blockcnt = 1; level > 0; level--)
467 		blockcnt *= MNINDIR(ump);
468 	return (blockcnt);
469 }
470 
471 int
ufs_bmap_seekdata(struct vnode * vp,off_t * offp)472 ufs_bmap_seekdata(struct vnode *vp, off_t *offp)
473 {
474 	struct buf *bp;
475 	struct indir a[UFS_NIADDR + 1], *ap;
476 	struct inode *ip;
477 	struct mount *mp;
478 	struct ufsmount *ump;
479 	ufs2_daddr_t bn, daddr, nextbn;
480 	uint64_t bsize;
481 	off_t numblks;
482 	int error, num, num1, off;
483 
484 	bp = NULL;
485 	error = 0;
486 	ip = VTOI(vp);
487 	mp = vp->v_mount;
488 	ump = VFSTOUFS(mp);
489 
490 	if (vp->v_type != VREG || IS_SNAPSHOT(ip))
491 		return (EINVAL);
492 	if (*offp < 0 || *offp >= ip->i_size)
493 		return (ENXIO);
494 
495 	/*
496 	 * We could have pages on the vnode' object queue which still
497 	 * do not have the data blocks allocated.  Convert all dirty
498 	 * pages into buffer writes to ensure that we see all
499 	 * allocated data.
500 	 */
501 	vnode_pager_clean_sync(vp);
502 
503 	bsize = mp->mnt_stat.f_iosize;
504 	for (bn = *offp / bsize, numblks = howmany(ip->i_size, bsize);
505 	    bn < numblks; bn = nextbn) {
506 		if (bn < UFS_NDADDR) {
507 			daddr = DIP(ip, i_db[bn]);
508 			if (daddr != 0)
509 				break;
510 			nextbn = bn + 1;
511 			continue;
512 		}
513 
514 		ap = a;
515 		error = ufs_getlbns(vp, bn, ap, &num);
516 		if (error != 0)
517 			break;
518 		MPASS(num >= 2);
519 		daddr = DIP(ip, i_ib[ap->in_off]);
520 		ap++, num--;
521 		for (nextbn = UFS_NDADDR, num1 = num - 1; num1 > 0; num1--)
522 			nextbn += lbn_count(ump, num1);
523 		if (daddr == 0) {
524 			nextbn += lbn_count(ump, num);
525 			continue;
526 		}
527 
528 		for (; daddr != 0 && num > 0; ap++, num--) {
529 			if (bp != NULL)
530 				bqrelse(bp);
531 			error = readindir(vp, ap->in_lbn, daddr, false, &bp);
532 			if (error != 0)
533 				return (error);
534 
535 			/*
536 			 * Scan the indirect block until we find a non-zero
537 			 * pointer.
538 			 */
539 			off = ap->in_off;
540 			do {
541 				daddr = I_IS_UFS1(ip) ?
542 				    ((ufs1_daddr_t *)bp->b_data)[off] :
543 				    ((ufs2_daddr_t *)bp->b_data)[off];
544 			} while (daddr == 0 && ++off < MNINDIR(ump));
545 			nextbn += off * lbn_count(ump, num - 1);
546 
547 			/*
548 			 * We need to recompute the LBNs of indirect
549 			 * blocks, so restart with the updated block offset.
550 			 */
551 			if (off != ap->in_off)
552 				break;
553 		}
554 		if (num == 0) {
555 			/*
556 			 * We found a data block.
557 			 */
558 			bn = nextbn;
559 			break;
560 		}
561 	}
562 	if (bp != NULL)
563 		bqrelse(bp);
564 	if (bn >= numblks)
565 		error = ENXIO;
566 	if (error == 0 && *offp < bn * bsize)
567 		*offp = bn * bsize;
568 	return (error);
569 }
570 
571 /*
572  * Create an array of logical block number/offset pairs which represent the
573  * path of indirect blocks required to access a data block.  The first "pair"
574  * contains the logical block number of the appropriate single, double or
575  * triple indirect block and the offset into the inode indirect block array.
576  * Note, the logical block number of the inode single/double/triple indirect
577  * block appears twice in the array, once with the offset into the i_ib and
578  * once with the offset into the page itself.
579  */
580 int
ufs_getlbns(struct vnode * vp,ufs2_daddr_t bn,struct indir * ap,int * nump)581 ufs_getlbns(struct vnode *vp,
582 	ufs2_daddr_t bn,
583 	struct indir *ap,
584 	int *nump)
585 {
586 	ufs2_daddr_t blockcnt;
587 	ufs_lbn_t metalbn, realbn;
588 	struct ufsmount *ump;
589 	int i, numlevels, off;
590 
591 	ump = VFSTOUFS(vp->v_mount);
592 	if (nump)
593 		*nump = 0;
594 	numlevels = 0;
595 	realbn = bn;
596 	if (bn < 0)
597 		bn = -bn;
598 
599 	/* The first UFS_NDADDR blocks are direct blocks. */
600 	if (bn < UFS_NDADDR)
601 		return (0);
602 
603 	/*
604 	 * Determine the number of levels of indirection.  After this loop
605 	 * is done, blockcnt indicates the number of data blocks possible
606 	 * at the previous level of indirection, and UFS_NIADDR - i is the
607 	 * number of levels of indirection needed to locate the requested block.
608 	 */
609 	for (blockcnt = 1, i = UFS_NIADDR, bn -= UFS_NDADDR; ;
610 	    i--, bn -= blockcnt) {
611 		if (i == 0)
612 			return (EFBIG);
613 		blockcnt *= MNINDIR(ump);
614 		if (bn < blockcnt)
615 			break;
616 	}
617 
618 	/* Calculate the address of the first meta-block. */
619 	if (realbn >= 0)
620 		metalbn = -(realbn - bn + UFS_NIADDR - i);
621 	else
622 		metalbn = -(-realbn - bn + UFS_NIADDR - i);
623 
624 	/*
625 	 * At each iteration, off is the offset into the bap array which is
626 	 * an array of disk addresses at the current level of indirection.
627 	 * The logical block number and the offset in that block are stored
628 	 * into the argument array.
629 	 */
630 	ap->in_lbn = metalbn;
631 	ap->in_off = off = UFS_NIADDR - i;
632 	ap++;
633 	for (++numlevels; i <= UFS_NIADDR; i++) {
634 		/* If searching for a meta-data block, quit when found. */
635 		if (metalbn == realbn)
636 			break;
637 
638 		blockcnt /= MNINDIR(ump);
639 		off = (bn / blockcnt) % MNINDIR(ump);
640 
641 		++numlevels;
642 		ap->in_lbn = metalbn;
643 		ap->in_off = off;
644 		++ap;
645 
646 		metalbn -= -1 + off * blockcnt;
647 	}
648 	if (nump)
649 		*nump = numlevels;
650 	return (0);
651 }
652