xref: /freebsd/sys/fs/ext2fs/ext2_bmap.c (revision 3cbb4cc200f8a0ad7ed08233425ea54524a21f1c)
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  *	@(#)ufs_bmap.c	8.7 (Berkeley) 3/21/95
37  * $FreeBSD$
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/bio.h>
43 #include <sys/buf.h>
44 #include <sys/endian.h>
45 #include <sys/proc.h>
46 #include <sys/vnode.h>
47 #include <sys/mount.h>
48 #include <sys/racct.h>
49 #include <sys/resourcevar.h>
50 #include <sys/stat.h>
51 
52 #include <fs/ext2fs/fs.h>
53 #include <fs/ext2fs/inode.h>
54 #include <fs/ext2fs/ext2fs.h>
55 #include <fs/ext2fs/ext2_dinode.h>
56 #include <fs/ext2fs/ext2_extern.h>
57 #include <fs/ext2fs/ext2_mount.h>
58 
59 /*
60  * Bmap converts the logical block number of a file to its physical block
61  * number on the disk. The conversion is done by using the logical block
62  * number to index into the array of block pointers described by the dinode.
63  */
64 int
65 ext2_bmap(struct vop_bmap_args *ap)
66 {
67 	daddr_t blkno;
68 	int error;
69 
70 	/*
71 	 * Check for underlying vnode requests and ensure that logical
72 	 * to physical mapping is requested.
73 	 */
74 	if (ap->a_bop != NULL)
75 		*ap->a_bop = &VTOI(ap->a_vp)->i_devvp->v_bufobj;
76 	if (ap->a_bnp == NULL)
77 		return (0);
78 
79 	if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS)
80 		error = ext4_bmapext(ap->a_vp, ap->a_bn, &blkno,
81 		    ap->a_runp, ap->a_runb);
82 	else
83 		error = ext2_bmaparray(ap->a_vp, ap->a_bn, &blkno,
84 		    ap->a_runp, ap->a_runb);
85 	*ap->a_bnp = blkno;
86 	return (error);
87 }
88 
89 /*
90  * Convert the logical block number of a file to its physical block number
91  * on the disk within ext4 extents.
92  */
93 int
94 ext4_bmapext(struct vnode *vp, int32_t bn, int64_t *bnp, int *runp, int *runb)
95 {
96 	struct inode *ip;
97 	struct m_ext2fs *fs;
98 	struct mount *mp;
99 	struct ext2mount *ump;
100 	struct ext4_extent_header *ehp;
101 	struct ext4_extent *ep;
102 	struct ext4_extent_path *path = NULL;
103 	daddr_t lbn;
104 	int error, depth, maxrun = 0, bsize;
105 
106 	ip = VTOI(vp);
107 	fs = ip->i_e2fs;
108 	mp = vp->v_mount;
109 	ump = VFSTOEXT2(mp);
110 	lbn = bn;
111 	ehp = (struct ext4_extent_header *)ip->i_data;
112 	depth = le16toh(ehp->eh_depth);
113 	bsize = EXT2_BLOCK_SIZE(ump->um_e2fs);
114 
115 	*bnp = -1;
116 	if (runp != NULL) {
117 		maxrun = mp->mnt_iosize_max / bsize - 1;
118 		*runp = 0;
119 	}
120 	if (runb != NULL)
121 		*runb = 0;
122 
123 	error = ext4_ext_find_extent(ip, lbn, &path);
124 	if (error)
125 		return (error);
126 
127 	ep = path[depth].ep_ext;
128 	if(ep) {
129 		if (lbn < le32toh(ep->e_blk)) {
130 			if (runp != NULL) {
131 				*runp = min(maxrun, le32toh(ep->e_blk) - lbn - 1);
132 			}
133 		} else if (le32toh(ep->e_blk) <= lbn &&
134 			    lbn < le32toh(ep->e_blk) + le16toh(ep->e_len)) {
135 			*bnp = fsbtodb(fs, lbn - le32toh(ep->e_blk) +
136 			    (le32toh(ep->e_start_lo) |
137 			    (daddr_t)le16toh(ep->e_start_hi) << 32));
138 			if (runp != NULL) {
139 				*runp = min(maxrun,
140 				    le16toh(ep->e_len) -
141 				    (lbn - le32toh(ep->e_blk)) - 1);
142 			}
143 			if (runb != NULL)
144 				*runb = min(maxrun, lbn - le32toh(ep->e_blk));
145 		} else {
146 			if (runb != NULL)
147 				*runb = min(maxrun, le32toh(ep->e_blk) + lbn -
148 				    le16toh(ep->e_len));
149 		}
150 	}
151 
152 	ext4_ext_path_free(path);
153 
154 	return (error);
155 }
156 
157 static int
158 readindir(struct vnode *vp, e2fs_lbn_t lbn, e2fs_daddr_t daddr, struct buf **bpp)
159 {
160 	struct buf *bp;
161 	struct mount *mp;
162 	struct ext2mount *ump;
163 	int error;
164 
165 	mp = vp->v_mount;
166 	ump = VFSTOEXT2(mp);
167 
168 	bp = getblk(vp, lbn, mp->mnt_stat.f_iosize, 0, 0, 0);
169 	if ((bp->b_flags & B_CACHE) == 0) {
170 		KASSERT(daddr != 0,
171 		    ("readindir: indirect block not in cache"));
172 
173 		bp->b_blkno = blkptrtodb(ump, daddr);
174 		bp->b_iocmd = BIO_READ;
175 		bp->b_flags &= ~B_INVAL;
176 		bp->b_ioflags &= ~BIO_ERROR;
177 		vfs_busy_pages(bp, 0);
178 		bp->b_iooffset = dbtob(bp->b_blkno);
179 		bstrategy(bp);
180 #ifdef RACCT
181 		if (racct_enable) {
182 			PROC_LOCK(curproc);
183 			racct_add_buf(curproc, bp, 0);
184 			PROC_UNLOCK(curproc);
185 		}
186 #endif
187 		curthread->td_ru.ru_inblock++;
188 		error = bufwait(bp);
189 		if (error != 0) {
190 			brelse(bp);
191 			return (error);
192 		}
193 	}
194 	*bpp = bp;
195 	return (0);
196 }
197 
198 /*
199  * Indirect blocks are now on the vnode for the file.  They are given negative
200  * logical block numbers.  Indirect blocks are addressed by the negative
201  * address of the first data block to which they point.  Double indirect blocks
202  * are addressed by one less than the address of the first indirect block to
203  * which they point.  Triple indirect blocks are addressed by one less than
204  * the address of the first double indirect block to which they point.
205  *
206  * ext2_bmaparray does the bmap conversion, and if requested returns the
207  * array of logical blocks which must be traversed to get to a block.
208  * Each entry contains the offset into that block that gets you to the
209  * next block and the disk address of the block (if it is assigned).
210  */
211 
212 int
213 ext2_bmaparray(struct vnode *vp, daddr_t bn, daddr_t *bnp, int *runp, int *runb)
214 {
215 	struct inode *ip;
216 	struct buf *bp;
217 	struct ext2mount *ump;
218 	struct mount *mp;
219 	struct indir a[EXT2_NIADDR + 1], *ap;
220 	daddr_t daddr;
221 	e2fs_lbn_t metalbn;
222 	int error, num, maxrun = 0, bsize;
223 	int *nump;
224 
225 	ap = NULL;
226 	ip = VTOI(vp);
227 	mp = vp->v_mount;
228 	ump = VFSTOEXT2(mp);
229 
230 	bsize = EXT2_BLOCK_SIZE(ump->um_e2fs);
231 
232 	if (runp) {
233 		maxrun = mp->mnt_iosize_max / bsize - 1;
234 		*runp = 0;
235 	}
236 	if (runb)
237 		*runb = 0;
238 
239 	ap = a;
240 	nump = &num;
241 	error = ext2_getlbns(vp, bn, ap, nump);
242 	if (error)
243 		return (error);
244 
245 	num = *nump;
246 	if (num == 0) {
247 		*bnp = blkptrtodb(ump, ip->i_db[bn]);
248 		if (*bnp == 0) {
249 			*bnp = -1;
250 		} else if (runp) {
251 			daddr_t bnb = bn;
252 
253 			for (++bn; bn < EXT2_NDADDR && *runp < maxrun &&
254 			    is_sequential(ump, ip->i_db[bn - 1], ip->i_db[bn]);
255 			    ++bn, ++*runp);
256 			bn = bnb;
257 			if (runb && (bn > 0)) {
258 				for (--bn; (bn >= 0) && (*runb < maxrun) &&
259 					is_sequential(ump, ip->i_db[bn],
260 						ip->i_db[bn + 1]);
261 						--bn, ++*runb);
262 			}
263 		}
264 		return (0);
265 	}
266 
267 	/* Get disk address out of indirect block array */
268 	daddr = ip->i_ib[ap->in_off];
269 
270 	for (bp = NULL, ++ap; --num; ++ap) {
271 		/*
272 		 * Exit the loop if there is no disk address assigned yet and
273 		 * the indirect block isn't in the cache, or if we were
274 		 * looking for an indirect block and we've found it.
275 		 */
276 
277 		metalbn = ap->in_lbn;
278 		if ((daddr == 0 && !incore(&vp->v_bufobj, metalbn)) || metalbn == bn)
279 			break;
280 		/*
281 		 * If we get here, we've either got the block in the cache
282 		 * or we have a disk address for it, go fetch it.
283 		 */
284 		if (bp)
285 			bqrelse(bp);
286 		error = readindir(vp, metalbn, daddr, &bp);
287 		if (error != 0)
288 			return (error);
289 
290 		daddr = le32toh(((e2fs_daddr_t *)bp->b_data)[ap->in_off]);
291 		if (num == 1 && daddr && runp) {
292 			for (bn = ap->in_off + 1;
293 			    bn < MNINDIR(ump) && *runp < maxrun &&
294 			    is_sequential(ump,
295 			    ((e2fs_daddr_t *)bp->b_data)[bn - 1],
296 			    ((e2fs_daddr_t *)bp->b_data)[bn]);
297 			    ++bn, ++*runp);
298 			bn = ap->in_off;
299 			if (runb && bn) {
300 				for (--bn; bn >= 0 && *runb < maxrun &&
301 					is_sequential(ump,
302 					((e2fs_daddr_t *)bp->b_data)[bn],
303 					((e2fs_daddr_t *)bp->b_data)[bn + 1]);
304 					--bn, ++*runb);
305 			}
306 		}
307 	}
308 	if (bp)
309 		bqrelse(bp);
310 
311 	/*
312 	 * Since this is FFS independent code, we are out of scope for the
313 	 * definitions of BLK_NOCOPY and BLK_SNAP, but we do know that they
314 	 * will fall in the range 1..um_seqinc, so we use that test and
315 	 * return a request for a zeroed out buffer if attempts are made
316 	 * to read a BLK_NOCOPY or BLK_SNAP block.
317 	 */
318 	if ((ip->i_flags & SF_SNAPSHOT) && daddr > 0 && daddr < ump->um_seqinc) {
319 		*bnp = -1;
320 		return (0);
321 	}
322 	*bnp = blkptrtodb(ump, daddr);
323 	if (*bnp == 0) {
324 		*bnp = -1;
325 	}
326 	return (0);
327 }
328 
329 static e2fs_lbn_t
330 lbn_count(struct ext2mount *ump, int level)
331 
332 {
333 	e2fs_lbn_t blockcnt;
334 
335 	for (blockcnt = 1; level > 0; level--)
336 		blockcnt *= MNINDIR(ump);
337 	return (blockcnt);
338 }
339 
340 int
341 ext2_bmap_seekdata(struct vnode *vp, off_t *offp)
342 {
343 	struct buf *bp;
344 	struct indir a[EXT2_NIADDR + 1], *ap;
345 	struct inode *ip;
346 	struct mount *mp;
347 	struct ext2mount *ump;
348 	e2fs_daddr_t bn, daddr, nextbn;
349 	uint64_t bsize;
350 	off_t numblks;
351 	int error, num, num1, off;
352 
353 	bp = NULL;
354 	error = 0;
355 	ip = VTOI(vp);
356 	mp = vp->v_mount;
357 	ump = VFSTOEXT2(mp);
358 
359 	if (vp->v_type != VREG || (ip->i_flags & SF_SNAPSHOT) != 0)
360 		return (EINVAL);
361 	if (*offp < 0 || *offp >= ip->i_size)
362 		return (ENXIO);
363 
364 	bsize = mp->mnt_stat.f_iosize;
365 	for (bn = *offp / bsize, numblks = howmany(ip->i_size, bsize);
366 	    bn < numblks; bn = nextbn) {
367 		if (bn < EXT2_NDADDR) {
368 			daddr = ip->i_db[bn];
369 			if (daddr != 0)
370 				break;
371 			nextbn = bn + 1;
372 			continue;
373 		}
374 
375 		ap = a;
376 		error = ext2_getlbns(vp, bn, ap, &num);
377 		if (error != 0)
378 			break;
379 		MPASS(num >= 2);
380 		daddr = ip->i_ib[ap->in_off];
381 		ap++, num--;
382 		for (nextbn = EXT2_NDADDR, num1 = num - 1; num1 > 0; num1--)
383 			nextbn += lbn_count(ump, num1);
384 		if (daddr == 0) {
385 			nextbn += lbn_count(ump, num);
386 			continue;
387 		}
388 
389 		for (; daddr != 0 && num > 0; ap++, num--) {
390 			if (bp != NULL)
391 				bqrelse(bp);
392 			error = readindir(vp, ap->in_lbn, daddr, &bp);
393 			if (error != 0)
394 				return (error);
395 
396 			/*
397 			 * Scan the indirect block until we find a non-zero
398 			 * pointer.
399 			 */
400 			off = ap->in_off;
401 			do {
402 				daddr = le32toh(((e2fs_daddr_t *)bp->b_data)[off]);
403 			} while (daddr == 0 && ++off < MNINDIR(ump));
404 			nextbn += off * lbn_count(ump, num - 1);
405 
406 			/*
407 			 * We need to recompute the LBNs of indirect
408 			 * blocks, so restart with the updated block offset.
409 			 */
410 			if (off != ap->in_off)
411 				break;
412 		}
413 		if (num == 0) {
414 			/*
415 			 * We found a data block.
416 			 */
417 			bn = nextbn;
418 			break;
419 		}
420 	}
421 	if (bp != NULL)
422 		bqrelse(bp);
423 	if (bn >= numblks)
424 		error = ENXIO;
425 	if (error == 0 && *offp < bn * bsize)
426 		*offp = bn * bsize;
427 	return (error);
428 }
429 
430 /*
431  * Create an array of logical block number/offset pairs which represent the
432  * path of indirect blocks required to access a data block.  The first "pair"
433  * contains the logical block number of the appropriate single, double or
434  * triple indirect block and the offset into the inode indirect block array.
435  * Note, the logical block number of the inode single/double/triple indirect
436  * block appears twice in the array, once with the offset into the i_ib and
437  * once with the offset into the page itself.
438  */
439 int
440 ext2_getlbns(struct vnode *vp, daddr_t bn, struct indir *ap, int *nump)
441 {
442 	long blockcnt;
443 	e2fs_lbn_t metalbn, realbn;
444 	struct ext2mount *ump;
445 	int i, numlevels, off;
446 	int64_t qblockcnt;
447 
448 	ump = VFSTOEXT2(vp->v_mount);
449 	if (nump)
450 		*nump = 0;
451 	numlevels = 0;
452 	realbn = bn;
453 	if ((long)bn < 0)
454 		bn = -(long)bn;
455 
456 	/* The first EXT2_NDADDR blocks are direct blocks. */
457 	if (bn < EXT2_NDADDR)
458 		return (0);
459 
460 	/*
461 	 * Determine the number of levels of indirection.  After this loop
462 	 * is done, blockcnt indicates the number of data blocks possible
463 	 * at the previous level of indirection, and EXT2_NIADDR - i is the
464 	 * number of levels of indirection needed to locate the requested block.
465 	 */
466 	for (blockcnt = 1, i = EXT2_NIADDR, bn -= EXT2_NDADDR; ;
467 	    i--, bn -= blockcnt) {
468 		if (i == 0)
469 			return (EFBIG);
470 		/*
471 		 * Use int64_t's here to avoid overflow for triple indirect
472 		 * blocks when longs have 32 bits and the block size is more
473 		 * than 4K.
474 		 */
475 		qblockcnt = (int64_t)blockcnt * MNINDIR(ump);
476 		if (bn < qblockcnt)
477 			break;
478 		blockcnt = qblockcnt;
479 	}
480 
481 	/* Calculate the address of the first meta-block. */
482 	if (realbn >= 0)
483 		metalbn = -(realbn - bn + EXT2_NIADDR - i);
484 	else
485 		metalbn = -(-realbn - bn + EXT2_NIADDR - i);
486 
487 	/*
488 	 * At each iteration, off is the offset into the bap array which is
489 	 * an array of disk addresses at the current level of indirection.
490 	 * The logical block number and the offset in that block are stored
491 	 * into the argument array.
492 	 */
493 	ap->in_lbn = metalbn;
494 	ap->in_off = off = EXT2_NIADDR - i;
495 	ap++;
496 	for (++numlevels; i <= EXT2_NIADDR; i++) {
497 		/* If searching for a meta-data block, quit when found. */
498 		if (metalbn == realbn)
499 			break;
500 
501 		off = (bn / blockcnt) % MNINDIR(ump);
502 
503 		++numlevels;
504 		ap->in_lbn = metalbn;
505 		ap->in_off = off;
506 		++ap;
507 
508 		metalbn -= -1 + off * blockcnt;
509 		blockcnt /= MNINDIR(ump);
510 	}
511 	if (nump)
512 		*nump = numlevels;
513 	return (0);
514 }
515