xref: /netbsd/sys/ufs/ufs/ufs_bmap.c (revision fc045a9d)
1 /*	$NetBSD: ufs_bmap.c,v 1.54 2022/11/17 06:40:40 chs Exp $	*/
2 
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	8.8 (Berkeley) 8/11/95
37  */
38 
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: ufs_bmap.c,v 1.54 2022/11/17 06:40:40 chs Exp $");
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/buf.h>
45 #include <sys/proc.h>
46 #include <sys/vnode.h>
47 #include <sys/mount.h>
48 #include <sys/resourcevar.h>
49 #include <sys/trace.h>
50 
51 #include <miscfs/specfs/specdev.h>
52 
53 #include <ufs/ufs/inode.h>
54 #include <ufs/ufs/ufsmount.h>
55 #include <ufs/ufs/ufs_extern.h>
56 #include <ufs/ufs/ufs_bswap.h>
57 
58 static bool
ufs_issequential(const struct ufsmount * ump,daddr_t daddr0,daddr_t daddr1)59 ufs_issequential(const struct ufsmount *ump, daddr_t daddr0, daddr_t daddr1)
60 {
61 
62 	/* for ufs, blocks in a hole is not 'contiguous'. */
63 	if (daddr0 == 0)
64 		return false;
65 
66 	return (daddr0 + ump->um_seqinc == daddr1);
67 }
68 
69 /*
70  * Bmap converts the logical block number of a file to its physical block
71  * number on the disk. The conversion is done by using the logical block
72  * number to index into the array of block pointers described by the dinode.
73  */
74 int
ufs_bmap(void * v)75 ufs_bmap(void *v)
76 {
77 	struct vop_bmap_args /* {
78 		struct vnode *a_vp;
79 		daddr_t  a_bn;
80 		struct vnode **a_vpp;
81 		daddr_t *a_bnp;
82 		int *a_runp;
83 	} */ *ap = v;
84 	int error;
85 
86 	/*
87 	 * Check for underlying vnode requests and ensure that logical
88 	 * to physical mapping is requested.
89 	 */
90 	if (ap->a_vpp != NULL)
91 		*ap->a_vpp = VTOI(ap->a_vp)->i_devvp;
92 	if (ap->a_bnp == NULL)
93 		return (0);
94 
95 	error = ufs_bmaparray(ap->a_vp, ap->a_bn, ap->a_bnp, NULL, NULL,
96 	    ap->a_runp, ufs_issequential);
97 	return error;
98 }
99 
100 /*
101  * Indirect blocks are now on the vnode for the file.  They are given negative
102  * logical block numbers.  Indirect blocks are addressed by the negative
103  * address of the first data block to which they point.  Double indirect blocks
104  * are addressed by one less than the address of the first indirect block to
105  * which they point.  Triple indirect blocks are addressed by one less than
106  * the address of the first double indirect block to which they point.
107  *
108  * ufs_bmaparray does the bmap conversion, and if requested returns the
109  * array of logical blocks which must be traversed to get to a block.
110  * Each entry contains the offset into that block that gets you to the
111  * next block and the disk address of the block (if it is assigned).
112  */
113 
114 int
ufs_bmaparray(struct vnode * vp,daddr_t bn,daddr_t * bnp,struct indir * ap,int * nump,int * runp,ufs_issequential_callback_t is_sequential)115 ufs_bmaparray(struct vnode *vp, daddr_t bn, daddr_t *bnp, struct indir *ap,
116     int *nump, int *runp, ufs_issequential_callback_t is_sequential)
117 {
118 	struct inode *ip;
119 	struct buf *bp, *cbp;
120 	struct ufsmount *ump;
121 	struct mount *mp;
122 	struct indir a[UFS_NIADDR + 1], *xap;
123 	daddr_t daddr;
124 	daddr_t metalbn;
125 	int error, maxrun = 0, num;
126 
127 	ip = VTOI(vp);
128 	mp = vp->v_mount;
129 	ump = ip->i_ump;
130 	KASSERTMSG(((ap == NULL) == (nump == NULL)),
131 	    "ufs_bmaparray: invalid arguments: ap = %p, nump = %p", ap, nump);
132 
133 	if (runp) {
134 		/*
135 		 * XXX
136 		 * If MAXBSIZE is the largest transfer the disks can handle,
137 		 * we probably want maxrun to be 1 block less so that we
138 		 * don't create a block larger than the device can handle.
139 		 */
140 		*runp = 0;
141 		maxrun = MAXPHYS / mp->mnt_stat.f_iosize - 1;
142 	}
143 
144 	if (bn >= 0 && bn < UFS_NDADDR) {
145 		if (nump != NULL)
146 			*nump = 0;
147 		if (ump->um_fstype == UFS1)
148 			daddr = ufs_rw32(ip->i_ffs1_db[bn],
149 			    UFS_MPNEEDSWAP(ump));
150 		else
151 			daddr = ufs_rw64(ip->i_ffs2_db[bn],
152 			    UFS_MPNEEDSWAP(ump));
153 		*bnp = blkptrtodb(ump, daddr);
154 		/*
155 		 * Since this is FFS independent code, we are out of
156 		 * scope for the definitions of BLK_NOCOPY and
157 		 * BLK_SNAP, but we do know that they will fall in
158 		 * the range 1..um_seqinc, so we use that test and
159 		 * return a request for a zeroed out buffer if attempts
160 		 * are made to read a BLK_NOCOPY or BLK_SNAP block.
161 		 */
162 		if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL)) == SF_SNAPSHOT
163 		    && daddr > 0 &&
164 		    daddr < ump->um_seqinc) {
165 			*bnp = -1;
166 		} else if (*bnp == 0) {
167 			if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL))
168 			    == SF_SNAPSHOT) {
169 				*bnp = blkptrtodb(ump, bn * ump->um_seqinc);
170 			} else {
171 				*bnp = -1;
172 			}
173 		} else if (runp) {
174 			if (ump->um_fstype == UFS1) {
175 				for (++bn; bn < UFS_NDADDR && *runp < maxrun &&
176 				    is_sequential(ump,
177 				        ufs_rw32(ip->i_ffs1_db[bn - 1],
178 				            UFS_MPNEEDSWAP(ump)),
179 				        ufs_rw32(ip->i_ffs1_db[bn],
180 				            UFS_MPNEEDSWAP(ump)));
181 				    ++bn, ++*runp);
182 			} else {
183 				for (++bn; bn < UFS_NDADDR && *runp < maxrun &&
184 				    is_sequential(ump,
185 				        ufs_rw64(ip->i_ffs2_db[bn - 1],
186 				            UFS_MPNEEDSWAP(ump)),
187 				        ufs_rw64(ip->i_ffs2_db[bn],
188 				            UFS_MPNEEDSWAP(ump)));
189 				    ++bn, ++*runp);
190 			}
191 		}
192 		return (0);
193 	} else if (bn < 0 && bn >= -UFS_NXADDR) {
194 		KASSERT(ump->um_fstype == UFS2 && (ump->um_flags & UFS_EA) != 0);
195 		daddr = ufs_rw64(ip->i_ffs2_extb[-1 - bn], UFS_MPNEEDSWAP(ump));
196 		*bnp = blkptrtodb(ump, daddr);
197 		if (*bnp == 0)
198 			*bnp = -1;
199 		return 0;
200 	}
201 
202 	xap = ap == NULL ? a : ap;
203 	if (!nump)
204 		nump = &num;
205 	if ((error = ufs_getlbns(vp, bn, xap, nump)) != 0)
206 		return (error);
207 
208 	num = *nump;
209 
210 	/* Get disk address out of indirect block array */
211 	if (ump->um_fstype == UFS1)
212 		daddr = ufs_rw32(ip->i_ffs1_ib[xap->in_off],
213 		    UFS_MPNEEDSWAP(ump));
214 	else
215 		daddr = ufs_rw64(ip->i_ffs2_ib[xap->in_off],
216 		    UFS_MPNEEDSWAP(ump));
217 
218 	for (bp = NULL, ++xap; --num; ++xap) {
219 		/*
220 		 * Exit the loop if there is no disk address assigned yet and
221 		 * the indirect block isn't in the cache, or if we were
222 		 * looking for an indirect block and we've found it.
223 		 */
224 
225 		metalbn = xap->in_lbn;
226 		if (metalbn == bn)
227 			break;
228 		if (daddr == 0) {
229 			mutex_enter(&bufcache_lock);
230 			cbp = incore(vp, metalbn);
231 			mutex_exit(&bufcache_lock);
232 			if (cbp == NULL)
233 				break;
234 		}
235 
236 		/*
237 		 * If we get here, we've either got the block in the cache
238 		 * or we have a disk address for it, go fetch it.
239 		 */
240 		if (bp)
241 			brelse(bp, 0);
242 
243 		xap->in_exists = 1;
244 		bp = getblk(vp, metalbn, mp->mnt_stat.f_iosize, 0, 0);
245 		if (bp == NULL) {
246 
247 			/*
248 			 * getblk() above returns NULL only iff we are
249 			 * pagedaemon.  See the implementation of getblk
250 			 * for detail.
251 			 */
252 
253 			return (ENOMEM);
254 		}
255 		if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
256 			trace(TR_BREADHIT, pack(vp, size), metalbn);
257 		} else {
258 			KASSERTMSG((daddr != 0),
259 			    "ufs_bmaparray: indirect block not in cache");
260 			trace(TR_BREADMISS, pack(vp, size), metalbn);
261 			bp->b_blkno = blkptrtodb(ump, daddr);
262 			bp->b_flags |= B_READ;
263 			BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
264 			VOP_STRATEGY(vp, bp);
265 			curlwp->l_ru.ru_inblock++;	/* XXX */
266 			if ((error = biowait(bp)) != 0) {
267 				brelse(bp, 0);
268 				return (error);
269 			}
270 		}
271 		if (ump->um_fstype == UFS1) {
272 			daddr = ufs_rw32(((u_int32_t *)bp->b_data)[xap->in_off],
273 			    UFS_MPNEEDSWAP(ump));
274 			if (num == 1 && daddr && runp) {
275 				for (bn = xap->in_off + 1;
276 				    bn < MNINDIR(ump) && *runp < maxrun &&
277 				    is_sequential(ump,
278 				        ufs_rw32(((int32_t *)bp->b_data)[bn-1],
279 				            UFS_MPNEEDSWAP(ump)),
280 				        ufs_rw32(((int32_t *)bp->b_data)[bn],
281 				            UFS_MPNEEDSWAP(ump)));
282 				    ++bn, ++*runp);
283 			}
284 		} else {
285 			daddr = ufs_rw64(((u_int64_t *)bp->b_data)[xap->in_off],
286 			    UFS_MPNEEDSWAP(ump));
287 			if (num == 1 && daddr && runp) {
288 				for (bn = xap->in_off + 1;
289 				    bn < MNINDIR(ump) && *runp < maxrun &&
290 				    is_sequential(ump,
291 				        ufs_rw64(((int64_t *)bp->b_data)[bn-1],
292 				            UFS_MPNEEDSWAP(ump)),
293 				        ufs_rw64(((int64_t *)bp->b_data)[bn],
294 				            UFS_MPNEEDSWAP(ump)));
295 				    ++bn, ++*runp);
296 			}
297 		}
298 	}
299 	if (bp)
300 		brelse(bp, 0);
301 
302 	/*
303 	 * Since this is FFS independent code, we are out of scope for the
304 	 * definitions of BLK_NOCOPY and BLK_SNAP, but we do know that they
305 	 * will fall in the range 1..um_seqinc, so we use that test and
306 	 * return a request for a zeroed out buffer if attempts are made
307 	 * to read a BLK_NOCOPY or BLK_SNAP block.
308 	 */
309 	if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL)) == SF_SNAPSHOT
310 	    && daddr > 0 && daddr < ump->um_seqinc) {
311 		*bnp = -1;
312 		return (0);
313 	}
314 	*bnp = blkptrtodb(ump, daddr);
315 	if (*bnp == 0) {
316 		if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL))
317 		    == SF_SNAPSHOT) {
318 			*bnp = blkptrtodb(ump, bn * ump->um_seqinc);
319 		} else {
320 			*bnp = -1;
321 		}
322 	}
323 	return (0);
324 }
325 
326 /*
327  * Create an array of logical block number/offset pairs which represent the
328  * path of indirect blocks required to access a data block.  The first "pair"
329  * contains the logical block number of the appropriate single, double or
330  * triple indirect block and the offset into the inode indirect block array.
331  * Note, the logical block number of the inode single/double/triple indirect
332  * block appears twice in the array, once with the offset into the i_ffs1_ib and
333  * once with the offset into the page itself.
334  */
335 int
ufs_getlbns(struct vnode * vp,daddr_t bn,struct indir * ap,int * nump)336 ufs_getlbns(struct vnode *vp, daddr_t bn, struct indir *ap, int *nump)
337 {
338 	daddr_t metalbn, realbn;
339 	struct ufsmount *ump;
340 	int64_t blockcnt;
341 	int lbc;
342 	int i, numlevels, off;
343 
344 	ump = VFSTOUFS(vp->v_mount);
345 	if (nump)
346 		*nump = 0;
347 	numlevels = 0;
348 	realbn = bn;
349 	if (bn < 0)
350 		bn = -bn;
351 	KASSERT(bn >= UFS_NDADDR);
352 
353 	/*
354 	 * Determine the number of levels of indirection.  After this loop
355 	 * is done, blockcnt indicates the number of data blocks possible
356 	 * at the given level of indirection, and UFS_NIADDR - i is the number
357 	 * of levels of indirection needed to locate the requested block.
358 	 */
359 
360 	bn -= UFS_NDADDR;
361 	for (lbc = 0, i = UFS_NIADDR;; i--, bn -= blockcnt) {
362 		if (i == 0)
363 			return (EFBIG);
364 
365 		lbc += ump->um_lognindir;
366 		blockcnt = (int64_t)1 << lbc;
367 
368 		if (bn < blockcnt)
369 			break;
370 	}
371 
372 	/* Calculate the address of the first meta-block. */
373 	metalbn = -((realbn >= 0 ? realbn : -realbn) - bn + UFS_NIADDR - i);
374 
375 	/*
376 	 * At each iteration, off is the offset into the bap array which is
377 	 * an array of disk addresses at the current level of indirection.
378 	 * The logical block number and the offset in that block are stored
379 	 * into the argument array.
380 	 */
381 	ap->in_lbn = metalbn;
382 	ap->in_off = off = UFS_NIADDR - i;
383 	ap->in_exists = 0;
384 	ap++;
385 	for (++numlevels; i <= UFS_NIADDR; i++) {
386 		/* If searching for a meta-data block, quit when found. */
387 		if (metalbn == realbn)
388 			break;
389 
390 		lbc -= ump->um_lognindir;
391 		off = (bn >> lbc) & (MNINDIR(ump) - 1);
392 
393 		++numlevels;
394 		ap->in_lbn = metalbn;
395 		ap->in_off = off;
396 		ap->in_exists = 0;
397 		++ap;
398 
399 		metalbn -= -1 + ((int64_t)off << lbc);
400 	}
401 	if (nump)
402 		*nump = numlevels;
403 	return (0);
404 }
405