xref: /netbsd-src/sys/ufs/lfs/ulfs_bmap.c (revision 63ce83dc30296c74d35dadfbe7dd4db9d9dfd63e)
1 /*	$NetBSD: ulfs_bmap.c,v 1.9 2017/03/30 09:10:08 hannken Exp $	*/
2 /*  from NetBSD: ufs_bmap.c,v 1.50 2013/01/22 09:39:18 dholland Exp  */
3 
4 /*
5  * Copyright (c) 1989, 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)ufs_bmap.c	8.8 (Berkeley) 8/11/95
38  */
39 
40 #include <sys/cdefs.h>
41 __KERNEL_RCSID(0, "$NetBSD: ulfs_bmap.c,v 1.9 2017/03/30 09:10:08 hannken Exp $");
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/stat.h>
46 #include <sys/buf.h>
47 #include <sys/proc.h>
48 #include <sys/vnode.h>
49 #include <sys/mount.h>
50 #include <sys/resourcevar.h>
51 #include <sys/trace.h>
52 
53 #include <miscfs/specfs/specdev.h>
54 
55 #include <ufs/lfs/ulfs_inode.h>
56 #include <ufs/lfs/ulfsmount.h>
57 #include <ufs/lfs/ulfs_extern.h>
58 #include <ufs/lfs/ulfs_bswap.h>
59 
60 static bool
ulfs_issequential(const struct lfs * fs,daddr_t daddr0,daddr_t daddr1)61 ulfs_issequential(const struct lfs *fs, daddr_t daddr0, daddr_t daddr1)
62 {
63 
64 	/* for ulfs, blocks in a hole is not 'contiguous'. */
65 	if (daddr0 == 0)
66 		return false;
67 
68 	return (daddr0 + fs->um_seqinc == daddr1);
69 }
70 
71 /*
72  * This is used for block pointers in inodes and elsewhere, which can
73  * contain the magic value UNWRITTEN, which is -2. This is mishandled
74  * by u32 -> u64 promotion unless special-cased.
75  *
76  * XXX this should be rolled into better inode accessors and go away.
77  */
78 static inline uint64_t
ulfs_fix_unwritten(uint32_t val)79 ulfs_fix_unwritten(uint32_t val)
80 {
81 	if (val == (uint32_t)UNWRITTEN) {
82 		return (uint64_t)(int64_t)UNWRITTEN;
83 	} else {
84 		return val;
85 	}
86 }
87 
88 
89 /*
90  * Bmap converts the logical block number of a file to its physical block
91  * number on the disk. The conversion is done by using the logical block
92  * number to index into the array of block pointers described by the dinode.
93  */
94 int
ulfs_bmap(void * v)95 ulfs_bmap(void *v)
96 {
97 	struct vop_bmap_args /* {
98 		struct vnode *a_vp;
99 		daddr_t  a_bn;
100 		struct vnode **a_vpp;
101 		daddr_t *a_bnp;
102 		int *a_runp;
103 	} */ *ap = v;
104 	int error;
105 
106 	/*
107 	 * Check for underlying vnode requests and ensure that logical
108 	 * to physical mapping is requested.
109 	 */
110 	if (ap->a_vpp != NULL)
111 		*ap->a_vpp = VTOI(ap->a_vp)->i_devvp;
112 	if (ap->a_bnp == NULL)
113 		return (0);
114 
115 	error = ulfs_bmaparray(ap->a_vp, ap->a_bn, ap->a_bnp, NULL, NULL,
116 	    ap->a_runp, ulfs_issequential);
117 	return error;
118 }
119 
120 /*
121  * Indirect blocks are now on the vnode for the file.  They are given negative
122  * logical block numbers.  Indirect blocks are addressed by the negative
123  * address of the first data block to which they point.  Double indirect blocks
124  * are addressed by one less than the address of the first indirect block to
125  * which they point.  Triple indirect blocks are addressed by one less than
126  * the address of the first double indirect block to which they point.
127  *
128  * ulfs_bmaparray does the bmap conversion, and if requested returns the
129  * array of logical blocks which must be traversed to get to a block.
130  * Each entry contains the offset into that block that gets you to the
131  * next block and the disk address of the block (if it is assigned).
132  */
133 
134 int
ulfs_bmaparray(struct vnode * vp,daddr_t bn,daddr_t * bnp,struct indir * ap,int * nump,int * runp,ulfs_issequential_callback_t is_sequential)135 ulfs_bmaparray(struct vnode *vp, daddr_t bn, daddr_t *bnp, struct indir *ap,
136     int *nump, int *runp, ulfs_issequential_callback_t is_sequential)
137 {
138 	struct inode *ip;
139 	struct buf *bp, *cbp;
140 	struct ulfsmount *ump;
141 	struct lfs *fs;
142 	struct mount *mp;
143 	struct indir a[ULFS_NIADDR + 1], *xap;
144 	daddr_t daddr;
145 	daddr_t metalbn;
146 	int error, maxrun = 0, num;
147 
148 	ip = VTOI(vp);
149 	mp = vp->v_mount;
150 	ump = ip->i_ump;
151 	fs = ip->i_lfs;
152 	KASSERTMSG(((ap == NULL) == (nump == NULL)),
153 	    "ulfs_bmaparray: invalid arguments: ap=%p, nump=%p", ap, nump);
154 
155 	if (runp) {
156 		/*
157 		 * XXX
158 		 * If MAXBSIZE is the largest transfer the disks can handle,
159 		 * we probably want maxrun to be 1 block less so that we
160 		 * don't create a block larger than the device can handle.
161 		 */
162 		*runp = 0;
163 		maxrun = MAXPHYS / mp->mnt_stat.f_iosize - 1;
164 	}
165 
166 	if (bn >= 0 && bn < ULFS_NDADDR) {
167 		if (nump != NULL)
168 			*nump = 0;
169 		if (ump->um_fstype == ULFS1)
170 			daddr = ulfs_fix_unwritten(ulfs_rw32(ip->i_din->u_32.di_db[bn],
171 			    ULFS_MPNEEDSWAP(fs)));
172 		else
173 			daddr = ulfs_rw64(ip->i_din->u_64.di_db[bn],
174 			    ULFS_MPNEEDSWAP(fs));
175 		*bnp = blkptrtodb(fs, daddr);
176 		/*
177 		 * Since this is FFS independent code, we are out of
178 		 * scope for the definitions of BLK_NOCOPY and
179 		 * BLK_SNAP, but we do know that they will fall in
180 		 * the range 1..um_seqinc, so we use that test and
181 		 * return a request for a zeroed out buffer if attempts
182 		 * are made to read a BLK_NOCOPY or BLK_SNAP block.
183 		 */
184 		if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL)) == SF_SNAPSHOT
185 		    && daddr > 0 &&
186 		    daddr < fs->um_seqinc) {
187 			*bnp = -1;
188 		} else if (*bnp == 0) {
189 			if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL))
190 			    == SF_SNAPSHOT) {
191 				*bnp = blkptrtodb(fs, bn * fs->um_seqinc);
192 			} else {
193 				*bnp = -1;
194 			}
195 		} else if (runp) {
196 			if (ump->um_fstype == ULFS1) {
197 				for (++bn; bn < ULFS_NDADDR && *runp < maxrun &&
198 				    is_sequential(fs,
199 				        ulfs_fix_unwritten(ulfs_rw32(ip->i_din->u_32.di_db[bn - 1],
200 				            ULFS_MPNEEDSWAP(fs))),
201 				        ulfs_fix_unwritten(ulfs_rw32(ip->i_din->u_32.di_db[bn],
202 				            ULFS_MPNEEDSWAP(fs))));
203 				    ++bn, ++*runp);
204 			} else {
205 				for (++bn; bn < ULFS_NDADDR && *runp < maxrun &&
206 				    is_sequential(fs,
207 				        ulfs_rw64(ip->i_din->u_64.di_db[bn - 1],
208 				            ULFS_MPNEEDSWAP(fs)),
209 				        ulfs_rw64(ip->i_din->u_64.di_db[bn],
210 				            ULFS_MPNEEDSWAP(fs)));
211 				    ++bn, ++*runp);
212 			}
213 		}
214 		return (0);
215 	}
216 
217 	xap = ap == NULL ? a : ap;
218 	if (!nump)
219 		nump = &num;
220 	if ((error = ulfs_getlbns(vp, bn, xap, nump)) != 0)
221 		return (error);
222 
223 	num = *nump;
224 
225 	/* Get disk address out of indirect block array */
226 	// XXX clean this up
227 	if (ump->um_fstype == ULFS1)
228 		daddr = ulfs_fix_unwritten(ulfs_rw32(ip->i_din->u_32.di_ib[xap->in_off],
229 		    ULFS_MPNEEDSWAP(fs)));
230 	else
231 		daddr = ulfs_rw64(ip->i_din->u_64.di_ib[xap->in_off],
232 		    ULFS_MPNEEDSWAP(fs));
233 
234 	for (bp = NULL, ++xap; --num; ++xap) {
235 		/*
236 		 * Exit the loop if there is no disk address assigned yet and
237 		 * the indirect block isn't in the cache, or if we were
238 		 * looking for an indirect block and we've found it.
239 		 */
240 
241 		metalbn = xap->in_lbn;
242 		if (metalbn == bn)
243 			break;
244 		if (daddr == 0) {
245 			mutex_enter(&bufcache_lock);
246 			cbp = incore(vp, metalbn);
247 			mutex_exit(&bufcache_lock);
248 			if (cbp == NULL)
249 				break;
250 		}
251 
252 		/*
253 		 * If we get here, we've either got the block in the cache
254 		 * or we have a disk address for it, go fetch it.
255 		 */
256 		if (bp)
257 			brelse(bp, 0);
258 
259 		xap->in_exists = 1;
260 		bp = getblk(vp, metalbn, mp->mnt_stat.f_iosize, 0, 0);
261 		if (bp == NULL) {
262 
263 			/*
264 			 * getblk() above returns NULL only iff we are
265 			 * pagedaemon.  See the implementation of getblk
266 			 * for detail.
267 			 */
268 
269 			return (ENOMEM);
270 		}
271 		if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
272 			trace(TR_BREADHIT, pack(vp, size), metalbn);
273 		} else {
274 			KASSERTMSG(daddr,
275 			    "ulfs_bmaparray: indirect block not in cache");
276 			trace(TR_BREADMISS, pack(vp, size), metalbn);
277 			bp->b_blkno = blkptrtodb(fs, daddr);
278 			bp->b_flags |= B_READ;
279 			BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
280 			VOP_STRATEGY(vp, bp);
281 			curlwp->l_ru.ru_inblock++;	/* XXX */
282 			if ((error = biowait(bp)) != 0) {
283 				brelse(bp, 0);
284 				return (error);
285 			}
286 		}
287 		if (ump->um_fstype == ULFS1) {
288 			daddr = ulfs_fix_unwritten(ulfs_rw32(((u_int32_t *)bp->b_data)[xap->in_off],
289 			    ULFS_MPNEEDSWAP(fs)));
290 			if (num == 1 && daddr && runp) {
291 				for (bn = xap->in_off + 1;
292 				    bn < MNINDIR(fs) && *runp < maxrun &&
293 				    is_sequential(fs,
294 				        ulfs_fix_unwritten(ulfs_rw32(((int32_t *)bp->b_data)[bn-1],
295 				            ULFS_MPNEEDSWAP(fs))),
296 				        ulfs_fix_unwritten(ulfs_rw32(((int32_t *)bp->b_data)[bn],
297 				            ULFS_MPNEEDSWAP(fs))));
298 				    ++bn, ++*runp);
299 			}
300 		} else {
301 			daddr = ulfs_rw64(((u_int64_t *)bp->b_data)[xap->in_off],
302 			    ULFS_MPNEEDSWAP(fs));
303 			if (num == 1 && daddr && runp) {
304 				for (bn = xap->in_off + 1;
305 				    bn < MNINDIR(fs) && *runp < maxrun &&
306 				    is_sequential(fs,
307 				        ulfs_rw64(((int64_t *)bp->b_data)[bn-1],
308 				            ULFS_MPNEEDSWAP(fs)),
309 				        ulfs_rw64(((int64_t *)bp->b_data)[bn],
310 				            ULFS_MPNEEDSWAP(fs)));
311 				    ++bn, ++*runp);
312 			}
313 		}
314 	}
315 	if (bp)
316 		brelse(bp, 0);
317 
318 	/*
319 	 * Since this is FFS independent code, we are out of scope for the
320 	 * definitions of BLK_NOCOPY and BLK_SNAP, but we do know that they
321 	 * will fall in the range 1..um_seqinc, so we use that test and
322 	 * return a request for a zeroed out buffer if attempts are made
323 	 * to read a BLK_NOCOPY or BLK_SNAP block.
324 	 */
325 	if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL)) == SF_SNAPSHOT
326 	    && daddr > 0 && daddr < fs->um_seqinc) {
327 		*bnp = -1;
328 		return (0);
329 	}
330 	*bnp = blkptrtodb(fs, daddr);
331 	if (*bnp == 0) {
332 		if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL))
333 		    == SF_SNAPSHOT) {
334 			*bnp = blkptrtodb(fs, bn * fs->um_seqinc);
335 		} else {
336 			*bnp = -1;
337 		}
338 	}
339 	return (0);
340 }
341 
342 /*
343  * Create an array of logical block number/offset pairs which represent the
344  * path of indirect blocks required to access a data block.  The first "pair"
345  * contains the logical block number of the appropriate single, double or
346  * triple indirect block and the offset into the inode indirect block array.
347  * Note, the logical block number of the inode single/double/triple indirect
348  * block appears twice in the array, once with the offset into the i_ffs1_ib and
349  * once with the offset into the page itself.
350  */
351 int
ulfs_getlbns(struct vnode * vp,daddr_t bn,struct indir * ap,int * nump)352 ulfs_getlbns(struct vnode *vp, daddr_t bn, struct indir *ap, int *nump)
353 {
354 	daddr_t metalbn, realbn;
355 	struct ulfsmount *ump;
356 	struct lfs *fs;
357 	int64_t blockcnt;
358 	int lbc;
359 	int i, numlevels, off;
360 
361 	ump = VFSTOULFS(vp->v_mount);
362 	fs = ump->um_lfs;
363 	if (nump)
364 		*nump = 0;
365 	numlevels = 0;
366 	realbn = bn;
367 	if (bn < 0)
368 		bn = -bn;
369 	KASSERT(bn >= ULFS_NDADDR);
370 
371 	/*
372 	 * Determine the number of levels of indirection.  After this loop
373 	 * is done, blockcnt indicates the number of data blocks possible
374 	 * at the given level of indirection, and ULFS_NIADDR - i is the number
375 	 * of levels of indirection needed to locate the requested block.
376 	 */
377 
378 	bn -= ULFS_NDADDR;
379 	for (lbc = 0, i = ULFS_NIADDR;; i--, bn -= blockcnt) {
380 		if (i == 0)
381 			return (EFBIG);
382 
383 		lbc += fs->um_lognindir;
384 		blockcnt = (int64_t)1 << lbc;
385 
386 		if (bn < blockcnt)
387 			break;
388 	}
389 
390 	/* Calculate the address of the first meta-block. */
391 	metalbn = -((realbn >= 0 ? realbn : -realbn) - bn + ULFS_NIADDR - i);
392 
393 	/*
394 	 * At each iteration, off is the offset into the bap array which is
395 	 * an array of disk addresses at the current level of indirection.
396 	 * The logical block number and the offset in that block are stored
397 	 * into the argument array.
398 	 */
399 	ap->in_lbn = metalbn;
400 	ap->in_off = off = ULFS_NIADDR - i;
401 	ap->in_exists = 0;
402 	ap++;
403 	for (++numlevels; i <= ULFS_NIADDR; i++) {
404 		/* If searching for a meta-data block, quit when found. */
405 		if (metalbn == realbn)
406 			break;
407 
408 		lbc -= fs->um_lognindir;
409 		off = (bn >> lbc) & (MNINDIR(fs) - 1);
410 
411 		++numlevels;
412 		ap->in_lbn = metalbn;
413 		ap->in_off = off;
414 		ap->in_exists = 0;
415 		++ap;
416 
417 		metalbn -= -1 + ((int64_t)off << lbc);
418 	}
419 	if (nump)
420 		*nump = numlevels;
421 	return (0);
422 }
423