xref: /netbsd-src/sys/ufs/lfs/lfs_balloc.c (revision 481fca6e59249d8ffcf24fef7cfbe7b131bfb080)
1 /*	$NetBSD: lfs_balloc.c,v 1.24 2000/07/04 22:30:37 perseant Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *      This product includes software developed by the NetBSD
21  *      Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 /*
39  * Copyright (c) 1989, 1991, 1993
40  *	The Regents of the University of California.  All rights reserved.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. All advertising materials mentioning features or use of this software
51  *    must display the following acknowledgement:
52  *	This product includes software developed by the University of
53  *	California, Berkeley and its contributors.
54  * 4. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)lfs_balloc.c	8.4 (Berkeley) 5/8/95
71  */
72 
73 #if defined(_KERNEL) && !defined(_LKM)
74 #include "opt_quota.h"
75 #endif
76 
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/buf.h>
80 #include <sys/proc.h>
81 #include <sys/vnode.h>
82 #include <sys/mount.h>
83 #include <sys/resourcevar.h>
84 #include <sys/trace.h>
85 
86 #include <miscfs/specfs/specdev.h>
87 
88 #include <ufs/ufs/quota.h>
89 #include <ufs/ufs/inode.h>
90 #include <ufs/ufs/ufsmount.h>
91 #include <ufs/ufs/ufs_extern.h>
92 
93 #include <ufs/lfs/lfs.h>
94 #include <ufs/lfs/lfs_extern.h>
95 
96 int lfs_fragextend __P((struct vnode *, int, int, ufs_daddr_t, struct buf **));
97 
98 /*
99  * Allocate a block, and to inode and filesystem block accounting for it
100  * and for any indirect blocks the may need to be created in order for
101  * this block to be created.
102  *
103  * Blocks which have never been accounted for (i.e., which "do not exist")
104  * have disk address 0, which is translated by ufs_bmap to the special value
105  * UNASSIGNED == -1, as in the historical UFS.
106  *
107  * Blocks which have been accounted for but which have not yet been written
108  * to disk are given the new special disk address UNWRITTEN == -2, so that
109  * they can be differentiated from completely new blocks.
110  */
111 int
112 lfs_balloc(v)
113 	void *v;
114 {
115 	struct vop_balloc_args /* {
116 		struct vnode *a_vp;
117 		off_t a_startoffset;
118 		int a_size;
119 		struct ucred *a_cred;
120 		int a_flags;
121 		struct buf *a_bpp;
122 	} */ *ap = v;
123 	struct vnode *vp;
124 	int offset;
125 	u_long iosize;
126 	daddr_t daddr, idaddr;
127 	struct buf *ibp, *bp;
128 	struct inode *ip;
129 	struct lfs *fs;
130 	struct indir indirs[NIADDR+2], *idp;
131 	ufs_daddr_t	lbn, lastblock;
132 	int bb, bcount;
133 	int error, frags, i, nsize, osize, num;
134 
135 	vp = ap->a_vp;
136 	ip = VTOI(vp);
137 	fs = ip->i_lfs;
138 	offset = blkoff(fs, ap->a_startoffset);
139 	iosize = ap->a_size;
140 	lbn = lblkno(fs, ap->a_startoffset);
141 	(void)lfs_check(vp, lbn, 0);
142 
143 	/*
144 	 * Three cases: it's a block beyond the end of file, it's a block in
145 	 * the file that may or may not have been assigned a disk address or
146 	 * we're writing an entire block.
147 	 *
148 	 * Note, if the daddr is UNWRITTEN, the block already exists in
149 	 * the cache (it was read or written earlier).  If so, make sure
150 	 * we don't count it as a new block or zero out its contents. If
151 	 * it did not, make sure we allocate any necessary indirect
152 	 * blocks.
153 	 *
154 	 * If we are writing a block beyond the end of the file, we need to
155 	 * check if the old last block was a fragment.	If it was, we need
156 	 * to rewrite it.
157 	 */
158 
159 	*ap->a_bpp = NULL;
160 
161 	/* Check for block beyond end of file and fragment extension needed. */
162 	lastblock = lblkno(fs, ip->i_ffs_size);
163 	if (lastblock < NDADDR && lastblock < lbn) {
164 		osize = blksize(fs, ip, lastblock);
165 		if (osize < fs->lfs_bsize && osize > 0) {
166 			if ((error = lfs_fragextend(vp, osize, fs->lfs_bsize,
167 						    lastblock, &bp)))
168 				return(error);
169 			ip->i_ffs_size = (lastblock + 1) * fs->lfs_bsize;
170 			uvm_vnp_setsize(vp, ip->i_ffs_size);
171 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
172 			VOP_BWRITE(bp);
173 		}
174 	}
175 
176 	/*
177 	 * If the block we are writing is a direct block, it's the last
178 	 * block in the file, and offset + iosize is less than a full
179 	 * block, we can write one or more fragments.  There are two cases:
180 	 * the block is brand new and we should allocate it the correct
181 	 * size or it already exists and contains some fragments and
182 	 * may need to extend it.
183 	 */
184 	if (lbn < NDADDR && lblkno(fs, ip->i_ffs_size) <= lbn) {
185 		osize = blksize(fs, ip, lbn);
186 		nsize = fragroundup(fs, offset + iosize);
187 		if (lblktosize(fs, lbn) >= ip->i_ffs_size) {
188 			/* Brand new block or fragment */
189 			frags = numfrags(fs, nsize);
190 			bb = fragstodb(fs, frags);
191 			*ap->a_bpp = bp = getblk(vp, lbn, nsize, 0, 0);
192 			ip->i_lfs_effnblks += bb;
193 			ip->i_lfs->lfs_bfree -= bb;
194 			ip->i_ffs_db[lbn] = bp->b_blkno = UNWRITTEN;
195 		} else {
196 			if (nsize <= osize) {
197 				/* No need to extend */
198 				if ((error = bread(vp, lbn, osize, NOCRED, &bp)))
199 					return error;
200 			} else {
201 				/* Extend existing block */
202 				if ((error =
203 				     lfs_fragextend(vp, osize, nsize, lbn, &bp)))
204 					return error;
205 			}
206 			*ap->a_bpp = bp;
207 		}
208 		return 0;
209 	}
210 
211 	error = ufs_bmaparray(vp, lbn, &daddr, &indirs[0], &num, NULL );
212 	if (error)
213 		return (error);
214 	/*
215 	 * Do byte accounting all at once, so we can gracefully fail *before*
216 	 * we start assigning blocks.
217 	 */
218 	bb = VFSTOUFS(vp->v_mount)->um_seqinc;
219 	bcount = 0;
220 	if (daddr == UNASSIGNED) {
221 		bcount = bb;
222 	}
223 	for (i = 1; i < num; ++i) {
224 		if (!indirs[i].in_exists) {
225 			bcount += bb;
226 		}
227 	}
228 	if (ISSPACE(fs, bcount, ap->a_cred)) {
229 		ip->i_lfs->lfs_bfree -= bcount;
230 		ip->i_lfs_effnblks += bcount;
231 	} else {
232 		return ENOSPC;
233 	}
234 
235 	if (daddr == UNASSIGNED) {
236 		if (num > 0 && ip->i_ffs_ib[indirs[0].in_off] == 0) {
237 			ip->i_ffs_ib[indirs[0].in_off] = UNWRITTEN;
238 		}
239 
240 		/*
241 		 * Create new indirect blocks if necessary
242 		 */
243 		if (num > 1)
244 			idaddr = ip->i_ffs_ib[indirs[0].in_off];
245 		for (i = 1; i < num; ++i) {
246 			ibp = getblk(vp, indirs[i].in_lbn, fs->lfs_bsize, 0,0);
247 			if (!indirs[i].in_exists) {
248 				clrbuf(ibp);
249 				ibp->b_blkno = UNWRITTEN;
250 			} else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) {
251 				ibp->b_blkno = idaddr;
252 				ibp->b_flags |= B_READ;
253 				VOP_STRATEGY(ibp);
254 				biowait(ibp);
255 			}
256 			/*
257 			 * This block exists, but the next one may not.
258 			 * If that is the case mark it UNWRITTEN to keep
259 			 * the accounting straight.
260 			 */
261 			if (((daddr_t *)ibp->b_data)[indirs[i].in_off]==0)
262 				((daddr_t *)ibp->b_data)[indirs[i].in_off] =
263 					UNWRITTEN;
264 			idaddr = ((daddr_t *)ibp->b_data)[indirs[i].in_off];
265 			if ((error = VOP_BWRITE(ibp))) {
266 				return error;
267 			}
268 		}
269 	}
270 
271 
272 	/*
273 	 * Get the existing block from the cache.
274 	 */
275 	frags = dbtofrags(fs, bb);
276 	*ap->a_bpp = bp = getblk(vp, lbn, blksize(fs, ip, lbn), 0, 0);
277 
278 	/*
279 	 * The block we are writing may be a brand new block
280 	 * in which case we need to do accounting.
281 	 *
282 	 * We can tell a truly new block because ufs_bmaparray will say
283 	 * it is UNASSIGNED.  Once we allocate it we will assign it the
284 	 * disk address UNWRITTEN.
285 	 */
286 	if (daddr == UNASSIGNED) {
287 		if (iosize != fs->lfs_bsize)
288 			clrbuf(bp);
289 
290 		/* Note the new address */
291 		bp->b_blkno = UNWRITTEN;
292 
293 		switch (num) {
294 		    case 0:
295 			ip->i_ffs_db[lbn] = UNWRITTEN;
296 			break;
297 		    case 1:
298 			ip->i_ffs_ib[indirs[0].in_off] = UNWRITTEN;
299 			break;
300 		    default:
301 			idp = &indirs[num - 1];
302 			if (bread(vp, idp->in_lbn, fs->lfs_bsize, NOCRED,
303 				  &ibp))
304 				panic("lfs_balloc: bread bno %d", idp->in_lbn);
305 			((ufs_daddr_t *)ibp->b_data)[idp->in_off] = UNWRITTEN;
306 			VOP_BWRITE(ibp);
307 		}
308 	} else if (!(bp->b_flags & (B_DONE|B_DELWRI))) {
309 		/*
310 		 * Not a brand new block, also not in the cache;
311 		 * read it in from disk.
312 		 */
313 		if (iosize == fs->lfs_bsize)
314 			/* Optimization: I/O is unnecessary. */
315 			bp->b_blkno = daddr;
316 		else {
317 			/*
318 			 * We need to read the block to preserve the
319 			 * existing bytes.
320 			 */
321 			bp->b_blkno = daddr;
322 			bp->b_flags |= B_READ;
323 			VOP_STRATEGY(bp);
324 			return(biowait(bp));
325 		}
326 	}
327 
328 	return (0);
329 }
330 
331 int
332 lfs_fragextend(vp, osize, nsize, lbn, bpp)
333 	struct vnode *vp;
334 	int osize;
335 	int nsize;
336 	ufs_daddr_t lbn;
337 	struct buf **bpp;
338 {
339 	struct inode *ip;
340 	struct lfs *fs;
341 	long bb;
342 	int error;
343 	extern long locked_queue_bytes;
344 	struct buf *ibp;
345 	SEGUSE *sup;
346 
347 	ip = VTOI(vp);
348 	fs = ip->i_lfs;
349 	bb = (long)fragstodb(fs, numfrags(fs, nsize - osize));
350 	error = 0;
351 
352 	/*
353 	 * Get the seglock so we don't enlarge blocks or change the segment
354 	 * accounting information while a segment is being written.
355 	 */
356 	lfs_seglock(fs, SEGM_PROT);
357 
358 	if (!ISSPACE(fs, bb, curproc->p_ucred)) {
359 		error = ENOSPC;
360 		goto out;
361 	}
362 	if ((error = bread(vp, lbn, osize, NOCRED, bpp))) {
363 		brelse(*bpp);
364 		goto out;
365 	}
366 #ifdef QUOTA
367 	if ((error = chkdq(ip, bb, curproc->p_ucred, 0))) {
368 		brelse(*bpp);
369 		goto out;
370 	}
371 #endif
372 	/*
373  	 * Fix the allocation for this fragment so that it looks like the
374          * source segment contained a block of the new size.  This overcounts;
375 	 * but the overcount only lasts until the block in question
376 	 * is written, so the on-disk live bytes count is always correct.
377 	 */
378 	if ((*bpp)->b_blkno > 0) {
379 		LFS_SEGENTRY(sup, fs, datosn(fs,(*bpp)->b_blkno), ibp);
380 		sup->su_nbytes += (nsize-osize);
381 		VOP_BWRITE(ibp);
382 		ip->i_ffs_blocks += bb;
383 	}
384 	fs->lfs_bfree -= bb;
385 	ip->i_lfs_effnblks += bb;
386 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
387 	if((*bpp)->b_flags & B_LOCKED)
388 		locked_queue_bytes += (nsize - osize);
389 	allocbuf(*bpp, nsize);
390 	bzero((char *)((*bpp)->b_data) + osize, (u_int)(nsize - osize));
391 
392     out:
393 	lfs_segunlock(fs);
394 	return (error);
395 }
396