Clearly I need to wakeup all the way before starting to make changes.
[dragonfly.git] / sys / vfs / ufs / ffs_balloc.c
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *      The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *      This product includes software developed by the University of
16  *      California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *      @(#)ffs_balloc.c        8.8 (Berkeley) 6/16/95
34  * $FreeBSD: src/sys/ufs/ffs/ffs_balloc.c,v 1.26.2.1 2002/10/10 19:48:20 dillon Exp $
35  * $DragonFly: src/sys/vfs/ufs/ffs_balloc.c,v 1.11 2004/08/24 16:32:11 drhodus Exp $
36  */
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/proc.h>
41 #include <sys/buf.h>
42 #include <sys/lock.h>
43 #include <sys/mount.h>
44 #include <sys/vnode.h>
45
46 #include "quota.h"
47 #include "inode.h"
48 #include "ufs_extern.h"
49
50 #include "fs.h"
51 #include "ffs_extern.h"
52
53 /*
54  * Balloc defines the structure of filesystem storage
55  * by allocating the physical blocks on a device given
56  * the inode and the logical block number in a file.
57  *
58  * ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size,
59  *            struct ucred *a_cred, int a_flags, struct buf *a_bpp)
60  */
61 int
62 ffs_balloc(struct vop_balloc_args *ap)
63 {
64         struct inode *ip;
65         ufs_daddr_t lbn;
66         int size;
67         struct ucred *cred;
68         int flags;
69         struct fs *fs;
70         ufs_daddr_t nb;
71         struct buf *bp, *nbp;
72         struct vnode *vp;
73         struct indir indirs[NIADDR + 2];
74         ufs_daddr_t newb, *bap, pref;
75         int deallocated, osize, nsize, num, i, error;
76         ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
77         int unwindidx = -1;
78         struct thread *td = curthread;  /* XXX */
79
80         vp = ap->a_vp;
81         ip = VTOI(vp);
82         fs = ip->i_fs;
83         lbn = lblkno(fs, ap->a_startoffset);
84         size = blkoff(fs, ap->a_startoffset) + ap->a_size;
85         if (size > fs->fs_bsize)
86                 panic("ffs_balloc: blk too big");
87         *ap->a_bpp = NULL;
88         if (lbn < 0)
89                 return (EFBIG);
90         cred = ap->a_cred;
91         flags = ap->a_flags;
92
93         /*
94          * If the next write will extend the file into a new block,
95          * and the file is currently composed of a fragment
96          * this fragment has to be extended to be a full block.
97          */
98         nb = lblkno(fs, ip->i_size);
99         if (nb < NDADDR && nb < lbn) {
100                 /*
101                  * The filesize prior to this write can fit in direct
102                  * blocks (ex. fragmentation is possibly done)
103                  * we are now extending the file write beyond
104                  * the block which has end of the file prior to this write.
105                  */
106                 osize = blksize(fs, ip, nb);
107                 /*
108                  * osize gives disk allocated size in the last block. It is
109                  * either in fragments or a file system block size.
110                  */
111                 if (osize < fs->fs_bsize && osize > 0) {
112                         /* A few fragments are already allocated, since the
113                          * current extends beyond this block allocated the
114                          * complete block as fragments are on in last block.
115                          */
116                         error = ffs_realloccg(ip, nb,
117                                 ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
118                                 osize, (int)fs->fs_bsize, cred, &bp);
119                         if (error)
120                                 return (error);
121                         if (DOINGSOFTDEP(vp))
122                                 softdep_setup_allocdirect(ip, nb,
123                                     dbtofsb(fs, bp->b_blkno), ip->i_db[nb],
124                                     fs->fs_bsize, osize, bp);
125                         /* adjust the inode size, we just grew */
126                         ip->i_size = smalllblktosize(fs, nb + 1);
127                         ip->i_db[nb] = dbtofsb(fs, bp->b_blkno);
128                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
129                         if (flags & B_SYNC)
130                                 bwrite(bp);
131                         else
132                                 bawrite(bp);
133                         /* bp is already released here */
134                 }
135         }
136         /*
137          * The first NDADDR blocks are direct blocks
138          */
139         if (lbn < NDADDR) {
140                 nb = ip->i_db[lbn];
141                 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
142                         error = bread(vp, lbn, fs->fs_bsize, &bp);
143                         if (error) {
144                                 brelse(bp);
145                                 return (error);
146                         }
147                         bp->b_blkno = fsbtodb(fs, nb);
148                         *ap->a_bpp = bp;
149                         return (0);
150                 }
151                 if (nb != 0) {
152                         /*
153                          * Consider need to reallocate a fragment.
154                          */
155                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
156                         nsize = fragroundup(fs, size);
157                         if (nsize <= osize) {
158                                 error = bread(vp, lbn, osize, &bp);
159                                 if (error) {
160                                         brelse(bp);
161                                         return (error);
162                                 }
163                                 bp->b_blkno = fsbtodb(fs, nb);
164                         } else {
165                                 error = ffs_realloccg(ip, lbn,
166                                     ffs_blkpref(ip, lbn, (int)lbn,
167                                         &ip->i_db[0]), osize, nsize, cred, &bp);
168                                 if (error)
169                                         return (error);
170                                 if (DOINGSOFTDEP(vp))
171                                         softdep_setup_allocdirect(ip, lbn,
172                                             dbtofsb(fs, bp->b_blkno), nb,
173                                             nsize, osize, bp);
174                         }
175                 } else {
176                         if (ip->i_size < smalllblktosize(fs, lbn + 1))
177                                 nsize = fragroundup(fs, size);
178                         else
179                                 nsize = fs->fs_bsize;
180                         error = ffs_alloc(ip, lbn,
181                             ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
182                             nsize, cred, &newb);
183                         if (error)
184                                 return (error);
185                         bp = getblk(vp, lbn, nsize, 0, 0);
186                         bp->b_blkno = fsbtodb(fs, newb);
187                         if (flags & B_CLRBUF)
188                                 vfs_bio_clrbuf(bp);
189                         if (DOINGSOFTDEP(vp))
190                                 softdep_setup_allocdirect(ip, lbn, newb, 0,
191                                     nsize, 0, bp);
192                 }
193                 ip->i_db[lbn] = dbtofsb(fs, bp->b_blkno);
194                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
195                 *ap->a_bpp = bp;
196                 return (0);
197         }
198         /*
199          * Determine the number of levels of indirection.
200          */
201         pref = 0;
202         if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
203                 return(error);
204 #ifdef DIAGNOSTIC
205         if (num < 1)
206                 panic ("ffs_balloc: ufs_bmaparray returned indirect block");
207 #endif
208         /*
209          * Fetch the first indirect block allocating if necessary.
210          */
211         --num;
212         nb = ip->i_ib[indirs[0].in_off];
213         allocib = NULL;
214         allocblk = allociblk;
215         if (nb == 0) {
216                 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
217                 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
218                     cred, &newb)) != 0)
219                         return (error);
220                 nb = newb;
221                 *allocblk++ = nb;
222                 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
223                 bp->b_blkno = fsbtodb(fs, nb);
224                 vfs_bio_clrbuf(bp);
225                 if (DOINGSOFTDEP(vp)) {
226                         softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
227                             newb, 0, fs->fs_bsize, 0, bp);
228                         bdwrite(bp);
229                 } else {
230                         /*
231                          * Write synchronously so that indirect blocks
232                          * never point at garbage.
233                          */
234                         if (DOINGASYNC(vp))
235                                 bdwrite(bp);
236                         else if ((error = bwrite(bp)) != 0)
237                                 goto fail;
238                 }
239                 allocib = &ip->i_ib[indirs[0].in_off];
240                 *allocib = nb;
241                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
242         }
243         /*
244          * Fetch through the indirect blocks, allocating as necessary.
245          */
246         for (i = 1;;) {
247                 error = bread(vp, indirs[i].in_lbn, (int)fs->fs_bsize, &bp);
248                 if (error) {
249                         brelse(bp);
250                         goto fail;
251                 }
252                 bap = (ufs_daddr_t *)bp->b_data;
253                 nb = bap[indirs[i].in_off];
254                 if (i == num)
255                         break;
256                 i += 1;
257                 if (nb != 0) {
258                         bqrelse(bp);
259                         continue;
260                 }
261                 if (pref == 0)
262                         pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
263                 if ((error =
264                     ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
265                         brelse(bp);
266                         goto fail;
267                 }
268                 nb = newb;
269                 *allocblk++ = nb;
270                 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
271                 nbp->b_blkno = fsbtodb(fs, nb);
272                 vfs_bio_clrbuf(nbp);
273                 if (DOINGSOFTDEP(vp)) {
274                         softdep_setup_allocindir_meta(nbp, ip, bp,
275                             indirs[i - 1].in_off, nb);
276                         bdwrite(nbp);
277                 } else {
278                         /*
279                          * Write synchronously so that indirect blocks
280                          * never point at garbage.
281                          */
282                         if ((error = bwrite(nbp)) != 0) {
283                                 brelse(bp);
284                                 goto fail;
285                         }
286                 }
287                 bap[indirs[i - 1].in_off] = nb;
288                 if (allocib == NULL && unwindidx < 0)
289                         unwindidx = i - 1;
290                 /*
291                  * If required, write synchronously, otherwise use
292                  * delayed write.
293                  */
294                 if (flags & B_SYNC) {
295                         bwrite(bp);
296                 } else {
297                         if (bp->b_bufsize == fs->fs_bsize)
298                                 bp->b_flags |= B_CLUSTEROK;
299                         bdwrite(bp);
300                 }
301         }
302         /*
303          * Get the data block, allocating if necessary.
304          */
305         if (nb == 0) {
306                 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
307                 error = ffs_alloc(ip,
308                     lbn, pref, (int)fs->fs_bsize, cred, &newb);
309                 if (error) {
310                         brelse(bp);
311                         goto fail;
312                 }
313                 nb = newb;
314                 *allocblk++ = nb;
315                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
316                 nbp->b_blkno = fsbtodb(fs, nb);
317                 if (flags & B_CLRBUF)
318                         vfs_bio_clrbuf(nbp);
319                 if (DOINGSOFTDEP(vp))
320                         softdep_setup_allocindir_page(ip, lbn, bp,
321                             indirs[i].in_off, nb, 0, nbp);
322                 bap[indirs[i].in_off] = nb;
323                 /*
324                  * If required, write synchronously, otherwise use
325                  * delayed write.
326                  */
327                 if (flags & B_SYNC) {
328                         bwrite(bp);
329                 } else {
330                         if (bp->b_bufsize == fs->fs_bsize)
331                                 bp->b_flags |= B_CLUSTEROK;
332                         bdwrite(bp);
333                 }
334                 *ap->a_bpp = nbp;
335                 return (0);
336         }
337         brelse(bp);
338         /*
339          * If requested clear invalid portions of the buffer.  If we 
340          * have to do a read-before-write (typical if B_CLRBUF is set),
341          * try to do some read-ahead in the sequential case to reduce
342          * the number of I/O transactions.
343          */
344         if (flags & B_CLRBUF) {
345                 int seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
346                 if (seqcount &&
347                     (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
348                         error = cluster_read(vp, ip->i_size, lbn,
349                                     (int)fs->fs_bsize, 
350                                     MAXBSIZE, seqcount, &nbp);
351                 } else {
352                         error = bread(vp, lbn, (int)fs->fs_bsize, &nbp);
353                 }
354                 if (error) {
355                         brelse(nbp);
356                         goto fail;
357                 }
358         } else {
359                 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
360                 nbp->b_blkno = fsbtodb(fs, nb);
361         }
362         *ap->a_bpp = nbp;
363         return (0);
364 fail:
365         /*
366          * If we have failed part way through block allocation, we
367          * have to deallocate any indirect blocks that we have allocated.
368          * We have to fsync the file before we start to get rid of all
369          * of its dependencies so that we do not leave them dangling.
370          * We have to sync it at the end so that the soft updates code
371          * does not find any untracked changes. Although this is really
372          * slow, running out of disk space is not expected to be a common
373          * occurence. The error return from fsync is ignored as we already
374          * have an error to return to the user.
375          */
376         (void) VOP_FSYNC(vp, MNT_WAIT, td);
377         for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
378                 ffs_blkfree(ip, *blkp, fs->fs_bsize);
379                 deallocated += fs->fs_bsize;
380         }
381         if (allocib != NULL) {
382                 *allocib = 0;
383         } else if (unwindidx >= 0) {
384                 int r;
385
386                 r = bread(vp, indirs[unwindidx].in_lbn, (int)fs->fs_bsize, &bp);
387                 if (r) {
388                         panic("Could not unwind indirect block, error %d", r);
389                         brelse(bp);
390                 } else {
391                         bap = (ufs_daddr_t *)bp->b_data;
392                         bap[indirs[unwindidx].in_off] = 0;
393                         if (flags & B_SYNC) {
394                                 bwrite(bp);
395                         } else {
396                                 if (bp->b_bufsize == fs->fs_bsize)
397                                         bp->b_flags |= B_CLUSTEROK;
398                                 bdwrite(bp);
399                         }
400                 }
401         }
402         if (deallocated) {
403 #ifdef QUOTA
404                 /*
405                  * Restore user's disk quota because allocation failed.
406                  */
407                 (void) chkdq(ip, (long)-btodb(deallocated), cred, FORCE);
408 #endif
409                 ip->i_blocks -= btodb(deallocated);
410                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
411         }
412         (void) VOP_FSYNC(vp, MNT_WAIT, td);
413         return (error);
414 }