Merge branch 'mini2440-dev-unlikely' into mini2440-dev
[linux-2.6/mini2440.git] / fs / jfs / jfs_extent.c
blob41d6045dbeb08b5db952e9945a054bf722470451
1 /*
2 * Copyright (C) International Business Machines Corp., 2000-2004
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 #include <linux/fs.h>
20 #include <linux/quotaops.h>
21 #include "jfs_incore.h"
22 #include "jfs_inode.h"
23 #include "jfs_superblock.h"
24 #include "jfs_dmap.h"
25 #include "jfs_extent.h"
26 #include "jfs_debug.h"
29 * forward references
31 static int extBalloc(struct inode *, s64, s64 *, s64 *);
32 #ifdef _NOTYET
33 static int extBrealloc(struct inode *, s64, s64, s64 *, s64 *);
34 #endif
35 static s64 extRoundDown(s64 nb);
37 #define DPD(a) (printk("(a): %d\n",(a)))
38 #define DPC(a) (printk("(a): %c\n",(a)))
39 #define DPL1(a) \
40 { \
41 if ((a) >> 32) \
42 printk("(a): %x%08x ",(a)); \
43 else \
44 printk("(a): %x ",(a) << 32); \
46 #define DPL(a) \
47 { \
48 if ((a) >> 32) \
49 printk("(a): %x%08x\n",(a)); \
50 else \
51 printk("(a): %x\n",(a) << 32); \
54 #define DPD1(a) (printk("(a): %d ",(a)))
55 #define DPX(a) (printk("(a): %08x\n",(a)))
56 #define DPX1(a) (printk("(a): %08x ",(a)))
57 #define DPS(a) (printk("%s\n",(a)))
58 #define DPE(a) (printk("\nENTERING: %s\n",(a)))
59 #define DPE1(a) (printk("\nENTERING: %s",(a)))
60 #define DPS1(a) (printk(" %s ",(a)))
64 * NAME: extAlloc()
66 * FUNCTION: allocate an extent for a specified page range within a
67 * file.
69 * PARAMETERS:
70 * ip - the inode of the file.
71 * xlen - requested extent length.
72 * pno - the starting page number with the file.
73 * xp - pointer to an xad. on entry, xad describes an
74 * extent that is used as an allocation hint if the
75 * xaddr of the xad is non-zero. on successful exit,
76 * the xad describes the newly allocated extent.
77 * abnr - bool indicating whether the newly allocated extent
78 * should be marked as allocated but not recorded.
80 * RETURN VALUES:
81 * 0 - success
82 * -EIO - i/o error.
83 * -ENOSPC - insufficient disk resources.
85 int
86 extAlloc(struct inode *ip, s64 xlen, s64 pno, xad_t * xp, bool abnr)
88 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
89 s64 nxlen, nxaddr, xoff, hint, xaddr = 0;
90 int rc;
91 int xflag;
93 /* This blocks if we are low on resources */
94 txBeginAnon(ip->i_sb);
96 /* Avoid race with jfs_commit_inode() */
97 mutex_lock(&JFS_IP(ip)->commit_mutex);
99 /* validate extent length */
100 if (xlen > MAXXLEN)
101 xlen = MAXXLEN;
103 /* get the page's starting extent offset */
104 xoff = pno << sbi->l2nbperpage;
106 /* check if an allocation hint was provided */
107 if ((hint = addressXAD(xp))) {
108 /* get the size of the extent described by the hint */
109 nxlen = lengthXAD(xp);
111 /* check if the hint is for the portion of the file
112 * immediately previous to the current allocation
113 * request and if hint extent has the same abnr
114 * value as the current request. if so, we can
115 * extend the hint extent to include the current
116 * extent if we can allocate the blocks immediately
117 * following the hint extent.
119 if (offsetXAD(xp) + nxlen == xoff &&
120 abnr == ((xp->flag & XAD_NOTRECORDED) ? true : false))
121 xaddr = hint + nxlen;
123 /* adjust the hint to the last block of the extent */
124 hint += (nxlen - 1);
127 /* allocate the disk blocks for the extent. initially, extBalloc()
128 * will try to allocate disk blocks for the requested size (xlen).
129 * if this fails (xlen contiguous free blocks not avaliable), it'll
130 * try to allocate a smaller number of blocks (producing a smaller
131 * extent), with this smaller number of blocks consisting of the
132 * requested number of blocks rounded down to the next smaller
133 * power of 2 number (i.e. 16 -> 8). it'll continue to round down
134 * and retry the allocation until the number of blocks to allocate
135 * is smaller than the number of blocks per page.
137 nxlen = xlen;
138 if ((rc = extBalloc(ip, hint ? hint : INOHINT(ip), &nxlen, &nxaddr))) {
139 mutex_unlock(&JFS_IP(ip)->commit_mutex);
140 return (rc);
143 /* Allocate blocks to quota. */
144 if (vfs_dq_alloc_block(ip, nxlen)) {
145 dbFree(ip, nxaddr, (s64) nxlen);
146 mutex_unlock(&JFS_IP(ip)->commit_mutex);
147 return -EDQUOT;
150 /* determine the value of the extent flag */
151 xflag = abnr ? XAD_NOTRECORDED : 0;
153 /* if we can extend the hint extent to cover the current request,
154 * extend it. otherwise, insert a new extent to
155 * cover the current request.
157 if (xaddr && xaddr == nxaddr)
158 rc = xtExtend(0, ip, xoff, (int) nxlen, 0);
159 else
160 rc = xtInsert(0, ip, xflag, xoff, (int) nxlen, &nxaddr, 0);
162 /* if the extend or insert failed,
163 * free the newly allocated blocks and return the error.
165 if (rc) {
166 dbFree(ip, nxaddr, nxlen);
167 vfs_dq_free_block(ip, nxlen);
168 mutex_unlock(&JFS_IP(ip)->commit_mutex);
169 return (rc);
172 /* set the results of the extent allocation */
173 XADaddress(xp, nxaddr);
174 XADlength(xp, nxlen);
175 XADoffset(xp, xoff);
176 xp->flag = xflag;
178 mark_inode_dirty(ip);
180 mutex_unlock(&JFS_IP(ip)->commit_mutex);
182 * COMMIT_SyncList flags an anonymous tlock on page that is on
183 * sync list.
184 * We need to commit the inode to get the page written disk.
186 if (test_and_clear_cflag(COMMIT_Synclist,ip))
187 jfs_commit_inode(ip, 0);
189 return (0);
193 #ifdef _NOTYET
195 * NAME: extRealloc()
197 * FUNCTION: extend the allocation of a file extent containing a
198 * partial back last page.
200 * PARAMETERS:
201 * ip - the inode of the file.
202 * cp - cbuf for the partial backed last page.
203 * xlen - request size of the resulting extent.
204 * xp - pointer to an xad. on successful exit, the xad
205 * describes the newly allocated extent.
206 * abnr - bool indicating whether the newly allocated extent
207 * should be marked as allocated but not recorded.
209 * RETURN VALUES:
210 * 0 - success
211 * -EIO - i/o error.
212 * -ENOSPC - insufficient disk resources.
214 int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, bool abnr)
216 struct super_block *sb = ip->i_sb;
217 s64 xaddr, xlen, nxaddr, delta, xoff;
218 s64 ntail, nextend, ninsert;
219 int rc, nbperpage = JFS_SBI(sb)->nbperpage;
220 int xflag;
222 /* This blocks if we are low on resources */
223 txBeginAnon(ip->i_sb);
225 mutex_lock(&JFS_IP(ip)->commit_mutex);
226 /* validate extent length */
227 if (nxlen > MAXXLEN)
228 nxlen = MAXXLEN;
230 /* get the extend (partial) page's disk block address and
231 * number of blocks.
233 xaddr = addressXAD(xp);
234 xlen = lengthXAD(xp);
235 xoff = offsetXAD(xp);
237 /* if the extend page is abnr and if the request is for
238 * the extent to be allocated and recorded,
239 * make the page allocated and recorded.
241 if ((xp->flag & XAD_NOTRECORDED) && !abnr) {
242 xp->flag = 0;
243 if ((rc = xtUpdate(0, ip, xp)))
244 goto exit;
247 /* try to allocated the request number of blocks for the
248 * extent. dbRealloc() first tries to satisfy the request
249 * by extending the allocation in place. otherwise, it will
250 * try to allocate a new set of blocks large enough for the
251 * request. in satisfying a request, dbReAlloc() may allocate
252 * less than what was request but will always allocate enough
253 * space as to satisfy the extend page.
255 if ((rc = extBrealloc(ip, xaddr, xlen, &nxlen, &nxaddr)))
256 goto exit;
258 /* Allocat blocks to quota. */
259 if (vfs_dq_alloc_block(ip, nxlen)) {
260 dbFree(ip, nxaddr, (s64) nxlen);
261 mutex_unlock(&JFS_IP(ip)->commit_mutex);
262 return -EDQUOT;
265 delta = nxlen - xlen;
267 /* check if the extend page is not abnr but the request is abnr
268 * and the allocated disk space is for more than one page. if this
269 * is the case, there is a miss match of abnr between the extend page
270 * and the one or more pages following the extend page. as a result,
271 * two extents will have to be manipulated. the first will be that
272 * of the extent of the extend page and will be manipulated thru
273 * an xtExtend() or an xtTailgate(), depending upon whether the
274 * disk allocation occurred as an inplace extension. the second
275 * extent will be manipulated (created) through an xtInsert() and
276 * will be for the pages following the extend page.
278 if (abnr && (!(xp->flag & XAD_NOTRECORDED)) && (nxlen > nbperpage)) {
279 ntail = nbperpage;
280 nextend = ntail - xlen;
281 ninsert = nxlen - nbperpage;
283 xflag = XAD_NOTRECORDED;
284 } else {
285 ntail = nxlen;
286 nextend = delta;
287 ninsert = 0;
289 xflag = xp->flag;
292 /* if we were able to extend the disk allocation in place,
293 * extend the extent. otherwise, move the extent to a
294 * new disk location.
296 if (xaddr == nxaddr) {
297 /* extend the extent */
298 if ((rc = xtExtend(0, ip, xoff + xlen, (int) nextend, 0))) {
299 dbFree(ip, xaddr + xlen, delta);
300 vfs_dq_free_block(ip, nxlen);
301 goto exit;
303 } else {
305 * move the extent to a new location:
307 * xtTailgate() accounts for relocated tail extent;
309 if ((rc = xtTailgate(0, ip, xoff, (int) ntail, nxaddr, 0))) {
310 dbFree(ip, nxaddr, nxlen);
311 vfs_dq_free_block(ip, nxlen);
312 goto exit;
317 /* check if we need to also insert a new extent */
318 if (ninsert) {
319 /* perform the insert. if it fails, free the blocks
320 * to be inserted and make it appear that we only did
321 * the xtExtend() or xtTailgate() above.
323 xaddr = nxaddr + ntail;
324 if (xtInsert (0, ip, xflag, xoff + ntail, (int) ninsert,
325 &xaddr, 0)) {
326 dbFree(ip, xaddr, (s64) ninsert);
327 delta = nextend;
328 nxlen = ntail;
329 xflag = 0;
333 /* set the return results */
334 XADaddress(xp, nxaddr);
335 XADlength(xp, nxlen);
336 XADoffset(xp, xoff);
337 xp->flag = xflag;
339 mark_inode_dirty(ip);
340 exit:
341 mutex_unlock(&JFS_IP(ip)->commit_mutex);
342 return (rc);
344 #endif /* _NOTYET */
348 * NAME: extHint()
350 * FUNCTION: produce an extent allocation hint for a file offset.
352 * PARAMETERS:
353 * ip - the inode of the file.
354 * offset - file offset for which the hint is needed.
355 * xp - pointer to the xad that is to be filled in with
356 * the hint.
358 * RETURN VALUES:
359 * 0 - success
360 * -EIO - i/o error.
362 int extHint(struct inode *ip, s64 offset, xad_t * xp)
364 struct super_block *sb = ip->i_sb;
365 int nbperpage = JFS_SBI(sb)->nbperpage;
366 s64 prev;
367 int rc = 0;
368 s64 xaddr;
369 int xlen;
370 int xflag;
372 /* init the hint as "no hint provided" */
373 XADaddress(xp, 0);
375 /* determine the starting extent offset of the page previous
376 * to the page containing the offset.
378 prev = ((offset & ~POFFSET) >> JFS_SBI(sb)->l2bsize) - nbperpage;
380 /* if the offset is in the first page of the file, no hint provided.
382 if (prev < 0)
383 goto out;
385 rc = xtLookup(ip, prev, nbperpage, &xflag, &xaddr, &xlen, 0);
387 if ((rc == 0) && xlen) {
388 if (xlen != nbperpage) {
389 jfs_error(ip->i_sb, "extHint: corrupt xtree");
390 rc = -EIO;
392 XADaddress(xp, xaddr);
393 XADlength(xp, xlen);
394 XADoffset(xp, prev);
396 * only preserve the abnr flag within the xad flags
397 * of the returned hint.
399 xp->flag = xflag & XAD_NOTRECORDED;
400 } else
401 rc = 0;
403 out:
404 return (rc);
409 * NAME: extRecord()
411 * FUNCTION: change a page with a file from not recorded to recorded.
413 * PARAMETERS:
414 * ip - inode of the file.
415 * cp - cbuf of the file page.
417 * RETURN VALUES:
418 * 0 - success
419 * -EIO - i/o error.
420 * -ENOSPC - insufficient disk resources.
422 int extRecord(struct inode *ip, xad_t * xp)
424 int rc;
426 txBeginAnon(ip->i_sb);
428 mutex_lock(&JFS_IP(ip)->commit_mutex);
430 /* update the extent */
431 rc = xtUpdate(0, ip, xp);
433 mutex_unlock(&JFS_IP(ip)->commit_mutex);
434 return rc;
438 #ifdef _NOTYET
440 * NAME: extFill()
442 * FUNCTION: allocate disk space for a file page that represents
443 * a file hole.
445 * PARAMETERS:
446 * ip - the inode of the file.
447 * cp - cbuf of the file page represent the hole.
449 * RETURN VALUES:
450 * 0 - success
451 * -EIO - i/o error.
452 * -ENOSPC - insufficient disk resources.
454 int extFill(struct inode *ip, xad_t * xp)
456 int rc, nbperpage = JFS_SBI(ip->i_sb)->nbperpage;
457 s64 blkno = offsetXAD(xp) >> ip->i_blkbits;
459 // assert(ISSPARSE(ip));
461 /* initialize the extent allocation hint */
462 XADaddress(xp, 0);
464 /* allocate an extent to fill the hole */
465 if ((rc = extAlloc(ip, nbperpage, blkno, xp, false)))
466 return (rc);
468 assert(lengthPXD(xp) == nbperpage);
470 return (0);
472 #endif /* _NOTYET */
476 * NAME: extBalloc()
478 * FUNCTION: allocate disk blocks to form an extent.
480 * initially, we will try to allocate disk blocks for the
481 * requested size (nblocks). if this fails (nblocks
482 * contiguous free blocks not avaliable), we'll try to allocate
483 * a smaller number of blocks (producing a smaller extent), with
484 * this smaller number of blocks consisting of the requested
485 * number of blocks rounded down to the next smaller power of 2
486 * number (i.e. 16 -> 8). we'll continue to round down and
487 * retry the allocation until the number of blocks to allocate
488 * is smaller than the number of blocks per page.
490 * PARAMETERS:
491 * ip - the inode of the file.
492 * hint - disk block number to be used as an allocation hint.
493 * *nblocks - pointer to an s64 value. on entry, this value specifies
494 * the desired number of block to be allocated. on successful
495 * exit, this value is set to the number of blocks actually
496 * allocated.
497 * blkno - pointer to a block address that is filled in on successful
498 * return with the starting block number of the newly
499 * allocated block range.
501 * RETURN VALUES:
502 * 0 - success
503 * -EIO - i/o error.
504 * -ENOSPC - insufficient disk resources.
506 static int
507 extBalloc(struct inode *ip, s64 hint, s64 * nblocks, s64 * blkno)
509 struct jfs_inode_info *ji = JFS_IP(ip);
510 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
511 s64 nb, nblks, daddr, max;
512 int rc, nbperpage = sbi->nbperpage;
513 struct bmap *bmp = sbi->bmap;
514 int ag;
516 /* get the number of blocks to initially attempt to allocate.
517 * we'll first try the number of blocks requested unless this
518 * number is greater than the maximum number of contiguous free
519 * blocks in the map. in that case, we'll start off with the
520 * maximum free.
522 max = (s64) 1 << bmp->db_maxfreebud;
523 if (*nblocks >= max && *nblocks > nbperpage)
524 nb = nblks = (max > nbperpage) ? max : nbperpage;
525 else
526 nb = nblks = *nblocks;
528 /* try to allocate blocks */
529 while ((rc = dbAlloc(ip, hint, nb, &daddr)) != 0) {
530 /* if something other than an out of space error,
531 * stop and return this error.
533 if (rc != -ENOSPC)
534 return (rc);
536 /* decrease the allocation request size */
537 nb = min(nblks, extRoundDown(nb));
539 /* give up if we cannot cover a page */
540 if (nb < nbperpage)
541 return (rc);
544 *nblocks = nb;
545 *blkno = daddr;
547 if (S_ISREG(ip->i_mode) && (ji->fileset == FILESYSTEM_I)) {
548 ag = BLKTOAG(daddr, sbi);
549 spin_lock_irq(&ji->ag_lock);
550 if (ji->active_ag == -1) {
551 atomic_inc(&bmp->db_active[ag]);
552 ji->active_ag = ag;
553 } else if (ji->active_ag != ag) {
554 atomic_dec(&bmp->db_active[ji->active_ag]);
555 atomic_inc(&bmp->db_active[ag]);
556 ji->active_ag = ag;
558 spin_unlock_irq(&ji->ag_lock);
561 return (0);
565 #ifdef _NOTYET
567 * NAME: extBrealloc()
569 * FUNCTION: attempt to extend an extent's allocation.
571 * Initially, we will try to extend the extent's allocation
572 * in place. If this fails, we'll try to move the extent
573 * to a new set of blocks. If moving the extent, we initially
574 * will try to allocate disk blocks for the requested size
575 * (newnblks). if this fails (new contiguous free blocks not
576 * avaliable), we'll try to allocate a smaller number of
577 * blocks (producing a smaller extent), with this smaller
578 * number of blocks consisting of the requested number of
579 * blocks rounded down to the next smaller power of 2
580 * number (i.e. 16 -> 8). We'll continue to round down and
581 * retry the allocation until the number of blocks to allocate
582 * is smaller than the number of blocks per page.
584 * PARAMETERS:
585 * ip - the inode of the file.
586 * blkno - starting block number of the extents current allocation.
587 * nblks - number of blocks within the extents current allocation.
588 * newnblks - pointer to a s64 value. on entry, this value is the
589 * the new desired extent size (number of blocks). on
590 * successful exit, this value is set to the extent's actual
591 * new size (new number of blocks).
592 * newblkno - the starting block number of the extents new allocation.
594 * RETURN VALUES:
595 * 0 - success
596 * -EIO - i/o error.
597 * -ENOSPC - insufficient disk resources.
599 static int
600 extBrealloc(struct inode *ip,
601 s64 blkno, s64 nblks, s64 * newnblks, s64 * newblkno)
603 int rc;
605 /* try to extend in place */
606 if ((rc = dbExtend(ip, blkno, nblks, *newnblks - nblks)) == 0) {
607 *newblkno = blkno;
608 return (0);
609 } else {
610 if (rc != -ENOSPC)
611 return (rc);
614 /* in place extension not possible.
615 * try to move the extent to a new set of blocks.
617 return (extBalloc(ip, blkno, newnblks, newblkno));
619 #endif /* _NOTYET */
623 * NAME: extRoundDown()
625 * FUNCTION: round down a specified number of blocks to the next
626 * smallest power of 2 number.
628 * PARAMETERS:
629 * nb - the inode of the file.
631 * RETURN VALUES:
632 * next smallest power of 2 number.
634 static s64 extRoundDown(s64 nb)
636 int i;
637 u64 m, k;
639 for (i = 0, m = (u64) 1 << 63; i < 64; i++, m >>= 1) {
640 if (m & nb)
641 break;
644 i = 63 - i;
645 k = (u64) 1 << i;
646 k = ((k - 1) & nb) ? k : k >> 1;
648 return (k);