2 * Copyright (c) International Business Machines Corp., 2000-2002
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
20 #include "jfs_incore.h"
22 #include "jfs_extent.h"
23 #include "jfs_debug.h"
28 static int extBalloc(struct inode
*, s64
, s64
*, s64
*);
30 static int extBrealloc(struct inode
*, s64
, s64
, s64
*, s64
*);
32 static s64
extRoundDown(s64 nb
);
37 extern int dbExtend(struct inode
*, s64
, s64
, s64
);
38 extern int jfs_commit_inode(struct inode
*, int);
41 #define DPD(a) (printk("(a): %d\n",(a)))
42 #define DPC(a) (printk("(a): %c\n",(a)))
46 printk("(a): %x%08x ",(a)); \
48 printk("(a): %x ",(a) << 32); \
53 printk("(a): %x%08x\n",(a)); \
55 printk("(a): %x\n",(a) << 32); \
58 #define DPD1(a) (printk("(a): %d ",(a)))
59 #define DPX(a) (printk("(a): %08x\n",(a)))
60 #define DPX1(a) (printk("(a): %08x ",(a)))
61 #define DPS(a) (printk("%s\n",(a)))
62 #define DPE(a) (printk("\nENTERING: %s\n",(a)))
63 #define DPE1(a) (printk("\nENTERING: %s",(a)))
64 #define DPS1(a) (printk(" %s ",(a)))
70 * FUNCTION: allocate an extent for a specified page range within a
74 * ip - the inode of the file.
75 * xlen - requested extent length.
76 * pno - the starting page number with the file.
77 * xp - pointer to an xad. on entry, xad describes an
78 * extent that is used as an allocation hint if the
79 * xaddr of the xad is non-zero. on successful exit,
80 * the xad describes the newly allocated extent.
81 * abnr - boolean_t indicating whether the newly allocated extent
82 * should be marked as allocated but not recorded.
87 * -ENOSPC - insufficient disk resources.
90 extAlloc(struct inode
*ip
, s64 xlen
, s64 pno
, xad_t
* xp
, boolean_t abnr
)
92 struct jfs_sb_info
*sbi
= JFS_SBI(ip
->i_sb
);
93 s64 nxlen
, nxaddr
, xoff
, hint
, xaddr
= 0;
97 /* This blocks if we are low on resources */
98 txBeginAnon(ip
->i_sb
);
100 /* Avoid race with jfs_commit_inode() */
101 down(&JFS_IP(ip
)->commit_sem
);
103 /* validate extent length */
107 /* get the page's starting extent offset */
108 xoff
= pno
<< sbi
->l2nbperpage
;
110 /* check if an allocation hint was provided */
111 if ((hint
= addressXAD(xp
))) {
112 /* get the size of the extent described by the hint */
113 nxlen
= lengthXAD(xp
);
115 /* check if the hint is for the portion of the file
116 * immediately previous to the current allocation
117 * request and if hint extent has the same abnr
118 * value as the current request. if so, we can
119 * extend the hint extent to include the current
120 * extent if we can allocate the blocks immediately
121 * following the hint extent.
123 if (offsetXAD(xp
) + nxlen
== xoff
&&
124 abnr
== ((xp
->flag
& XAD_NOTRECORDED
) ? TRUE
: FALSE
))
125 xaddr
= hint
+ nxlen
;
127 /* adjust the hint to the last block of the extent */
131 /* allocate the disk blocks for the extent. initially, extBalloc()
132 * will try to allocate disk blocks for the requested size (xlen).
133 * if this fails (xlen contigious free blocks not avaliable), it'll
134 * try to allocate a smaller number of blocks (producing a smaller
135 * extent), with this smaller number of blocks consisting of the
136 * requested number of blocks rounded down to the next smaller
137 * power of 2 number (i.e. 16 -> 8). it'll continue to round down
138 * and retry the allocation until the number of blocks to allocate
139 * is smaller than the number of blocks per page.
142 if ((rc
= extBalloc(ip
, hint
? hint
: INOHINT(ip
), &nxlen
, &nxaddr
))) {
143 up(&JFS_IP(ip
)->commit_sem
);
147 /* determine the value of the extent flag */
148 xflag
= (abnr
== TRUE
) ? XAD_NOTRECORDED
: 0;
150 /* if we can extend the hint extent to cover the current request,
151 * extend it. otherwise, insert a new extent to
152 * cover the current request.
154 if (xaddr
&& xaddr
== nxaddr
)
155 rc
= xtExtend(0, ip
, xoff
, (int) nxlen
, 0);
157 rc
= xtInsert(0, ip
, xflag
, xoff
, (int) nxlen
, &nxaddr
, 0);
159 /* if the extend or insert failed,
160 * free the newly allocated blocks and return the error.
163 dbFree(ip
, nxaddr
, nxlen
);
164 up(&JFS_IP(ip
)->commit_sem
);
168 /* update the number of blocks allocated to the file */
169 ip
->i_blocks
+= LBLK2PBLK(ip
->i_sb
, nxlen
);
171 /* set the results of the extent allocation */
172 XADaddress(xp
, nxaddr
);
173 XADlength(xp
, nxlen
);
177 mark_inode_dirty(ip
);
179 up(&JFS_IP(ip
)->commit_sem
);
181 * COMMIT_SyncList flags an anonymous tlock on page that is on
183 * We need to commit the inode to get the page written disk.
185 if (test_and_clear_cflag(COMMIT_Synclist
,ip
))
186 jfs_commit_inode(ip
, 0);
196 * FUNCTION: extend the allocation of a file extent containing a
197 * partial back last page.
200 * ip - the inode of the file.
201 * cp - cbuf for the partial backed last page.
202 * xlen - request size of the resulting extent.
203 * xp - pointer to an xad. on successful exit, the xad
204 * describes the newly allocated extent.
205 * abnr - boolean_t indicating whether the newly allocated extent
206 * should be marked as allocated but not recorded.
211 * -ENOSPC - insufficient disk resources.
213 int extRealloc(struct inode
*ip
, s64 nxlen
, xad_t
* xp
, boolean_t abnr
)
215 struct super_block
*sb
= ip
->i_sb
;
216 s64 xaddr
, xlen
, nxaddr
, delta
, xoff
;
217 s64 ntail
, nextend
, ninsert
;
218 int rc
, nbperpage
= JFS_SBI(sb
)->nbperpage
;
221 /* This blocks if we are low on resources */
222 txBeginAnon(ip
->i_sb
);
224 down(&JFS_IP(ip
)->commit_sem
);
225 /* validate extent length */
229 /* get the extend (partial) page's disk block address and
232 xaddr
= addressXAD(xp
);
233 xlen
= lengthXAD(xp
);
234 xoff
= offsetXAD(xp
);
236 /* if the extend page is abnr and if the request is for
237 * the extent to be allocated and recorded,
238 * make the page allocated and recorded.
240 if ((xp
->flag
& XAD_NOTRECORDED
) && !abnr
) {
242 if ((rc
= xtUpdate(0, ip
, xp
)))
246 /* try to allocated the request number of blocks for the
247 * extent. dbRealloc() first tries to satisfy the request
248 * by extending the allocation in place. otherwise, it will
249 * try to allocate a new set of blocks large enough for the
250 * request. in satisfying a request, dbReAlloc() may allocate
251 * less than what was request but will always allocate enough
252 * space as to satisfy the extend page.
254 if ((rc
= extBrealloc(ip
, xaddr
, xlen
, &nxlen
, &nxaddr
)))
257 delta
= nxlen
- xlen
;
259 /* check if the extend page is not abnr but the request is abnr
260 * and the allocated disk space is for more than one page. if this
261 * is the case, there is a miss match of abnr between the extend page
262 * and the one or more pages following the extend page. as a result,
263 * two extents will have to be manipulated. the first will be that
264 * of the extent of the extend page and will be manipulated thru
265 * an xtExtend() or an xtTailgate(), depending upon whether the
266 * disk allocation occurred as an inplace extension. the second
267 * extent will be manipulated (created) through an xtInsert() and
268 * will be for the pages following the extend page.
270 if (abnr
&& (!(xp
->flag
& XAD_NOTRECORDED
)) && (nxlen
> nbperpage
)) {
272 nextend
= ntail
- xlen
;
273 ninsert
= nxlen
- nbperpage
;
275 xflag
= XAD_NOTRECORDED
;
284 /* if we were able to extend the disk allocation in place,
285 * extend the extent. otherwise, move the extent to a
288 if (xaddr
== nxaddr
) {
289 /* extend the extent */
290 if ((rc
= xtExtend(0, ip
, xoff
+ xlen
, (int) nextend
, 0))) {
291 dbFree(ip
, xaddr
+ xlen
, delta
);
296 * move the extent to a new location:
298 * xtTailgate() accounts for relocated tail extent;
300 if ((rc
= xtTailgate(0, ip
, xoff
, (int) ntail
, nxaddr
, 0))) {
301 dbFree(ip
, nxaddr
, nxlen
);
307 /* check if we need to also insert a new extent */
309 /* perform the insert. if it fails, free the blocks
310 * to be inserted and make it appear that we only did
311 * the xtExtend() or xtTailgate() above.
313 xaddr
= nxaddr
+ ntail
;
314 if (xtInsert (0, ip
, xflag
, xoff
+ ntail
, (int) ninsert
,
316 dbFree(ip
, xaddr
, (s64
) ninsert
);
323 /* update the inode with the number of blocks allocated */
324 ip
->i_blocks
+= LBLK2PBLK(sb
, delta
);
326 /* set the return results */
327 XADaddress(xp
, nxaddr
);
328 XADlength(xp
, nxlen
);
332 mark_inode_dirty(ip
);
334 up(&JFS_IP(ip
)->commit_sem
);
343 * FUNCTION: produce an extent allocation hint for a file offset.
346 * ip - the inode of the file.
347 * offset - file offset for which the hint is needed.
348 * xp - pointer to the xad that is to be filled in with
355 int extHint(struct inode
*ip
, s64 offset
, xad_t
* xp
)
357 struct super_block
*sb
= ip
->i_sb
;
362 int rc
, nbperpage
= JFS_SBI(sb
)->nbperpage
;
364 /* init the hint as "no hint provided" */
367 /* determine the starting extent offset of the page previous
368 * to the page containing the offset.
370 prev
= ((offset
& ~POFFSET
) >> JFS_SBI(sb
)->l2bsize
) - nbperpage
;
372 /* if the offsets in the first page of the file,
378 /* prepare to lookup the previous page's extent info */
382 LXDoffset(&lxd
, prev
)
383 LXDlength(&lxd
, nbperpage
);
389 /* perform the lookup */
390 if ((rc
= xtLookupList(ip
, &lxdl
, &xadl
, 0)))
393 /* check if not extent exists for the previous page.
394 * this is possible for sparse files.
396 if (xadl
.nxad
== 0) {
397 // assert(ISSPARSE(ip));
401 /* only preserve the abnr flag within the xad flags
402 * of the returned hint.
404 xp
->flag
&= XAD_NOTRECORDED
;
406 assert(xadl
.nxad
== 1);
407 assert(lengthXAD(xp
) == nbperpage
);
416 * FUNCTION: change a page with a file from not recorded to recorded.
419 * ip - inode of the file.
420 * cp - cbuf of the file page.
425 * -ENOSPC - insufficient disk resources.
427 int extRecord(struct inode
*ip
, xad_t
* xp
)
431 txBeginAnon(ip
->i_sb
);
433 down(&JFS_IP(ip
)->commit_sem
);
435 /* update the extent */
436 rc
= xtUpdate(0, ip
, xp
);
438 up(&JFS_IP(ip
)->commit_sem
);
447 * FUNCTION: allocate disk space for a file page that represents
451 * ip - the inode of the file.
452 * cp - cbuf of the file page represent the hole.
457 * -ENOSPC - insufficient disk resources.
459 int extFill(struct inode
*ip
, xad_t
* xp
)
461 int rc
, nbperpage
= JFS_SBI(ip
->i_sb
)->nbperpage
;
462 s64 blkno
= offsetXAD(xp
) >> ip
->i_blksize
;
464 // assert(ISSPARSE(ip));
466 /* initialize the extent allocation hint */
469 /* allocate an extent to fill the hole */
470 if ((rc
= extAlloc(ip
, nbperpage
, blkno
, xp
, FALSE
)))
473 assert(lengthPXD(xp
) == nbperpage
);
483 * FUNCTION: allocate disk blocks to form an extent.
485 * initially, we will try to allocate disk blocks for the
486 * requested size (nblocks). if this fails (nblocks
487 * contigious free blocks not avaliable), we'll try to allocate
488 * a smaller number of blocks (producing a smaller extent), with
489 * this smaller number of blocks consisting of the requested
490 * number of blocks rounded down to the next smaller power of 2
491 * number (i.e. 16 -> 8). we'll continue to round down and
492 * retry the allocation until the number of blocks to allocate
493 * is smaller than the number of blocks per page.
496 * ip - the inode of the file.
497 * hint - disk block number to be used as an allocation hint.
498 * *nblocks - pointer to an s64 value. on entry, this value specifies
499 * the desired number of block to be allocated. on successful
500 * exit, this value is set to the number of blocks actually
502 * blkno - pointer to a block address that is filled in on successful
503 * return with the starting block number of the newly
504 * allocated block range.
509 * -ENOSPC - insufficient disk resources.
512 extBalloc(struct inode
*ip
, s64 hint
, s64
* nblocks
, s64
* blkno
)
514 struct jfs_inode_info
*ji
= JFS_IP(ip
);
515 struct jfs_sb_info
*sbi
= JFS_SBI(ip
->i_sb
);
516 s64 nb
, nblks
, daddr
, max
;
517 int rc
, nbperpage
= sbi
->nbperpage
;
518 struct bmap
*bmp
= sbi
->bmap
;
521 /* get the number of blocks to initially attempt to allocate.
522 * we'll first try the number of blocks requested unless this
523 * number is greater than the maximum number of contigious free
524 * blocks in the map. in that case, we'll start off with the
527 max
= (s64
) 1 << bmp
->db_maxfreebud
;
528 if (*nblocks
>= max
&& *nblocks
> nbperpage
)
529 nb
= nblks
= (max
> nbperpage
) ? max
: nbperpage
;
531 nb
= nblks
= *nblocks
;
533 /* try to allocate blocks */
534 while ((rc
= dbAlloc(ip
, hint
, nb
, &daddr
))) {
535 /* if something other than an out of space error,
536 * stop and return this error.
541 /* decrease the allocation request size */
542 nb
= min(nblks
, extRoundDown(nb
));
544 /* give up if we cannot cover a page */
552 if (S_ISREG(ip
->i_mode
) && (ji
->fileset
== FILESYSTEM_I
)) {
553 ag
= BLKTOAG(daddr
, sbi
);
554 if (ji
->active_ag
== -1) {
555 atomic_inc(&bmp
->db_active
[ag
]);
557 } else if (ji
->active_ag
!= ag
) {
558 atomic_dec(&bmp
->db_active
[ji
->active_ag
]);
559 atomic_inc(&bmp
->db_active
[ag
]);
570 * NAME: extBrealloc()
572 * FUNCTION: attempt to extend an extent's allocation.
574 * initially, we will try to extend the extent's allocation
575 * in place. if this fails, we'll try to move the extent
576 * to a new set of blocks. if moving the extent, we initially
577 * will try to allocate disk blocks for the requested size
578 * (nnew). if this fails (nnew contigious free blocks not
579 * avaliable), we'll try to allocate a smaller number of
580 * blocks (producing a smaller extent), with this smaller
581 * number of blocks consisting of the requested number of
582 * blocks rounded down to the next smaller power of 2
583 * number (i.e. 16 -> 8). we'll continue to round down and
584 * retry the allocation until the number of blocks to allocate
585 * is smaller than the number of blocks per page.
588 * ip - the inode of the file.
589 * blkno - starting block number of the extents current allocation.
590 * nblks - number of blocks within the extents current allocation.
591 * newnblks - pointer to a s64 value. on entry, this value is the
592 * the new desired extent size (number of blocks). on
593 * successful exit, this value is set to the extent's actual
594 * new size (new number of blocks).
595 * newblkno - the starting block number of the extents new allocation.
600 * -ENOSPC - insufficient disk resources.
603 extBrealloc(struct inode
*ip
,
604 s64 blkno
, s64 nblks
, s64
* newnblks
, s64
* newblkno
)
608 /* try to extend in place */
609 if ((rc
= dbExtend(ip
, blkno
, nblks
, *newnblks
- nblks
)) == 0) {
617 /* in place extension not possible.
618 * try to move the extent to a new set of blocks.
620 return (extBalloc(ip
, blkno
, newnblks
, newblkno
));
626 * NAME: extRoundDown()
628 * FUNCTION: round down a specified number of blocks to the next
629 * smallest power of 2 number.
632 * nb - the inode of the file.
635 * next smallest power of 2 number.
637 static s64
extRoundDown(s64 nb
)
642 for (i
= 0, m
= (u64
) 1 << 63; i
< 64; i
++, m
>>= 1) {
649 k
= ((k
- 1) & nb
) ? k
: k
>> 1;