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
20 #include <linux/quotaops.h>
21 #include "jfs_incore.h"
22 #include "jfs_inode.h"
23 #include "jfs_superblock.h"
25 #include "jfs_extent.h"
26 #include "jfs_debug.h"
31 static int extBalloc(struct inode *, s64, s64 *, s64 *);
33 static int extBrealloc(struct inode *, s64, s64, s64 *, s64 *);
35 static s64 extRoundDown(s64 nb);
37 #define DPD(a) (printk("(a): %d\n",(a)))
38 #define DPC(a) (printk("(a): %c\n",(a)))
42 printk("(a): %x%08x ",(a)); \
44 printk("(a): %x ",(a) << 32); \
49 printk("(a): %x%08x\n",(a)); \
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)))
66 * FUNCTION: allocate an extent for a specified page range within a
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.
83 * -ENOSPC - insufficient disk resources.
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;
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 */
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 */
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.
138 if ((rc = extBalloc(ip, hint ? hint : INOHINT(ip), &nxlen, &nxaddr))) {
139 mutex_unlock(&JFS_IP(ip)->commit_mutex);
143 /* Allocate blocks to quota. */
144 if (DQUOT_ALLOC_BLOCK(ip, nxlen)) {
145 dbFree(ip, nxaddr, (s64) nxlen);
146 mutex_unlock(&JFS_IP(ip)->commit_mutex);
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);
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.
166 dbFree(ip, nxaddr, nxlen);
167 DQUOT_FREE_BLOCK(ip, nxlen);
168 mutex_unlock(&JFS_IP(ip)->commit_mutex);
172 /* set the results of the extent allocation */
173 XADaddress(xp, nxaddr);
174 XADlength(xp, nxlen);
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
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);
197 * FUNCTION: extend the allocation of a file extent containing a
198 * partial back last page.
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.
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;
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 */
230 /* get the extend (partial) page's disk block address and
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) {
243 if ((rc = xtUpdate(0, ip, xp)))
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)))
258 /* Allocat blocks to quota. */
259 if (DQUOT_ALLOC_BLOCK(ip, nxlen)) {
260 dbFree(ip, nxaddr, (s64) nxlen);
261 mutex_unlock(&JFS_IP(ip)->commit_mutex);
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)) {
280 nextend = ntail - xlen;
281 ninsert = nxlen - nbperpage;
283 xflag = XAD_NOTRECORDED;
292 /* if we were able to extend the disk allocation in place,
293 * extend the extent. otherwise, move the extent to a
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 DQUOT_FREE_BLOCK(ip, nxlen);
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 DQUOT_FREE_BLOCK(ip, nxlen);
317 /* check if we need to also insert a new extent */
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,
326 dbFree(ip, xaddr, (s64) ninsert);
333 /* set the return results */
334 XADaddress(xp, nxaddr);
335 XADlength(xp, nxlen);
339 mark_inode_dirty(ip);
341 mutex_unlock(&JFS_IP(ip)->commit_mutex);
350 * FUNCTION: produce an extent allocation hint for a file offset.
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
362 int extHint(struct inode *ip, s64 offset, xad_t * xp)
364 struct super_block *sb = ip->i_sb;
369 int rc, nbperpage = JFS_SBI(sb)->nbperpage;
371 /* init the hint as "no hint provided" */
374 /* determine the starting extent offset of the page previous
375 * to the page containing the offset.
377 prev = ((offset & ~POFFSET) >> JFS_SBI(sb)->l2bsize) - nbperpage;
379 /* if the offsets in the first page of the file,
385 /* prepare to lookup the previous page's extent info */
389 LXDoffset(&lxd, prev)
390 LXDlength(&lxd, nbperpage);
396 /* perform the lookup */
397 if ((rc = xtLookupList(ip, &lxdl, &xadl, 0)))
400 /* check if no extent exists for the previous page.
401 * this is possible for sparse files.
403 if (xadl.nxad == 0) {
404 // assert(ISSPARSE(ip));
408 /* only preserve the abnr flag within the xad flags
409 * of the returned hint.
411 xp->flag &= XAD_NOTRECORDED;
413 if(xadl.nxad != 1 || lengthXAD(xp) != nbperpage) {
414 jfs_error(ip->i_sb, "extHint: corrupt xtree");
425 * FUNCTION: change a page with a file from not recorded to recorded.
428 * ip - inode of the file.
429 * cp - cbuf of the file page.
434 * -ENOSPC - insufficient disk resources.
436 int extRecord(struct inode *ip, xad_t * xp)
440 txBeginAnon(ip->i_sb);
442 mutex_lock(&JFS_IP(ip)->commit_mutex);
444 /* update the extent */
445 rc = xtUpdate(0, ip, xp);
447 mutex_unlock(&JFS_IP(ip)->commit_mutex);
456 * FUNCTION: allocate disk space for a file page that represents
460 * ip - the inode of the file.
461 * cp - cbuf of the file page represent the hole.
466 * -ENOSPC - insufficient disk resources.
468 int extFill(struct inode *ip, xad_t * xp)
470 int rc, nbperpage = JFS_SBI(ip->i_sb)->nbperpage;
471 s64 blkno = offsetXAD(xp) >> ip->i_blkbits;
473 // assert(ISSPARSE(ip));
475 /* initialize the extent allocation hint */
478 /* allocate an extent to fill the hole */
479 if ((rc = extAlloc(ip, nbperpage, blkno, xp, false)))
482 assert(lengthPXD(xp) == nbperpage);
492 * FUNCTION: allocate disk blocks to form an extent.
494 * initially, we will try to allocate disk blocks for the
495 * requested size (nblocks). if this fails (nblocks
496 * contiguous free blocks not avaliable), we'll try to allocate
497 * a smaller number of blocks (producing a smaller extent), with
498 * this smaller number of blocks consisting of the requested
499 * number of blocks rounded down to the next smaller power of 2
500 * number (i.e. 16 -> 8). we'll continue to round down and
501 * retry the allocation until the number of blocks to allocate
502 * is smaller than the number of blocks per page.
505 * ip - the inode of the file.
506 * hint - disk block number to be used as an allocation hint.
507 * *nblocks - pointer to an s64 value. on entry, this value specifies
508 * the desired number of block to be allocated. on successful
509 * exit, this value is set to the number of blocks actually
511 * blkno - pointer to a block address that is filled in on successful
512 * return with the starting block number of the newly
513 * allocated block range.
518 * -ENOSPC - insufficient disk resources.
521 extBalloc(struct inode *ip, s64 hint, s64 * nblocks, s64 * blkno)
523 struct jfs_inode_info *ji = JFS_IP(ip);
524 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
525 s64 nb, nblks, daddr, max;
526 int rc, nbperpage = sbi->nbperpage;
527 struct bmap *bmp = sbi->bmap;
530 /* get the number of blocks to initially attempt to allocate.
531 * we'll first try the number of blocks requested unless this
532 * number is greater than the maximum number of contiguous free
533 * blocks in the map. in that case, we'll start off with the
536 max = (s64) 1 << bmp->db_maxfreebud;
537 if (*nblocks >= max && *nblocks > nbperpage)
538 nb = nblks = (max > nbperpage) ? max : nbperpage;
540 nb = nblks = *nblocks;
542 /* try to allocate blocks */
543 while ((rc = dbAlloc(ip, hint, nb, &daddr)) != 0) {
544 /* if something other than an out of space error,
545 * stop and return this error.
550 /* decrease the allocation request size */
551 nb = min(nblks, extRoundDown(nb));
553 /* give up if we cannot cover a page */
561 if (S_ISREG(ip->i_mode) && (ji->fileset == FILESYSTEM_I)) {
562 ag = BLKTOAG(daddr, sbi);
563 spin_lock_irq(&ji->ag_lock);
564 if (ji->active_ag == -1) {
565 atomic_inc(&bmp->db_active[ag]);
567 } else if (ji->active_ag != ag) {
568 atomic_dec(&bmp->db_active[ji->active_ag]);
569 atomic_inc(&bmp->db_active[ag]);
572 spin_unlock_irq(&ji->ag_lock);
581 * NAME: extBrealloc()
583 * FUNCTION: attempt to extend an extent's allocation.
585 * Initially, we will try to extend the extent's allocation
586 * in place. If this fails, we'll try to move the extent
587 * to a new set of blocks. If moving the extent, we initially
588 * will try to allocate disk blocks for the requested size
589 * (newnblks). if this fails (new contiguous free blocks not
590 * avaliable), we'll try to allocate a smaller number of
591 * blocks (producing a smaller extent), with this smaller
592 * number of blocks consisting of the requested number of
593 * blocks rounded down to the next smaller power of 2
594 * number (i.e. 16 -> 8). We'll continue to round down and
595 * retry the allocation until the number of blocks to allocate
596 * is smaller than the number of blocks per page.
599 * ip - the inode of the file.
600 * blkno - starting block number of the extents current allocation.
601 * nblks - number of blocks within the extents current allocation.
602 * newnblks - pointer to a s64 value. on entry, this value is the
603 * the new desired extent size (number of blocks). on
604 * successful exit, this value is set to the extent's actual
605 * new size (new number of blocks).
606 * newblkno - the starting block number of the extents new allocation.
611 * -ENOSPC - insufficient disk resources.
614 extBrealloc(struct inode *ip,
615 s64 blkno, s64 nblks, s64 * newnblks, s64 * newblkno)
619 /* try to extend in place */
620 if ((rc = dbExtend(ip, blkno, nblks, *newnblks - nblks)) == 0) {
628 /* in place extension not possible.
629 * try to move the extent to a new set of blocks.
631 return (extBalloc(ip, blkno, newnblks, newblkno));
637 * NAME: extRoundDown()
639 * FUNCTION: round down a specified number of blocks to the next
640 * smallest power of 2 number.
643 * nb - the inode of the file.
646 * next smallest power of 2 number.
648 static s64 extRoundDown(s64 nb)
653 for (i = 0, m = (u64) 1 << 63; i < 64; i++, m >>= 1) {
660 k = ((k - 1) & nb) ? k : k >> 1;