2 * Copyright (c) International Business Machines Corp., 2000-2003
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/buffer_head.h>
21 #include "jfs_incore.h"
22 #include "jfs_filsys.h"
23 #include "jfs_metapage.h"
24 #include "jfs_dinode.h"
27 #include "jfs_superblock.h"
28 #include "jfs_txnmgr.h"
29 #include "jfs_debug.h"
31 #define BITSPERPAGE (PSIZE << 3)
33 #define MEGABYTE (1 << L2MEGABYTE)
34 #define MEGABYTE32 (MEGABYTE << 5)
36 /* convert block number to bmap file page number */
37 #define BLKTODMAPN(b)\
38 (((b) >> 13) + ((b) >> 23) + ((b) >> 33) + 3 + 1)
43 * function: extend file system;
45 * |-------------------------------|----------|----------|
46 * file system space fsck inline log
50 * new LVSize: in LV blocks (required)
51 * new LogSize: in LV blocks (optional)
52 * new FSSize: in LV blocks (optional)
55 * 1. set new LogSize as specified or default from new LVSize;
56 * 2. compute new FSCKSize from new LVSize;
57 * 3. set new FSSize as MIN(FSSize, LVSize-(LogSize+FSCKSize)) where
58 * assert(new FSSize >= old FSSize),
59 * i.e., file system must not be shrinked;
61 int jfs_extendfs(struct super_block
*sb
, s64 newLVSize
, int newLogSize
)
64 struct jfs_sb_info
*sbi
= JFS_SBI(sb
);
65 struct inode
*ipbmap
= sbi
->ipbmap
;
66 struct inode
*ipbmap2
;
67 struct inode
*ipimap
= sbi
->ipimap
;
68 struct jfs_log
*log
= sbi
->log
;
69 struct bmap
*bmp
= sbi
->bmap
;
70 s64 newLogAddress
, newFSCKAddress
;
72 s64 newMapSize
= 0, mapSize
;
73 s64 XAddress
, XSize
, nblocks
, xoff
, xaddr
, t64
;
77 int newNpages
= 0, nPages
, newPage
, xlen
, t32
;
79 int log_formatted
= 0;
80 struct inode
*iplist
[1];
81 struct jfs_superblock
*j_sb
, *j_sb2
;
83 struct buffer_head
*bh
, *bh2
;
85 /* If the volume hasn't grown, get out now */
87 if (sbi
->mntflag
& JFS_INLINELOG
)
88 oldLVSize
= addressPXD(&sbi
->logpxd
) + lengthPXD(&sbi
->logpxd
);
90 oldLVSize
= addressPXD(&sbi
->fsckpxd
) +
91 lengthPXD(&sbi
->fsckpxd
);
93 if (oldLVSize
>= newLVSize
) {
95 "jfs_extendfs: volume hasn't grown, returning\n");
99 VolumeSize
= sb
->s_bdev
->bd_inode
->i_size
>> sb
->s_blocksize_bits
;
102 if (newLVSize
> VolumeSize
) {
103 printk(KERN_WARNING
"jfs_extendfs: invalid size\n");
108 /* check the device */
109 bh
= sb_bread(sb
, newLVSize
- 1);
111 printk(KERN_WARNING
"jfs_extendfs: invalid size\n");
118 /* Can't extend write-protected drive */
120 if (isReadOnly(ipbmap
)) {
121 printk(KERN_WARNING
"jfs_extendfs: read-only file system\n");
127 * reconfigure LV spaces
128 * ---------------------
130 * validate new size, or, if not specified, determine new size
134 * reconfigure inline log space:
136 if ((sbi
->mntflag
& JFS_INLINELOG
)) {
137 if (newLogSize
== 0) {
139 * no size specified: default to 1/256 of aggregate
140 * size; rounded up to a megabyte boundary;
142 newLogSize
= newLVSize
>> 8;
143 t32
= (1 << (20 - sbi
->l2bsize
)) - 1;
144 newLogSize
= (newLogSize
+ t32
) & ~t32
;
146 min(newLogSize
, MEGABYTE32
>> sbi
->l2bsize
);
149 * convert the newLogSize to fs blocks.
151 * Since this is given in megabytes, it will always be
152 * an even number of pages.
154 newLogSize
= (newLogSize
* MEGABYTE
) >> sbi
->l2bsize
;
160 newLogAddress
= newLVSize
- newLogSize
;
163 * reconfigure fsck work space:
165 * configure it to the end of the logical volume regardless of
166 * whether file system extends to the end of the aggregate;
167 * Need enough 4k pages to cover:
168 * - 1 bit per block in aggregate rounded up to BPERDMAP boundary
169 * - 1 extra page to handle control page and intermediate level pages
170 * - 50 extra pages for the chkdsk service log
172 t64
= ((newLVSize
- newLogSize
+ BPERDMAP
- 1) >> L2BPERDMAP
)
174 t32
= ((t64
+ (BITSPERPAGE
- 1)) / BITSPERPAGE
) + 1 + 50;
175 newFSCKSize
= t32
<< sbi
->l2nbperpage
;
176 newFSCKAddress
= newLogAddress
- newFSCKSize
;
179 * compute new file system space;
181 newFSSize
= newLVSize
- newLogSize
- newFSCKSize
;
183 /* file system cannot be shrinked */
184 if (newFSSize
< bmp
->db_mapsize
) {
190 * If we're expanding enough that the inline log does not overlap
191 * the old one, we can format the new log before we quiesce the
194 if ((sbi
->mntflag
& JFS_INLINELOG
) && (newLogAddress
> oldLVSize
)) {
195 if ((rc
= lmLogFormat(log
, newLogAddress
, newLogSize
)))
200 * quiesce file system
202 * (prepare to move the inline log and to prevent map update)
204 * block any new transactions and wait for completion of
205 * all wip transactions and flush modified pages s.t.
206 * on-disk file system is in consistent state and
207 * log is not required for recovery.
211 if (sbi
->mntflag
& JFS_INLINELOG
) {
213 * deactivate old inline log
218 * mark on-disk super block for fs in transition;
220 * update on-disk superblock for the new space configuration
221 * of inline log space and fsck work space descriptors:
222 * N.B. FS descriptor is NOT updated;
225 * logredo(): if FM_EXTENDFS, return to fsck() for cleanup;
226 * fsck(): if FM_EXTENDFS, reformat inline log and fsck
227 * workspace from superblock inline log descriptor and fsck
228 * workspace descriptor;
231 /* read in superblock */
232 if ((rc
= readSuper(sb
, &bh
)))
234 j_sb
= (struct jfs_superblock
*)bh
->b_data
;
236 /* mark extendfs() in progress */
237 j_sb
->s_state
|= cpu_to_le32(FM_EXTENDFS
);
238 j_sb
->s_xsize
= cpu_to_le64(newFSSize
);
239 PXDaddress(&j_sb
->s_xfsckpxd
, newFSCKAddress
);
240 PXDlength(&j_sb
->s_xfsckpxd
, newFSCKSize
);
241 PXDaddress(&j_sb
->s_xlogpxd
, newLogAddress
);
242 PXDlength(&j_sb
->s_xlogpxd
, newLogSize
);
244 /* synchronously update superblock */
245 mark_buffer_dirty(bh
);
246 sync_dirty_buffer(bh
);
250 * format new inline log synchronously;
252 * crash recovery: if log move in progress,
253 * reformat log and exit success;
256 if ((rc
= lmLogFormat(log
, newLogAddress
, newLogSize
)))
262 log
->base
= newLogAddress
;
263 log
->size
= newLogSize
>> (L2LOGPSIZE
- sb
->s_blocksize_bits
);
264 if ((rc
= lmLogInit(log
)))
269 * extend block allocation map
270 * ---------------------------
272 * extendfs() for new extension, retry after crash recovery;
274 * note: both logredo() and fsck() rebuild map from
275 * the bitmap and configuration parameter from superblock
276 * (disregarding all other control information in the map);
279 * s_size: aggregate size in physical blocks;
282 * compute the new block allocation map configuration
285 * di_size: map file size in byte;
286 * di_nblocks: number of blocks allocated for map file;
287 * di_mapsize: number of blocks in aggregate (covered by map);
289 * db_mapsize: number of blocks in aggregate (covered by map);
291 newMapSize
= newFSSize
;
292 /* number of data pages of new bmap file:
293 * roundup new size to full dmap page boundary and
294 * add 1 extra dmap page for next extendfs()
296 t64
= (newMapSize
- 1) + BPERDMAP
;
297 newNpages
= BLKTODMAPN(t64
) + 1;
300 * extend map from current map (WITHOUT growing mapfile)
302 * map new extension with unmapped part of the last partial
303 * dmap page, if applicable, and extra page(s) allocated
304 * at end of bmap by mkfs() or previous extendfs();
307 /* compute number of blocks requested to extend */
308 mapSize
= bmp
->db_mapsize
;
309 XAddress
= mapSize
; /* eXtension Address */
310 XSize
= newMapSize
- mapSize
; /* eXtension Size */
311 old_agsize
= bmp
->db_agsize
; /* We need to know if this changes */
313 /* compute number of blocks that can be extended by current mapfile */
314 t64
= dbMapFileSizeToMapSize(ipbmap
);
316 printk(KERN_ERR
"jfs_extendfs: mapSize (0x%Lx) > t64 (0x%Lx)\n",
317 (long long) mapSize
, (long long) t64
);
321 nblocks
= min(t64
- mapSize
, XSize
);
324 * update map pages for new extension:
326 * update/init dmap and bubble up the control hierarchy
327 * incrementally fold up dmaps into upper levels;
328 * update bmap control page;
330 if ((rc
= dbExtendFS(ipbmap
, XAddress
, nblocks
)))
333 * the map now has extended to cover additional nblocks:
334 * dn_mapsize = oldMapsize + nblocks;
336 /* ipbmap->i_mapsize += nblocks; */
340 * grow map file to cover remaining extension
341 * and/or one extra dmap page for next extendfs();
343 * allocate new map pages and its backing blocks, and
344 * update map file xtree
346 /* compute number of data pages of current bmap file */
347 nPages
= ipbmap
->i_size
>> L2PSIZE
;
349 /* need to grow map file ? */
350 if (nPages
== newNpages
)
354 * grow bmap file for the new map pages required:
356 * allocate growth at the start of newly extended region;
357 * bmap file only grows sequentially, i.e., both data pages
358 * and possibly xtree index pages may grow in append mode,
359 * s.t. logredo() can reconstruct pre-extension state
360 * by washing away bmap file of pages outside s_size boundary;
363 * journal map file growth as if a regular file growth:
364 * (note: bmap is created with di_mode = IFJOURNAL|IFREG);
366 * journaling of bmap file growth is not required since
367 * logredo() do/can not use log records of bmap file growth
368 * but it provides careful write semantics, pmap update, etc.;
370 /* synchronous write of data pages: bmap data pages are
371 * cached in meta-data cache, and not written out
374 filemap_fdatawait(ipbmap
->i_mapping
);
375 filemap_fdatawrite(ipbmap
->i_mapping
);
376 filemap_fdatawait(ipbmap
->i_mapping
);
377 diWriteSpecial(ipbmap
, 0);
379 newPage
= nPages
; /* first new page number */
380 xoff
= newPage
<< sbi
->l2nbperpage
;
381 xlen
= (newNpages
- nPages
) << sbi
->l2nbperpage
;
382 xlen
= min(xlen
, (int) nblocks
) & ~(sbi
->nbperpage
- 1);
385 tid
= txBegin(sb
, COMMIT_FORCE
);
387 if ((rc
= xtAppend(tid
, ipbmap
, 0, xoff
, nblocks
, &xlen
, &xaddr
, 0))) {
391 /* update bmap file size */
392 ipbmap
->i_size
+= xlen
<< sbi
->l2bsize
;
393 ipbmap
->i_blocks
+= LBLK2PBLK(sb
, xlen
);
396 rc
= txCommit(tid
, 1, &iplist
[0], COMMIT_FORCE
);
404 * map file has been grown now to cover extension to further out;
405 * di_size = new map file size;
407 * if huge extension, the previous extension based on previous
408 * map file size may not have been sufficient to cover whole extension
409 * (it could have been used up for new map pages),
410 * but the newly grown map file now covers lot bigger new free space
411 * available for further extension of map;
413 /* any more blocks to extend ? */
419 dbFinalizeBmap(ipbmap
);
422 * update inode allocation map
423 * ---------------------------
425 * move iag lists from old to new iag;
426 * agstart field is not updated for logredo() to reconstruct
427 * iag lists if system crash occurs.
428 * (computation of ag number from agstart based on agsize
429 * will correctly identify the new ag);
431 /* if new AG size the same as old AG size, done! */
432 if (bmp
->db_agsize
!= old_agsize
) {
433 if ((rc
= diExtendFS(ipimap
, ipbmap
)))
437 if ((rc
= diSync(ipimap
)))
445 * extension is committed when on-disk super block is
446 * updated with new descriptors: logredo will recover
447 * crash before it to pre-extension state;
450 /* sync log to skip log replay of bmap file growth transaction; */
451 /* lmLogSync(log, 1); */
454 * synchronous write bmap global control page;
455 * for crash before completion of write
456 * logredo() will recover to pre-extendfs state;
457 * for crash after completion of write,
458 * logredo() will recover post-extendfs state;
460 if ((rc
= dbSync(ipbmap
)))
464 * copy primary bmap inode to secondary bmap inode
467 ipbmap2
= diReadSpecial(sb
, BMAP_I
, 1);
468 if (ipbmap2
== NULL
) {
469 printk(KERN_ERR
"jfs_extendfs: diReadSpecial(bmap) failed\n");
472 memcpy(&JFS_IP(ipbmap2
)->i_xtroot
, &JFS_IP(ipbmap
)->i_xtroot
, 288);
473 ipbmap2
->i_size
= ipbmap
->i_size
;
474 ipbmap2
->i_blocks
= ipbmap
->i_blocks
;
476 diWriteSpecial(ipbmap2
, 1);
477 diFreeSpecial(ipbmap2
);
482 if ((rc
= readSuper(sb
, &bh
)))
484 j_sb
= (struct jfs_superblock
*)bh
->b_data
;
486 /* mark extendfs() completion */
487 j_sb
->s_state
&= cpu_to_le32(~FM_EXTENDFS
);
488 j_sb
->s_size
= cpu_to_le64(bmp
->db_mapsize
<<
489 le16_to_cpu(j_sb
->s_l2bfactor
));
490 j_sb
->s_agsize
= cpu_to_le32(bmp
->db_agsize
);
492 /* update inline log space descriptor */
493 if (sbi
->mntflag
& JFS_INLINELOG
) {
494 PXDaddress(&(j_sb
->s_logpxd
), newLogAddress
);
495 PXDlength(&(j_sb
->s_logpxd
), newLogSize
);
498 /* record log's mount serial number */
499 j_sb
->s_logserial
= cpu_to_le32(log
->serial
);
501 /* update fsck work space descriptor */
502 PXDaddress(&(j_sb
->s_fsckpxd
), newFSCKAddress
);
503 PXDlength(&(j_sb
->s_fsckpxd
), newFSCKSize
);
505 /* sb->s_fsckloglen remains the same */
507 /* Update secondary superblock */
508 bh2
= sb_bread(sb
, SUPER2_OFF
>> sb
->s_blocksize_bits
);
510 j_sb2
= (struct jfs_superblock
*)bh2
->b_data
;
511 memcpy(j_sb2
, j_sb
, sizeof (struct jfs_superblock
));
513 mark_buffer_dirty(bh
);
514 sync_dirty_buffer(bh2
);
518 /* write primary superblock */
519 mark_buffer_dirty(bh
);
520 sync_dirty_buffer(bh
);
526 updateSuper(sb
, FM_DIRTY
);
530 * resume file system transactions