1 /*
2 * Copyright (C) International Business Machines Corp., 2000-2003
3 *
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.
8 *
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.
13 *
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
17 */
18
19 #include <linux/fs.h>
20 #include <linux/locks.h>
21
22 #include "jfs_incore.h"
23 #include "jfs_filsys.h"
24 #include "jfs_metapage.h"
25 #include "jfs_dinode.h"
26 #include "jfs_imap.h"
27 #include "jfs_dmap.h"
28 #include "jfs_superblock.h"
29 #include "jfs_txnmgr.h"
30 #include "jfs_debug.h"
31
32 extern s64 jfs_get_volume_size(struct super_block *);
33
34 #define BITSPERPAGE (PSIZE << 3)
35 #define L2MEGABYTE 20
36 #define MEGABYTE (1 << L2MEGABYTE)
37 #define MEGABYTE32 (MEGABYTE << 5)
38
39 /* convert block number to bmap file page number */
40 #define BLKTODMAPN(b)\
41 (((b) >> 13) + ((b) >> 23) + ((b) >> 33) + 3 + 1)
42
43 /*
44 * jfs_extendfs()
45 *
46 * function: extend file system;
47 *
48 * |-------------------------------|----------|----------|
49 * file system space fsck inline log
50 * workspace space
51 *
52 * input:
53 * new LVSize: in LV blocks (required)
54 * new LogSize: in LV blocks (optional)
55 * new FSSize: in LV blocks (optional)
56 *
57 * new configuration:
58 * 1. set new LogSize as specified or default from new LVSize;
59 * 2. compute new FSCKSize from new LVSize;
60 * 3. set new FSSize as MIN(FSSize, LVSize-(LogSize+FSCKSize)) where
61 * assert(new FSSize >= old FSSize),
62 * i.e., file system must not be shrinked;
63 */
jfs_extendfs(struct super_block * sb,s64 newLVSize,int newLogSize)64 int jfs_extendfs(struct super_block *sb, s64 newLVSize, int newLogSize)
65 {
66 int rc = 0;
67 struct jfs_sb_info *sbi = JFS_SBI(sb);
68 struct inode *ipbmap = sbi->ipbmap;
69 struct inode *ipbmap2;
70 struct inode *ipimap = sbi->ipimap;
71 struct jfs_log *log = sbi->log;
72 struct bmap *bmp = sbi->bmap;
73 s64 newLogAddress, newFSCKAddress;
74 int newFSCKSize;
75 s64 newMapSize = 0, mapSize;
76 s64 XAddress, XSize, nblocks, xoff, xaddr, t64;
77 s64 oldLVSize;
78 s64 newFSSize;
79 s64 VolumeSize;
80 int newNpages = 0, nPages, newPage, xlen, t32;
81 int tid;
82 int log_formatted = 0;
83 struct inode *iplist[1];
84 struct jfs_superblock *j_sb, *j_sb2;
85 uint old_agsize;
86 struct buffer_head *bh, *bh2;
87
88 /* If the volume hasn't grown, get out now */
89
90 if (sbi->mntflag & JFS_INLINELOG)
91 oldLVSize = addressPXD(&sbi->logpxd) + lengthPXD(&sbi->logpxd);
92 else
93 oldLVSize = addressPXD(&sbi->fsckpxd) +
94 lengthPXD(&sbi->fsckpxd);
95
96 if (oldLVSize >= newLVSize) {
97 printk(KERN_WARNING
98 "jfs_extendfs: volume hasn't grown, returning\n");
99 goto out;
100 }
101
102 VolumeSize = jfs_get_volume_size(sb);
103 if (VolumeSize) {
104 if (newLVSize > VolumeSize) {
105 printk(KERN_WARNING "jfs_extendfs: invalid size\n");
106 rc = -EINVAL;
107 goto out;
108 }
109 } else {
110 /* check the device */
111 bh = sb_bread(sb, newLVSize - 1);
112 if (!bh) {
113 printk(KERN_WARNING "jfs_extendfs: invalid size\n");
114 rc = -EINVAL;
115 goto out;
116 }
117 bforget(bh);
118 }
119
120 /* Can't extend write-protected drive */
121
122 if (isReadOnly(ipbmap)) {
123 printk(KERN_WARNING "jfs_extendfs: read-only file system\n");
124 rc = -EROFS;
125 goto out;
126 }
127
128 /*
129 * reconfigure LV spaces
130 * ---------------------
131 *
132 * validate new size, or, if not specified, determine new size
133 */
134
135 /*
136 * reconfigure inline log space:
137 */
138 if ((sbi->mntflag & JFS_INLINELOG)) {
139 if (newLogSize == 0) {
140 /*
141 * no size specified: default to 1/256 of aggregate
142 * size; rounded up to a megabyte boundary;
143 */
144 newLogSize = newLVSize >> 8;
145 t32 = (1 << (20 - sbi->l2bsize)) - 1;
146 newLogSize = (newLogSize + t32) & ~t32;
147 newLogSize =
148 min(newLogSize, MEGABYTE32 >> sbi->l2bsize);
149 } else {
150 /*
151 * convert the newLogSize to fs blocks.
152 *
153 * Since this is given in megabytes, it will always be
154 * an even number of pages.
155 */
156 newLogSize = (newLogSize * MEGABYTE) >> sbi->l2bsize;
157 }
158
159 } else
160 newLogSize = 0;
161
162 newLogAddress = newLVSize - newLogSize;
163
164 /*
165 * reconfigure fsck work space:
166 *
167 * configure it to the end of the logical volume regardless of
168 * whether file system extends to the end of the aggregate;
169 * Need enough 4k pages to cover:
170 * - 1 bit per block in aggregate rounded up to BPERDMAP boundary
171 * - 1 extra page to handle control page and intermediate level pages
172 * - 50 extra pages for the chkdsk service log
173 */
174 t64 = ((newLVSize - newLogSize + BPERDMAP - 1) >> L2BPERDMAP)
175 << L2BPERDMAP;
176 t32 = ((t64 + (BITSPERPAGE - 1)) / BITSPERPAGE) + 1 + 50;
177 newFSCKSize = t32 << sbi->l2nbperpage;
178 newFSCKAddress = newLogAddress - newFSCKSize;
179
180 /*
181 * compute new file system space;
182 */
183 newFSSize = newLVSize - newLogSize - newFSCKSize;
184
185 /* file system cannot be shrinked */
186 if (newFSSize < bmp->db_mapsize) {
187 rc = -EINVAL;
188 goto out;
189 }
190
191 /*
192 * If we're expanding enough that the inline log does not overlap
193 * the old one, we can format the new log before we quiesce the
194 * filesystem.
195 */
196 if ((sbi->mntflag & JFS_INLINELOG) && (newLogAddress > oldLVSize)) {
197 if ((rc = lmLogFormat(log, newLogAddress, newLogSize)))
198 goto out;
199 log_formatted = 1;
200 }
201 /*
202 * quiesce file system
203 *
204 * (prepare to move the inline log and to prevent map update)
205 *
206 * block any new transactions and wait for completion of
207 * all wip transactions and flush modified pages s.t.
208 * on-disk file system is in consistent state and
209 * log is not required for recovery.
210 */
211 txQuiesce(sb);
212
213 if (sbi->mntflag & JFS_INLINELOG) {
214 /*
215 * deactivate old inline log
216 */
217 lmLogShutdown(log);
218
219 /*
220 * mark on-disk super block for fs in transition;
221 *
222 * update on-disk superblock for the new space configuration
223 * of inline log space and fsck work space descriptors:
224 * N.B. FS descriptor is NOT updated;
225 *
226 * crash recovery:
227 * logredo(): if FM_EXTENDFS, return to fsck() for cleanup;
228 * fsck(): if FM_EXTENDFS, reformat inline log and fsck
229 * workspace from superblock inline log descriptor and fsck
230 * workspace descriptor;
231 */
232
233 /* read in superblock */
234 if ((rc = readSuper(sb, &bh)))
235 goto error_out;
236 j_sb = (struct jfs_superblock *)bh->b_data;
237
238 /* mark extendfs() in progress */
239 j_sb->s_state |= cpu_to_le32(FM_EXTENDFS);
240 j_sb->s_xsize = cpu_to_le64(newFSSize);
241 PXDaddress(&j_sb->s_xfsckpxd, newFSCKAddress);
242 PXDlength(&j_sb->s_xfsckpxd, newFSCKSize);
243 PXDaddress(&j_sb->s_xlogpxd, newLogAddress);
244 PXDlength(&j_sb->s_xlogpxd, newLogSize);
245
246 /* synchronously update superblock */
247 mark_buffer_dirty(bh);
248 ll_rw_block(WRITE, 1, &bh);
249 wait_on_buffer(bh);
250 brelse(bh);
251
252 /*
253 * format new inline log synchronously;
254 *
255 * crash recovery: if log move in progress,
256 * reformat log and exit success;
257 */
258 if (!log_formatted)
259 if ((rc = lmLogFormat(log, newLogAddress, newLogSize)))
260 goto error_out;
261
262 /*
263 * activate new log
264 */
265 log->base = newLogAddress;
266 log->size = newLogSize >> (L2LOGPSIZE - sb->s_blocksize_bits);
267 if ((rc = lmLogInit(log)))
268 goto error_out;
269 }
270
271 /*
272 * extend block allocation map
273 * ---------------------------
274 *
275 * extendfs() for new extension, retry after crash recovery;
276 *
277 * note: both logredo() and fsck() rebuild map from
278 * the bitmap and configuration parameter from superblock
279 * (disregarding all other control information in the map);
280 *
281 * superblock:
282 * s_size: aggregate size in physical blocks;
283 */
284 /*
285 * compute the new block allocation map configuration
286 *
287 * map dinode:
288 * di_size: map file size in byte;
289 * di_nblocks: number of blocks allocated for map file;
290 * di_mapsize: number of blocks in aggregate (covered by map);
291 * map control page:
292 * db_mapsize: number of blocks in aggregate (covered by map);
293 */
294 newMapSize = newFSSize;
295 /* number of data pages of new bmap file:
296 * roundup new size to full dmap page boundary and
297 * add 1 extra dmap page for next extendfs()
298 */
299 t64 = (newMapSize - 1) + BPERDMAP;
300 newNpages = BLKTODMAPN(t64) + 1;
301
302 /*
303 * extend map from current map (WITHOUT growing mapfile)
304 *
305 * map new extension with unmapped part of the last partial
306 * dmap page, if applicable, and extra page(s) allocated
307 * at end of bmap by mkfs() or previous extendfs();
308 */
309 extendBmap:
310 /* compute number of blocks requested to extend */
311 mapSize = bmp->db_mapsize;
312 XAddress = mapSize; /* eXtension Address */
313 XSize = newMapSize - mapSize; /* eXtension Size */
314 old_agsize = bmp->db_agsize; /* We need to know if this changes */
315
316 /* compute number of blocks that can be extended by current mapfile */
317 t64 = dbMapFileSizeToMapSize(ipbmap);
318 if (mapSize > t64) {
319 printk(KERN_ERR "jfs_extendfs: mapSize (0x%Lx) > t64 (0x%Lx)\n",
320 (long long) mapSize, (long long) t64);
321 rc = -EIO;
322 goto error_out;
323 }
324 nblocks = min(t64 - mapSize, XSize);
325
326 /*
327 * update map pages for new extension:
328 *
329 * update/init dmap and bubble up the control hierarchy
330 * incrementally fold up dmaps into upper levels;
331 * update bmap control page;
332 */
333 if ((rc = dbExtendFS(ipbmap, XAddress, nblocks)))
334 goto error_out;
335 /*
336 * the map now has extended to cover additional nblocks:
337 * dn_mapsize = oldMapsize + nblocks;
338 */
339 /* ipbmap->i_mapsize += nblocks; */
340 XSize -= nblocks;
341
342 /*
343 * grow map file to cover remaining extension
344 * and/or one extra dmap page for next extendfs();
345 *
346 * allocate new map pages and its backing blocks, and
347 * update map file xtree
348 */
349 /* compute number of data pages of current bmap file */
350 nPages = ipbmap->i_size >> L2PSIZE;
351
352 /* need to grow map file ? */
353 if (nPages == newNpages)
354 goto finalizeBmap;
355
356 /*
357 * grow bmap file for the new map pages required:
358 *
359 * allocate growth at the start of newly extended region;
360 * bmap file only grows sequentially, i.e., both data pages
361 * and possibly xtree index pages may grow in append mode,
362 * s.t. logredo() can reconstruct pre-extension state
363 * by washing away bmap file of pages outside s_size boundary;
364 */
365 /*
366 * journal map file growth as if a regular file growth:
367 * (note: bmap is created with di_mode = IFJOURNAL|IFREG);
368 *
369 * journaling of bmap file growth is not required since
370 * logredo() do/can not use log records of bmap file growth
371 * but it provides careful write semantics, pmap update, etc.;
372 */
373 /* synchronous write of data pages: bmap data pages are
374 * cached in meta-data cache, and not written out
375 * by txCommit();
376 */
377 fsync_inode_data_buffers(ipbmap);
378 diWriteSpecial(ipbmap, 0);
379
380 newPage = nPages; /* first new page number */
381 xoff = newPage << sbi->l2nbperpage;
382 xlen = (newNpages - nPages) << sbi->l2nbperpage;
383 xlen = min(xlen, (int) nblocks) & ~(sbi->nbperpage - 1);
384 xaddr = XAddress;
385
386 tid = txBegin(sb, COMMIT_FORCE);
387
388 if ((rc = xtAppend(tid, ipbmap, 0, xoff, nblocks, &xlen, &xaddr, 0))) {
389 txEnd(tid);
390 goto error_out;
391 }
392 /* update bmap file size */
393 ipbmap->i_size += xlen << sbi->l2bsize;
394 ipbmap->i_blocks += LBLK2PBLK(sb, xlen);
395
396 iplist[0] = ipbmap;
397 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
398
399 txEnd(tid);
400
401 if (rc)
402 goto error_out;
403
404 /*
405 * map file has been grown now to cover extension to further out;
406 * di_size = new map file size;
407 *
408 * if huge extension, the previous extension based on previous
409 * map file size may not have been sufficient to cover whole extension
410 * (it could have been used up for new map pages),
411 * but the newly grown map file now covers lot bigger new free space
412 * available for further extension of map;
413 */
414 /* any more blocks to extend ? */
415 if (XSize)
416 goto extendBmap;
417
418 finalizeBmap:
419 /* finalize bmap */
420 dbFinalizeBmap(ipbmap);
421
422 /*
423 * update inode allocation map
424 * ---------------------------
425 *
426 * move iag lists from old to new iag;
427 * agstart field is not updated for logredo() to reconstruct
428 * iag lists if system crash occurs.
429 * (computation of ag number from agstart based on agsize
430 * will correctly identify the new ag);
431 */
432 /* if new AG size the same as old AG size, done! */
433 if (bmp->db_agsize != old_agsize) {
434 if ((rc = diExtendFS(ipimap, ipbmap)))
435 goto error_out;
436
437 /* finalize imap */
438 if ((rc = diSync(ipimap)))
439 goto error_out;
440 }
441
442 /*
443 * finalize
444 * --------
445 *
446 * extension is committed when on-disk super block is
447 * updated with new descriptors: logredo will recover
448 * crash before it to pre-extension state;
449 */
450
451 /* sync log to skip log replay of bmap file growth transaction; */
452 /* lmLogSync(log, 1); */
453
454 /*
455 * synchronous write bmap global control page;
456 * for crash before completion of write
457 * logredo() will recover to pre-extendfs state;
458 * for crash after completion of write,
459 * logredo() will recover post-extendfs state;
460 */
461 if ((rc = dbSync(ipbmap)))
462 goto error_out;
463
464 /*
465 * copy primary bmap inode to secondary bmap inode
466 */
467
468 ipbmap2 = diReadSpecial(sb, BMAP_I, 1);
469 if (ipbmap2 == NULL) {
470 printk(KERN_ERR "jfs_extendfs: diReadSpecial(bmap) failed\n");
471 goto error_out;
472 }
473 memcpy(&JFS_IP(ipbmap2)->i_xtroot, &JFS_IP(ipbmap)->i_xtroot, 288);
474 ipbmap2->i_size = ipbmap->i_size;
475 ipbmap2->i_blocks = ipbmap->i_blocks;
476
477 diWriteSpecial(ipbmap2, 1);
478 diFreeSpecial(ipbmap2);
479
480 /*
481 * update superblock
482 */
483 if ((rc = readSuper(sb, &bh)))
484 goto error_out;
485 j_sb = (struct jfs_superblock *)bh->b_data;
486
487 /* mark extendfs() completion */
488 j_sb->s_state &= cpu_to_le32(~FM_EXTENDFS);
489 j_sb->s_size = cpu_to_le64(bmp->db_mapsize <<
490 le16_to_cpu(j_sb->s_l2bfactor));
491 j_sb->s_agsize = cpu_to_le32(bmp->db_agsize);
492
493 /* update inline log space descriptor */
494 if (sbi->mntflag & JFS_INLINELOG) {
495 PXDaddress(&(j_sb->s_logpxd), newLogAddress);
496 PXDlength(&(j_sb->s_logpxd), newLogSize);
497 }
498
499 /* record log's mount serial number */
500 j_sb->s_logserial = cpu_to_le32(log->serial);
501
502 /* update fsck work space descriptor */
503 PXDaddress(&(j_sb->s_fsckpxd), newFSCKAddress);
504 PXDlength(&(j_sb->s_fsckpxd), newFSCKSize);
505 j_sb->s_fscklog = 1;
506 /* sb->s_fsckloglen remains the same */
507
508 /* Update secondary superblock */
509 bh2 = sb_bread(sb, SUPER2_OFF >> sb->s_blocksize_bits);
510 if (bh2) {
511 j_sb2 = (struct jfs_superblock *)bh2->b_data;
512 memcpy(j_sb2, j_sb, sizeof (struct jfs_superblock));
513
514 mark_buffer_dirty(bh);
515 ll_rw_block(WRITE, 1, &bh2);
516 wait_on_buffer(bh2);
517 brelse(bh2);
518 }
519
520 /* write primary superblock */
521 mark_buffer_dirty(bh);
522 ll_rw_block(WRITE, 1, &bh);
523 wait_on_buffer(bh);
524 brelse(bh);
525
526 goto resume;
527
528 error_out:
529 jfs_error(sb, "jfs_extendfs");
530
531 resume:
532 /*
533 * resume file system transactions
534 */
535 txResume(sb);
536
537 out:
538 return rc;
539 }
540