Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
[linux-2.6] / fs / ext4 / resize.c
1 /*
2  *  linux/fs/ext4/resize.c
3  *
4  * Support for resizing an ext4 filesystem while it is mounted.
5  *
6  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
7  *
8  * This could probably be made into a module, because it is not often in use.
9  */
10
11
12 #define EXT4FS_DEBUG
13
14 #include <linux/ext4_jbd2.h>
15
16 #include <linux/errno.h>
17 #include <linux/slab.h>
18
19 #include "group.h"
20
21 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
22 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
23
24 static int verify_group_input(struct super_block *sb,
25                               struct ext4_new_group_data *input)
26 {
27         struct ext4_sb_info *sbi = EXT4_SB(sb);
28         struct ext4_super_block *es = sbi->s_es;
29         ext4_fsblk_t start = ext4_blocks_count(es);
30         ext4_fsblk_t end = start + input->blocks_count;
31         ext4_group_t group = input->group;
32         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
33         unsigned overhead = ext4_bg_has_super(sb, group) ?
34                 (1 + ext4_bg_num_gdb(sb, group) +
35                  le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
36         ext4_fsblk_t metaend = start + overhead;
37         struct buffer_head *bh = NULL;
38         ext4_grpblk_t free_blocks_count, offset;
39         int err = -EINVAL;
40
41         input->free_blocks_count = free_blocks_count =
42                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
43
44         if (test_opt(sb, DEBUG))
45                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
46                        "(%d free, %u reserved)\n",
47                        ext4_bg_has_super(sb, input->group) ? "normal" :
48                        "no-super", input->group, input->blocks_count,
49                        free_blocks_count, input->reserved_blocks);
50
51         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
52         if (group != sbi->s_groups_count)
53                 ext4_warning(sb, __FUNCTION__,
54                              "Cannot add at group %u (only %lu groups)",
55                              input->group, sbi->s_groups_count);
56         else if (offset != 0)
57                         ext4_warning(sb, __FUNCTION__, "Last group not full");
58         else if (input->reserved_blocks > input->blocks_count / 5)
59                 ext4_warning(sb, __FUNCTION__, "Reserved blocks too high (%u)",
60                              input->reserved_blocks);
61         else if (free_blocks_count < 0)
62                 ext4_warning(sb, __FUNCTION__, "Bad blocks count %u",
63                              input->blocks_count);
64         else if (!(bh = sb_bread(sb, end - 1)))
65                 ext4_warning(sb, __FUNCTION__,
66                              "Cannot read last block (%llu)",
67                              end - 1);
68         else if (outside(input->block_bitmap, start, end))
69                 ext4_warning(sb, __FUNCTION__,
70                              "Block bitmap not in group (block %llu)",
71                              (unsigned long long)input->block_bitmap);
72         else if (outside(input->inode_bitmap, start, end))
73                 ext4_warning(sb, __FUNCTION__,
74                              "Inode bitmap not in group (block %llu)",
75                              (unsigned long long)input->inode_bitmap);
76         else if (outside(input->inode_table, start, end) ||
77                  outside(itend - 1, start, end))
78                 ext4_warning(sb, __FUNCTION__,
79                              "Inode table not in group (blocks %llu-%llu)",
80                              (unsigned long long)input->inode_table, itend - 1);
81         else if (input->inode_bitmap == input->block_bitmap)
82                 ext4_warning(sb, __FUNCTION__,
83                              "Block bitmap same as inode bitmap (%llu)",
84                              (unsigned long long)input->block_bitmap);
85         else if (inside(input->block_bitmap, input->inode_table, itend))
86                 ext4_warning(sb, __FUNCTION__,
87                              "Block bitmap (%llu) in inode table (%llu-%llu)",
88                              (unsigned long long)input->block_bitmap,
89                              (unsigned long long)input->inode_table, itend - 1);
90         else if (inside(input->inode_bitmap, input->inode_table, itend))
91                 ext4_warning(sb, __FUNCTION__,
92                              "Inode bitmap (%llu) in inode table (%llu-%llu)",
93                              (unsigned long long)input->inode_bitmap,
94                              (unsigned long long)input->inode_table, itend - 1);
95         else if (inside(input->block_bitmap, start, metaend))
96                 ext4_warning(sb, __FUNCTION__,
97                              "Block bitmap (%llu) in GDT table"
98                              " (%llu-%llu)",
99                              (unsigned long long)input->block_bitmap,
100                              start, metaend - 1);
101         else if (inside(input->inode_bitmap, start, metaend))
102                 ext4_warning(sb, __FUNCTION__,
103                              "Inode bitmap (%llu) in GDT table"
104                              " (%llu-%llu)",
105                              (unsigned long long)input->inode_bitmap,
106                              start, metaend - 1);
107         else if (inside(input->inode_table, start, metaend) ||
108                  inside(itend - 1, start, metaend))
109                 ext4_warning(sb, __FUNCTION__,
110                              "Inode table (%llu-%llu) overlaps"
111                              "GDT table (%llu-%llu)",
112                              (unsigned long long)input->inode_table,
113                              itend - 1, start, metaend - 1);
114         else
115                 err = 0;
116         brelse(bh);
117
118         return err;
119 }
120
121 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
122                                   ext4_fsblk_t blk)
123 {
124         struct buffer_head *bh;
125         int err;
126
127         bh = sb_getblk(sb, blk);
128         if (!bh)
129                 return ERR_PTR(-EIO);
130         if ((err = ext4_journal_get_write_access(handle, bh))) {
131                 brelse(bh);
132                 bh = ERR_PTR(err);
133         } else {
134                 lock_buffer(bh);
135                 memset(bh->b_data, 0, sb->s_blocksize);
136                 set_buffer_uptodate(bh);
137                 unlock_buffer(bh);
138         }
139
140         return bh;
141 }
142
143 /*
144  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
145  * If that fails, restart the transaction & regain write access for the
146  * buffer head which is used for block_bitmap modifications.
147  */
148 static int extend_or_restart_transaction(handle_t *handle, int thresh,
149                                          struct buffer_head *bh)
150 {
151         int err;
152
153         if (handle->h_buffer_credits >= thresh)
154                 return 0;
155
156         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
157         if (err < 0)
158                 return err;
159         if (err) {
160                 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
161                         return err;
162                 if ((err = ext4_journal_get_write_access(handle, bh)))
163                         return err;
164         }
165
166         return 0;
167 }
168
169 /*
170  * Set up the block and inode bitmaps, and the inode table for the new group.
171  * This doesn't need to be part of the main transaction, since we are only
172  * changing blocks outside the actual filesystem.  We still do journaling to
173  * ensure the recovery is correct in case of a failure just after resize.
174  * If any part of this fails, we simply abort the resize.
175  */
176 static int setup_new_group_blocks(struct super_block *sb,
177                                   struct ext4_new_group_data *input)
178 {
179         struct ext4_sb_info *sbi = EXT4_SB(sb);
180         ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
181         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
182                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
183         unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
184         struct buffer_head *bh;
185         handle_t *handle;
186         ext4_fsblk_t block;
187         ext4_grpblk_t bit;
188         int i;
189         int err = 0, err2;
190
191         /* This transaction may be extended/restarted along the way */
192         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
193
194         if (IS_ERR(handle))
195                 return PTR_ERR(handle);
196
197         lock_super(sb);
198         if (input->group != sbi->s_groups_count) {
199                 err = -EBUSY;
200                 goto exit_journal;
201         }
202
203         if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
204                 err = PTR_ERR(bh);
205                 goto exit_journal;
206         }
207
208         if (ext4_bg_has_super(sb, input->group)) {
209                 ext4_debug("mark backup superblock %#04llx (+0)\n", start);
210                 ext4_set_bit(0, bh->b_data);
211         }
212
213         /* Copy all of the GDT blocks into the backup in this group */
214         for (i = 0, bit = 1, block = start + 1;
215              i < gdblocks; i++, block++, bit++) {
216                 struct buffer_head *gdb;
217
218                 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
219
220                 if ((err = extend_or_restart_transaction(handle, 1, bh)))
221                         goto exit_bh;
222
223                 gdb = sb_getblk(sb, block);
224                 if (!gdb) {
225                         err = -EIO;
226                         goto exit_bh;
227                 }
228                 if ((err = ext4_journal_get_write_access(handle, gdb))) {
229                         brelse(gdb);
230                         goto exit_bh;
231                 }
232                 lock_buffer(gdb);
233                 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
234                 set_buffer_uptodate(gdb);
235                 unlock_buffer(gdb);
236                 ext4_journal_dirty_metadata(handle, gdb);
237                 ext4_set_bit(bit, bh->b_data);
238                 brelse(gdb);
239         }
240
241         /* Zero out all of the reserved backup group descriptor table blocks */
242         for (i = 0, bit = gdblocks + 1, block = start + bit;
243              i < reserved_gdb; i++, block++, bit++) {
244                 struct buffer_head *gdb;
245
246                 ext4_debug("clear reserved block %#04llx (+%d)\n", block, bit);
247
248                 if ((err = extend_or_restart_transaction(handle, 1, bh)))
249                         goto exit_bh;
250
251                 if (IS_ERR(gdb = bclean(handle, sb, block))) {
252                         err = PTR_ERR(bh);
253                         goto exit_bh;
254                 }
255                 ext4_journal_dirty_metadata(handle, gdb);
256                 ext4_set_bit(bit, bh->b_data);
257                 brelse(gdb);
258         }
259         ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
260                    input->block_bitmap - start);
261         ext4_set_bit(input->block_bitmap - start, bh->b_data);
262         ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
263                    input->inode_bitmap - start);
264         ext4_set_bit(input->inode_bitmap - start, bh->b_data);
265
266         /* Zero out all of the inode table blocks */
267         for (i = 0, block = input->inode_table, bit = block - start;
268              i < sbi->s_itb_per_group; i++, bit++, block++) {
269                 struct buffer_head *it;
270
271                 ext4_debug("clear inode block %#04llx (+%d)\n", block, bit);
272
273                 if ((err = extend_or_restart_transaction(handle, 1, bh)))
274                         goto exit_bh;
275
276                 if (IS_ERR(it = bclean(handle, sb, block))) {
277                         err = PTR_ERR(it);
278                         goto exit_bh;
279                 }
280                 ext4_journal_dirty_metadata(handle, it);
281                 brelse(it);
282                 ext4_set_bit(bit, bh->b_data);
283         }
284
285         if ((err = extend_or_restart_transaction(handle, 2, bh)))
286                 goto exit_bh;
287
288         mark_bitmap_end(input->blocks_count, EXT4_BLOCKS_PER_GROUP(sb),
289                         bh->b_data);
290         ext4_journal_dirty_metadata(handle, bh);
291         brelse(bh);
292
293         /* Mark unused entries in inode bitmap used */
294         ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
295                    input->inode_bitmap, input->inode_bitmap - start);
296         if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
297                 err = PTR_ERR(bh);
298                 goto exit_journal;
299         }
300
301         mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb),
302                         bh->b_data);
303         ext4_journal_dirty_metadata(handle, bh);
304 exit_bh:
305         brelse(bh);
306
307 exit_journal:
308         unlock_super(sb);
309         if ((err2 = ext4_journal_stop(handle)) && !err)
310                 err = err2;
311
312         return err;
313 }
314
315 /*
316  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
317  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
318  * calling this for the first time.  In a sparse filesystem it will be the
319  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
320  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
321  */
322 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
323                                   unsigned *five, unsigned *seven)
324 {
325         unsigned *min = three;
326         int mult = 3;
327         unsigned ret;
328
329         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
330                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
331                 ret = *min;
332                 *min += 1;
333                 return ret;
334         }
335
336         if (*five < *min) {
337                 min = five;
338                 mult = 5;
339         }
340         if (*seven < *min) {
341                 min = seven;
342                 mult = 7;
343         }
344
345         ret = *min;
346         *min *= mult;
347
348         return ret;
349 }
350
351 /*
352  * Check that all of the backup GDT blocks are held in the primary GDT block.
353  * It is assumed that they are stored in group order.  Returns the number of
354  * groups in current filesystem that have BACKUPS, or -ve error code.
355  */
356 static int verify_reserved_gdb(struct super_block *sb,
357                                struct buffer_head *primary)
358 {
359         const ext4_fsblk_t blk = primary->b_blocknr;
360         const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
361         unsigned three = 1;
362         unsigned five = 5;
363         unsigned seven = 7;
364         unsigned grp;
365         __le32 *p = (__le32 *)primary->b_data;
366         int gdbackups = 0;
367
368         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
369                 if (le32_to_cpu(*p++) !=
370                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
371                         ext4_warning(sb, __FUNCTION__,
372                                      "reserved GDT %llu"
373                                      " missing grp %d (%llu)",
374                                      blk, grp,
375                                      grp *
376                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
377                                      blk);
378                         return -EINVAL;
379                 }
380                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
381                         return -EFBIG;
382         }
383
384         return gdbackups;
385 }
386
387 /*
388  * Called when we need to bring a reserved group descriptor table block into
389  * use from the resize inode.  The primary copy of the new GDT block currently
390  * is an indirect block (under the double indirect block in the resize inode).
391  * The new backup GDT blocks will be stored as leaf blocks in this indirect
392  * block, in group order.  Even though we know all the block numbers we need,
393  * we check to ensure that the resize inode has actually reserved these blocks.
394  *
395  * Don't need to update the block bitmaps because the blocks are still in use.
396  *
397  * We get all of the error cases out of the way, so that we are sure to not
398  * fail once we start modifying the data on disk, because JBD has no rollback.
399  */
400 static int add_new_gdb(handle_t *handle, struct inode *inode,
401                        struct ext4_new_group_data *input,
402                        struct buffer_head **primary)
403 {
404         struct super_block *sb = inode->i_sb;
405         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
406         unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
407         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
408         struct buffer_head **o_group_desc, **n_group_desc;
409         struct buffer_head *dind;
410         int gdbackups;
411         struct ext4_iloc iloc;
412         __le32 *data;
413         int err;
414
415         if (test_opt(sb, DEBUG))
416                 printk(KERN_DEBUG
417                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
418                        gdb_num);
419
420         /*
421          * If we are not using the primary superblock/GDT copy don't resize,
422          * because the user tools have no way of handling this.  Probably a
423          * bad time to do it anyways.
424          */
425         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
426             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
427                 ext4_warning(sb, __FUNCTION__,
428                         "won't resize using backup superblock at %llu",
429                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
430                 return -EPERM;
431         }
432
433         *primary = sb_bread(sb, gdblock);
434         if (!*primary)
435                 return -EIO;
436
437         if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
438                 err = gdbackups;
439                 goto exit_bh;
440         }
441
442         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
443         dind = sb_bread(sb, le32_to_cpu(*data));
444         if (!dind) {
445                 err = -EIO;
446                 goto exit_bh;
447         }
448
449         data = (__le32 *)dind->b_data;
450         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
451                 ext4_warning(sb, __FUNCTION__,
452                              "new group %u GDT block %llu not reserved",
453                              input->group, gdblock);
454                 err = -EINVAL;
455                 goto exit_dind;
456         }
457
458         if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh)))
459                 goto exit_dind;
460
461         if ((err = ext4_journal_get_write_access(handle, *primary)))
462                 goto exit_sbh;
463
464         if ((err = ext4_journal_get_write_access(handle, dind)))
465                 goto exit_primary;
466
467         /* ext4_reserve_inode_write() gets a reference on the iloc */
468         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
469                 goto exit_dindj;
470
471         n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
472                         GFP_KERNEL);
473         if (!n_group_desc) {
474                 err = -ENOMEM;
475                 ext4_warning (sb, __FUNCTION__,
476                               "not enough memory for %lu groups", gdb_num + 1);
477                 goto exit_inode;
478         }
479
480         /*
481          * Finally, we have all of the possible failures behind us...
482          *
483          * Remove new GDT block from inode double-indirect block and clear out
484          * the new GDT block for use (which also "frees" the backup GDT blocks
485          * from the reserved inode).  We don't need to change the bitmaps for
486          * these blocks, because they are marked as in-use from being in the
487          * reserved inode, and will become GDT blocks (primary and backup).
488          */
489         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
490         ext4_journal_dirty_metadata(handle, dind);
491         brelse(dind);
492         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
493         ext4_mark_iloc_dirty(handle, inode, &iloc);
494         memset((*primary)->b_data, 0, sb->s_blocksize);
495         ext4_journal_dirty_metadata(handle, *primary);
496
497         o_group_desc = EXT4_SB(sb)->s_group_desc;
498         memcpy(n_group_desc, o_group_desc,
499                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
500         n_group_desc[gdb_num] = *primary;
501         EXT4_SB(sb)->s_group_desc = n_group_desc;
502         EXT4_SB(sb)->s_gdb_count++;
503         kfree(o_group_desc);
504
505         es->s_reserved_gdt_blocks =
506                 cpu_to_le16(le16_to_cpu(es->s_reserved_gdt_blocks) - 1);
507         ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
508
509         return 0;
510
511 exit_inode:
512         //ext4_journal_release_buffer(handle, iloc.bh);
513         brelse(iloc.bh);
514 exit_dindj:
515         //ext4_journal_release_buffer(handle, dind);
516 exit_primary:
517         //ext4_journal_release_buffer(handle, *primary);
518 exit_sbh:
519         //ext4_journal_release_buffer(handle, *primary);
520 exit_dind:
521         brelse(dind);
522 exit_bh:
523         brelse(*primary);
524
525         ext4_debug("leaving with error %d\n", err);
526         return err;
527 }
528
529 /*
530  * Called when we are adding a new group which has a backup copy of each of
531  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
532  * We need to add these reserved backup GDT blocks to the resize inode, so
533  * that they are kept for future resizing and not allocated to files.
534  *
535  * Each reserved backup GDT block will go into a different indirect block.
536  * The indirect blocks are actually the primary reserved GDT blocks,
537  * so we know in advance what their block numbers are.  We only get the
538  * double-indirect block to verify it is pointing to the primary reserved
539  * GDT blocks so we don't overwrite a data block by accident.  The reserved
540  * backup GDT blocks are stored in their reserved primary GDT block.
541  */
542 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
543                               struct ext4_new_group_data *input)
544 {
545         struct super_block *sb = inode->i_sb;
546         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
547         struct buffer_head **primary;
548         struct buffer_head *dind;
549         struct ext4_iloc iloc;
550         ext4_fsblk_t blk;
551         __le32 *data, *end;
552         int gdbackups = 0;
553         int res, i;
554         int err;
555
556         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_KERNEL);
557         if (!primary)
558                 return -ENOMEM;
559
560         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
561         dind = sb_bread(sb, le32_to_cpu(*data));
562         if (!dind) {
563                 err = -EIO;
564                 goto exit_free;
565         }
566
567         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
568         data = (__le32 *)dind->b_data + EXT4_SB(sb)->s_gdb_count;
569         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
570
571         /* Get each reserved primary GDT block and verify it holds backups */
572         for (res = 0; res < reserved_gdb; res++, blk++) {
573                 if (le32_to_cpu(*data) != blk) {
574                         ext4_warning(sb, __FUNCTION__,
575                                      "reserved block %llu"
576                                      " not at offset %ld",
577                                      blk,
578                                      (long)(data - (__le32 *)dind->b_data));
579                         err = -EINVAL;
580                         goto exit_bh;
581                 }
582                 primary[res] = sb_bread(sb, blk);
583                 if (!primary[res]) {
584                         err = -EIO;
585                         goto exit_bh;
586                 }
587                 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
588                         brelse(primary[res]);
589                         err = gdbackups;
590                         goto exit_bh;
591                 }
592                 if (++data >= end)
593                         data = (__le32 *)dind->b_data;
594         }
595
596         for (i = 0; i < reserved_gdb; i++) {
597                 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
598                         /*
599                         int j;
600                         for (j = 0; j < i; j++)
601                                 ext4_journal_release_buffer(handle, primary[j]);
602                          */
603                         goto exit_bh;
604                 }
605         }
606
607         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
608                 goto exit_bh;
609
610         /*
611          * Finally we can add each of the reserved backup GDT blocks from
612          * the new group to its reserved primary GDT block.
613          */
614         blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
615         for (i = 0; i < reserved_gdb; i++) {
616                 int err2;
617                 data = (__le32 *)primary[i]->b_data;
618                 /* printk("reserving backup %lu[%u] = %lu\n",
619                        primary[i]->b_blocknr, gdbackups,
620                        blk + primary[i]->b_blocknr); */
621                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
622                 err2 = ext4_journal_dirty_metadata(handle, primary[i]);
623                 if (!err)
624                         err = err2;
625         }
626         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
627         ext4_mark_iloc_dirty(handle, inode, &iloc);
628
629 exit_bh:
630         while (--res >= 0)
631                 brelse(primary[res]);
632         brelse(dind);
633
634 exit_free:
635         kfree(primary);
636
637         return err;
638 }
639
640 /*
641  * Update the backup copies of the ext4 metadata.  These don't need to be part
642  * of the main resize transaction, because e2fsck will re-write them if there
643  * is a problem (basically only OOM will cause a problem).  However, we
644  * _should_ update the backups if possible, in case the primary gets trashed
645  * for some reason and we need to run e2fsck from a backup superblock.  The
646  * important part is that the new block and inode counts are in the backup
647  * superblocks, and the location of the new group metadata in the GDT backups.
648  *
649  * We do not need lock_super() for this, because these blocks are not
650  * otherwise touched by the filesystem code when it is mounted.  We don't
651  * need to worry about last changing from sbi->s_groups_count, because the
652  * worst that can happen is that we do not copy the full number of backups
653  * at this time.  The resize which changed s_groups_count will backup again.
654  */
655 static void update_backups(struct super_block *sb,
656                            int blk_off, char *data, int size)
657 {
658         struct ext4_sb_info *sbi = EXT4_SB(sb);
659         const ext4_group_t last = sbi->s_groups_count;
660         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
661         unsigned three = 1;
662         unsigned five = 5;
663         unsigned seven = 7;
664         ext4_group_t group;
665         int rest = sb->s_blocksize - size;
666         handle_t *handle;
667         int err = 0, err2;
668
669         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
670         if (IS_ERR(handle)) {
671                 group = 1;
672                 err = PTR_ERR(handle);
673                 goto exit_err;
674         }
675
676         while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
677                 struct buffer_head *bh;
678
679                 /* Out of journal space, and can't get more - abort - so sad */
680                 if (handle->h_buffer_credits == 0 &&
681                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
682                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
683                         break;
684
685                 bh = sb_getblk(sb, group * bpg + blk_off);
686                 if (!bh) {
687                         err = -EIO;
688                         break;
689                 }
690                 ext4_debug("update metadata backup %#04lx\n",
691                           (unsigned long)bh->b_blocknr);
692                 if ((err = ext4_journal_get_write_access(handle, bh)))
693                         break;
694                 lock_buffer(bh);
695                 memcpy(bh->b_data, data, size);
696                 if (rest)
697                         memset(bh->b_data + size, 0, rest);
698                 set_buffer_uptodate(bh);
699                 unlock_buffer(bh);
700                 ext4_journal_dirty_metadata(handle, bh);
701                 brelse(bh);
702         }
703         if ((err2 = ext4_journal_stop(handle)) && !err)
704                 err = err2;
705
706         /*
707          * Ugh! Need to have e2fsck write the backup copies.  It is too
708          * late to revert the resize, we shouldn't fail just because of
709          * the backup copies (they are only needed in case of corruption).
710          *
711          * However, if we got here we have a journal problem too, so we
712          * can't really start a transaction to mark the superblock.
713          * Chicken out and just set the flag on the hope it will be written
714          * to disk, and if not - we will simply wait until next fsck.
715          */
716 exit_err:
717         if (err) {
718                 ext4_warning(sb, __FUNCTION__,
719                              "can't update backup for group %lu (err %d), "
720                              "forcing fsck on next reboot", group, err);
721                 sbi->s_mount_state &= ~EXT4_VALID_FS;
722                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
723                 mark_buffer_dirty(sbi->s_sbh);
724         }
725 }
726
727 /* Add group descriptor data to an existing or new group descriptor block.
728  * Ensure we handle all possible error conditions _before_ we start modifying
729  * the filesystem, because we cannot abort the transaction and not have it
730  * write the data to disk.
731  *
732  * If we are on a GDT block boundary, we need to get the reserved GDT block.
733  * Otherwise, we may need to add backup GDT blocks for a sparse group.
734  *
735  * We only need to hold the superblock lock while we are actually adding
736  * in the new group's counts to the superblock.  Prior to that we have
737  * not really "added" the group at all.  We re-check that we are still
738  * adding in the last group in case things have changed since verifying.
739  */
740 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
741 {
742         struct ext4_sb_info *sbi = EXT4_SB(sb);
743         struct ext4_super_block *es = sbi->s_es;
744         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
745                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
746         struct buffer_head *primary = NULL;
747         struct ext4_group_desc *gdp;
748         struct inode *inode = NULL;
749         handle_t *handle;
750         int gdb_off, gdb_num;
751         int err, err2;
752
753         gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
754         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
755
756         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
757                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
758                 ext4_warning(sb, __FUNCTION__,
759                              "Can't resize non-sparse filesystem further");
760                 return -EPERM;
761         }
762
763         if (ext4_blocks_count(es) + input->blocks_count <
764             ext4_blocks_count(es)) {
765                 ext4_warning(sb, __FUNCTION__, "blocks_count overflow\n");
766                 return -EINVAL;
767         }
768
769         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
770             le32_to_cpu(es->s_inodes_count)) {
771                 ext4_warning(sb, __FUNCTION__, "inodes_count overflow\n");
772                 return -EINVAL;
773         }
774
775         if (reserved_gdb || gdb_off == 0) {
776                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
777                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)){
778                         ext4_warning(sb, __FUNCTION__,
779                                      "No reserved GDT blocks, can't resize");
780                         return -EPERM;
781                 }
782                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
783                 if (IS_ERR(inode)) {
784                         ext4_warning(sb, __FUNCTION__,
785                                      "Error opening resize inode");
786                         return PTR_ERR(inode);
787                 }
788         }
789
790         if ((err = verify_group_input(sb, input)))
791                 goto exit_put;
792
793         if ((err = setup_new_group_blocks(sb, input)))
794                 goto exit_put;
795
796         /*
797          * We will always be modifying at least the superblock and a GDT
798          * block.  If we are adding a group past the last current GDT block,
799          * we will also modify the inode and the dindirect block.  If we
800          * are adding a group with superblock/GDT backups  we will also
801          * modify each of the reserved GDT dindirect blocks.
802          */
803         handle = ext4_journal_start_sb(sb,
804                                        ext4_bg_has_super(sb, input->group) ?
805                                        3 + reserved_gdb : 4);
806         if (IS_ERR(handle)) {
807                 err = PTR_ERR(handle);
808                 goto exit_put;
809         }
810
811         lock_super(sb);
812         if (input->group != sbi->s_groups_count) {
813                 ext4_warning(sb, __FUNCTION__,
814                              "multiple resizers run on filesystem!");
815                 err = -EBUSY;
816                 goto exit_journal;
817         }
818
819         if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
820                 goto exit_journal;
821
822         /*
823          * We will only either add reserved group blocks to a backup group
824          * or remove reserved blocks for the first group in a new group block.
825          * Doing both would be mean more complex code, and sane people don't
826          * use non-sparse filesystems anymore.  This is already checked above.
827          */
828         if (gdb_off) {
829                 primary = sbi->s_group_desc[gdb_num];
830                 if ((err = ext4_journal_get_write_access(handle, primary)))
831                         goto exit_journal;
832
833                 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
834                     (err = reserve_backup_gdb(handle, inode, input)))
835                         goto exit_journal;
836         } else if ((err = add_new_gdb(handle, inode, input, &primary)))
837                 goto exit_journal;
838
839         /*
840          * OK, now we've set up the new group.  Time to make it active.
841          *
842          * Current kernels don't lock all allocations via lock_super(),
843          * so we have to be safe wrt. concurrent accesses the group
844          * data.  So we need to be careful to set all of the relevant
845          * group descriptor data etc. *before* we enable the group.
846          *
847          * The key field here is sbi->s_groups_count: as long as
848          * that retains its old value, nobody is going to access the new
849          * group.
850          *
851          * So first we update all the descriptor metadata for the new
852          * group; then we update the total disk blocks count; then we
853          * update the groups count to enable the group; then finally we
854          * update the free space counts so that the system can start
855          * using the new disk blocks.
856          */
857
858         /* Update group descriptor block for new group */
859         gdp = (struct ext4_group_desc *)primary->b_data + gdb_off;
860
861         ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
862         ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
863         ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
864         gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
865         gdp->bg_free_inodes_count = cpu_to_le16(EXT4_INODES_PER_GROUP(sb));
866         gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
867
868         /*
869          * Make the new blocks and inodes valid next.  We do this before
870          * increasing the group count so that once the group is enabled,
871          * all of its blocks and inodes are already valid.
872          *
873          * We always allocate group-by-group, then block-by-block or
874          * inode-by-inode within a group, so enabling these
875          * blocks/inodes before the group is live won't actually let us
876          * allocate the new space yet.
877          */
878         ext4_blocks_count_set(es, ext4_blocks_count(es) +
879                 input->blocks_count);
880         es->s_inodes_count = cpu_to_le32(le32_to_cpu(es->s_inodes_count) +
881                 EXT4_INODES_PER_GROUP(sb));
882
883         /*
884          * We need to protect s_groups_count against other CPUs seeing
885          * inconsistent state in the superblock.
886          *
887          * The precise rules we use are:
888          *
889          * * Writers of s_groups_count *must* hold lock_super
890          * AND
891          * * Writers must perform a smp_wmb() after updating all dependent
892          *   data and before modifying the groups count
893          *
894          * * Readers must hold lock_super() over the access
895          * OR
896          * * Readers must perform an smp_rmb() after reading the groups count
897          *   and before reading any dependent data.
898          *
899          * NB. These rules can be relaxed when checking the group count
900          * while freeing data, as we can only allocate from a block
901          * group after serialising against the group count, and we can
902          * only then free after serialising in turn against that
903          * allocation.
904          */
905         smp_wmb();
906
907         /* Update the global fs size fields */
908         sbi->s_groups_count++;
909
910         ext4_journal_dirty_metadata(handle, primary);
911
912         /* Update the reserved block counts only once the new group is
913          * active. */
914         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
915                 input->reserved_blocks);
916
917         /* Update the free space counts */
918         percpu_counter_add(&sbi->s_freeblocks_counter,
919                            input->free_blocks_count);
920         percpu_counter_add(&sbi->s_freeinodes_counter,
921                            EXT4_INODES_PER_GROUP(sb));
922
923         ext4_journal_dirty_metadata(handle, sbi->s_sbh);
924         sb->s_dirt = 1;
925
926 exit_journal:
927         unlock_super(sb);
928         if ((err2 = ext4_journal_stop(handle)) && !err)
929                 err = err2;
930         if (!err) {
931                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
932                                sizeof(struct ext4_super_block));
933                 update_backups(sb, primary->b_blocknr, primary->b_data,
934                                primary->b_size);
935         }
936 exit_put:
937         iput(inode);
938         return err;
939 } /* ext4_group_add */
940
941 /* Extend the filesystem to the new number of blocks specified.  This entry
942  * point is only used to extend the current filesystem to the end of the last
943  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
944  * for emergencies (because it has no dependencies on reserved blocks).
945  *
946  * If we _really_ wanted, we could use default values to call ext4_group_add()
947  * allow the "remount" trick to work for arbitrary resizing, assuming enough
948  * GDT blocks are reserved to grow to the desired size.
949  */
950 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
951                       ext4_fsblk_t n_blocks_count)
952 {
953         ext4_fsblk_t o_blocks_count;
954         ext4_group_t o_groups_count;
955         ext4_grpblk_t last;
956         ext4_grpblk_t add;
957         struct buffer_head * bh;
958         handle_t *handle;
959         int err;
960         unsigned long freed_blocks;
961
962         /* We don't need to worry about locking wrt other resizers just
963          * yet: we're going to revalidate es->s_blocks_count after
964          * taking lock_super() below. */
965         o_blocks_count = ext4_blocks_count(es);
966         o_groups_count = EXT4_SB(sb)->s_groups_count;
967
968         if (test_opt(sb, DEBUG))
969                 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
970                        o_blocks_count, n_blocks_count);
971
972         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
973                 return 0;
974
975         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
976                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
977                         " too large to resize to %llu blocks safely\n",
978                         sb->s_id, n_blocks_count);
979                 if (sizeof(sector_t) < 8)
980                         ext4_warning(sb, __FUNCTION__,
981                         "CONFIG_LBD not enabled\n");
982                 return -EINVAL;
983         }
984
985         if (n_blocks_count < o_blocks_count) {
986                 ext4_warning(sb, __FUNCTION__,
987                              "can't shrink FS - resize aborted");
988                 return -EBUSY;
989         }
990
991         /* Handle the remaining blocks in the last group only. */
992         ext4_get_group_no_and_offset(sb, o_blocks_count, NULL, &last);
993
994         if (last == 0) {
995                 ext4_warning(sb, __FUNCTION__,
996                              "need to use ext2online to resize further");
997                 return -EPERM;
998         }
999
1000         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1001
1002         if (o_blocks_count + add < o_blocks_count) {
1003                 ext4_warning(sb, __FUNCTION__, "blocks_count overflow");
1004                 return -EINVAL;
1005         }
1006
1007         if (o_blocks_count + add > n_blocks_count)
1008                 add = n_blocks_count - o_blocks_count;
1009
1010         if (o_blocks_count + add < n_blocks_count)
1011                 ext4_warning(sb, __FUNCTION__,
1012                              "will only finish group (%llu"
1013                              " blocks, %u new)",
1014                              o_blocks_count + add, add);
1015
1016         /* See if the device is actually as big as what was requested */
1017         bh = sb_bread(sb, o_blocks_count + add -1);
1018         if (!bh) {
1019                 ext4_warning(sb, __FUNCTION__,
1020                              "can't read last block, resize aborted");
1021                 return -ENOSPC;
1022         }
1023         brelse(bh);
1024
1025         /* We will update the superblock, one block bitmap, and
1026          * one group descriptor via ext4_free_blocks().
1027          */
1028         handle = ext4_journal_start_sb(sb, 3);
1029         if (IS_ERR(handle)) {
1030                 err = PTR_ERR(handle);
1031                 ext4_warning(sb, __FUNCTION__, "error %d on journal start",err);
1032                 goto exit_put;
1033         }
1034
1035         lock_super(sb);
1036         if (o_blocks_count != ext4_blocks_count(es)) {
1037                 ext4_warning(sb, __FUNCTION__,
1038                              "multiple resizers run on filesystem!");
1039                 unlock_super(sb);
1040                 ext4_journal_stop(handle);
1041                 err = -EBUSY;
1042                 goto exit_put;
1043         }
1044
1045         if ((err = ext4_journal_get_write_access(handle,
1046                                                  EXT4_SB(sb)->s_sbh))) {
1047                 ext4_warning(sb, __FUNCTION__,
1048                              "error %d on journal write access", err);
1049                 unlock_super(sb);
1050                 ext4_journal_stop(handle);
1051                 goto exit_put;
1052         }
1053         ext4_blocks_count_set(es, o_blocks_count + add);
1054         ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1055         sb->s_dirt = 1;
1056         unlock_super(sb);
1057         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1058                    o_blocks_count + add);
1059         ext4_free_blocks_sb(handle, sb, o_blocks_count, add, &freed_blocks);
1060         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1061                    o_blocks_count + add);
1062         if ((err = ext4_journal_stop(handle)))
1063                 goto exit_put;
1064         if (test_opt(sb, DEBUG))
1065                 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1066                        ext4_blocks_count(es));
1067         update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1068                        sizeof(struct ext4_super_block));
1069 exit_put:
1070         return err;
1071 } /* ext4_group_extend */