blob: 12eace0965460d0f2f9dc172f55c15791a8131be [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/resize.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
Mingming Cao617ba132006-10-11 01:20:53 -07004 * Support for resizing an ext4 filesystem while it is mounted.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005 *
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
Mingming Cao617ba132006-10-11 01:20:53 -070012#define EXT4FS_DEBUG
Dave Kleikampac27a0e2006-10-11 01:20:50 -070013
Dave Kleikampac27a0e2006-10-11 01:20:50 -070014#include <linux/errno.h>
15#include <linux/slab.h>
16
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040017#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070018
Yongqiang Yang8f82f842011-07-26 21:35:44 -040019int ext4_resize_begin(struct super_block *sb)
20{
21 int ret = 0;
22
23 if (!capable(CAP_SYS_RESOURCE))
24 return -EPERM;
25
Yongqiang Yangce723c32011-07-26 21:39:09 -040026 /*
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
29 */
30 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
31 ext4_warning(sb, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
33 return -EPERM;
34 }
35
Yongqiang Yang8f82f842011-07-26 21:35:44 -040036 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
37 ret = -EBUSY;
38
39 return ret;
40}
41
42void ext4_resize_end(struct super_block *sb)
43{
44 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
45 smp_mb__after_clear_bit();
46}
47
Dave Kleikampac27a0e2006-10-11 01:20:50 -070048#define outside(b, first, last) ((b) < (first) || (b) >= (last))
49#define inside(b, first, last) ((b) >= (first) && (b) < (last))
50
51static int verify_group_input(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -070052 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070053{
Mingming Cao617ba132006-10-11 01:20:53 -070054 struct ext4_sb_info *sbi = EXT4_SB(sb);
55 struct ext4_super_block *es = sbi->s_es;
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070056 ext4_fsblk_t start = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -070057 ext4_fsblk_t end = start + input->blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -050058 ext4_group_t group = input->group;
Mingming Cao617ba132006-10-11 01:20:53 -070059 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
60 unsigned overhead = ext4_bg_has_super(sb, group) ?
61 (1 + ext4_bg_num_gdb(sb, group) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -070062 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -070063 ext4_fsblk_t metaend = start + overhead;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070064 struct buffer_head *bh = NULL;
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070065 ext4_grpblk_t free_blocks_count, offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070066 int err = -EINVAL;
67
68 input->free_blocks_count = free_blocks_count =
69 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
70
71 if (test_opt(sb, DEBUG))
Mingming Cao617ba132006-10-11 01:20:53 -070072 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
Dave Kleikampac27a0e2006-10-11 01:20:50 -070073 "(%d free, %u reserved)\n",
Mingming Cao617ba132006-10-11 01:20:53 -070074 ext4_bg_has_super(sb, input->group) ? "normal" :
Dave Kleikampac27a0e2006-10-11 01:20:50 -070075 "no-super", input->group, input->blocks_count,
76 free_blocks_count, input->reserved_blocks);
77
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070078 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070079 if (group != sbi->s_groups_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -050080 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070081 input->group, sbi->s_groups_count);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070082 else if (offset != 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -050083 ext4_warning(sb, "Last group not full");
Dave Kleikampac27a0e2006-10-11 01:20:50 -070084 else if (input->reserved_blocks > input->blocks_count / 5)
Eric Sandeen12062dd2010-02-15 14:19:27 -050085 ext4_warning(sb, "Reserved blocks too high (%u)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070086 input->reserved_blocks);
87 else if (free_blocks_count < 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -050088 ext4_warning(sb, "Bad blocks count %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070089 input->blocks_count);
90 else if (!(bh = sb_bread(sb, end - 1)))
Eric Sandeen12062dd2010-02-15 14:19:27 -050091 ext4_warning(sb, "Cannot read last block (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070092 end - 1);
93 else if (outside(input->block_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -050094 ext4_warning(sb, "Block bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070095 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096 else if (outside(input->inode_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -050097 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070098 (unsigned long long)input->inode_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070099 else if (outside(input->inode_table, start, end) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400100 outside(itend - 1, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500101 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700102 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700103 else if (input->inode_bitmap == input->block_bitmap)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500104 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700105 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106 else if (inside(input->block_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500107 ext4_warning(sb, "Block bitmap (%llu) in inode table "
108 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700109 (unsigned long long)input->block_bitmap,
110 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700111 else if (inside(input->inode_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500112 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
113 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700114 (unsigned long long)input->inode_bitmap,
115 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700116 else if (inside(input->block_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500117 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700118 (unsigned long long)input->block_bitmap,
119 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700120 else if (inside(input->inode_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500121 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700122 (unsigned long long)input->inode_bitmap,
123 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700124 else if (inside(input->inode_table, start, metaend) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400125 inside(itend - 1, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500126 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
127 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700128 (unsigned long long)input->inode_table,
129 itend - 1, start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700130 else
131 err = 0;
132 brelse(bh);
133
134 return err;
135}
136
Yongqiang Yang28c7bac2012-01-03 23:22:50 -0500137/*
138 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
139 * group each time.
140 */
141struct ext4_new_flex_group_data {
142 struct ext4_new_group_data *groups; /* new_group_data for groups
143 in the flex group */
144 __u16 *bg_flags; /* block group flags of groups
145 in @groups */
146 ext4_group_t count; /* number of groups in @groups
147 */
148};
149
150/*
151 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
152 * @flexbg_size.
153 *
154 * Returns NULL on failure otherwise address of the allocated structure.
155 */
156static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
157{
158 struct ext4_new_flex_group_data *flex_gd;
159
160 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
161 if (flex_gd == NULL)
162 goto out3;
163
164 flex_gd->count = flexbg_size;
165
166 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
167 flexbg_size, GFP_NOFS);
168 if (flex_gd->groups == NULL)
169 goto out2;
170
171 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
172 if (flex_gd->bg_flags == NULL)
173 goto out1;
174
175 return flex_gd;
176
177out1:
178 kfree(flex_gd->groups);
179out2:
180 kfree(flex_gd);
181out3:
182 return NULL;
183}
184
185static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
186{
187 kfree(flex_gd->bg_flags);
188 kfree(flex_gd->groups);
189 kfree(flex_gd);
190}
191
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700192static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700193 ext4_fsblk_t blk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700194{
195 struct buffer_head *bh;
196 int err;
197
198 bh = sb_getblk(sb, blk);
199 if (!bh)
200 return ERR_PTR(-EIO);
Mingming Cao617ba132006-10-11 01:20:53 -0700201 if ((err = ext4_journal_get_write_access(handle, bh))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700202 brelse(bh);
203 bh = ERR_PTR(err);
204 } else {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700205 memset(bh->b_data, 0, sb->s_blocksize);
206 set_buffer_uptodate(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700207 }
208
209 return bh;
210}
211
212/*
Eric Sandeen14904102007-10-16 18:38:25 -0400213 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
214 * If that fails, restart the transaction & regain write access for the
215 * buffer head which is used for block_bitmap modifications.
216 */
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400217static int extend_or_restart_transaction(handle_t *handle, int thresh)
Eric Sandeen14904102007-10-16 18:38:25 -0400218{
219 int err;
220
Frank Mayhar03901312009-01-07 00:06:22 -0500221 if (ext4_handle_has_enough_credits(handle, thresh))
Eric Sandeen14904102007-10-16 18:38:25 -0400222 return 0;
223
224 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
225 if (err < 0)
226 return err;
227 if (err) {
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400228 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
229 if (err)
Eric Sandeen14904102007-10-16 18:38:25 -0400230 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400231 }
Eric Sandeen14904102007-10-16 18:38:25 -0400232
233 return 0;
234}
235
236/*
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500237 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
238 *
239 * Helper function for ext4_setup_new_group_blocks() which set .
240 *
241 * @sb: super block
242 * @handle: journal handle
243 * @flex_gd: flex group data
244 */
245static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
246 struct ext4_new_flex_group_data *flex_gd,
247 ext4_fsblk_t block, ext4_group_t count)
248{
249 ext4_group_t count2;
250
251 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
252 for (count2 = count; count > 0; count -= count2, block += count2) {
253 ext4_fsblk_t start;
254 struct buffer_head *bh;
255 ext4_group_t group;
256 int err;
257
258 ext4_get_group_no_and_offset(sb, block, &group, NULL);
259 start = ext4_group_first_block_no(sb, group);
260 group -= flex_gd->groups[0].group;
261
262 count2 = sb->s_blocksize * 8 - (block - start);
263 if (count2 > count)
264 count2 = count;
265
266 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
267 BUG_ON(flex_gd->count > 1);
268 continue;
269 }
270
271 err = extend_or_restart_transaction(handle, 1);
272 if (err)
273 return err;
274
275 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
276 if (!bh)
277 return -EIO;
278
279 err = ext4_journal_get_write_access(handle, bh);
280 if (err)
281 return err;
282 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
283 block - start, count2);
284 ext4_set_bits(bh->b_data, block - start, count2);
285
286 err = ext4_handle_dirty_metadata(handle, NULL, bh);
287 if (unlikely(err))
288 return err;
289 brelse(bh);
290 }
291
292 return 0;
293}
294
295/*
296 * Set up the block and inode bitmaps, and the inode table for the new groups.
297 * This doesn't need to be part of the main transaction, since we are only
298 * changing blocks outside the actual filesystem. We still do journaling to
299 * ensure the recovery is correct in case of a failure just after resize.
300 * If any part of this fails, we simply abort the resize.
301 *
302 * setup_new_flex_group_blocks handles a flex group as follow:
303 * 1. copy super block and GDT, and initialize group tables if necessary.
304 * In this step, we only set bits in blocks bitmaps for blocks taken by
305 * super block and GDT.
306 * 2. allocate group tables in block bitmaps, that is, set bits in block
307 * bitmap for blocks taken by group tables.
308 */
309static int setup_new_flex_group_blocks(struct super_block *sb,
310 struct ext4_new_flex_group_data *flex_gd)
311{
312 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
313 ext4_fsblk_t start;
314 ext4_fsblk_t block;
315 struct ext4_sb_info *sbi = EXT4_SB(sb);
316 struct ext4_super_block *es = sbi->s_es;
317 struct ext4_new_group_data *group_data = flex_gd->groups;
318 __u16 *bg_flags = flex_gd->bg_flags;
319 handle_t *handle;
320 ext4_group_t group, count;
321 struct buffer_head *bh = NULL;
322 int reserved_gdb, i, j, err = 0, err2;
323
324 BUG_ON(!flex_gd->count || !group_data ||
325 group_data[0].group != sbi->s_groups_count);
326
327 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
328
329 /* This transaction may be extended/restarted along the way */
330 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
331 if (IS_ERR(handle))
332 return PTR_ERR(handle);
333
334 group = group_data[0].group;
335 for (i = 0; i < flex_gd->count; i++, group++) {
336 unsigned long gdblocks;
337
338 gdblocks = ext4_bg_num_gdb(sb, group);
339 start = ext4_group_first_block_no(sb, group);
340
341 /* Copy all of the GDT blocks into the backup in this group */
342 for (j = 0, block = start + 1; j < gdblocks; j++, block++) {
343 struct buffer_head *gdb;
344
345 ext4_debug("update backup group %#04llx\n", block);
346 err = extend_or_restart_transaction(handle, 1);
347 if (err)
348 goto out;
349
350 gdb = sb_getblk(sb, block);
351 if (!gdb) {
352 err = -EIO;
353 goto out;
354 }
355
356 err = ext4_journal_get_write_access(handle, gdb);
357 if (err) {
358 brelse(gdb);
359 goto out;
360 }
361 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
362 gdb->b_size);
363 set_buffer_uptodate(gdb);
364
365 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
366 if (unlikely(err)) {
367 brelse(gdb);
368 goto out;
369 }
370 brelse(gdb);
371 }
372
373 /* Zero out all of the reserved backup group descriptor
374 * table blocks
375 */
376 if (ext4_bg_has_super(sb, group)) {
377 err = sb_issue_zeroout(sb, gdblocks + start + 1,
378 reserved_gdb, GFP_NOFS);
379 if (err)
380 goto out;
381 }
382
383 /* Initialize group tables of the grop @group */
384 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
385 goto handle_bb;
386
387 /* Zero out all of the inode table blocks */
388 block = group_data[i].inode_table;
389 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
390 block, sbi->s_itb_per_group);
391 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
392 GFP_NOFS);
393 if (err)
394 goto out;
395
396handle_bb:
397 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
398 goto handle_ib;
399
400 /* Initialize block bitmap of the @group */
401 block = group_data[i].block_bitmap;
402 err = extend_or_restart_transaction(handle, 1);
403 if (err)
404 goto out;
405
406 bh = bclean(handle, sb, block);
407 if (IS_ERR(bh)) {
408 err = PTR_ERR(bh);
409 goto out;
410 }
411 if (ext4_bg_has_super(sb, group)) {
412 ext4_debug("mark backup superblock %#04llx (+0)\n",
413 start);
414 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb +
415 1);
416 }
417 ext4_mark_bitmap_end(group_data[i].blocks_count,
418 sb->s_blocksize * 8, bh->b_data);
419 err = ext4_handle_dirty_metadata(handle, NULL, bh);
420 if (err)
421 goto out;
422 brelse(bh);
423
424handle_ib:
425 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
426 continue;
427
428 /* Initialize inode bitmap of the @group */
429 block = group_data[i].inode_bitmap;
430 err = extend_or_restart_transaction(handle, 1);
431 if (err)
432 goto out;
433 /* Mark unused entries in inode bitmap used */
434 bh = bclean(handle, sb, block);
435 if (IS_ERR(bh)) {
436 err = PTR_ERR(bh);
437 goto out;
438 }
439
440 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
441 sb->s_blocksize * 8, bh->b_data);
442 err = ext4_handle_dirty_metadata(handle, NULL, bh);
443 if (err)
444 goto out;
445 brelse(bh);
446 }
447 bh = NULL;
448
449 /* Mark group tables in block bitmap */
450 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
451 count = group_table_count[j];
452 start = (&group_data[0].block_bitmap)[j];
453 block = start;
454 for (i = 1; i < flex_gd->count; i++) {
455 block += group_table_count[j];
456 if (block == (&group_data[i].block_bitmap)[j]) {
457 count += group_table_count[j];
458 continue;
459 }
460 err = set_flexbg_block_bitmap(sb, handle,
461 flex_gd, start, count);
462 if (err)
463 goto out;
464 count = group_table_count[j];
465 start = group_data[i].block_bitmap;
466 block = start;
467 }
468
469 if (count) {
470 err = set_flexbg_block_bitmap(sb, handle,
471 flex_gd, start, count);
472 if (err)
473 goto out;
474 }
475 }
476
477out:
478 brelse(bh);
479 err2 = ext4_journal_stop(handle);
480 if (err2 && !err)
481 err = err2;
482
483 return err;
484}
485
486/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700487 * Set up the block and inode bitmaps, and the inode table for the new group.
488 * This doesn't need to be part of the main transaction, since we are only
489 * changing blocks outside the actual filesystem. We still do journaling to
490 * ensure the recovery is correct in case of a failure just after resize.
491 * If any part of this fails, we simply abort the resize.
492 */
493static int setup_new_group_blocks(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700494 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700495{
Mingming Cao617ba132006-10-11 01:20:53 -0700496 struct ext4_sb_info *sbi = EXT4_SB(sb);
497 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
498 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700499 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700500 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700501 struct buffer_head *bh;
502 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -0700503 ext4_fsblk_t block;
504 ext4_grpblk_t bit;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700505 int i;
506 int err = 0, err2;
507
Eric Sandeen14904102007-10-16 18:38:25 -0400508 /* This transaction may be extended/restarted along the way */
509 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
510
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700511 if (IS_ERR(handle))
512 return PTR_ERR(handle);
513
Yongqiang Yang8f82f842011-07-26 21:35:44 -0400514 BUG_ON(input->group != sbi->s_groups_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700515
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700516 /* Copy all of the GDT blocks into the backup in this group */
517 for (i = 0, bit = 1, block = start + 1;
518 i < gdblocks; i++, block++, bit++) {
519 struct buffer_head *gdb;
520
Eric Sandeenc549a952008-01-28 23:58:27 -0500521 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400522 err = extend_or_restart_transaction(handle, 1);
523 if (err)
524 goto exit_journal;
Eric Sandeen14904102007-10-16 18:38:25 -0400525
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700526 gdb = sb_getblk(sb, block);
527 if (!gdb) {
528 err = -EIO;
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400529 goto exit_journal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700530 }
Mingming Cao617ba132006-10-11 01:20:53 -0700531 if ((err = ext4_journal_get_write_access(handle, gdb))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700532 brelse(gdb);
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400533 goto exit_journal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700534 }
Eric Sandeen5b615282007-10-16 18:38:25 -0400535 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700536 set_buffer_uptodate(gdb);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500537 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
538 if (unlikely(err)) {
539 brelse(gdb);
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400540 goto exit_journal;
Theodore Ts'ob4097142011-01-10 12:46:59 -0500541 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700542 brelse(gdb);
543 }
544
545 /* Zero out all of the reserved backup group descriptor table blocks */
Theodore Ts'oda4889452011-02-21 20:39:58 -0500546 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
Lukas Czernera31437b2010-10-27 21:30:05 -0400547 block, sbi->s_itb_per_group);
548 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
Theodore Ts'oa107e5a2010-10-27 23:44:47 -0400549 GFP_NOFS);
Lukas Czernera31437b2010-10-27 21:30:05 -0400550 if (err)
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400551 goto exit_journal;
552
553 err = extend_or_restart_transaction(handle, 2);
554 if (err)
555 goto exit_journal;
556
557 bh = bclean(handle, sb, input->block_bitmap);
558 if (IS_ERR(bh)) {
559 err = PTR_ERR(bh);
560 goto exit_journal;
561 }
Yongqiang Yangc3e94d12011-07-26 22:05:53 -0400562
563 if (ext4_bg_has_super(sb, input->group)) {
564 ext4_debug("mark backup group tables %#04llx (+0)\n", start);
565 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1);
566 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700567
Eric Sandeenc549a952008-01-28 23:58:27 -0500568 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700569 input->block_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700570 ext4_set_bit(input->block_bitmap - start, bh->b_data);
Eric Sandeenc549a952008-01-28 23:58:27 -0500571 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700572 input->inode_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700573 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700574
575 /* Zero out all of the inode table blocks */
Lukas Czernera31437b2010-10-27 21:30:05 -0400576 block = input->inode_table;
Theodore Ts'oda4889452011-02-21 20:39:58 -0500577 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
Lukas Czernera31437b2010-10-27 21:30:05 -0400578 block, sbi->s_itb_per_group);
Theodore Ts'oa107e5a2010-10-27 23:44:47 -0400579 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
Lukas Czernera31437b2010-10-27 21:30:05 -0400580 if (err)
581 goto exit_bh;
Yongqiang Yangc3e94d12011-07-26 22:05:53 -0400582 ext4_set_bits(bh->b_data, input->inode_table - start,
583 sbi->s_itb_per_group);
Eric Sandeen14904102007-10-16 18:38:25 -0400584
Eric Sandeen14904102007-10-16 18:38:25 -0400585
Theodore Ts'o61d08672010-10-27 21:30:15 -0400586 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
587 bh->b_data);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500588 err = ext4_handle_dirty_metadata(handle, NULL, bh);
589 if (unlikely(err)) {
590 ext4_std_error(sb, err);
591 goto exit_bh;
592 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700593 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700594 /* Mark unused entries in inode bitmap used */
Eric Sandeenc549a952008-01-28 23:58:27 -0500595 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700596 input->inode_bitmap, input->inode_bitmap - start);
597 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
598 err = PTR_ERR(bh);
599 goto exit_journal;
600 }
601
Theodore Ts'o61d08672010-10-27 21:30:15 -0400602 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
603 bh->b_data);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500604 err = ext4_handle_dirty_metadata(handle, NULL, bh);
605 if (unlikely(err))
606 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700607exit_bh:
608 brelse(bh);
609
610exit_journal:
Mingming Cao617ba132006-10-11 01:20:53 -0700611 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700612 err = err2;
613
614 return err;
615}
616
617/*
618 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
Mingming Cao617ba132006-10-11 01:20:53 -0700619 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700620 * calling this for the first time. In a sparse filesystem it will be the
621 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
622 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
623 */
Mingming Cao617ba132006-10-11 01:20:53 -0700624static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700625 unsigned *five, unsigned *seven)
626{
627 unsigned *min = three;
628 int mult = 3;
629 unsigned ret;
630
Mingming Cao617ba132006-10-11 01:20:53 -0700631 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
632 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700633 ret = *min;
634 *min += 1;
635 return ret;
636 }
637
638 if (*five < *min) {
639 min = five;
640 mult = 5;
641 }
642 if (*seven < *min) {
643 min = seven;
644 mult = 7;
645 }
646
647 ret = *min;
648 *min *= mult;
649
650 return ret;
651}
652
653/*
654 * Check that all of the backup GDT blocks are held in the primary GDT block.
655 * It is assumed that they are stored in group order. Returns the number of
656 * groups in current filesystem that have BACKUPS, or -ve error code.
657 */
658static int verify_reserved_gdb(struct super_block *sb,
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500659 ext4_group_t end,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700660 struct buffer_head *primary)
661{
Mingming Cao617ba132006-10-11 01:20:53 -0700662 const ext4_fsblk_t blk = primary->b_blocknr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700663 unsigned three = 1;
664 unsigned five = 5;
665 unsigned seven = 7;
666 unsigned grp;
667 __le32 *p = (__le32 *)primary->b_data;
668 int gdbackups = 0;
669
Mingming Cao617ba132006-10-11 01:20:53 -0700670 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700671 if (le32_to_cpu(*p++) !=
672 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
Eric Sandeen12062dd2010-02-15 14:19:27 -0500673 ext4_warning(sb, "reserved GDT %llu"
Mingming Cao2ae02102006-10-11 01:21:11 -0700674 " missing grp %d (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700675 blk, grp,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700676 grp *
677 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
678 blk);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700679 return -EINVAL;
680 }
Mingming Cao617ba132006-10-11 01:20:53 -0700681 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700682 return -EFBIG;
683 }
684
685 return gdbackups;
686}
687
688/*
689 * Called when we need to bring a reserved group descriptor table block into
690 * use from the resize inode. The primary copy of the new GDT block currently
691 * is an indirect block (under the double indirect block in the resize inode).
692 * The new backup GDT blocks will be stored as leaf blocks in this indirect
693 * block, in group order. Even though we know all the block numbers we need,
694 * we check to ensure that the resize inode has actually reserved these blocks.
695 *
696 * Don't need to update the block bitmaps because the blocks are still in use.
697 *
698 * We get all of the error cases out of the way, so that we are sure to not
699 * fail once we start modifying the data on disk, because JBD has no rollback.
700 */
701static int add_new_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang2f919712011-07-27 21:16:33 -0400702 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700703{
704 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700705 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400706 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700707 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700708 struct buffer_head **o_group_desc, **n_group_desc;
709 struct buffer_head *dind;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400710 struct buffer_head *gdb_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700711 int gdbackups;
Mingming Cao617ba132006-10-11 01:20:53 -0700712 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700713 __le32 *data;
714 int err;
715
716 if (test_opt(sb, DEBUG))
717 printk(KERN_DEBUG
Mingming Cao617ba132006-10-11 01:20:53 -0700718 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700719 gdb_num);
720
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400721 /*
722 * If we are not using the primary superblock/GDT copy don't resize,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400723 * because the user tools have no way of handling this. Probably a
724 * bad time to do it anyways.
725 */
Mingming Cao617ba132006-10-11 01:20:53 -0700726 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
727 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500728 ext4_warning(sb, "won't resize using backup superblock at %llu",
Mingming Cao617ba132006-10-11 01:20:53 -0700729 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700730 return -EPERM;
731 }
732
Yongqiang Yang2f919712011-07-27 21:16:33 -0400733 gdb_bh = sb_bread(sb, gdblock);
734 if (!gdb_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700735 return -EIO;
736
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500737 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
Yongqiang Yang2f919712011-07-27 21:16:33 -0400738 if (gdbackups < 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700739 err = gdbackups;
740 goto exit_bh;
741 }
742
Mingming Cao617ba132006-10-11 01:20:53 -0700743 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700744 dind = sb_bread(sb, le32_to_cpu(*data));
745 if (!dind) {
746 err = -EIO;
747 goto exit_bh;
748 }
749
750 data = (__le32 *)dind->b_data;
Mingming Cao617ba132006-10-11 01:20:53 -0700751 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500752 ext4_warning(sb, "new group %u GDT block %llu not reserved",
Yongqiang Yang2f919712011-07-27 21:16:33 -0400753 group, gdblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700754 err = -EINVAL;
755 goto exit_dind;
756 }
757
Theodore Ts'ob4097142011-01-10 12:46:59 -0500758 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
759 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700760 goto exit_dind;
761
Yongqiang Yang2f919712011-07-27 21:16:33 -0400762 err = ext4_journal_get_write_access(handle, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500763 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700764 goto exit_sbh;
765
Theodore Ts'ob4097142011-01-10 12:46:59 -0500766 err = ext4_journal_get_write_access(handle, dind);
767 if (unlikely(err))
768 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700769
Mingming Cao617ba132006-10-11 01:20:53 -0700770 /* ext4_reserve_inode_write() gets a reference on the iloc */
Theodore Ts'ob4097142011-01-10 12:46:59 -0500771 err = ext4_reserve_inode_write(handle, inode, &iloc);
772 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700773 goto exit_dindj;
774
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400775 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
776 sizeof(struct buffer_head *),
777 GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700778 if (!n_group_desc) {
779 err = -ENOMEM;
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400780 ext4_warning(sb, "not enough memory for %lu groups",
781 gdb_num + 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700782 goto exit_inode;
783 }
784
785 /*
786 * Finally, we have all of the possible failures behind us...
787 *
788 * Remove new GDT block from inode double-indirect block and clear out
789 * the new GDT block for use (which also "frees" the backup GDT blocks
790 * from the reserved inode). We don't need to change the bitmaps for
791 * these blocks, because they are marked as in-use from being in the
792 * reserved inode, and will become GDT blocks (primary and backup).
793 */
Mingming Cao617ba132006-10-11 01:20:53 -0700794 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
Theodore Ts'ob4097142011-01-10 12:46:59 -0500795 err = ext4_handle_dirty_metadata(handle, NULL, dind);
796 if (unlikely(err)) {
797 ext4_std_error(sb, err);
798 goto exit_inode;
799 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700800 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700801 ext4_mark_iloc_dirty(handle, inode, &iloc);
Yongqiang Yang2f919712011-07-27 21:16:33 -0400802 memset(gdb_bh->b_data, 0, sb->s_blocksize);
803 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500804 if (unlikely(err)) {
805 ext4_std_error(sb, err);
806 goto exit_inode;
807 }
808 brelse(dind);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700809
Mingming Cao617ba132006-10-11 01:20:53 -0700810 o_group_desc = EXT4_SB(sb)->s_group_desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700811 memcpy(n_group_desc, o_group_desc,
Mingming Cao617ba132006-10-11 01:20:53 -0700812 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Yongqiang Yang2f919712011-07-27 21:16:33 -0400813 n_group_desc[gdb_num] = gdb_bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700814 EXT4_SB(sb)->s_group_desc = n_group_desc;
815 EXT4_SB(sb)->s_gdb_count++;
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400816 ext4_kvfree(o_group_desc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700817
Marcin Slusarze8546d02008-04-17 10:38:59 -0400818 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500819 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
820 if (err)
821 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700822
Theodore Ts'ob4097142011-01-10 12:46:59 -0500823 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700824
825exit_inode:
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400826 ext4_kvfree(n_group_desc);
Amir Goldstein537a0312011-03-20 22:57:02 -0400827 /* ext4_handle_release_buffer(handle, iloc.bh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700828 brelse(iloc.bh);
829exit_dindj:
Amir Goldstein537a0312011-03-20 22:57:02 -0400830 /* ext4_handle_release_buffer(handle, dind); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831exit_sbh:
Amir Goldstein537a0312011-03-20 22:57:02 -0400832 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700833exit_dind:
834 brelse(dind);
835exit_bh:
Yongqiang Yang2f919712011-07-27 21:16:33 -0400836 brelse(gdb_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700837
Mingming Cao617ba132006-10-11 01:20:53 -0700838 ext4_debug("leaving with error %d\n", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700839 return err;
840}
841
842/*
843 * Called when we are adding a new group which has a backup copy of each of
844 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
845 * We need to add these reserved backup GDT blocks to the resize inode, so
846 * that they are kept for future resizing and not allocated to files.
847 *
848 * Each reserved backup GDT block will go into a different indirect block.
849 * The indirect blocks are actually the primary reserved GDT blocks,
850 * so we know in advance what their block numbers are. We only get the
851 * double-indirect block to verify it is pointing to the primary reserved
852 * GDT blocks so we don't overwrite a data block by accident. The reserved
853 * backup GDT blocks are stored in their reserved primary GDT block.
854 */
855static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang668f4dc2011-07-27 21:23:13 -0400856 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700857{
858 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700859 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700860 struct buffer_head **primary;
861 struct buffer_head *dind;
Mingming Cao617ba132006-10-11 01:20:53 -0700862 struct ext4_iloc iloc;
863 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700864 __le32 *data, *end;
865 int gdbackups = 0;
866 int res, i;
867 int err;
868
Josef Bacik216553c2008-04-29 22:02:02 -0400869 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700870 if (!primary)
871 return -ENOMEM;
872
Mingming Cao617ba132006-10-11 01:20:53 -0700873 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700874 dind = sb_bread(sb, le32_to_cpu(*data));
875 if (!dind) {
876 err = -EIO;
877 goto exit_free;
878 }
879
Mingming Cao617ba132006-10-11 01:20:53 -0700880 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
Josef Bacik94460092008-06-06 18:05:52 -0400881 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
882 EXT4_ADDR_PER_BLOCK(sb));
Mingming Cao617ba132006-10-11 01:20:53 -0700883 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700884
885 /* Get each reserved primary GDT block and verify it holds backups */
886 for (res = 0; res < reserved_gdb; res++, blk++) {
887 if (le32_to_cpu(*data) != blk) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500888 ext4_warning(sb, "reserved block %llu"
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700889 " not at offset %ld",
890 blk,
891 (long)(data - (__le32 *)dind->b_data));
892 err = -EINVAL;
893 goto exit_bh;
894 }
895 primary[res] = sb_bread(sb, blk);
896 if (!primary[res]) {
897 err = -EIO;
898 goto exit_bh;
899 }
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500900 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
901 if (gdbackups < 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700902 brelse(primary[res]);
903 err = gdbackups;
904 goto exit_bh;
905 }
906 if (++data >= end)
907 data = (__le32 *)dind->b_data;
908 }
909
910 for (i = 0; i < reserved_gdb; i++) {
Mingming Cao617ba132006-10-11 01:20:53 -0700911 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700912 /*
913 int j;
914 for (j = 0; j < i; j++)
Amir Goldstein537a0312011-03-20 22:57:02 -0400915 ext4_handle_release_buffer(handle, primary[j]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700916 */
917 goto exit_bh;
918 }
919 }
920
Mingming Cao617ba132006-10-11 01:20:53 -0700921 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700922 goto exit_bh;
923
924 /*
925 * Finally we can add each of the reserved backup GDT blocks from
926 * the new group to its reserved primary GDT block.
927 */
Yongqiang Yang668f4dc2011-07-27 21:23:13 -0400928 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700929 for (i = 0; i < reserved_gdb; i++) {
930 int err2;
931 data = (__le32 *)primary[i]->b_data;
932 /* printk("reserving backup %lu[%u] = %lu\n",
933 primary[i]->b_blocknr, gdbackups,
934 blk + primary[i]->b_blocknr); */
935 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
Frank Mayhar03901312009-01-07 00:06:22 -0500936 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700937 if (!err)
938 err = err2;
939 }
940 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700941 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700942
943exit_bh:
944 while (--res >= 0)
945 brelse(primary[res]);
946 brelse(dind);
947
948exit_free:
949 kfree(primary);
950
951 return err;
952}
953
954/*
Mingming Cao617ba132006-10-11 01:20:53 -0700955 * Update the backup copies of the ext4 metadata. These don't need to be part
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700956 * of the main resize transaction, because e2fsck will re-write them if there
957 * is a problem (basically only OOM will cause a problem). However, we
958 * _should_ update the backups if possible, in case the primary gets trashed
959 * for some reason and we need to run e2fsck from a backup superblock. The
960 * important part is that the new block and inode counts are in the backup
961 * superblocks, and the location of the new group metadata in the GDT backups.
962 *
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400963 * We do not need take the s_resize_lock for this, because these
964 * blocks are not otherwise touched by the filesystem code when it is
965 * mounted. We don't need to worry about last changing from
966 * sbi->s_groups_count, because the worst that can happen is that we
967 * do not copy the full number of backups at this time. The resize
968 * which changed s_groups_count will backup again.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700969 */
970static void update_backups(struct super_block *sb,
971 int blk_off, char *data, int size)
972{
Mingming Cao617ba132006-10-11 01:20:53 -0700973 struct ext4_sb_info *sbi = EXT4_SB(sb);
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500974 const ext4_group_t last = sbi->s_groups_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700975 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700976 unsigned three = 1;
977 unsigned five = 5;
978 unsigned seven = 7;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500979 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700980 int rest = sb->s_blocksize - size;
981 handle_t *handle;
982 int err = 0, err2;
983
Mingming Cao617ba132006-10-11 01:20:53 -0700984 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700985 if (IS_ERR(handle)) {
986 group = 1;
987 err = PTR_ERR(handle);
988 goto exit_err;
989 }
990
Mingming Cao617ba132006-10-11 01:20:53 -0700991 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700992 struct buffer_head *bh;
993
994 /* Out of journal space, and can't get more - abort - so sad */
Frank Mayhar03901312009-01-07 00:06:22 -0500995 if (ext4_handle_valid(handle) &&
996 handle->h_buffer_credits == 0 &&
Mingming Cao617ba132006-10-11 01:20:53 -0700997 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
998 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700999 break;
1000
1001 bh = sb_getblk(sb, group * bpg + blk_off);
1002 if (!bh) {
1003 err = -EIO;
1004 break;
1005 }
Mingming Cao617ba132006-10-11 01:20:53 -07001006 ext4_debug("update metadata backup %#04lx\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001007 (unsigned long)bh->b_blocknr);
Mingming Cao617ba132006-10-11 01:20:53 -07001008 if ((err = ext4_journal_get_write_access(handle, bh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001009 break;
1010 lock_buffer(bh);
1011 memcpy(bh->b_data, data, size);
1012 if (rest)
1013 memset(bh->b_data + size, 0, rest);
1014 set_buffer_uptodate(bh);
1015 unlock_buffer(bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -05001016 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1017 if (unlikely(err))
1018 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001019 brelse(bh);
1020 }
Mingming Cao617ba132006-10-11 01:20:53 -07001021 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001022 err = err2;
1023
1024 /*
1025 * Ugh! Need to have e2fsck write the backup copies. It is too
1026 * late to revert the resize, we shouldn't fail just because of
1027 * the backup copies (they are only needed in case of corruption).
1028 *
1029 * However, if we got here we have a journal problem too, so we
1030 * can't really start a transaction to mark the superblock.
1031 * Chicken out and just set the flag on the hope it will be written
1032 * to disk, and if not - we will simply wait until next fsck.
1033 */
1034exit_err:
1035 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001036 ext4_warning(sb, "can't update backup for group %u (err %d), "
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001037 "forcing fsck on next reboot", group, err);
Mingming Cao617ba132006-10-11 01:20:53 -07001038 sbi->s_mount_state &= ~EXT4_VALID_FS;
1039 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001040 mark_buffer_dirty(sbi->s_sbh);
1041 }
1042}
1043
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001044/*
1045 * ext4_add_new_descs() adds @count group descriptor of groups
1046 * starting at @group
1047 *
1048 * @handle: journal handle
1049 * @sb: super block
1050 * @group: the group no. of the first group desc to be added
1051 * @resize_inode: the resize inode
1052 * @count: number of group descriptors to be added
1053 */
1054static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1055 ext4_group_t group, struct inode *resize_inode,
1056 ext4_group_t count)
1057{
1058 struct ext4_sb_info *sbi = EXT4_SB(sb);
1059 struct ext4_super_block *es = sbi->s_es;
1060 struct buffer_head *gdb_bh;
1061 int i, gdb_off, gdb_num, err = 0;
1062
1063 for (i = 0; i < count; i++, group++) {
1064 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1065 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1066
1067 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1068 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1069
1070 /*
1071 * We will only either add reserved group blocks to a backup group
1072 * or remove reserved blocks for the first group in a new group block.
1073 * Doing both would be mean more complex code, and sane people don't
1074 * use non-sparse filesystems anymore. This is already checked above.
1075 */
1076 if (gdb_off) {
1077 gdb_bh = sbi->s_group_desc[gdb_num];
1078 err = ext4_journal_get_write_access(handle, gdb_bh);
1079
1080 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1081 err = reserve_backup_gdb(handle, resize_inode, group);
1082 } else
1083 err = add_new_gdb(handle, resize_inode, group);
1084 if (err)
1085 break;
1086 }
1087 return err;
1088}
1089
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001090/*
1091 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1092 */
1093static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1094 struct ext4_new_flex_group_data *flex_gd)
1095{
1096 struct ext4_new_group_data *group_data = flex_gd->groups;
1097 struct ext4_group_desc *gdp;
1098 struct ext4_sb_info *sbi = EXT4_SB(sb);
1099 struct buffer_head *gdb_bh;
1100 ext4_group_t group;
1101 __u16 *bg_flags = flex_gd->bg_flags;
1102 int i, gdb_off, gdb_num, err = 0;
1103
1104
1105 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1106 group = group_data->group;
1107
1108 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1109 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1110
1111 /*
1112 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1113 */
1114 gdb_bh = sbi->s_group_desc[gdb_num];
1115 /* Update group descriptor block for new group */
1116 gdp = (struct ext4_group_desc *)((char *)gdb_bh->b_data +
1117 gdb_off * EXT4_DESC_SIZE(sb));
1118
1119 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1120 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1121 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1122 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1123 ext4_free_group_clusters_set(sb, gdp,
1124 EXT4_B2C(sbi, group_data->free_blocks_count));
1125 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1126 gdp->bg_flags = cpu_to_le16(*bg_flags);
1127 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1128
1129 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1130 if (unlikely(err)) {
1131 ext4_std_error(sb, err);
1132 break;
1133 }
1134
1135 /*
1136 * We can allocate memory for mb_alloc based on the new group
1137 * descriptor
1138 */
1139 err = ext4_mb_add_groupinfo(sb, group, gdp);
1140 if (err)
1141 break;
1142 }
1143 return err;
1144}
1145
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001146/*
1147 * ext4_update_super() updates the super block so that the newly added
1148 * groups can be seen by the filesystem.
1149 *
1150 * @sb: super block
1151 * @flex_gd: new added groups
1152 */
1153static void ext4_update_super(struct super_block *sb,
1154 struct ext4_new_flex_group_data *flex_gd)
1155{
1156 ext4_fsblk_t blocks_count = 0;
1157 ext4_fsblk_t free_blocks = 0;
1158 ext4_fsblk_t reserved_blocks = 0;
1159 struct ext4_new_group_data *group_data = flex_gd->groups;
1160 struct ext4_sb_info *sbi = EXT4_SB(sb);
1161 struct ext4_super_block *es = sbi->s_es;
1162 int i;
1163
1164 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1165 /*
1166 * Make the new blocks and inodes valid next. We do this before
1167 * increasing the group count so that once the group is enabled,
1168 * all of its blocks and inodes are already valid.
1169 *
1170 * We always allocate group-by-group, then block-by-block or
1171 * inode-by-inode within a group, so enabling these
1172 * blocks/inodes before the group is live won't actually let us
1173 * allocate the new space yet.
1174 */
1175 for (i = 0; i < flex_gd->count; i++) {
1176 blocks_count += group_data[i].blocks_count;
1177 free_blocks += group_data[i].free_blocks_count;
1178 }
1179
1180 reserved_blocks = ext4_r_blocks_count(es) * 100;
1181 do_div(reserved_blocks, ext4_blocks_count(es));
1182 reserved_blocks *= blocks_count;
1183 do_div(reserved_blocks, 100);
1184
1185 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1186 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1187 flex_gd->count);
1188
1189 /*
1190 * We need to protect s_groups_count against other CPUs seeing
1191 * inconsistent state in the superblock.
1192 *
1193 * The precise rules we use are:
1194 *
1195 * * Writers must perform a smp_wmb() after updating all
1196 * dependent data and before modifying the groups count
1197 *
1198 * * Readers must perform an smp_rmb() after reading the groups
1199 * count and before reading any dependent data.
1200 *
1201 * NB. These rules can be relaxed when checking the group count
1202 * while freeing data, as we can only allocate from a block
1203 * group after serialising against the group count, and we can
1204 * only then free after serialising in turn against that
1205 * allocation.
1206 */
1207 smp_wmb();
1208
1209 /* Update the global fs size fields */
1210 sbi->s_groups_count += flex_gd->count;
1211
1212 /* Update the reserved block counts only once the new group is
1213 * active. */
1214 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1215 reserved_blocks);
1216
1217 /* Update the free space counts */
1218 percpu_counter_add(&sbi->s_freeclusters_counter,
1219 EXT4_B2C(sbi, free_blocks));
1220 percpu_counter_add(&sbi->s_freeinodes_counter,
1221 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1222
1223 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1224 EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1225 sbi->s_log_groups_per_flex) {
1226 ext4_group_t flex_group;
1227 flex_group = ext4_flex_group(sbi, group_data[0].group);
1228 atomic_add(EXT4_B2C(sbi, free_blocks),
1229 &sbi->s_flex_groups[flex_group].free_clusters);
1230 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1231 &sbi->s_flex_groups[flex_group].free_inodes);
1232 }
1233
1234 if (test_opt(sb, DEBUG))
1235 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1236 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1237 blocks_count, free_blocks, reserved_blocks);
1238}
1239
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001240/* Add group descriptor data to an existing or new group descriptor block.
1241 * Ensure we handle all possible error conditions _before_ we start modifying
1242 * the filesystem, because we cannot abort the transaction and not have it
1243 * write the data to disk.
1244 *
1245 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1246 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1247 *
1248 * We only need to hold the superblock lock while we are actually adding
1249 * in the new group's counts to the superblock. Prior to that we have
1250 * not really "added" the group at all. We re-check that we are still
1251 * adding in the last group in case things have changed since verifying.
1252 */
Mingming Cao617ba132006-10-11 01:20:53 -07001253int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001254{
Mingming Cao617ba132006-10-11 01:20:53 -07001255 struct ext4_sb_info *sbi = EXT4_SB(sb);
1256 struct ext4_super_block *es = sbi->s_es;
1257 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001258 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1259 struct buffer_head *primary = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001260 struct ext4_group_desc *gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001261 struct inode *inode = NULL;
1262 handle_t *handle;
1263 int gdb_off, gdb_num;
1264 int err, err2;
1265
Mingming Cao617ba132006-10-11 01:20:53 -07001266 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
1267 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001268
Mingming Cao617ba132006-10-11 01:20:53 -07001269 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1270 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001271 ext4_warning(sb, "Can't resize non-sparse filesystem further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001272 return -EPERM;
1273 }
1274
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001275 if (ext4_blocks_count(es) + input->blocks_count <
1276 ext4_blocks_count(es)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001277 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001278 return -EINVAL;
1279 }
1280
Mingming Cao617ba132006-10-11 01:20:53 -07001281 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001282 le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001283 ext4_warning(sb, "inodes_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001284 return -EINVAL;
1285 }
1286
1287 if (reserved_gdb || gdb_off == 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001288 if (!EXT4_HAS_COMPAT_FEATURE(sb,
Josef Bacik37609fd2008-08-19 22:13:41 -04001289 EXT4_FEATURE_COMPAT_RESIZE_INODE)
1290 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001291 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001292 "No reserved GDT blocks, can't resize");
1293 return -EPERM;
1294 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001295 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1296 if (IS_ERR(inode)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001297 ext4_warning(sb, "Error opening resize inode");
David Howells1d1fe1e2008-02-07 00:15:37 -08001298 return PTR_ERR(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001299 }
1300 }
1301
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -05001302
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001303 if ((err = verify_group_input(sb, input)))
1304 goto exit_put;
1305
1306 if ((err = setup_new_group_blocks(sb, input)))
1307 goto exit_put;
1308
1309 /*
1310 * We will always be modifying at least the superblock and a GDT
1311 * block. If we are adding a group past the last current GDT block,
1312 * we will also modify the inode and the dindirect block. If we
1313 * are adding a group with superblock/GDT backups we will also
1314 * modify each of the reserved GDT dindirect blocks.
1315 */
Mingming Cao617ba132006-10-11 01:20:53 -07001316 handle = ext4_journal_start_sb(sb,
1317 ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001318 3 + reserved_gdb : 4);
1319 if (IS_ERR(handle)) {
1320 err = PTR_ERR(handle);
1321 goto exit_put;
1322 }
1323
Mingming Cao617ba132006-10-11 01:20:53 -07001324 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001325 goto exit_journal;
1326
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001327 /*
1328 * We will only either add reserved group blocks to a backup group
1329 * or remove reserved blocks for the first group in a new group block.
1330 * Doing both would be mean more complex code, and sane people don't
1331 * use non-sparse filesystems anymore. This is already checked above.
1332 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001333 if (gdb_off) {
1334 primary = sbi->s_group_desc[gdb_num];
Mingming Cao617ba132006-10-11 01:20:53 -07001335 if ((err = ext4_journal_get_write_access(handle, primary)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001336 goto exit_journal;
1337
Yongqiang Yang668f4dc2011-07-27 21:23:13 -04001338 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group)) {
1339 err = reserve_backup_gdb(handle, inode, input->group);
1340 if (err)
1341 goto exit_journal;
1342 }
Yongqiang Yang2f919712011-07-27 21:16:33 -04001343 } else {
1344 /*
1345 * Note that we can access new group descriptor block safely
1346 * only if add_new_gdb() succeeds.
1347 */
1348 err = add_new_gdb(handle, inode, input->group);
1349 if (err)
1350 goto exit_journal;
1351 primary = sbi->s_group_desc[gdb_num];
1352 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001353
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001354 /*
1355 * OK, now we've set up the new group. Time to make it active.
1356 *
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001357 * so we have to be safe wrt. concurrent accesses the group
1358 * data. So we need to be careful to set all of the relevant
1359 * group descriptor data etc. *before* we enable the group.
1360 *
1361 * The key field here is sbi->s_groups_count: as long as
1362 * that retains its old value, nobody is going to access the new
1363 * group.
1364 *
1365 * So first we update all the descriptor metadata for the new
1366 * group; then we update the total disk blocks count; then we
1367 * update the groups count to enable the group; then finally we
1368 * update the free space counts so that the system can start
1369 * using the new disk blocks.
1370 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001371
1372 /* Update group descriptor block for new group */
Frederic Bohe28569222008-06-20 11:48:48 -04001373 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
1374 gdb_off * EXT4_DESC_SIZE(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001375
Theodore Ts'ofdff73f2009-01-26 19:06:41 -05001376 memset(gdp, 0, EXT4_DESC_SIZE(sb));
Alexandre Ratchov8fadc142006-10-11 01:21:15 -07001377 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
1378 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
1379 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
Theodore Ts'o021b65b2011-09-09 19:08:51 -04001380 ext4_free_group_clusters_set(sb, gdp, input->free_blocks_count);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001381 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
Theodore Ts'ofdff73f2009-01-26 19:06:41 -05001382 gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
Andreas Dilger717d50e2007-10-16 18:38:25 -04001383 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001384
1385 /*
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001386 * We can allocate memory for mb_alloc based on the new group
1387 * descriptor
1388 */
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -05001389 err = ext4_mb_add_groupinfo(sb, input->group, gdp);
Aneesh Kumar K.V08c3a812009-09-09 23:50:17 -04001390 if (err)
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04001391 goto exit_journal;
1392
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001393 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001394 * Make the new blocks and inodes valid next. We do this before
1395 * increasing the group count so that once the group is enabled,
1396 * all of its blocks and inodes are already valid.
1397 *
1398 * We always allocate group-by-group, then block-by-block or
1399 * inode-by-inode within a group, so enabling these
1400 * blocks/inodes before the group is live won't actually let us
1401 * allocate the new space yet.
1402 */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001403 ext4_blocks_count_set(es, ext4_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001404 input->blocks_count);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001405 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001406
1407 /*
1408 * We need to protect s_groups_count against other CPUs seeing
1409 * inconsistent state in the superblock.
1410 *
1411 * The precise rules we use are:
1412 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001413 * * Writers must perform a smp_wmb() after updating all dependent
1414 * data and before modifying the groups count
1415 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001416 * * Readers must perform an smp_rmb() after reading the groups count
1417 * and before reading any dependent data.
1418 *
1419 * NB. These rules can be relaxed when checking the group count
1420 * while freeing data, as we can only allocate from a block
1421 * group after serialising against the group count, and we can
1422 * only then free after serialising in turn against that
1423 * allocation.
1424 */
1425 smp_wmb();
1426
1427 /* Update the global fs size fields */
1428 sbi->s_groups_count++;
1429
Theodore Ts'ob4097142011-01-10 12:46:59 -05001430 err = ext4_handle_dirty_metadata(handle, NULL, primary);
1431 if (unlikely(err)) {
1432 ext4_std_error(sb, err);
1433 goto exit_journal;
1434 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001435
1436 /* Update the reserved block counts only once the new group is
1437 * active. */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001438 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001439 input->reserved_blocks);
1440
1441 /* Update the free space counts */
Theodore Ts'o57042652011-09-09 18:56:51 -04001442 percpu_counter_add(&sbi->s_freeclusters_counter,
1443 EXT4_B2C(sbi, input->free_blocks_count));
Peter Zijlstraaa0dff22007-10-16 23:25:42 -07001444 percpu_counter_add(&sbi->s_freeinodes_counter,
Mingming Cao617ba132006-10-11 01:20:53 -07001445 EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001446
Eric Sandeen42007ef2010-05-16 01:00:00 -04001447 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1448 sbi->s_log_groups_per_flex) {
Frederic Bohec62a11f2008-09-08 10:20:24 -04001449 ext4_group_t flex_group;
1450 flex_group = ext4_flex_group(sbi, input->group);
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -04001451 atomic_add(EXT4_B2C(sbi, input->free_blocks_count),
1452 &sbi->s_flex_groups[flex_group].free_clusters);
Theodore Ts'o9f24e422009-03-04 19:09:10 -05001453 atomic_add(EXT4_INODES_PER_GROUP(sb),
1454 &sbi->s_flex_groups[flex_group].free_inodes);
Frederic Bohec62a11f2008-09-08 10:20:24 -04001455 }
1456
Theodore Ts'oa0375152010-06-11 23:14:04 -04001457 ext4_handle_dirty_super(handle, sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001458
1459exit_journal:
Mingming Cao617ba132006-10-11 01:20:53 -07001460 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001461 err = err2;
Yongqiang Yang2f919712011-07-27 21:16:33 -04001462 if (!err && primary) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001463 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
Mingming Cao617ba132006-10-11 01:20:53 -07001464 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001465 update_backups(sb, primary->b_blocknr, primary->b_data,
1466 primary->b_size);
1467 }
1468exit_put:
1469 iput(inode);
1470 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001471} /* ext4_group_add */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001472
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001473/*
Yongqiang Yang18e31432012-01-03 23:18:50 -05001474 * extend a group without checking assuming that checking has been done.
1475 */
1476static int ext4_group_extend_no_check(struct super_block *sb,
1477 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1478{
1479 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1480 handle_t *handle;
1481 int err = 0, err2;
1482
1483 /* We will update the superblock, one block bitmap, and
1484 * one group descriptor via ext4_group_add_blocks().
1485 */
1486 handle = ext4_journal_start_sb(sb, 3);
1487 if (IS_ERR(handle)) {
1488 err = PTR_ERR(handle);
1489 ext4_warning(sb, "error %d on journal start", err);
1490 return err;
1491 }
1492
1493 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1494 if (err) {
1495 ext4_warning(sb, "error %d on journal write access", err);
1496 goto errout;
1497 }
1498
1499 ext4_blocks_count_set(es, o_blocks_count + add);
1500 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1501 o_blocks_count + add);
1502 /* We add the blocks to the bitmap and set the group need init bit */
1503 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1504 if (err)
1505 goto errout;
1506 ext4_handle_dirty_super(handle, sb);
1507 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1508 o_blocks_count + add);
1509errout:
1510 err2 = ext4_journal_stop(handle);
1511 if (err2 && !err)
1512 err = err2;
1513
1514 if (!err) {
1515 if (test_opt(sb, DEBUG))
1516 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1517 "blocks\n", ext4_blocks_count(es));
1518 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1519 sizeof(struct ext4_super_block));
1520 }
1521 return err;
1522}
1523
1524/*
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001525 * Extend the filesystem to the new number of blocks specified. This entry
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001526 * point is only used to extend the current filesystem to the end of the last
1527 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1528 * for emergencies (because it has no dependencies on reserved blocks).
1529 *
Mingming Cao617ba132006-10-11 01:20:53 -07001530 * If we _really_ wanted, we could use default values to call ext4_group_add()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001531 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1532 * GDT blocks are reserved to grow to the desired size.
1533 */
Mingming Cao617ba132006-10-11 01:20:53 -07001534int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1535 ext4_fsblk_t n_blocks_count)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001536{
Mingming Cao617ba132006-10-11 01:20:53 -07001537 ext4_fsblk_t o_blocks_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001538 ext4_grpblk_t last;
1539 ext4_grpblk_t add;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001540 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001541 handle_t *handle;
Yongqiang Yangcc7365d2011-07-26 21:46:07 -04001542 int err, err2;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001543 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001544
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001545 o_blocks_count = ext4_blocks_count(es);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001546
1547 if (test_opt(sb, DEBUG))
Yongqiang Yang2b79b092011-07-26 21:53:35 -04001548 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001549 o_blocks_count, n_blocks_count);
1550
1551 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1552 return 0;
1553
1554 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001555 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
Mingming Cao2ae02102006-10-11 01:21:11 -07001556 " too large to resize to %llu blocks safely\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001557 sb->s_id, n_blocks_count);
1558 if (sizeof(sector_t) < 8)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001559 ext4_warning(sb, "CONFIG_LBDAF not enabled");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001560 return -EINVAL;
1561 }
1562
1563 if (n_blocks_count < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001564 ext4_warning(sb, "can't shrink FS - resize aborted");
Yongqiang Yang8f82f842011-07-26 21:35:44 -04001565 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001566 }
1567
1568 /* Handle the remaining blocks in the last group only. */
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001569 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001570
1571 if (last == 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001572 ext4_warning(sb, "need to use ext2online to resize further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001573 return -EPERM;
1574 }
1575
Mingming Cao617ba132006-10-11 01:20:53 -07001576 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001577
1578 if (o_blocks_count + add < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001579 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001580 return -EINVAL;
1581 }
1582
1583 if (o_blocks_count + add > n_blocks_count)
1584 add = n_blocks_count - o_blocks_count;
1585
1586 if (o_blocks_count + add < n_blocks_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001587 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001588 o_blocks_count + add, add);
1589
1590 /* See if the device is actually as big as what was requested */
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001591 bh = sb_bread(sb, o_blocks_count + add - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001592 if (!bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001593 ext4_warning(sb, "can't read last block, resize aborted");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001594 return -ENOSPC;
1595 }
1596 brelse(bh);
1597
1598 /* We will update the superblock, one block bitmap, and
Mingming Cao617ba132006-10-11 01:20:53 -07001599 * one group descriptor via ext4_free_blocks().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001600 */
Mingming Cao617ba132006-10-11 01:20:53 -07001601 handle = ext4_journal_start_sb(sb, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001602 if (IS_ERR(handle)) {
1603 err = PTR_ERR(handle);
Eric Sandeen12062dd2010-02-15 14:19:27 -05001604 ext4_warning(sb, "error %d on journal start", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001605 goto exit_put;
1606 }
1607
Mingming Cao617ba132006-10-11 01:20:53 -07001608 if ((err = ext4_journal_get_write_access(handle,
1609 EXT4_SB(sb)->s_sbh))) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001610 ext4_warning(sb, "error %d on journal write access", err);
Mingming Cao617ba132006-10-11 01:20:53 -07001611 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001612 goto exit_put;
1613 }
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001614 ext4_blocks_count_set(es, o_blocks_count + add);
Eric Sandeenc549a952008-01-28 23:58:27 -05001615 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001616 o_blocks_count + add);
Aneesh Kumar K.Ve21675d2009-01-05 21:36:02 -05001617 /* We add the blocks to the bitmap and set the group need init bit */
Yongqiang Yangcc7365d2011-07-26 21:46:07 -04001618 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
Theodore Ts'oa0375152010-06-11 23:14:04 -04001619 ext4_handle_dirty_super(handle, sb);
Mingming Cao2ae02102006-10-11 01:21:11 -07001620 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001621 o_blocks_count + add);
Yongqiang Yangcc7365d2011-07-26 21:46:07 -04001622 err2 = ext4_journal_stop(handle);
1623 if (!err && err2)
1624 err = err2;
1625
1626 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001627 goto exit_put;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001628
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001629 if (test_opt(sb, DEBUG))
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001630 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1631 ext4_blocks_count(es));
Mingming Cao617ba132006-10-11 01:20:53 -07001632 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1633 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001634exit_put:
1635 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001636} /* ext4_group_extend */