Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2 | * linux/fs/ext4/balloc.c |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3 | * |
| 4 | * Copyright (C) 1992, 1993, 1994, 1995 |
| 5 | * Remy Card (card@masi.ibp.fr) |
| 6 | * Laboratoire MASI - Institut Blaise Pascal |
| 7 | * Universite Pierre et Marie Curie (Paris VI) |
| 8 | * |
| 9 | * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993 |
| 10 | * Big-endian to little-endian byte-swapping/bitmaps by |
| 11 | * David S. Miller (davem@caip.rutgers.edu), 1995 |
| 12 | */ |
| 13 | |
| 14 | #include <linux/time.h> |
| 15 | #include <linux/capability.h> |
| 16 | #include <linux/fs.h> |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 17 | #include <linux/jbd2.h> |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 18 | #include <linux/quotaops.h> |
| 19 | #include <linux/buffer_head.h> |
Christoph Hellwig | 3dcf545 | 2008-04-29 18:13:32 -0400 | [diff] [blame] | 20 | #include "ext4.h" |
| 21 | #include "ext4_jbd2.h" |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 22 | #include "group.h" |
Christoph Hellwig | 3dcf545 | 2008-04-29 18:13:32 -0400 | [diff] [blame] | 23 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 24 | /* |
| 25 | * balloc.c contains the blocks allocation and deallocation routines |
| 26 | */ |
| 27 | |
| 28 | /* |
Andrew Morton | 72b64b5 | 2006-10-11 01:21:18 -0700 | [diff] [blame] | 29 | * Calculate the block group number and offset, given a block number |
| 30 | */ |
| 31 | void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr, |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 32 | ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp) |
Andrew Morton | 72b64b5 | 2006-10-11 01:21:18 -0700 | [diff] [blame] | 33 | { |
Dave Kleikamp | 8c55e20 | 2007-05-24 13:04:54 -0400 | [diff] [blame] | 34 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
Andrew Morton | 72b64b5 | 2006-10-11 01:21:18 -0700 | [diff] [blame] | 35 | ext4_grpblk_t offset; |
| 36 | |
Dave Kleikamp | 8c55e20 | 2007-05-24 13:04:54 -0400 | [diff] [blame] | 37 | blocknr = blocknr - le32_to_cpu(es->s_first_data_block); |
Andrew Morton | f4e5bc2 | 2006-10-11 01:21:19 -0700 | [diff] [blame] | 38 | offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)); |
Andrew Morton | 72b64b5 | 2006-10-11 01:21:18 -0700 | [diff] [blame] | 39 | if (offsetp) |
| 40 | *offsetp = offset; |
| 41 | if (blockgrpp) |
Dave Kleikamp | 8c55e20 | 2007-05-24 13:04:54 -0400 | [diff] [blame] | 42 | *blockgrpp = blocknr; |
Andrew Morton | 72b64b5 | 2006-10-11 01:21:18 -0700 | [diff] [blame] | 43 | |
| 44 | } |
| 45 | |
Jose R. Santos | 0bf7e83 | 2008-06-03 14:07:29 -0400 | [diff] [blame] | 46 | static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block, |
| 47 | ext4_group_t block_group) |
| 48 | { |
| 49 | ext4_group_t actual_group; |
Aneesh Kumar K.V | 7477827 | 2008-07-11 19:27:31 -0400 | [diff] [blame^] | 50 | ext4_get_group_no_and_offset(sb, block, &actual_group, NULL); |
Jose R. Santos | 0bf7e83 | 2008-06-03 14:07:29 -0400 | [diff] [blame] | 51 | if (actual_group == block_group) |
| 52 | return 1; |
| 53 | return 0; |
| 54 | } |
| 55 | |
| 56 | static int ext4_group_used_meta_blocks(struct super_block *sb, |
| 57 | ext4_group_t block_group) |
| 58 | { |
| 59 | ext4_fsblk_t tmp; |
| 60 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 61 | /* block bitmap, inode bitmap, and inode table blocks */ |
| 62 | int used_blocks = sbi->s_itb_per_group + 2; |
| 63 | |
| 64 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) { |
| 65 | struct ext4_group_desc *gdp; |
| 66 | struct buffer_head *bh; |
| 67 | |
| 68 | gdp = ext4_get_group_desc(sb, block_group, &bh); |
| 69 | if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), |
| 70 | block_group)) |
| 71 | used_blocks--; |
| 72 | |
| 73 | if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), |
| 74 | block_group)) |
| 75 | used_blocks--; |
| 76 | |
| 77 | tmp = ext4_inode_table(sb, gdp); |
| 78 | for (; tmp < ext4_inode_table(sb, gdp) + |
| 79 | sbi->s_itb_per_group; tmp++) { |
| 80 | if (!ext4_block_in_group(sb, tmp, block_group)) |
| 81 | used_blocks -= 1; |
| 82 | } |
| 83 | } |
| 84 | return used_blocks; |
| 85 | } |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 86 | /* Initializes an uninitialized block bitmap if given, and returns the |
| 87 | * number of blocks free in the group. */ |
| 88 | unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh, |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 89 | ext4_group_t block_group, struct ext4_group_desc *gdp) |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 90 | { |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 91 | int bit, bit_max; |
| 92 | unsigned free_blocks, group_blocks; |
| 93 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 94 | |
| 95 | if (bh) { |
| 96 | J_ASSERT_BH(bh, buffer_locked(bh)); |
| 97 | |
| 98 | /* If checksum is bad mark all blocks used to prevent allocation |
| 99 | * essentially implementing a per-group read-only flag. */ |
| 100 | if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) { |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 101 | ext4_error(sb, __func__, |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 102 | "Checksum bad for group %lu\n", block_group); |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 103 | gdp->bg_free_blocks_count = 0; |
| 104 | gdp->bg_free_inodes_count = 0; |
| 105 | gdp->bg_itable_unused = 0; |
| 106 | memset(bh->b_data, 0xff, sb->s_blocksize); |
| 107 | return 0; |
| 108 | } |
| 109 | memset(bh->b_data, 0, sb->s_blocksize); |
| 110 | } |
| 111 | |
| 112 | /* Check for superblock and gdt backups in this group */ |
| 113 | bit_max = ext4_bg_has_super(sb, block_group); |
| 114 | |
| 115 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) || |
| 116 | block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) * |
| 117 | sbi->s_desc_per_block) { |
| 118 | if (bit_max) { |
| 119 | bit_max += ext4_bg_num_gdb(sb, block_group); |
| 120 | bit_max += |
| 121 | le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); |
| 122 | } |
| 123 | } else { /* For META_BG_BLOCK_GROUPS */ |
| 124 | int group_rel = (block_group - |
| 125 | le32_to_cpu(sbi->s_es->s_first_meta_bg)) % |
| 126 | EXT4_DESC_PER_BLOCK(sb); |
| 127 | if (group_rel == 0 || group_rel == 1 || |
| 128 | (group_rel == EXT4_DESC_PER_BLOCK(sb) - 1)) |
| 129 | bit_max += 1; |
| 130 | } |
| 131 | |
| 132 | if (block_group == sbi->s_groups_count - 1) { |
| 133 | /* |
| 134 | * Even though mke2fs always initialize first and last group |
| 135 | * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need |
| 136 | * to make sure we calculate the right free blocks |
| 137 | */ |
| 138 | group_blocks = ext4_blocks_count(sbi->s_es) - |
| 139 | le32_to_cpu(sbi->s_es->s_first_data_block) - |
| 140 | (EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count -1)); |
| 141 | } else { |
| 142 | group_blocks = EXT4_BLOCKS_PER_GROUP(sb); |
| 143 | } |
| 144 | |
| 145 | free_blocks = group_blocks - bit_max; |
| 146 | |
| 147 | if (bh) { |
Jose R. Santos | 0bf7e83 | 2008-06-03 14:07:29 -0400 | [diff] [blame] | 148 | ext4_fsblk_t start, tmp; |
| 149 | int flex_bg = 0; |
Akinobu Mita | d00a6d7 | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 150 | |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 151 | for (bit = 0; bit < bit_max; bit++) |
| 152 | ext4_set_bit(bit, bh->b_data); |
| 153 | |
Akinobu Mita | d00a6d7 | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 154 | start = ext4_group_first_block_no(sb, block_group); |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 155 | |
Jose R. Santos | 0bf7e83 | 2008-06-03 14:07:29 -0400 | [diff] [blame] | 156 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, |
| 157 | EXT4_FEATURE_INCOMPAT_FLEX_BG)) |
| 158 | flex_bg = 1; |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 159 | |
Jose R. Santos | 0bf7e83 | 2008-06-03 14:07:29 -0400 | [diff] [blame] | 160 | /* Set bits for block and inode bitmaps, and inode table */ |
| 161 | tmp = ext4_block_bitmap(sb, gdp); |
| 162 | if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) |
| 163 | ext4_set_bit(tmp - start, bh->b_data); |
| 164 | |
| 165 | tmp = ext4_inode_bitmap(sb, gdp); |
| 166 | if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) |
| 167 | ext4_set_bit(tmp - start, bh->b_data); |
| 168 | |
| 169 | tmp = ext4_inode_table(sb, gdp); |
| 170 | for (; tmp < ext4_inode_table(sb, gdp) + |
| 171 | sbi->s_itb_per_group; tmp++) { |
| 172 | if (!flex_bg || |
| 173 | ext4_block_in_group(sb, tmp, block_group)) |
| 174 | ext4_set_bit(tmp - start, bh->b_data); |
| 175 | } |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 176 | /* |
| 177 | * Also if the number of blocks within the group is |
| 178 | * less than the blocksize * 8 ( which is the size |
| 179 | * of bitmap ), set rest of the block bitmap to 1 |
| 180 | */ |
| 181 | mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data); |
| 182 | } |
Jose R. Santos | 0bf7e83 | 2008-06-03 14:07:29 -0400 | [diff] [blame] | 183 | return free_blocks - ext4_group_used_meta_blocks(sb, block_group); |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 184 | } |
| 185 | |
| 186 | |
Andrew Morton | 72b64b5 | 2006-10-11 01:21:18 -0700 | [diff] [blame] | 187 | /* |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 188 | * The free blocks are managed by bitmaps. A file system contains several |
| 189 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap |
| 190 | * block for inodes, N blocks for the inode table and data blocks. |
| 191 | * |
| 192 | * The file system contains group descriptors which are located after the |
| 193 | * super block. Each descriptor contains the number of the bitmap block and |
| 194 | * the free blocks count in the block. The descriptors are loaded in memory |
Aneesh Kumar K.V | e627432 | 2007-02-20 13:57:58 -0800 | [diff] [blame] | 195 | * when a file system is mounted (see ext4_fill_super). |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 196 | */ |
| 197 | |
| 198 | |
| 199 | #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1) |
| 200 | |
| 201 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 202 | * ext4_get_group_desc() -- load group descriptor from disk |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 203 | * @sb: super block |
| 204 | * @block_group: given block group |
| 205 | * @bh: pointer to the buffer head to store the block |
| 206 | * group descriptor |
| 207 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 208 | struct ext4_group_desc * ext4_get_group_desc(struct super_block * sb, |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 209 | ext4_group_t block_group, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 210 | struct buffer_head ** bh) |
| 211 | { |
| 212 | unsigned long group_desc; |
| 213 | unsigned long offset; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 214 | struct ext4_group_desc * desc; |
| 215 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 216 | |
| 217 | if (block_group >= sbi->s_groups_count) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 218 | ext4_error (sb, "ext4_get_group_desc", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 219 | "block_group >= groups_count - " |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 220 | "block_group = %lu, groups_count = %lu", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 221 | block_group, sbi->s_groups_count); |
| 222 | |
| 223 | return NULL; |
| 224 | } |
| 225 | smp_rmb(); |
| 226 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 227 | group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); |
| 228 | offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 229 | if (!sbi->s_group_desc[group_desc]) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 230 | ext4_error (sb, "ext4_get_group_desc", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 231 | "Group descriptor not loaded - " |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 232 | "block_group = %lu, group_desc = %lu, desc = %lu", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 233 | block_group, group_desc, offset); |
| 234 | return NULL; |
| 235 | } |
| 236 | |
Alexandre Ratchov | 0d1ee42 | 2006-10-11 01:21:14 -0700 | [diff] [blame] | 237 | desc = (struct ext4_group_desc *)( |
| 238 | (__u8 *)sbi->s_group_desc[group_desc]->b_data + |
| 239 | offset * EXT4_DESC_SIZE(sb)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 240 | if (bh) |
| 241 | *bh = sbi->s_group_desc[group_desc]; |
Alexandre Ratchov | 0d1ee42 | 2006-10-11 01:21:14 -0700 | [diff] [blame] | 242 | return desc; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 243 | } |
| 244 | |
Aneesh Kumar K.V | abcb294 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 245 | static int ext4_valid_block_bitmap(struct super_block *sb, |
| 246 | struct ext4_group_desc *desc, |
| 247 | unsigned int block_group, |
| 248 | struct buffer_head *bh) |
| 249 | { |
| 250 | ext4_grpblk_t offset; |
| 251 | ext4_grpblk_t next_zero_bit; |
| 252 | ext4_fsblk_t bitmap_blk; |
| 253 | ext4_fsblk_t group_first_block; |
| 254 | |
| 255 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) { |
| 256 | /* with FLEX_BG, the inode/block bitmaps and itable |
| 257 | * blocks may not be in the group at all |
| 258 | * so the bitmap validation will be skipped for those groups |
| 259 | * or it has to also read the block group where the bitmaps |
| 260 | * are located to verify they are set. |
| 261 | */ |
| 262 | return 1; |
| 263 | } |
| 264 | group_first_block = ext4_group_first_block_no(sb, block_group); |
| 265 | |
| 266 | /* check whether block bitmap block number is set */ |
| 267 | bitmap_blk = ext4_block_bitmap(sb, desc); |
| 268 | offset = bitmap_blk - group_first_block; |
| 269 | if (!ext4_test_bit(offset, bh->b_data)) |
| 270 | /* bad block bitmap */ |
| 271 | goto err_out; |
| 272 | |
| 273 | /* check whether the inode bitmap block number is set */ |
| 274 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
| 275 | offset = bitmap_blk - group_first_block; |
| 276 | if (!ext4_test_bit(offset, bh->b_data)) |
| 277 | /* bad block bitmap */ |
| 278 | goto err_out; |
| 279 | |
| 280 | /* check whether the inode table block number is set */ |
| 281 | bitmap_blk = ext4_inode_table(sb, desc); |
| 282 | offset = bitmap_blk - group_first_block; |
| 283 | next_zero_bit = ext4_find_next_zero_bit(bh->b_data, |
| 284 | offset + EXT4_SB(sb)->s_itb_per_group, |
| 285 | offset); |
| 286 | if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group) |
| 287 | /* good bitmap for inode tables */ |
| 288 | return 1; |
| 289 | |
| 290 | err_out: |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 291 | ext4_error(sb, __func__, |
Aneesh Kumar K.V | abcb294 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 292 | "Invalid block bitmap - " |
| 293 | "block_group = %d, block = %llu", |
| 294 | block_group, bitmap_blk); |
| 295 | return 0; |
| 296 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 297 | /** |
Theodore Ts'o | 574ca17 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 298 | * ext4_read_block_bitmap() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 299 | * @sb: super block |
| 300 | * @block_group: given block group |
| 301 | * |
Aneesh Kumar K.V | abcb294 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 302 | * Read the bitmap for a given block_group,and validate the |
| 303 | * bits for block/inode/inode tables are set in the bitmaps |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 304 | * |
| 305 | * Return buffer_head on success or NULL in case of failure. |
| 306 | */ |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 307 | struct buffer_head * |
Theodore Ts'o | 574ca17 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 308 | ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 309 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 310 | struct ext4_group_desc * desc; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 311 | struct buffer_head * bh = NULL; |
Aneesh Kumar K.V | 7c9e69f | 2007-10-16 23:27:02 -0700 | [diff] [blame] | 312 | ext4_fsblk_t bitmap_blk; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 313 | |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 314 | desc = ext4_get_group_desc(sb, block_group, NULL); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 315 | if (!desc) |
Aneesh Kumar K.V | 7c9e69f | 2007-10-16 23:27:02 -0700 | [diff] [blame] | 316 | return NULL; |
| 317 | bitmap_blk = ext4_block_bitmap(sb, desc); |
Aneesh Kumar K.V | abcb294 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 318 | bh = sb_getblk(sb, bitmap_blk); |
| 319 | if (unlikely(!bh)) { |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 320 | ext4_error(sb, __func__, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 321 | "Cannot read block bitmap - " |
Aneesh Kumar K.V | abcb294 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 322 | "block_group = %d, block_bitmap = %llu", |
| 323 | (int)block_group, (unsigned long long)bitmap_blk); |
| 324 | return NULL; |
| 325 | } |
| 326 | if (bh_uptodate_or_lock(bh)) |
| 327 | return bh; |
| 328 | |
| 329 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
| 330 | ext4_init_block_bitmap(sb, bh, block_group, desc); |
| 331 | set_buffer_uptodate(bh); |
| 332 | unlock_buffer(bh); |
| 333 | return bh; |
| 334 | } |
| 335 | if (bh_submit_read(bh) < 0) { |
| 336 | put_bh(bh); |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 337 | ext4_error(sb, __func__, |
Aneesh Kumar K.V | abcb294 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 338 | "Cannot read block bitmap - " |
| 339 | "block_group = %d, block_bitmap = %llu", |
| 340 | (int)block_group, (unsigned long long)bitmap_blk); |
| 341 | return NULL; |
| 342 | } |
Aneesh Kumar K.V | 519deca | 2008-05-15 14:43:20 -0400 | [diff] [blame] | 343 | ext4_valid_block_bitmap(sb, desc, block_group, bh); |
| 344 | /* |
| 345 | * file system mounted not to panic on error, |
| 346 | * continue with corrupt bitmap |
| 347 | */ |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 348 | return bh; |
| 349 | } |
| 350 | /* |
| 351 | * The reservation window structure operations |
| 352 | * -------------------------------------------- |
| 353 | * Operations include: |
| 354 | * dump, find, add, remove, is_empty, find_next_reservable_window, etc. |
| 355 | * |
| 356 | * We use a red-black tree to represent per-filesystem reservation |
| 357 | * windows. |
| 358 | * |
| 359 | */ |
| 360 | |
| 361 | /** |
| 362 | * __rsv_window_dump() -- Dump the filesystem block allocation reservation map |
| 363 | * @rb_root: root of per-filesystem reservation rb tree |
| 364 | * @verbose: verbose mode |
| 365 | * @fn: function which wishes to dump the reservation map |
| 366 | * |
| 367 | * If verbose is turned on, it will print the whole block reservation |
| 368 | * windows(start, end). Otherwise, it will only print out the "bad" windows, |
| 369 | * those windows that overlap with their immediate neighbors. |
| 370 | */ |
| 371 | #if 1 |
| 372 | static void __rsv_window_dump(struct rb_root *root, int verbose, |
| 373 | const char *fn) |
| 374 | { |
| 375 | struct rb_node *n; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 376 | struct ext4_reserve_window_node *rsv, *prev; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 377 | int bad; |
| 378 | |
| 379 | restart: |
| 380 | n = rb_first(root); |
| 381 | bad = 0; |
| 382 | prev = NULL; |
| 383 | |
| 384 | printk("Block Allocation Reservation Windows Map (%s):\n", fn); |
| 385 | while (n) { |
Hugh Dickins | b78a657 | 2006-12-06 20:39:21 -0800 | [diff] [blame] | 386 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 387 | if (verbose) |
| 388 | printk("reservation window 0x%p " |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 389 | "start: %llu, end: %llu\n", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 390 | rsv, rsv->rsv_start, rsv->rsv_end); |
| 391 | if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) { |
| 392 | printk("Bad reservation %p (start >= end)\n", |
| 393 | rsv); |
| 394 | bad = 1; |
| 395 | } |
| 396 | if (prev && prev->rsv_end >= rsv->rsv_start) { |
| 397 | printk("Bad reservation %p (prev->end >= start)\n", |
| 398 | rsv); |
| 399 | bad = 1; |
| 400 | } |
| 401 | if (bad) { |
| 402 | if (!verbose) { |
| 403 | printk("Restarting reservation walk in verbose mode\n"); |
| 404 | verbose = 1; |
| 405 | goto restart; |
| 406 | } |
| 407 | } |
| 408 | n = rb_next(n); |
| 409 | prev = rsv; |
| 410 | } |
| 411 | printk("Window map complete.\n"); |
Julia Lawall | 07d45f1 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 412 | BUG_ON(bad); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 413 | } |
| 414 | #define rsv_window_dump(root, verbose) \ |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 415 | __rsv_window_dump((root), (verbose), __func__) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 416 | #else |
| 417 | #define rsv_window_dump(root, verbose) do {} while (0) |
| 418 | #endif |
| 419 | |
| 420 | /** |
| 421 | * goal_in_my_reservation() |
| 422 | * @rsv: inode's reservation window |
| 423 | * @grp_goal: given goal block relative to the allocation block group |
| 424 | * @group: the current allocation block group |
| 425 | * @sb: filesystem super block |
| 426 | * |
| 427 | * Test if the given goal block (group relative) is within the file's |
| 428 | * own block reservation window range. |
| 429 | * |
| 430 | * If the reservation window is outside the goal allocation group, return 0; |
| 431 | * grp_goal (given goal block) could be -1, which means no specific |
| 432 | * goal block. In this case, always return 1. |
| 433 | * If the goal block is within the reservation window, return 1; |
| 434 | * otherwise, return 0; |
| 435 | */ |
| 436 | static int |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 437 | goal_in_my_reservation(struct ext4_reserve_window *rsv, ext4_grpblk_t grp_goal, |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 438 | ext4_group_t group, struct super_block *sb) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 439 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 440 | ext4_fsblk_t group_first_block, group_last_block; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 441 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 442 | group_first_block = ext4_group_first_block_no(sb, group); |
| 443 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 444 | |
| 445 | if ((rsv->_rsv_start > group_last_block) || |
| 446 | (rsv->_rsv_end < group_first_block)) |
| 447 | return 0; |
| 448 | if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start) |
| 449 | || (grp_goal + group_first_block > rsv->_rsv_end))) |
| 450 | return 0; |
| 451 | return 1; |
| 452 | } |
| 453 | |
| 454 | /** |
| 455 | * search_reserve_window() |
| 456 | * @rb_root: root of reservation tree |
| 457 | * @goal: target allocation block |
| 458 | * |
| 459 | * Find the reserved window which includes the goal, or the previous one |
| 460 | * if the goal is not in any window. |
| 461 | * Returns NULL if there are no windows or if all windows start after the goal. |
| 462 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 463 | static struct ext4_reserve_window_node * |
| 464 | search_reserve_window(struct rb_root *root, ext4_fsblk_t goal) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 465 | { |
| 466 | struct rb_node *n = root->rb_node; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 467 | struct ext4_reserve_window_node *rsv; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 468 | |
| 469 | if (!n) |
| 470 | return NULL; |
| 471 | |
| 472 | do { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 473 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 474 | |
| 475 | if (goal < rsv->rsv_start) |
| 476 | n = n->rb_left; |
| 477 | else if (goal > rsv->rsv_end) |
| 478 | n = n->rb_right; |
| 479 | else |
| 480 | return rsv; |
| 481 | } while (n); |
| 482 | /* |
| 483 | * We've fallen off the end of the tree: the goal wasn't inside |
| 484 | * any particular node. OK, the previous node must be to one |
| 485 | * side of the interval containing the goal. If it's the RHS, |
| 486 | * we need to back up one. |
| 487 | */ |
| 488 | if (rsv->rsv_start > goal) { |
| 489 | n = rb_prev(&rsv->rsv_node); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 490 | rsv = rb_entry(n, struct ext4_reserve_window_node, rsv_node); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 491 | } |
| 492 | return rsv; |
| 493 | } |
| 494 | |
| 495 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 496 | * ext4_rsv_window_add() -- Insert a window to the block reservation rb tree. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 497 | * @sb: super block |
| 498 | * @rsv: reservation window to add |
| 499 | * |
| 500 | * Must be called with rsv_lock hold. |
| 501 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 502 | void ext4_rsv_window_add(struct super_block *sb, |
| 503 | struct ext4_reserve_window_node *rsv) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 504 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 505 | struct rb_root *root = &EXT4_SB(sb)->s_rsv_window_root; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 506 | struct rb_node *node = &rsv->rsv_node; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 507 | ext4_fsblk_t start = rsv->rsv_start; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 508 | |
| 509 | struct rb_node ** p = &root->rb_node; |
| 510 | struct rb_node * parent = NULL; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 511 | struct ext4_reserve_window_node *this; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 512 | |
| 513 | while (*p) |
| 514 | { |
| 515 | parent = *p; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 516 | this = rb_entry(parent, struct ext4_reserve_window_node, rsv_node); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 517 | |
| 518 | if (start < this->rsv_start) |
| 519 | p = &(*p)->rb_left; |
| 520 | else if (start > this->rsv_end) |
| 521 | p = &(*p)->rb_right; |
| 522 | else { |
| 523 | rsv_window_dump(root, 1); |
| 524 | BUG(); |
| 525 | } |
| 526 | } |
| 527 | |
| 528 | rb_link_node(node, parent, p); |
| 529 | rb_insert_color(node, root); |
| 530 | } |
| 531 | |
| 532 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 533 | * ext4_rsv_window_remove() -- unlink a window from the reservation rb tree |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 534 | * @sb: super block |
| 535 | * @rsv: reservation window to remove |
| 536 | * |
| 537 | * Mark the block reservation window as not allocated, and unlink it |
| 538 | * from the filesystem reservation window rb tree. Must be called with |
| 539 | * rsv_lock hold. |
| 540 | */ |
| 541 | static void rsv_window_remove(struct super_block *sb, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 542 | struct ext4_reserve_window_node *rsv) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 543 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 544 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
| 545 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 546 | rsv->rsv_alloc_hit = 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 547 | rb_erase(&rsv->rsv_node, &EXT4_SB(sb)->s_rsv_window_root); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 548 | } |
| 549 | |
| 550 | /* |
| 551 | * rsv_is_empty() -- Check if the reservation window is allocated. |
| 552 | * @rsv: given reservation window to check |
| 553 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 554 | * returns 1 if the end block is EXT4_RESERVE_WINDOW_NOT_ALLOCATED. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 555 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 556 | static inline int rsv_is_empty(struct ext4_reserve_window *rsv) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 557 | { |
| 558 | /* a valid reservation end block could not be 0 */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 559 | return rsv->_rsv_end == EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 560 | } |
| 561 | |
| 562 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 563 | * ext4_init_block_alloc_info() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 564 | * @inode: file inode structure |
| 565 | * |
| 566 | * Allocate and initialize the reservation window structure, and |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 567 | * link the window to the ext4 inode structure at last |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 568 | * |
| 569 | * The reservation window structure is only dynamically allocated |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 570 | * and linked to ext4 inode the first time the open file |
| 571 | * needs a new block. So, before every ext4_new_block(s) call, for |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 572 | * regular files, we should check whether the reservation window |
| 573 | * structure exists or not. In the latter case, this function is called. |
| 574 | * Fail to do so will result in block reservation being turned off for that |
| 575 | * open file. |
| 576 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 577 | * This function is called from ext4_get_blocks_handle(), also called |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 578 | * when setting the reservation window size through ioctl before the file |
| 579 | * is open for write (needs block allocation). |
| 580 | * |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 581 | * Needs down_write(i_data_sem) protection prior to call this function. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 582 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 583 | void ext4_init_block_alloc_info(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 584 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 585 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 586 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 587 | struct super_block *sb = inode->i_sb; |
| 588 | |
| 589 | block_i = kmalloc(sizeof(*block_i), GFP_NOFS); |
| 590 | if (block_i) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 591 | struct ext4_reserve_window_node *rsv = &block_i->rsv_window_node; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 592 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 593 | rsv->rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
| 594 | rsv->rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 595 | |
| 596 | /* |
| 597 | * if filesystem is mounted with NORESERVATION, the goal |
| 598 | * reservation window size is set to zero to indicate |
| 599 | * block reservation is off |
| 600 | */ |
| 601 | if (!test_opt(sb, RESERVATION)) |
| 602 | rsv->rsv_goal_size = 0; |
| 603 | else |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 604 | rsv->rsv_goal_size = EXT4_DEFAULT_RESERVE_BLOCKS; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 605 | rsv->rsv_alloc_hit = 0; |
| 606 | block_i->last_alloc_logical_block = 0; |
| 607 | block_i->last_alloc_physical_block = 0; |
| 608 | } |
| 609 | ei->i_block_alloc_info = block_i; |
| 610 | } |
| 611 | |
| 612 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 613 | * ext4_discard_reservation() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 614 | * @inode: inode |
| 615 | * |
| 616 | * Discard(free) block reservation window on last file close, or truncate |
| 617 | * or at last iput(). |
| 618 | * |
| 619 | * It is being called in three cases: |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 620 | * ext4_release_file(): last writer close the file |
| 621 | * ext4_clear_inode(): last iput(), when nobody link to this file. |
| 622 | * ext4_truncate(): when the block indirect map is about to change. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 623 | * |
| 624 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 625 | void ext4_discard_reservation(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 626 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 627 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 628 | struct ext4_block_alloc_info *block_i = ei->i_block_alloc_info; |
| 629 | struct ext4_reserve_window_node *rsv; |
| 630 | spinlock_t *rsv_lock = &EXT4_SB(inode->i_sb)->s_rsv_window_lock; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 631 | |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 632 | ext4_mb_discard_inode_preallocations(inode); |
| 633 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 634 | if (!block_i) |
| 635 | return; |
| 636 | |
| 637 | rsv = &block_i->rsv_window_node; |
| 638 | if (!rsv_is_empty(&rsv->rsv_window)) { |
| 639 | spin_lock(rsv_lock); |
| 640 | if (!rsv_is_empty(&rsv->rsv_window)) |
| 641 | rsv_window_remove(inode->i_sb, rsv); |
| 642 | spin_unlock(rsv_lock); |
| 643 | } |
| 644 | } |
| 645 | |
| 646 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 647 | * ext4_free_blocks_sb() -- Free given blocks and update quota |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 648 | * @handle: handle to this transaction |
| 649 | * @sb: super block |
| 650 | * @block: start physcial block to free |
| 651 | * @count: number of blocks to free |
| 652 | * @pdquot_freed_blocks: pointer to quota |
| 653 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 654 | void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb, |
| 655 | ext4_fsblk_t block, unsigned long count, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 656 | unsigned long *pdquot_freed_blocks) |
| 657 | { |
| 658 | struct buffer_head *bitmap_bh = NULL; |
| 659 | struct buffer_head *gd_bh; |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 660 | ext4_group_t block_group; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 661 | ext4_grpblk_t bit; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 662 | unsigned long i; |
| 663 | unsigned long overflow; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 664 | struct ext4_group_desc * desc; |
| 665 | struct ext4_super_block * es; |
| 666 | struct ext4_sb_info *sbi; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 667 | int err = 0, ret; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 668 | ext4_grpblk_t group_freed; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 669 | |
| 670 | *pdquot_freed_blocks = 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 671 | sbi = EXT4_SB(sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 672 | es = sbi->s_es; |
| 673 | if (block < le32_to_cpu(es->s_first_data_block) || |
| 674 | block + count < block || |
Laurent Vivier | bd81d8e | 2006-10-11 01:21:10 -0700 | [diff] [blame] | 675 | block + count > ext4_blocks_count(es)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 676 | ext4_error (sb, "ext4_free_blocks", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 677 | "Freeing blocks not in datazone - " |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 678 | "block = %llu, count = %lu", block, count); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 679 | goto error_return; |
| 680 | } |
| 681 | |
Laurent Vivier | bd81d8e | 2006-10-11 01:21:10 -0700 | [diff] [blame] | 682 | ext4_debug ("freeing block(s) %llu-%llu\n", block, block + count - 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 683 | |
| 684 | do_more: |
| 685 | overflow = 0; |
Mingming Cao | 3a5b2ec | 2006-10-11 01:21:05 -0700 | [diff] [blame] | 686 | ext4_get_group_no_and_offset(sb, block, &block_group, &bit); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 687 | /* |
| 688 | * Check to see if we are freeing blocks across a group |
| 689 | * boundary. |
| 690 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 691 | if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) { |
| 692 | overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 693 | count -= overflow; |
| 694 | } |
| 695 | brelse(bitmap_bh); |
Theodore Ts'o | 574ca17 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 696 | bitmap_bh = ext4_read_block_bitmap(sb, block_group); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 697 | if (!bitmap_bh) |
| 698 | goto error_return; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 699 | desc = ext4_get_group_desc (sb, block_group, &gd_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 700 | if (!desc) |
| 701 | goto error_return; |
| 702 | |
Alexandre Ratchov | 8fadc14 | 2006-10-11 01:21:15 -0700 | [diff] [blame] | 703 | if (in_range(ext4_block_bitmap(sb, desc), block, count) || |
| 704 | in_range(ext4_inode_bitmap(sb, desc), block, count) || |
| 705 | in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) || |
| 706 | in_range(block + count - 1, ext4_inode_table(sb, desc), |
Aneesh Kumar K.V | cb47dce | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 707 | sbi->s_itb_per_group)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 708 | ext4_error (sb, "ext4_free_blocks", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 709 | "Freeing blocks in system zones - " |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 710 | "Block = %llu, count = %lu", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 711 | block, count); |
Aneesh Kumar K.V | cb47dce | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 712 | goto error_return; |
| 713 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 714 | |
| 715 | /* |
| 716 | * We are about to start releasing blocks in the bitmap, |
| 717 | * so we need undo access. |
| 718 | */ |
| 719 | /* @@@ check errors */ |
| 720 | BUFFER_TRACE(bitmap_bh, "getting undo access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 721 | err = ext4_journal_get_undo_access(handle, bitmap_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 722 | if (err) |
| 723 | goto error_return; |
| 724 | |
| 725 | /* |
| 726 | * We are about to modify some metadata. Call the journal APIs |
| 727 | * to unshare ->b_data if a currently-committing transaction is |
| 728 | * using it |
| 729 | */ |
| 730 | BUFFER_TRACE(gd_bh, "get_write_access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 731 | err = ext4_journal_get_write_access(handle, gd_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 732 | if (err) |
| 733 | goto error_return; |
| 734 | |
| 735 | jbd_lock_bh_state(bitmap_bh); |
| 736 | |
| 737 | for (i = 0, group_freed = 0; i < count; i++) { |
| 738 | /* |
| 739 | * An HJ special. This is expensive... |
| 740 | */ |
Jose R. Santos | e23291b | 2007-07-18 08:57:06 -0400 | [diff] [blame] | 741 | #ifdef CONFIG_JBD2_DEBUG |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 742 | jbd_unlock_bh_state(bitmap_bh); |
| 743 | { |
| 744 | struct buffer_head *debug_bh; |
| 745 | debug_bh = sb_find_get_block(sb, block + i); |
| 746 | if (debug_bh) { |
| 747 | BUFFER_TRACE(debug_bh, "Deleted!"); |
| 748 | if (!bh2jh(bitmap_bh)->b_committed_data) |
| 749 | BUFFER_TRACE(debug_bh, |
| 750 | "No commited data in bitmap"); |
| 751 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap"); |
| 752 | __brelse(debug_bh); |
| 753 | } |
| 754 | } |
| 755 | jbd_lock_bh_state(bitmap_bh); |
| 756 | #endif |
| 757 | if (need_resched()) { |
| 758 | jbd_unlock_bh_state(bitmap_bh); |
| 759 | cond_resched(); |
| 760 | jbd_lock_bh_state(bitmap_bh); |
| 761 | } |
| 762 | /* @@@ This prevents newly-allocated data from being |
| 763 | * freed and then reallocated within the same |
| 764 | * transaction. |
| 765 | * |
| 766 | * Ideally we would want to allow that to happen, but to |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 767 | * do so requires making jbd2_journal_forget() capable of |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 768 | * revoking the queued write of a data block, which |
| 769 | * implies blocking on the journal lock. *forget() |
| 770 | * cannot block due to truncate races. |
| 771 | * |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 772 | * Eventually we can fix this by making jbd2_journal_forget() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 773 | * return a status indicating whether or not it was able |
| 774 | * to revoke the buffer. On successful revoke, it is |
| 775 | * safe not to set the allocation bit in the committed |
| 776 | * bitmap, because we know that there is no outstanding |
| 777 | * activity on the buffer any more and so it is safe to |
| 778 | * reallocate it. |
| 779 | */ |
| 780 | BUFFER_TRACE(bitmap_bh, "set in b_committed_data"); |
| 781 | J_ASSERT_BH(bitmap_bh, |
| 782 | bh2jh(bitmap_bh)->b_committed_data != NULL); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 783 | ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 784 | bh2jh(bitmap_bh)->b_committed_data); |
| 785 | |
| 786 | /* |
| 787 | * We clear the bit in the bitmap after setting the committed |
| 788 | * data bit, because this is the reverse order to that which |
| 789 | * the allocator uses. |
| 790 | */ |
| 791 | BUFFER_TRACE(bitmap_bh, "clear bit"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 792 | if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 793 | bit + i, bitmap_bh->b_data)) { |
| 794 | jbd_unlock_bh_state(bitmap_bh); |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 795 | ext4_error(sb, __func__, |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 796 | "bit already cleared for block %llu", |
Laurent Vivier | bd81d8e | 2006-10-11 01:21:10 -0700 | [diff] [blame] | 797 | (ext4_fsblk_t)(block + i)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 798 | jbd_lock_bh_state(bitmap_bh); |
| 799 | BUFFER_TRACE(bitmap_bh, "bit already cleared"); |
| 800 | } else { |
| 801 | group_freed++; |
| 802 | } |
| 803 | } |
| 804 | jbd_unlock_bh_state(bitmap_bh); |
| 805 | |
| 806 | spin_lock(sb_bgl_lock(sbi, block_group)); |
Marcin Slusarz | e8546d0 | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 807 | le16_add_cpu(&desc->bg_free_blocks_count, group_freed); |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 808 | desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 809 | spin_unlock(sb_bgl_lock(sbi, block_group)); |
Peter Zijlstra | aa0dff2 | 2007-10-16 23:25:42 -0700 | [diff] [blame] | 810 | percpu_counter_add(&sbi->s_freeblocks_counter, count); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 811 | |
Jose R. Santos | 772cb7c | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 812 | if (sbi->s_log_groups_per_flex) { |
| 813 | ext4_group_t flex_group = ext4_flex_group(sbi, block_group); |
| 814 | spin_lock(sb_bgl_lock(sbi, flex_group)); |
| 815 | sbi->s_flex_groups[flex_group].free_blocks += count; |
| 816 | spin_unlock(sb_bgl_lock(sbi, flex_group)); |
| 817 | } |
| 818 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 819 | /* We dirtied the bitmap block */ |
| 820 | BUFFER_TRACE(bitmap_bh, "dirtied bitmap block"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 821 | err = ext4_journal_dirty_metadata(handle, bitmap_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 822 | |
| 823 | /* And the group descriptor block */ |
| 824 | BUFFER_TRACE(gd_bh, "dirtied group descriptor block"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 825 | ret = ext4_journal_dirty_metadata(handle, gd_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 826 | if (!err) err = ret; |
| 827 | *pdquot_freed_blocks += group_freed; |
| 828 | |
| 829 | if (overflow && !err) { |
| 830 | block += count; |
| 831 | count = overflow; |
| 832 | goto do_more; |
| 833 | } |
| 834 | sb->s_dirt = 1; |
| 835 | error_return: |
| 836 | brelse(bitmap_bh); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 837 | ext4_std_error(sb, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 838 | return; |
| 839 | } |
| 840 | |
| 841 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 842 | * ext4_free_blocks() -- Free given blocks and update quota |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 843 | * @handle: handle for this transaction |
| 844 | * @inode: inode |
| 845 | * @block: start physical block to free |
| 846 | * @count: number of blocks to count |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 847 | * @metadata: Are these metadata blocks |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 848 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 849 | void ext4_free_blocks(handle_t *handle, struct inode *inode, |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 850 | ext4_fsblk_t block, unsigned long count, |
| 851 | int metadata) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 852 | { |
| 853 | struct super_block * sb; |
| 854 | unsigned long dquot_freed_blocks; |
| 855 | |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 856 | /* this isn't the right place to decide whether block is metadata |
| 857 | * inode.c/extents.c knows better, but for safety ... */ |
| 858 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) || |
| 859 | ext4_should_journal_data(inode)) |
| 860 | metadata = 1; |
| 861 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 862 | sb = inode->i_sb; |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 863 | |
| 864 | if (!test_opt(sb, MBALLOC) || !EXT4_SB(sb)->s_group_info) |
| 865 | ext4_free_blocks_sb(handle, sb, block, count, |
| 866 | &dquot_freed_blocks); |
| 867 | else |
| 868 | ext4_mb_free_blocks(handle, inode, block, count, |
| 869 | metadata, &dquot_freed_blocks); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 870 | if (dquot_freed_blocks) |
| 871 | DQUOT_FREE_BLOCK(inode, dquot_freed_blocks); |
| 872 | return; |
| 873 | } |
| 874 | |
| 875 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 876 | * ext4_test_allocatable() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 877 | * @nr: given allocation block group |
| 878 | * @bh: bufferhead contains the bitmap of the given block group |
| 879 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 880 | * For ext4 allocations, we must not reuse any blocks which are |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 881 | * allocated in the bitmap buffer's "last committed data" copy. This |
| 882 | * prevents deletes from freeing up the page for reuse until we have |
| 883 | * committed the delete transaction. |
| 884 | * |
| 885 | * If we didn't do this, then deleting something and reallocating it as |
| 886 | * data would allow the old block to be overwritten before the |
| 887 | * transaction committed (because we force data to disk before commit). |
| 888 | * This would lead to corruption if we crashed between overwriting the |
| 889 | * data and committing the delete. |
| 890 | * |
| 891 | * @@@ We may want to make this allocation behaviour conditional on |
| 892 | * data-writes at some point, and disable it for metadata allocations or |
| 893 | * sync-data inodes. |
| 894 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 895 | static int ext4_test_allocatable(ext4_grpblk_t nr, struct buffer_head *bh) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 896 | { |
| 897 | int ret; |
| 898 | struct journal_head *jh = bh2jh(bh); |
| 899 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 900 | if (ext4_test_bit(nr, bh->b_data)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 901 | return 0; |
| 902 | |
| 903 | jbd_lock_bh_state(bh); |
| 904 | if (!jh->b_committed_data) |
| 905 | ret = 1; |
| 906 | else |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 907 | ret = !ext4_test_bit(nr, jh->b_committed_data); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 908 | jbd_unlock_bh_state(bh); |
| 909 | return ret; |
| 910 | } |
| 911 | |
| 912 | /** |
| 913 | * bitmap_search_next_usable_block() |
| 914 | * @start: the starting block (group relative) of the search |
| 915 | * @bh: bufferhead contains the block group bitmap |
| 916 | * @maxblocks: the ending block (group relative) of the reservation |
| 917 | * |
| 918 | * The bitmap search --- search forward alternately through the actual |
| 919 | * bitmap on disk and the last-committed copy in journal, until we find a |
| 920 | * bit free in both bitmaps. |
| 921 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 922 | static ext4_grpblk_t |
| 923 | bitmap_search_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, |
| 924 | ext4_grpblk_t maxblocks) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 925 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 926 | ext4_grpblk_t next; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 927 | struct journal_head *jh = bh2jh(bh); |
| 928 | |
| 929 | while (start < maxblocks) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 930 | next = ext4_find_next_zero_bit(bh->b_data, maxblocks, start); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 931 | if (next >= maxblocks) |
| 932 | return -1; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 933 | if (ext4_test_allocatable(next, bh)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 934 | return next; |
| 935 | jbd_lock_bh_state(bh); |
| 936 | if (jh->b_committed_data) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 937 | start = ext4_find_next_zero_bit(jh->b_committed_data, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 938 | maxblocks, next); |
| 939 | jbd_unlock_bh_state(bh); |
| 940 | } |
| 941 | return -1; |
| 942 | } |
| 943 | |
| 944 | /** |
| 945 | * find_next_usable_block() |
| 946 | * @start: the starting block (group relative) to find next |
| 947 | * allocatable block in bitmap. |
| 948 | * @bh: bufferhead contains the block group bitmap |
| 949 | * @maxblocks: the ending block (group relative) for the search |
| 950 | * |
| 951 | * Find an allocatable block in a bitmap. We honor both the bitmap and |
| 952 | * its last-committed copy (if that exists), and perform the "most |
| 953 | * appropriate allocation" algorithm of looking for a free block near |
| 954 | * the initial goal; then for a free byte somewhere in the bitmap; then |
| 955 | * for any free bit in the bitmap. |
| 956 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 957 | static ext4_grpblk_t |
| 958 | find_next_usable_block(ext4_grpblk_t start, struct buffer_head *bh, |
| 959 | ext4_grpblk_t maxblocks) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 960 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 961 | ext4_grpblk_t here, next; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 962 | char *p, *r; |
| 963 | |
| 964 | if (start > 0) { |
| 965 | /* |
| 966 | * The goal was occupied; search forward for a free |
| 967 | * block within the next XX blocks. |
| 968 | * |
| 969 | * end_goal is more or less random, but it has to be |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 970 | * less than EXT4_BLOCKS_PER_GROUP. Aligning up to the |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 971 | * next 64-bit boundary is simple.. |
| 972 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 973 | ext4_grpblk_t end_goal = (start + 63) & ~63; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 974 | if (end_goal > maxblocks) |
| 975 | end_goal = maxblocks; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 976 | here = ext4_find_next_zero_bit(bh->b_data, end_goal, start); |
| 977 | if (here < end_goal && ext4_test_allocatable(here, bh)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 978 | return here; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 979 | ext4_debug("Bit not found near goal\n"); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 980 | } |
| 981 | |
| 982 | here = start; |
| 983 | if (here < 0) |
| 984 | here = 0; |
| 985 | |
| 986 | p = ((char *)bh->b_data) + (here >> 3); |
Hugh Dickins | ec0837f | 2006-12-06 20:39:26 -0800 | [diff] [blame] | 987 | r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 988 | next = (r - ((char *)bh->b_data)) << 3; |
| 989 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 990 | if (next < maxblocks && next >= start && ext4_test_allocatable(next, bh)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 991 | return next; |
| 992 | |
| 993 | /* |
| 994 | * The bitmap search --- search forward alternately through the actual |
| 995 | * bitmap and the last-committed copy until we find a bit free in |
| 996 | * both |
| 997 | */ |
| 998 | here = bitmap_search_next_usable_block(here, bh, maxblocks); |
| 999 | return here; |
| 1000 | } |
| 1001 | |
| 1002 | /** |
| 1003 | * claim_block() |
| 1004 | * @block: the free block (group relative) to allocate |
| 1005 | * @bh: the bufferhead containts the block group bitmap |
| 1006 | * |
| 1007 | * We think we can allocate this block in this bitmap. Try to set the bit. |
| 1008 | * If that succeeds then check that nobody has allocated and then freed the |
| 1009 | * block since we saw that is was not marked in b_committed_data. If it _was_ |
| 1010 | * allocated and freed then clear the bit in the bitmap again and return |
| 1011 | * zero (failure). |
| 1012 | */ |
| 1013 | static inline int |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1014 | claim_block(spinlock_t *lock, ext4_grpblk_t block, struct buffer_head *bh) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1015 | { |
| 1016 | struct journal_head *jh = bh2jh(bh); |
| 1017 | int ret; |
| 1018 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1019 | if (ext4_set_bit_atomic(lock, block, bh->b_data)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1020 | return 0; |
| 1021 | jbd_lock_bh_state(bh); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1022 | if (jh->b_committed_data && ext4_test_bit(block,jh->b_committed_data)) { |
| 1023 | ext4_clear_bit_atomic(lock, block, bh->b_data); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1024 | ret = 0; |
| 1025 | } else { |
| 1026 | ret = 1; |
| 1027 | } |
| 1028 | jbd_unlock_bh_state(bh); |
| 1029 | return ret; |
| 1030 | } |
| 1031 | |
| 1032 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1033 | * ext4_try_to_allocate() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1034 | * @sb: superblock |
| 1035 | * @handle: handle to this transaction |
| 1036 | * @group: given allocation block group |
| 1037 | * @bitmap_bh: bufferhead holds the block bitmap |
| 1038 | * @grp_goal: given target block within the group |
| 1039 | * @count: target number of blocks to allocate |
| 1040 | * @my_rsv: reservation window |
| 1041 | * |
| 1042 | * Attempt to allocate blocks within a give range. Set the range of allocation |
| 1043 | * first, then find the first free bit(s) from the bitmap (within the range), |
| 1044 | * and at last, allocate the blocks by claiming the found free bit as allocated. |
| 1045 | * |
| 1046 | * To set the range of this allocation: |
| 1047 | * if there is a reservation window, only try to allocate block(s) from the |
| 1048 | * file's own reservation window; |
| 1049 | * Otherwise, the allocation range starts from the give goal block, ends at |
| 1050 | * the block group's last block. |
| 1051 | * |
| 1052 | * If we failed to allocate the desired block then we may end up crossing to a |
| 1053 | * new bitmap. In that case we must release write access to the old one via |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1054 | * ext4_journal_release_buffer(), else we'll run out of credits. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1055 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1056 | static ext4_grpblk_t |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1057 | ext4_try_to_allocate(struct super_block *sb, handle_t *handle, |
| 1058 | ext4_group_t group, struct buffer_head *bitmap_bh, |
| 1059 | ext4_grpblk_t grp_goal, unsigned long *count, |
| 1060 | struct ext4_reserve_window *my_rsv) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1061 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1062 | ext4_fsblk_t group_first_block; |
| 1063 | ext4_grpblk_t start, end; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1064 | unsigned long num = 0; |
| 1065 | |
| 1066 | /* we do allocation within the reservation window if we have a window */ |
| 1067 | if (my_rsv) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1068 | group_first_block = ext4_group_first_block_no(sb, group); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1069 | if (my_rsv->_rsv_start >= group_first_block) |
| 1070 | start = my_rsv->_rsv_start - group_first_block; |
| 1071 | else |
| 1072 | /* reservation window cross group boundary */ |
| 1073 | start = 0; |
| 1074 | end = my_rsv->_rsv_end - group_first_block + 1; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1075 | if (end > EXT4_BLOCKS_PER_GROUP(sb)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1076 | /* reservation window crosses group boundary */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1077 | end = EXT4_BLOCKS_PER_GROUP(sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1078 | if ((start <= grp_goal) && (grp_goal < end)) |
| 1079 | start = grp_goal; |
| 1080 | else |
| 1081 | grp_goal = -1; |
| 1082 | } else { |
| 1083 | if (grp_goal > 0) |
| 1084 | start = grp_goal; |
| 1085 | else |
| 1086 | start = 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1087 | end = EXT4_BLOCKS_PER_GROUP(sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1088 | } |
| 1089 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1090 | BUG_ON(start > EXT4_BLOCKS_PER_GROUP(sb)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1091 | |
| 1092 | repeat: |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1093 | if (grp_goal < 0 || !ext4_test_allocatable(grp_goal, bitmap_bh)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1094 | grp_goal = find_next_usable_block(start, bitmap_bh, end); |
| 1095 | if (grp_goal < 0) |
| 1096 | goto fail_access; |
| 1097 | if (!my_rsv) { |
| 1098 | int i; |
| 1099 | |
| 1100 | for (i = 0; i < 7 && grp_goal > start && |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1101 | ext4_test_allocatable(grp_goal - 1, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1102 | bitmap_bh); |
| 1103 | i++, grp_goal--) |
| 1104 | ; |
| 1105 | } |
| 1106 | } |
| 1107 | start = grp_goal; |
| 1108 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1109 | if (!claim_block(sb_bgl_lock(EXT4_SB(sb), group), |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1110 | grp_goal, bitmap_bh)) { |
| 1111 | /* |
| 1112 | * The block was allocated by another thread, or it was |
| 1113 | * allocated and then freed by another thread |
| 1114 | */ |
| 1115 | start++; |
| 1116 | grp_goal++; |
| 1117 | if (start >= end) |
| 1118 | goto fail_access; |
| 1119 | goto repeat; |
| 1120 | } |
| 1121 | num++; |
| 1122 | grp_goal++; |
| 1123 | while (num < *count && grp_goal < end |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1124 | && ext4_test_allocatable(grp_goal, bitmap_bh) |
| 1125 | && claim_block(sb_bgl_lock(EXT4_SB(sb), group), |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1126 | grp_goal, bitmap_bh)) { |
| 1127 | num++; |
| 1128 | grp_goal++; |
| 1129 | } |
| 1130 | *count = num; |
| 1131 | return grp_goal - num; |
| 1132 | fail_access: |
| 1133 | *count = num; |
| 1134 | return -1; |
| 1135 | } |
| 1136 | |
| 1137 | /** |
| 1138 | * find_next_reservable_window(): |
| 1139 | * find a reservable space within the given range. |
| 1140 | * It does not allocate the reservation window for now: |
| 1141 | * alloc_new_reservation() will do the work later. |
| 1142 | * |
| 1143 | * @search_head: the head of the searching list; |
| 1144 | * This is not necessarily the list head of the whole filesystem |
| 1145 | * |
| 1146 | * We have both head and start_block to assist the search |
| 1147 | * for the reservable space. The list starts from head, |
| 1148 | * but we will shift to the place where start_block is, |
| 1149 | * then start from there, when looking for a reservable space. |
| 1150 | * |
| 1151 | * @size: the target new reservation window size |
| 1152 | * |
| 1153 | * @group_first_block: the first block we consider to start |
| 1154 | * the real search from |
| 1155 | * |
| 1156 | * @last_block: |
| 1157 | * the maximum block number that our goal reservable space |
| 1158 | * could start from. This is normally the last block in this |
| 1159 | * group. The search will end when we found the start of next |
| 1160 | * possible reservable space is out of this boundary. |
| 1161 | * This could handle the cross boundary reservation window |
| 1162 | * request. |
| 1163 | * |
| 1164 | * basically we search from the given range, rather than the whole |
| 1165 | * reservation double linked list, (start_block, last_block) |
| 1166 | * to find a free region that is of my size and has not |
| 1167 | * been reserved. |
| 1168 | * |
| 1169 | */ |
| 1170 | static int find_next_reservable_window( |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1171 | struct ext4_reserve_window_node *search_head, |
| 1172 | struct ext4_reserve_window_node *my_rsv, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1173 | struct super_block * sb, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1174 | ext4_fsblk_t start_block, |
| 1175 | ext4_fsblk_t last_block) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1176 | { |
| 1177 | struct rb_node *next; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1178 | struct ext4_reserve_window_node *rsv, *prev; |
| 1179 | ext4_fsblk_t cur; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1180 | int size = my_rsv->rsv_goal_size; |
| 1181 | |
| 1182 | /* TODO: make the start of the reservation window byte-aligned */ |
| 1183 | /* cur = *start_block & ~7;*/ |
| 1184 | cur = start_block; |
| 1185 | rsv = search_head; |
| 1186 | if (!rsv) |
| 1187 | return -1; |
| 1188 | |
| 1189 | while (1) { |
| 1190 | if (cur <= rsv->rsv_end) |
| 1191 | cur = rsv->rsv_end + 1; |
| 1192 | |
| 1193 | /* TODO? |
| 1194 | * in the case we could not find a reservable space |
| 1195 | * that is what is expected, during the re-search, we could |
| 1196 | * remember what's the largest reservable space we could have |
| 1197 | * and return that one. |
| 1198 | * |
| 1199 | * For now it will fail if we could not find the reservable |
| 1200 | * space with expected-size (or more)... |
| 1201 | */ |
| 1202 | if (cur > last_block) |
| 1203 | return -1; /* fail */ |
| 1204 | |
| 1205 | prev = rsv; |
| 1206 | next = rb_next(&rsv->rsv_node); |
Hugh Dickins | b78a657 | 2006-12-06 20:39:21 -0800 | [diff] [blame] | 1207 | rsv = rb_entry(next,struct ext4_reserve_window_node,rsv_node); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1208 | |
| 1209 | /* |
| 1210 | * Reached the last reservation, we can just append to the |
| 1211 | * previous one. |
| 1212 | */ |
| 1213 | if (!next) |
| 1214 | break; |
| 1215 | |
| 1216 | if (cur + size <= rsv->rsv_start) { |
| 1217 | /* |
| 1218 | * Found a reserveable space big enough. We could |
| 1219 | * have a reservation across the group boundary here |
| 1220 | */ |
| 1221 | break; |
| 1222 | } |
| 1223 | } |
| 1224 | /* |
| 1225 | * we come here either : |
| 1226 | * when we reach the end of the whole list, |
| 1227 | * and there is empty reservable space after last entry in the list. |
| 1228 | * append it to the end of the list. |
| 1229 | * |
| 1230 | * or we found one reservable space in the middle of the list, |
| 1231 | * return the reservation window that we could append to. |
| 1232 | * succeed. |
| 1233 | */ |
| 1234 | |
| 1235 | if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window))) |
| 1236 | rsv_window_remove(sb, my_rsv); |
| 1237 | |
| 1238 | /* |
| 1239 | * Let's book the whole avaliable window for now. We will check the |
| 1240 | * disk bitmap later and then, if there are free blocks then we adjust |
| 1241 | * the window size if it's larger than requested. |
| 1242 | * Otherwise, we will remove this node from the tree next time |
| 1243 | * call find_next_reservable_window. |
| 1244 | */ |
| 1245 | my_rsv->rsv_start = cur; |
| 1246 | my_rsv->rsv_end = cur + size - 1; |
| 1247 | my_rsv->rsv_alloc_hit = 0; |
| 1248 | |
| 1249 | if (prev != my_rsv) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1250 | ext4_rsv_window_add(sb, my_rsv); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1251 | |
| 1252 | return 0; |
| 1253 | } |
| 1254 | |
| 1255 | /** |
| 1256 | * alloc_new_reservation()--allocate a new reservation window |
| 1257 | * |
| 1258 | * To make a new reservation, we search part of the filesystem |
| 1259 | * reservation list (the list that inside the group). We try to |
| 1260 | * allocate a new reservation window near the allocation goal, |
| 1261 | * or the beginning of the group, if there is no goal. |
| 1262 | * |
| 1263 | * We first find a reservable space after the goal, then from |
| 1264 | * there, we check the bitmap for the first free block after |
| 1265 | * it. If there is no free block until the end of group, then the |
| 1266 | * whole group is full, we failed. Otherwise, check if the free |
| 1267 | * block is inside the expected reservable space, if so, we |
| 1268 | * succeed. |
| 1269 | * If the first free block is outside the reservable space, then |
| 1270 | * start from the first free block, we search for next available |
| 1271 | * space, and go on. |
| 1272 | * |
| 1273 | * on succeed, a new reservation will be found and inserted into the list |
| 1274 | * It contains at least one free block, and it does not overlap with other |
| 1275 | * reservation windows. |
| 1276 | * |
| 1277 | * failed: we failed to find a reservation window in this group |
| 1278 | * |
| 1279 | * @rsv: the reservation |
| 1280 | * |
| 1281 | * @grp_goal: The goal (group-relative). It is where the search for a |
| 1282 | * free reservable space should start from. |
| 1283 | * if we have a grp_goal(grp_goal >0 ), then start from there, |
| 1284 | * no grp_goal(grp_goal = -1), we start from the first block |
| 1285 | * of the group. |
| 1286 | * |
| 1287 | * @sb: the super block |
| 1288 | * @group: the group we are trying to allocate in |
| 1289 | * @bitmap_bh: the block group block bitmap |
| 1290 | * |
| 1291 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1292 | static int alloc_new_reservation(struct ext4_reserve_window_node *my_rsv, |
| 1293 | ext4_grpblk_t grp_goal, struct super_block *sb, |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1294 | ext4_group_t group, struct buffer_head *bitmap_bh) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1295 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1296 | struct ext4_reserve_window_node *search_head; |
| 1297 | ext4_fsblk_t group_first_block, group_end_block, start_block; |
| 1298 | ext4_grpblk_t first_free_block; |
| 1299 | struct rb_root *fs_rsv_root = &EXT4_SB(sb)->s_rsv_window_root; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1300 | unsigned long size; |
| 1301 | int ret; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1302 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1303 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1304 | group_first_block = ext4_group_first_block_no(sb, group); |
| 1305 | group_end_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1306 | |
| 1307 | if (grp_goal < 0) |
| 1308 | start_block = group_first_block; |
| 1309 | else |
| 1310 | start_block = grp_goal + group_first_block; |
| 1311 | |
| 1312 | size = my_rsv->rsv_goal_size; |
| 1313 | |
| 1314 | if (!rsv_is_empty(&my_rsv->rsv_window)) { |
| 1315 | /* |
| 1316 | * if the old reservation is cross group boundary |
| 1317 | * and if the goal is inside the old reservation window, |
| 1318 | * we will come here when we just failed to allocate from |
| 1319 | * the first part of the window. We still have another part |
| 1320 | * that belongs to the next group. In this case, there is no |
| 1321 | * point to discard our window and try to allocate a new one |
| 1322 | * in this group(which will fail). we should |
| 1323 | * keep the reservation window, just simply move on. |
| 1324 | * |
| 1325 | * Maybe we could shift the start block of the reservation |
| 1326 | * window to the first block of next group. |
| 1327 | */ |
| 1328 | |
| 1329 | if ((my_rsv->rsv_start <= group_end_block) && |
| 1330 | (my_rsv->rsv_end > group_end_block) && |
| 1331 | (start_block >= my_rsv->rsv_start)) |
| 1332 | return -1; |
| 1333 | |
| 1334 | if ((my_rsv->rsv_alloc_hit > |
| 1335 | (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) { |
| 1336 | /* |
| 1337 | * if the previously allocation hit ratio is |
| 1338 | * greater than 1/2, then we double the size of |
| 1339 | * the reservation window the next time, |
| 1340 | * otherwise we keep the same size window |
| 1341 | */ |
| 1342 | size = size * 2; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1343 | if (size > EXT4_MAX_RESERVE_BLOCKS) |
| 1344 | size = EXT4_MAX_RESERVE_BLOCKS; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1345 | my_rsv->rsv_goal_size= size; |
| 1346 | } |
| 1347 | } |
| 1348 | |
| 1349 | spin_lock(rsv_lock); |
| 1350 | /* |
| 1351 | * shift the search start to the window near the goal block |
| 1352 | */ |
| 1353 | search_head = search_reserve_window(fs_rsv_root, start_block); |
| 1354 | |
| 1355 | /* |
| 1356 | * find_next_reservable_window() simply finds a reservable window |
| 1357 | * inside the given range(start_block, group_end_block). |
| 1358 | * |
| 1359 | * To make sure the reservation window has a free bit inside it, we |
| 1360 | * need to check the bitmap after we found a reservable window. |
| 1361 | */ |
| 1362 | retry: |
| 1363 | ret = find_next_reservable_window(search_head, my_rsv, sb, |
| 1364 | start_block, group_end_block); |
| 1365 | |
| 1366 | if (ret == -1) { |
| 1367 | if (!rsv_is_empty(&my_rsv->rsv_window)) |
| 1368 | rsv_window_remove(sb, my_rsv); |
| 1369 | spin_unlock(rsv_lock); |
| 1370 | return -1; |
| 1371 | } |
| 1372 | |
| 1373 | /* |
| 1374 | * On success, find_next_reservable_window() returns the |
| 1375 | * reservation window where there is a reservable space after it. |
| 1376 | * Before we reserve this reservable space, we need |
| 1377 | * to make sure there is at least a free block inside this region. |
| 1378 | * |
| 1379 | * searching the first free bit on the block bitmap and copy of |
| 1380 | * last committed bitmap alternatively, until we found a allocatable |
| 1381 | * block. Search start from the start block of the reservable space |
| 1382 | * we just found. |
| 1383 | */ |
| 1384 | spin_unlock(rsv_lock); |
| 1385 | first_free_block = bitmap_search_next_usable_block( |
| 1386 | my_rsv->rsv_start - group_first_block, |
| 1387 | bitmap_bh, group_end_block - group_first_block + 1); |
| 1388 | |
| 1389 | if (first_free_block < 0) { |
| 1390 | /* |
| 1391 | * no free block left on the bitmap, no point |
| 1392 | * to reserve the space. return failed. |
| 1393 | */ |
| 1394 | spin_lock(rsv_lock); |
| 1395 | if (!rsv_is_empty(&my_rsv->rsv_window)) |
| 1396 | rsv_window_remove(sb, my_rsv); |
| 1397 | spin_unlock(rsv_lock); |
| 1398 | return -1; /* failed */ |
| 1399 | } |
| 1400 | |
| 1401 | start_block = first_free_block + group_first_block; |
| 1402 | /* |
| 1403 | * check if the first free block is within the |
| 1404 | * free space we just reserved |
| 1405 | */ |
Hugh Dickins | b2f2c76 | 2006-12-06 20:39:20 -0800 | [diff] [blame] | 1406 | if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1407 | return 0; /* success */ |
| 1408 | /* |
| 1409 | * if the first free bit we found is out of the reservable space |
| 1410 | * continue search for next reservable space, |
| 1411 | * start from where the free block is, |
| 1412 | * we also shift the list head to where we stopped last time |
| 1413 | */ |
| 1414 | search_head = my_rsv; |
| 1415 | spin_lock(rsv_lock); |
| 1416 | goto retry; |
| 1417 | } |
| 1418 | |
| 1419 | /** |
| 1420 | * try_to_extend_reservation() |
| 1421 | * @my_rsv: given reservation window |
| 1422 | * @sb: super block |
| 1423 | * @size: the delta to extend |
| 1424 | * |
| 1425 | * Attempt to expand the reservation window large enough to have |
| 1426 | * required number of free blocks |
| 1427 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1428 | * Since ext4_try_to_allocate() will always allocate blocks within |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1429 | * the reservation window range, if the window size is too small, |
| 1430 | * multiple blocks allocation has to stop at the end of the reservation |
| 1431 | * window. To make this more efficient, given the total number of |
| 1432 | * blocks needed and the current size of the window, we try to |
| 1433 | * expand the reservation window size if necessary on a best-effort |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1434 | * basis before ext4_new_blocks() tries to allocate blocks, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1435 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1436 | static void try_to_extend_reservation(struct ext4_reserve_window_node *my_rsv, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1437 | struct super_block *sb, int size) |
| 1438 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1439 | struct ext4_reserve_window_node *next_rsv; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1440 | struct rb_node *next; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1441 | spinlock_t *rsv_lock = &EXT4_SB(sb)->s_rsv_window_lock; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1442 | |
| 1443 | if (!spin_trylock(rsv_lock)) |
| 1444 | return; |
| 1445 | |
| 1446 | next = rb_next(&my_rsv->rsv_node); |
| 1447 | |
| 1448 | if (!next) |
| 1449 | my_rsv->rsv_end += size; |
| 1450 | else { |
Hugh Dickins | b78a657 | 2006-12-06 20:39:21 -0800 | [diff] [blame] | 1451 | next_rsv = rb_entry(next, struct ext4_reserve_window_node, rsv_node); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1452 | |
| 1453 | if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size) |
| 1454 | my_rsv->rsv_end += size; |
| 1455 | else |
| 1456 | my_rsv->rsv_end = next_rsv->rsv_start - 1; |
| 1457 | } |
| 1458 | spin_unlock(rsv_lock); |
| 1459 | } |
| 1460 | |
| 1461 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1462 | * ext4_try_to_allocate_with_rsv() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1463 | * @sb: superblock |
| 1464 | * @handle: handle to this transaction |
| 1465 | * @group: given allocation block group |
| 1466 | * @bitmap_bh: bufferhead holds the block bitmap |
| 1467 | * @grp_goal: given target block within the group |
| 1468 | * @count: target number of blocks to allocate |
| 1469 | * @my_rsv: reservation window |
| 1470 | * @errp: pointer to store the error code |
| 1471 | * |
| 1472 | * This is the main function used to allocate a new block and its reservation |
| 1473 | * window. |
| 1474 | * |
| 1475 | * Each time when a new block allocation is need, first try to allocate from |
| 1476 | * its own reservation. If it does not have a reservation window, instead of |
| 1477 | * looking for a free bit on bitmap first, then look up the reservation list to |
| 1478 | * see if it is inside somebody else's reservation window, we try to allocate a |
| 1479 | * reservation window for it starting from the goal first. Then do the block |
| 1480 | * allocation within the reservation window. |
| 1481 | * |
| 1482 | * This will avoid keeping on searching the reservation list again and |
| 1483 | * again when somebody is looking for a free block (without |
| 1484 | * reservation), and there are lots of free blocks, but they are all |
| 1485 | * being reserved. |
| 1486 | * |
| 1487 | * We use a red-black tree for the per-filesystem reservation list. |
| 1488 | * |
| 1489 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1490 | static ext4_grpblk_t |
| 1491 | ext4_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle, |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1492 | ext4_group_t group, struct buffer_head *bitmap_bh, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1493 | ext4_grpblk_t grp_goal, |
| 1494 | struct ext4_reserve_window_node * my_rsv, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1495 | unsigned long *count, int *errp) |
| 1496 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1497 | ext4_fsblk_t group_first_block, group_last_block; |
| 1498 | ext4_grpblk_t ret = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1499 | int fatal; |
| 1500 | unsigned long num = *count; |
| 1501 | |
| 1502 | *errp = 0; |
| 1503 | |
| 1504 | /* |
| 1505 | * Make sure we use undo access for the bitmap, because it is critical |
| 1506 | * that we do the frozen_data COW on bitmap buffers in all cases even |
| 1507 | * if the buffer is in BJ_Forget state in the committing transaction. |
| 1508 | */ |
| 1509 | BUFFER_TRACE(bitmap_bh, "get undo access for new block"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1510 | fatal = ext4_journal_get_undo_access(handle, bitmap_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1511 | if (fatal) { |
| 1512 | *errp = fatal; |
| 1513 | return -1; |
| 1514 | } |
| 1515 | |
| 1516 | /* |
| 1517 | * we don't deal with reservation when |
| 1518 | * filesystem is mounted without reservation |
| 1519 | * or the file is not a regular file |
| 1520 | * or last attempt to allocate a block with reservation turned on failed |
| 1521 | */ |
| 1522 | if (my_rsv == NULL ) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1523 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1524 | grp_goal, count, NULL); |
| 1525 | goto out; |
| 1526 | } |
| 1527 | /* |
| 1528 | * grp_goal is a group relative block number (if there is a goal) |
Hugh Dickins | e7dc95d | 2006-12-06 20:39:19 -0800 | [diff] [blame] | 1529 | * 0 <= grp_goal < EXT4_BLOCKS_PER_GROUP(sb) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1530 | * first block is a filesystem wide block number |
| 1531 | * first block is the block number of the first block in this group |
| 1532 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1533 | group_first_block = ext4_group_first_block_no(sb, group); |
| 1534 | group_last_block = group_first_block + (EXT4_BLOCKS_PER_GROUP(sb) - 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1535 | |
| 1536 | /* |
| 1537 | * Basically we will allocate a new block from inode's reservation |
| 1538 | * window. |
| 1539 | * |
| 1540 | * We need to allocate a new reservation window, if: |
| 1541 | * a) inode does not have a reservation window; or |
| 1542 | * b) last attempt to allocate a block from existing reservation |
| 1543 | * failed; or |
| 1544 | * c) we come here with a goal and with a reservation window |
| 1545 | * |
| 1546 | * We do not need to allocate a new reservation window if we come here |
| 1547 | * at the beginning with a goal and the goal is inside the window, or |
| 1548 | * we don't have a goal but already have a reservation window. |
| 1549 | * then we could go to allocate from the reservation window directly. |
| 1550 | */ |
| 1551 | while (1) { |
| 1552 | if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) || |
| 1553 | !goal_in_my_reservation(&my_rsv->rsv_window, |
| 1554 | grp_goal, group, sb)) { |
| 1555 | if (my_rsv->rsv_goal_size < *count) |
| 1556 | my_rsv->rsv_goal_size = *count; |
| 1557 | ret = alloc_new_reservation(my_rsv, grp_goal, sb, |
| 1558 | group, bitmap_bh); |
| 1559 | if (ret < 0) |
| 1560 | break; /* failed */ |
| 1561 | |
| 1562 | if (!goal_in_my_reservation(&my_rsv->rsv_window, |
| 1563 | grp_goal, group, sb)) |
| 1564 | grp_goal = -1; |
Hugh Dickins | e7dc95d | 2006-12-06 20:39:19 -0800 | [diff] [blame] | 1565 | } else if (grp_goal >= 0) { |
Mingming Cao | 1df1e63 | 2006-12-06 20:38:19 -0800 | [diff] [blame] | 1566 | int curr = my_rsv->rsv_end - |
| 1567 | (grp_goal + group_first_block) + 1; |
| 1568 | |
| 1569 | if (curr < *count) |
| 1570 | try_to_extend_reservation(my_rsv, sb, |
| 1571 | *count - curr); |
| 1572 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1573 | |
| 1574 | if ((my_rsv->rsv_start > group_last_block) || |
| 1575 | (my_rsv->rsv_end < group_first_block)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1576 | rsv_window_dump(&EXT4_SB(sb)->s_rsv_window_root, 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1577 | BUG(); |
| 1578 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1579 | ret = ext4_try_to_allocate(sb, handle, group, bitmap_bh, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1580 | grp_goal, &num, &my_rsv->rsv_window); |
| 1581 | if (ret >= 0) { |
| 1582 | my_rsv->rsv_alloc_hit += num; |
| 1583 | *count = num; |
| 1584 | break; /* succeed */ |
| 1585 | } |
| 1586 | num = *count; |
| 1587 | } |
| 1588 | out: |
| 1589 | if (ret >= 0) { |
| 1590 | BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for " |
| 1591 | "bitmap block"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1592 | fatal = ext4_journal_dirty_metadata(handle, bitmap_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1593 | if (fatal) { |
| 1594 | *errp = fatal; |
| 1595 | return -1; |
| 1596 | } |
| 1597 | return ret; |
| 1598 | } |
| 1599 | |
| 1600 | BUFFER_TRACE(bitmap_bh, "journal_release_buffer"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1601 | ext4_journal_release_buffer(handle, bitmap_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1602 | return ret; |
| 1603 | } |
| 1604 | |
| 1605 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1606 | * ext4_has_free_blocks() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1607 | * @sbi: in-core super block structure. |
| 1608 | * |
| 1609 | * Check if filesystem has at least 1 free block available for allocation. |
| 1610 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1611 | static int ext4_has_free_blocks(struct ext4_sb_info *sbi) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1612 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1613 | ext4_fsblk_t free_blocks, root_blocks; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1614 | |
| 1615 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); |
Laurent Vivier | bd81d8e | 2006-10-11 01:21:10 -0700 | [diff] [blame] | 1616 | root_blocks = ext4_r_blocks_count(sbi->s_es); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1617 | if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) && |
| 1618 | sbi->s_resuid != current->fsuid && |
| 1619 | (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) { |
| 1620 | return 0; |
| 1621 | } |
| 1622 | return 1; |
| 1623 | } |
| 1624 | |
| 1625 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1626 | * ext4_should_retry_alloc() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1627 | * @sb: super block |
| 1628 | * @retries number of attemps has been made |
| 1629 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1630 | * ext4_should_retry_alloc() is called when ENOSPC is returned, and if |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1631 | * it is profitable to retry the operation, this function will wait |
| 1632 | * for the current or commiting transaction to complete, and then |
| 1633 | * return TRUE. |
| 1634 | * |
| 1635 | * if the total number of retries exceed three times, return FALSE. |
| 1636 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1637 | int ext4_should_retry_alloc(struct super_block *sb, int *retries) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1638 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1639 | if (!ext4_has_free_blocks(EXT4_SB(sb)) || (*retries)++ > 3) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1640 | return 0; |
| 1641 | |
| 1642 | jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id); |
| 1643 | |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 1644 | return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1645 | } |
| 1646 | |
| 1647 | /** |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 1648 | * ext4_new_blocks_old() -- core block(s) allocation function |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1649 | * @handle: handle to this transaction |
| 1650 | * @inode: file inode |
| 1651 | * @goal: given target block(filesystem wide) |
| 1652 | * @count: target number of blocks to allocate |
| 1653 | * @errp: error code |
| 1654 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1655 | * ext4_new_blocks uses a goal block to assist allocation. It tries to |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1656 | * allocate block(s) from the block group contains the goal block first. If that |
| 1657 | * fails, it will try to allocate block(s) from other block groups without |
| 1658 | * any specific goal block. |
| 1659 | * |
| 1660 | */ |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 1661 | ext4_fsblk_t ext4_new_blocks_old(handle_t *handle, struct inode *inode, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1662 | ext4_fsblk_t goal, unsigned long *count, int *errp) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1663 | { |
| 1664 | struct buffer_head *bitmap_bh = NULL; |
| 1665 | struct buffer_head *gdp_bh; |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1666 | ext4_group_t group_no; |
| 1667 | ext4_group_t goal_group; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1668 | ext4_grpblk_t grp_target_blk; /* blockgroup relative goal block */ |
| 1669 | ext4_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/ |
| 1670 | ext4_fsblk_t ret_block; /* filesyetem-wide allocated block */ |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1671 | ext4_group_t bgi; /* blockgroup iteration index */ |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1672 | int fatal = 0, err; |
| 1673 | int performed_allocation = 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1674 | ext4_grpblk_t free_blocks; /* number of free blocks in a group */ |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1675 | struct super_block *sb; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1676 | struct ext4_group_desc *gdp; |
| 1677 | struct ext4_super_block *es; |
| 1678 | struct ext4_sb_info *sbi; |
| 1679 | struct ext4_reserve_window_node *my_rsv = NULL; |
| 1680 | struct ext4_block_alloc_info *block_i; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1681 | unsigned short windowsz = 0; |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1682 | ext4_group_t ngroups; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1683 | unsigned long num = *count; |
| 1684 | |
| 1685 | *errp = -ENOSPC; |
| 1686 | sb = inode->i_sb; |
| 1687 | if (!sb) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1688 | printk("ext4_new_block: nonexistent device"); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1689 | return 0; |
| 1690 | } |
| 1691 | |
| 1692 | /* |
| 1693 | * Check quota for allocation of this block. |
| 1694 | */ |
| 1695 | if (DQUOT_ALLOC_BLOCK(inode, num)) { |
| 1696 | *errp = -EDQUOT; |
| 1697 | return 0; |
| 1698 | } |
| 1699 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1700 | sbi = EXT4_SB(sb); |
| 1701 | es = EXT4_SB(sb)->s_es; |
Eric Sandeen | c549a95 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1702 | ext4_debug("goal=%llu.\n", goal); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1703 | /* |
| 1704 | * Allocate a block from reservation only when |
| 1705 | * filesystem is mounted with reservation(default,-o reservation), and |
| 1706 | * it's a regular file, and |
| 1707 | * the desired window size is greater than 0 (One could use ioctl |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1708 | * command EXT4_IOC_SETRSVSZ to set the window size to 0 to turn off |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1709 | * reservation on that particular file) |
| 1710 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1711 | block_i = EXT4_I(inode)->i_block_alloc_info; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1712 | if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0)) |
| 1713 | my_rsv = &block_i->rsv_window_node; |
| 1714 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1715 | if (!ext4_has_free_blocks(sbi)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1716 | *errp = -ENOSPC; |
| 1717 | goto out; |
| 1718 | } |
| 1719 | |
| 1720 | /* |
| 1721 | * First, test whether the goal block is free. |
| 1722 | */ |
| 1723 | if (goal < le32_to_cpu(es->s_first_data_block) || |
Laurent Vivier | bd81d8e | 2006-10-11 01:21:10 -0700 | [diff] [blame] | 1724 | goal >= ext4_blocks_count(es)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1725 | goal = le32_to_cpu(es->s_first_data_block); |
Mingming Cao | 3a5b2ec | 2006-10-11 01:21:05 -0700 | [diff] [blame] | 1726 | ext4_get_group_no_and_offset(sb, goal, &group_no, &grp_target_blk); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1727 | goal_group = group_no; |
| 1728 | retry_alloc: |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1729 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1730 | if (!gdp) |
| 1731 | goto io_error; |
| 1732 | |
| 1733 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); |
| 1734 | /* |
| 1735 | * if there is not enough free blocks to make a new resevation |
| 1736 | * turn off reservation for this allocation |
| 1737 | */ |
| 1738 | if (my_rsv && (free_blocks < windowsz) |
| 1739 | && (rsv_is_empty(&my_rsv->rsv_window))) |
| 1740 | my_rsv = NULL; |
| 1741 | |
| 1742 | if (free_blocks > 0) { |
Theodore Ts'o | 574ca17 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1743 | bitmap_bh = ext4_read_block_bitmap(sb, group_no); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1744 | if (!bitmap_bh) |
| 1745 | goto io_error; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1746 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1747 | group_no, bitmap_bh, grp_target_blk, |
| 1748 | my_rsv, &num, &fatal); |
| 1749 | if (fatal) |
| 1750 | goto out; |
| 1751 | if (grp_alloc_blk >= 0) |
| 1752 | goto allocated; |
| 1753 | } |
| 1754 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1755 | ngroups = EXT4_SB(sb)->s_groups_count; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1756 | smp_rmb(); |
| 1757 | |
| 1758 | /* |
| 1759 | * Now search the rest of the groups. We assume that |
Akinobu Mita | 144704e | 2008-02-06 01:40:15 -0800 | [diff] [blame] | 1760 | * group_no and gdp correctly point to the last group visited. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1761 | */ |
| 1762 | for (bgi = 0; bgi < ngroups; bgi++) { |
| 1763 | group_no++; |
| 1764 | if (group_no >= ngroups) |
| 1765 | group_no = 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1766 | gdp = ext4_get_group_desc(sb, group_no, &gdp_bh); |
Hugh Dickins | 341cee4 | 2006-12-06 20:39:24 -0800 | [diff] [blame] | 1767 | if (!gdp) |
| 1768 | goto io_error; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1769 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); |
| 1770 | /* |
| 1771 | * skip this group if the number of |
| 1772 | * free blocks is less than half of the reservation |
| 1773 | * window size. |
| 1774 | */ |
| 1775 | if (free_blocks <= (windowsz/2)) |
| 1776 | continue; |
| 1777 | |
| 1778 | brelse(bitmap_bh); |
Theodore Ts'o | 574ca17 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1779 | bitmap_bh = ext4_read_block_bitmap(sb, group_no); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1780 | if (!bitmap_bh) |
| 1781 | goto io_error; |
| 1782 | /* |
| 1783 | * try to allocate block(s) from this group, without a goal(-1). |
| 1784 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1785 | grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1786 | group_no, bitmap_bh, -1, my_rsv, |
| 1787 | &num, &fatal); |
| 1788 | if (fatal) |
| 1789 | goto out; |
| 1790 | if (grp_alloc_blk >= 0) |
| 1791 | goto allocated; |
| 1792 | } |
| 1793 | /* |
| 1794 | * We may end up a bogus ealier ENOSPC error due to |
| 1795 | * filesystem is "full" of reservations, but |
| 1796 | * there maybe indeed free blocks avaliable on disk |
| 1797 | * In this case, we just forget about the reservations |
| 1798 | * just do block allocation as without reservations. |
| 1799 | */ |
| 1800 | if (my_rsv) { |
| 1801 | my_rsv = NULL; |
Hugh Dickins | cd16c8f | 2006-12-06 20:39:18 -0800 | [diff] [blame] | 1802 | windowsz = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1803 | group_no = goal_group; |
| 1804 | goto retry_alloc; |
| 1805 | } |
| 1806 | /* No space left on the device */ |
| 1807 | *errp = -ENOSPC; |
| 1808 | goto out; |
| 1809 | |
| 1810 | allocated: |
| 1811 | |
Eric Sandeen | c549a95 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1812 | ext4_debug("using block group %lu(%d)\n", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1813 | group_no, gdp->bg_free_blocks_count); |
| 1814 | |
| 1815 | BUFFER_TRACE(gdp_bh, "get_write_access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1816 | fatal = ext4_journal_get_write_access(handle, gdp_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1817 | if (fatal) |
| 1818 | goto out; |
| 1819 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1820 | ret_block = grp_alloc_blk + ext4_group_first_block_no(sb, group_no); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1821 | |
Alexandre Ratchov | 8fadc14 | 2006-10-11 01:21:15 -0700 | [diff] [blame] | 1822 | if (in_range(ext4_block_bitmap(sb, gdp), ret_block, num) || |
Toshiyuki Okajima | 29bc5b4 | 2007-07-15 23:41:22 -0700 | [diff] [blame] | 1823 | in_range(ext4_inode_bitmap(sb, gdp), ret_block, num) || |
Alexandre Ratchov | 8fadc14 | 2006-10-11 01:21:15 -0700 | [diff] [blame] | 1824 | in_range(ret_block, ext4_inode_table(sb, gdp), |
Laurent Vivier | bd81d8e | 2006-10-11 01:21:10 -0700 | [diff] [blame] | 1825 | EXT4_SB(sb)->s_itb_per_group) || |
Alexandre Ratchov | 8fadc14 | 2006-10-11 01:21:15 -0700 | [diff] [blame] | 1826 | in_range(ret_block + num - 1, ext4_inode_table(sb, gdp), |
Aneesh Kumar K.V | cb47dce | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1827 | EXT4_SB(sb)->s_itb_per_group)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1828 | ext4_error(sb, "ext4_new_block", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1829 | "Allocating block in system zone - " |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 1830 | "blocks from %llu, length %lu", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1831 | ret_block, num); |
Aneesh Kumar K.V | 519deca | 2008-05-15 14:43:20 -0400 | [diff] [blame] | 1832 | /* |
| 1833 | * claim_block marked the blocks we allocated |
| 1834 | * as in use. So we may want to selectively |
| 1835 | * mark some of the blocks as free |
| 1836 | */ |
| 1837 | goto retry_alloc; |
Aneesh Kumar K.V | cb47dce | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1838 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1839 | |
| 1840 | performed_allocation = 1; |
| 1841 | |
Jose R. Santos | e23291b | 2007-07-18 08:57:06 -0400 | [diff] [blame] | 1842 | #ifdef CONFIG_JBD2_DEBUG |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1843 | { |
| 1844 | struct buffer_head *debug_bh; |
| 1845 | |
| 1846 | /* Record bitmap buffer state in the newly allocated block */ |
| 1847 | debug_bh = sb_find_get_block(sb, ret_block); |
| 1848 | if (debug_bh) { |
| 1849 | BUFFER_TRACE(debug_bh, "state when allocated"); |
| 1850 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state"); |
| 1851 | brelse(debug_bh); |
| 1852 | } |
| 1853 | } |
| 1854 | jbd_lock_bh_state(bitmap_bh); |
| 1855 | spin_lock(sb_bgl_lock(sbi, group_no)); |
| 1856 | if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) { |
| 1857 | int i; |
| 1858 | |
| 1859 | for (i = 0; i < num; i++) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1860 | if (ext4_test_bit(grp_alloc_blk+i, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1861 | bh2jh(bitmap_bh)->b_committed_data)) { |
| 1862 | printk("%s: block was unexpectedly set in " |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 1863 | "b_committed_data\n", __func__); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1864 | } |
| 1865 | } |
| 1866 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1867 | ext4_debug("found bit %d\n", grp_alloc_blk); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1868 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
| 1869 | jbd_unlock_bh_state(bitmap_bh); |
| 1870 | #endif |
| 1871 | |
Laurent Vivier | bd81d8e | 2006-10-11 01:21:10 -0700 | [diff] [blame] | 1872 | if (ret_block + num - 1 >= ext4_blocks_count(es)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1873 | ext4_error(sb, "ext4_new_block", |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 1874 | "block(%llu) >= blocks count(%llu) - " |
Mingming Cao | 3a5b2ec | 2006-10-11 01:21:05 -0700 | [diff] [blame] | 1875 | "block_group = %lu, es == %p ", ret_block, |
Laurent Vivier | bd81d8e | 2006-10-11 01:21:10 -0700 | [diff] [blame] | 1876 | ext4_blocks_count(es), group_no, es); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1877 | goto out; |
| 1878 | } |
| 1879 | |
| 1880 | /* |
| 1881 | * It is up to the caller to add the new buffer to a journal |
| 1882 | * list of some description. We don't know in advance whether |
| 1883 | * the caller wants to use it as metadata or data. |
| 1884 | */ |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1885 | spin_lock(sb_bgl_lock(sbi, group_no)); |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 1886 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) |
| 1887 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
Marcin Slusarz | e8546d0 | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 1888 | le16_add_cpu(&gdp->bg_free_blocks_count, -num); |
Andreas Dilger | 717d50e | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 1889 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group_no, gdp); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1890 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
Peter Zijlstra | 3cb4f9f | 2007-10-16 23:25:42 -0700 | [diff] [blame] | 1891 | percpu_counter_sub(&sbi->s_freeblocks_counter, num); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1892 | |
Jose R. Santos | 772cb7c | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1893 | if (sbi->s_log_groups_per_flex) { |
| 1894 | ext4_group_t flex_group = ext4_flex_group(sbi, group_no); |
| 1895 | spin_lock(sb_bgl_lock(sbi, flex_group)); |
| 1896 | sbi->s_flex_groups[flex_group].free_blocks -= num; |
| 1897 | spin_unlock(sb_bgl_lock(sbi, flex_group)); |
| 1898 | } |
| 1899 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1900 | BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1901 | err = ext4_journal_dirty_metadata(handle, gdp_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1902 | if (!fatal) |
| 1903 | fatal = err; |
| 1904 | |
| 1905 | sb->s_dirt = 1; |
| 1906 | if (fatal) |
| 1907 | goto out; |
| 1908 | |
| 1909 | *errp = 0; |
| 1910 | brelse(bitmap_bh); |
| 1911 | DQUOT_FREE_BLOCK(inode, *count-num); |
| 1912 | *count = num; |
| 1913 | return ret_block; |
| 1914 | |
| 1915 | io_error: |
| 1916 | *errp = -EIO; |
| 1917 | out: |
| 1918 | if (fatal) { |
| 1919 | *errp = fatal; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1920 | ext4_std_error(sb, fatal); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1921 | } |
| 1922 | /* |
| 1923 | * Undo the block allocation |
| 1924 | */ |
| 1925 | if (!performed_allocation) |
| 1926 | DQUOT_FREE_BLOCK(inode, *count); |
| 1927 | brelse(bitmap_bh); |
| 1928 | return 0; |
| 1929 | } |
| 1930 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1931 | ext4_fsblk_t ext4_new_block(handle_t *handle, struct inode *inode, |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 1932 | ext4_fsblk_t goal, int *errp) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1933 | { |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 1934 | struct ext4_allocation_request ar; |
| 1935 | ext4_fsblk_t ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1936 | |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 1937 | if (!test_opt(inode->i_sb, MBALLOC)) { |
| 1938 | unsigned long count = 1; |
| 1939 | ret = ext4_new_blocks_old(handle, inode, goal, &count, errp); |
| 1940 | return ret; |
| 1941 | } |
| 1942 | |
| 1943 | memset(&ar, 0, sizeof(ar)); |
| 1944 | ar.inode = inode; |
| 1945 | ar.goal = goal; |
| 1946 | ar.len = 1; |
| 1947 | ret = ext4_mb_new_blocks(handle, &ar, errp); |
| 1948 | return ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1949 | } |
| 1950 | |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 1951 | ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode, |
| 1952 | ext4_fsblk_t goal, unsigned long *count, int *errp) |
| 1953 | { |
| 1954 | struct ext4_allocation_request ar; |
| 1955 | ext4_fsblk_t ret; |
| 1956 | |
| 1957 | if (!test_opt(inode->i_sb, MBALLOC)) { |
| 1958 | ret = ext4_new_blocks_old(handle, inode, goal, count, errp); |
| 1959 | return ret; |
| 1960 | } |
| 1961 | |
| 1962 | memset(&ar, 0, sizeof(ar)); |
| 1963 | ar.inode = inode; |
| 1964 | ar.goal = goal; |
| 1965 | ar.len = *count; |
| 1966 | ret = ext4_mb_new_blocks(handle, &ar, errp); |
| 1967 | *count = ar.len; |
| 1968 | return ret; |
| 1969 | } |
| 1970 | |
| 1971 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1972 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1973 | * ext4_count_free_blocks() -- count filesystem free blocks |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1974 | * @sb: superblock |
| 1975 | * |
| 1976 | * Adds up the number of free blocks from each block group. |
| 1977 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1978 | ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1979 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1980 | ext4_fsblk_t desc_count; |
| 1981 | struct ext4_group_desc *gdp; |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1982 | ext4_group_t i; |
| 1983 | ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1984 | #ifdef EXT4FS_DEBUG |
| 1985 | struct ext4_super_block *es; |
| 1986 | ext4_fsblk_t bitmap_count; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1987 | unsigned long x; |
| 1988 | struct buffer_head *bitmap_bh = NULL; |
| 1989 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1990 | es = EXT4_SB(sb)->s_es; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1991 | desc_count = 0; |
| 1992 | bitmap_count = 0; |
| 1993 | gdp = NULL; |
| 1994 | |
| 1995 | smp_rmb(); |
| 1996 | for (i = 0; i < ngroups; i++) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1997 | gdp = ext4_get_group_desc(sb, i, NULL); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1998 | if (!gdp) |
| 1999 | continue; |
| 2000 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); |
| 2001 | brelse(bitmap_bh); |
Theodore Ts'o | 574ca17 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2002 | bitmap_bh = ext4_read_block_bitmap(sb, i); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2003 | if (bitmap_bh == NULL) |
| 2004 | continue; |
| 2005 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2006 | x = ext4_count_free(bitmap_bh, sb->s_blocksize); |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2007 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2008 | i, le16_to_cpu(gdp->bg_free_blocks_count), x); |
| 2009 | bitmap_count += x; |
| 2010 | } |
| 2011 | brelse(bitmap_bh); |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 2012 | printk("ext4_count_free_blocks: stored = %llu" |
| 2013 | ", computed = %llu, %llu\n", |
Eric Sandeen | c549a95 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2014 | ext4_free_blocks_count(es), |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2015 | desc_count, bitmap_count); |
| 2016 | return bitmap_count; |
| 2017 | #else |
| 2018 | desc_count = 0; |
| 2019 | smp_rmb(); |
| 2020 | for (i = 0; i < ngroups; i++) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2021 | gdp = ext4_get_group_desc(sb, i, NULL); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2022 | if (!gdp) |
| 2023 | continue; |
| 2024 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); |
| 2025 | } |
| 2026 | |
| 2027 | return desc_count; |
| 2028 | #endif |
| 2029 | } |
| 2030 | |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2031 | static inline int test_root(ext4_group_t a, int b) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2032 | { |
| 2033 | int num = b; |
| 2034 | |
| 2035 | while (a > num) |
| 2036 | num *= b; |
| 2037 | return num == a; |
| 2038 | } |
| 2039 | |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2040 | static int ext4_group_sparse(ext4_group_t group) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2041 | { |
| 2042 | if (group <= 1) |
| 2043 | return 1; |
| 2044 | if (!(group & 1)) |
| 2045 | return 0; |
| 2046 | return (test_root(group, 7) || test_root(group, 5) || |
| 2047 | test_root(group, 3)); |
| 2048 | } |
| 2049 | |
| 2050 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2051 | * ext4_bg_has_super - number of blocks used by the superblock in group |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2052 | * @sb: superblock for filesystem |
| 2053 | * @group: group number to check |
| 2054 | * |
| 2055 | * Return the number of blocks used by the superblock (primary or backup) |
| 2056 | * in this group. Currently this will be only 0 or 1. |
| 2057 | */ |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2058 | int ext4_bg_has_super(struct super_block *sb, ext4_group_t group) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2059 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2060 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
| 2061 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && |
| 2062 | !ext4_group_sparse(group)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2063 | return 0; |
| 2064 | return 1; |
| 2065 | } |
| 2066 | |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2067 | static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, |
| 2068 | ext4_group_t group) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2069 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2070 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2071 | ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb); |
| 2072 | ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2073 | |
| 2074 | if (group == first || group == first + 1 || group == last) |
| 2075 | return 1; |
| 2076 | return 0; |
| 2077 | } |
| 2078 | |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2079 | static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, |
| 2080 | ext4_group_t group) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2081 | { |
Akinobu Mita | 859cb93 | 2008-02-06 01:40:17 -0800 | [diff] [blame] | 2082 | return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2083 | } |
| 2084 | |
| 2085 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2086 | * ext4_bg_num_gdb - number of blocks used by the group table in group |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2087 | * @sb: superblock for filesystem |
| 2088 | * @group: group number to check |
| 2089 | * |
| 2090 | * Return the number of blocks used by the group descriptor table |
| 2091 | * (primary or backup) in this group. In the future there may be a |
| 2092 | * different number of descriptor blocks in each group. |
| 2093 | */ |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2094 | unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2095 | { |
| 2096 | unsigned long first_meta_bg = |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2097 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); |
| 2098 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2099 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2100 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) || |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2101 | metagroup < first_meta_bg) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2102 | return ext4_bg_num_gdb_nometa(sb,group); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2103 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2104 | return ext4_bg_num_gdb_meta(sb,group); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2105 | |
| 2106 | } |