| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* | 
 | 2 |  *  linux/fs/ext3/ialloc.c | 
 | 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 |  *  BSD ufs-inspired inode and directory allocation by | 
 | 10 |  *  Stephen Tweedie (sct@redhat.com), 1993 | 
 | 11 |  *  Big-endian to little-endian byte-swapping/bitmaps by | 
 | 12 |  *        David S. Miller (davem@caip.rutgers.edu), 1995 | 
 | 13 |  */ | 
 | 14 |  | 
 | 15 | #include <linux/time.h> | 
 | 16 | #include <linux/fs.h> | 
 | 17 | #include <linux/jbd.h> | 
 | 18 | #include <linux/ext3_fs.h> | 
 | 19 | #include <linux/ext3_jbd.h> | 
 | 20 | #include <linux/stat.h> | 
 | 21 | #include <linux/string.h> | 
 | 22 | #include <linux/quotaops.h> | 
 | 23 | #include <linux/buffer_head.h> | 
 | 24 | #include <linux/random.h> | 
 | 25 | #include <linux/bitops.h> | 
 | 26 |  | 
 | 27 | #include <asm/byteorder.h> | 
 | 28 |  | 
 | 29 | #include "xattr.h" | 
 | 30 | #include "acl.h" | 
 | 31 |  | 
 | 32 | /* | 
 | 33 |  * ialloc.c contains the inodes allocation and deallocation routines | 
 | 34 |  */ | 
 | 35 |  | 
 | 36 | /* | 
 | 37 |  * The free inodes are managed by bitmaps.  A file system contains several | 
 | 38 |  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap | 
 | 39 |  * block for inodes, N blocks for the inode table and data blocks. | 
 | 40 |  * | 
 | 41 |  * The file system contains group descriptors which are located after the | 
 | 42 |  * super block.  Each descriptor contains the number of the bitmap block and | 
 | 43 |  * the free blocks count in the block. | 
 | 44 |  */ | 
 | 45 |  | 
 | 46 |  | 
 | 47 | /* | 
 | 48 |  * Read the inode allocation bitmap for a given block_group, reading | 
 | 49 |  * into the specified slot in the superblock's bitmap cache. | 
 | 50 |  * | 
 | 51 |  * Return buffer_head of bitmap on success or NULL. | 
 | 52 |  */ | 
 | 53 | static struct buffer_head * | 
 | 54 | read_inode_bitmap(struct super_block * sb, unsigned long block_group) | 
 | 55 | { | 
 | 56 | 	struct ext3_group_desc *desc; | 
 | 57 | 	struct buffer_head *bh = NULL; | 
 | 58 |  | 
 | 59 | 	desc = ext3_get_group_desc(sb, block_group, NULL); | 
 | 60 | 	if (!desc) | 
 | 61 | 		goto error_out; | 
 | 62 |  | 
 | 63 | 	bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap)); | 
 | 64 | 	if (!bh) | 
 | 65 | 		ext3_error(sb, "read_inode_bitmap", | 
 | 66 | 			    "Cannot read inode bitmap - " | 
 | 67 | 			    "block_group = %lu, inode_bitmap = %u", | 
 | 68 | 			    block_group, le32_to_cpu(desc->bg_inode_bitmap)); | 
 | 69 | error_out: | 
 | 70 | 	return bh; | 
 | 71 | } | 
 | 72 |  | 
 | 73 | /* | 
 | 74 |  * NOTE! When we get the inode, we're the only people | 
 | 75 |  * that have access to it, and as such there are no | 
 | 76 |  * race conditions we have to worry about. The inode | 
 | 77 |  * is not on the hash-lists, and it cannot be reached | 
 | 78 |  * through the filesystem because the directory entry | 
 | 79 |  * has been deleted earlier. | 
 | 80 |  * | 
 | 81 |  * HOWEVER: we must make sure that we get no aliases, | 
 | 82 |  * which means that we have to call "clear_inode()" | 
 | 83 |  * _before_ we mark the inode not in use in the inode | 
 | 84 |  * bitmaps. Otherwise a newly created file might use | 
 | 85 |  * the same inode number (not actually the same pointer | 
 | 86 |  * though), and then we'd have two inodes sharing the | 
 | 87 |  * same inode number and space on the harddisk. | 
 | 88 |  */ | 
 | 89 | void ext3_free_inode (handle_t *handle, struct inode * inode) | 
 | 90 | { | 
 | 91 | 	struct super_block * sb = inode->i_sb; | 
 | 92 | 	int is_directory; | 
 | 93 | 	unsigned long ino; | 
 | 94 | 	struct buffer_head *bitmap_bh = NULL; | 
 | 95 | 	struct buffer_head *bh2; | 
 | 96 | 	unsigned long block_group; | 
 | 97 | 	unsigned long bit; | 
 | 98 | 	struct ext3_group_desc * gdp; | 
 | 99 | 	struct ext3_super_block * es; | 
 | 100 | 	struct ext3_sb_info *sbi; | 
 | 101 | 	int fatal = 0, err; | 
 | 102 |  | 
 | 103 | 	if (atomic_read(&inode->i_count) > 1) { | 
 | 104 | 		printk ("ext3_free_inode: inode has count=%d\n", | 
 | 105 | 					atomic_read(&inode->i_count)); | 
 | 106 | 		return; | 
 | 107 | 	} | 
 | 108 | 	if (inode->i_nlink) { | 
 | 109 | 		printk ("ext3_free_inode: inode has nlink=%d\n", | 
 | 110 | 			inode->i_nlink); | 
 | 111 | 		return; | 
 | 112 | 	} | 
 | 113 | 	if (!sb) { | 
 | 114 | 		printk("ext3_free_inode: inode on nonexistent device\n"); | 
 | 115 | 		return; | 
 | 116 | 	} | 
 | 117 | 	sbi = EXT3_SB(sb); | 
 | 118 |  | 
 | 119 | 	ino = inode->i_ino; | 
 | 120 | 	ext3_debug ("freeing inode %lu\n", ino); | 
 | 121 |  | 
 | 122 | 	/* | 
 | 123 | 	 * Note: we must free any quota before locking the superblock, | 
 | 124 | 	 * as writing the quota to disk may need the lock as well. | 
 | 125 | 	 */ | 
 | 126 | 	DQUOT_INIT(inode); | 
 | 127 | 	ext3_xattr_delete_inode(handle, inode); | 
 | 128 | 	DQUOT_FREE_INODE(inode); | 
 | 129 | 	DQUOT_DROP(inode); | 
 | 130 |  | 
 | 131 | 	is_directory = S_ISDIR(inode->i_mode); | 
 | 132 |  | 
 | 133 | 	/* Do this BEFORE marking the inode not in use or returning an error */ | 
 | 134 | 	clear_inode (inode); | 
 | 135 |  | 
 | 136 | 	es = EXT3_SB(sb)->s_es; | 
 | 137 | 	if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | 
 | 138 | 		ext3_error (sb, "ext3_free_inode", | 
 | 139 | 			    "reserved or nonexistent inode %lu", ino); | 
 | 140 | 		goto error_return; | 
 | 141 | 	} | 
 | 142 | 	block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); | 
 | 143 | 	bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb); | 
 | 144 | 	bitmap_bh = read_inode_bitmap(sb, block_group); | 
 | 145 | 	if (!bitmap_bh) | 
 | 146 | 		goto error_return; | 
 | 147 |  | 
 | 148 | 	BUFFER_TRACE(bitmap_bh, "get_write_access"); | 
 | 149 | 	fatal = ext3_journal_get_write_access(handle, bitmap_bh); | 
 | 150 | 	if (fatal) | 
 | 151 | 		goto error_return; | 
 | 152 |  | 
 | 153 | 	/* Ok, now we can actually update the inode bitmaps.. */ | 
 | 154 | 	if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group), | 
 | 155 | 					bit, bitmap_bh->b_data)) | 
 | 156 | 		ext3_error (sb, "ext3_free_inode", | 
 | 157 | 			      "bit already cleared for inode %lu", ino); | 
 | 158 | 	else { | 
 | 159 | 		gdp = ext3_get_group_desc (sb, block_group, &bh2); | 
 | 160 |  | 
 | 161 | 		BUFFER_TRACE(bh2, "get_write_access"); | 
 | 162 | 		fatal = ext3_journal_get_write_access(handle, bh2); | 
 | 163 | 		if (fatal) goto error_return; | 
 | 164 |  | 
 | 165 | 		if (gdp) { | 
 | 166 | 			spin_lock(sb_bgl_lock(sbi, block_group)); | 
 | 167 | 			gdp->bg_free_inodes_count = cpu_to_le16( | 
 | 168 | 				le16_to_cpu(gdp->bg_free_inodes_count) + 1); | 
 | 169 | 			if (is_directory) | 
 | 170 | 				gdp->bg_used_dirs_count = cpu_to_le16( | 
 | 171 | 				  le16_to_cpu(gdp->bg_used_dirs_count) - 1); | 
 | 172 | 			spin_unlock(sb_bgl_lock(sbi, block_group)); | 
 | 173 | 			percpu_counter_inc(&sbi->s_freeinodes_counter); | 
 | 174 | 			if (is_directory) | 
 | 175 | 				percpu_counter_dec(&sbi->s_dirs_counter); | 
 | 176 |  | 
 | 177 | 		} | 
 | 178 | 		BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata"); | 
 | 179 | 		err = ext3_journal_dirty_metadata(handle, bh2); | 
 | 180 | 		if (!fatal) fatal = err; | 
 | 181 | 	} | 
 | 182 | 	BUFFER_TRACE(bitmap_bh, "call ext3_journal_dirty_metadata"); | 
 | 183 | 	err = ext3_journal_dirty_metadata(handle, bitmap_bh); | 
 | 184 | 	if (!fatal) | 
 | 185 | 		fatal = err; | 
 | 186 | 	sb->s_dirt = 1; | 
 | 187 | error_return: | 
 | 188 | 	brelse(bitmap_bh); | 
 | 189 | 	ext3_std_error(sb, fatal); | 
 | 190 | } | 
 | 191 |  | 
 | 192 | /* | 
 | 193 |  * There are two policies for allocating an inode.  If the new inode is | 
 | 194 |  * a directory, then a forward search is made for a block group with both | 
 | 195 |  * free space and a low directory-to-inode ratio; if that fails, then of | 
 | 196 |  * the groups with above-average free space, that group with the fewest | 
 | 197 |  * directories already is chosen. | 
 | 198 |  * | 
 | 199 |  * For other inodes, search forward from the parent directory\'s block | 
 | 200 |  * group to find a free inode. | 
 | 201 |  */ | 
 | 202 | static int find_group_dir(struct super_block *sb, struct inode *parent) | 
 | 203 | { | 
 | 204 | 	int ngroups = EXT3_SB(sb)->s_groups_count; | 
 | 205 | 	int freei, avefreei; | 
 | 206 | 	struct ext3_group_desc *desc, *best_desc = NULL; | 
 | 207 | 	struct buffer_head *bh; | 
 | 208 | 	int group, best_group = -1; | 
 | 209 |  | 
 | 210 | 	freei = percpu_counter_read_positive(&EXT3_SB(sb)->s_freeinodes_counter); | 
 | 211 | 	avefreei = freei / ngroups; | 
 | 212 |  | 
 | 213 | 	for (group = 0; group < ngroups; group++) { | 
 | 214 | 		desc = ext3_get_group_desc (sb, group, &bh); | 
 | 215 | 		if (!desc || !desc->bg_free_inodes_count) | 
 | 216 | 			continue; | 
 | 217 | 		if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | 
 | 218 | 			continue; | 
 | 219 | 		if (!best_desc ||  | 
 | 220 | 		    (le16_to_cpu(desc->bg_free_blocks_count) > | 
 | 221 | 		     le16_to_cpu(best_desc->bg_free_blocks_count))) { | 
 | 222 | 			best_group = group; | 
 | 223 | 			best_desc = desc; | 
 | 224 | 		} | 
 | 225 | 	} | 
 | 226 | 	return best_group; | 
 | 227 | } | 
 | 228 |  | 
 | 229 | /*  | 
 | 230 |  * Orlov's allocator for directories.  | 
 | 231 |  *  | 
 | 232 |  * We always try to spread first-level directories. | 
 | 233 |  * | 
 | 234 |  * If there are blockgroups with both free inodes and free blocks counts  | 
 | 235 |  * not worse than average we return one with smallest directory count.  | 
 | 236 |  * Otherwise we simply return a random group.  | 
 | 237 |  *  | 
 | 238 |  * For the rest rules look so:  | 
 | 239 |  *  | 
 | 240 |  * It's OK to put directory into a group unless  | 
 | 241 |  * it has too many directories already (max_dirs) or  | 
 | 242 |  * it has too few free inodes left (min_inodes) or  | 
 | 243 |  * it has too few free blocks left (min_blocks) or  | 
 | 244 |  * it's already running too large debt (max_debt).  | 
 | 245 |  * Parent's group is prefered, if it doesn't satisfy these  | 
 | 246 |  * conditions we search cyclically through the rest. If none  | 
 | 247 |  * of the groups look good we just look for a group with more  | 
 | 248 |  * free inodes than average (starting at parent's group).  | 
 | 249 |  *  | 
 | 250 |  * Debt is incremented each time we allocate a directory and decremented  | 
 | 251 |  * when we allocate an inode, within 0--255.  | 
 | 252 |  */  | 
 | 253 |  | 
 | 254 | #define INODE_COST 64 | 
 | 255 | #define BLOCK_COST 256 | 
 | 256 |  | 
 | 257 | static int find_group_orlov(struct super_block *sb, struct inode *parent) | 
 | 258 | { | 
 | 259 | 	int parent_group = EXT3_I(parent)->i_block_group; | 
 | 260 | 	struct ext3_sb_info *sbi = EXT3_SB(sb); | 
 | 261 | 	struct ext3_super_block *es = sbi->s_es; | 
 | 262 | 	int ngroups = sbi->s_groups_count; | 
 | 263 | 	int inodes_per_group = EXT3_INODES_PER_GROUP(sb); | 
 | 264 | 	int freei, avefreei; | 
| Mingming Cao | 1c2bf37 | 2006-06-25 05:48:06 -0700 | [diff] [blame] | 265 | 	ext3_fsblk_t freeb, avefreeb; | 
 | 266 | 	ext3_fsblk_t blocks_per_dir; | 
 | 267 | 	int ndirs; | 
 | 268 | 	int max_debt, max_dirs, min_inodes; | 
 | 269 | 	ext3_grpblk_t min_blocks; | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 270 | 	int group = -1, i; | 
 | 271 | 	struct ext3_group_desc *desc; | 
 | 272 | 	struct buffer_head *bh; | 
 | 273 |  | 
 | 274 | 	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | 
 | 275 | 	avefreei = freei / ngroups; | 
 | 276 | 	freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | 
 | 277 | 	avefreeb = freeb / ngroups; | 
 | 278 | 	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); | 
 | 279 |  | 
 | 280 | 	if ((parent == sb->s_root->d_inode) || | 
 | 281 | 	    (EXT3_I(parent)->i_flags & EXT3_TOPDIR_FL)) { | 
 | 282 | 		int best_ndir = inodes_per_group; | 
 | 283 | 		int best_group = -1; | 
 | 284 |  | 
 | 285 | 		get_random_bytes(&group, sizeof(group)); | 
 | 286 | 		parent_group = (unsigned)group % ngroups; | 
 | 287 | 		for (i = 0; i < ngroups; i++) { | 
 | 288 | 			group = (parent_group + i) % ngroups; | 
 | 289 | 			desc = ext3_get_group_desc (sb, group, &bh); | 
 | 290 | 			if (!desc || !desc->bg_free_inodes_count) | 
 | 291 | 				continue; | 
 | 292 | 			if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) | 
 | 293 | 				continue; | 
 | 294 | 			if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | 
 | 295 | 				continue; | 
 | 296 | 			if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) | 
 | 297 | 				continue; | 
 | 298 | 			best_group = group; | 
 | 299 | 			best_ndir = le16_to_cpu(desc->bg_used_dirs_count); | 
 | 300 | 		} | 
 | 301 | 		if (best_group >= 0) | 
 | 302 | 			return best_group; | 
 | 303 | 		goto fallback; | 
 | 304 | 	} | 
 | 305 |  | 
 | 306 | 	blocks_per_dir = (le32_to_cpu(es->s_blocks_count) - freeb) / ndirs; | 
 | 307 |  | 
 | 308 | 	max_dirs = ndirs / ngroups + inodes_per_group / 16; | 
 | 309 | 	min_inodes = avefreei - inodes_per_group / 4; | 
 | 310 | 	min_blocks = avefreeb - EXT3_BLOCKS_PER_GROUP(sb) / 4; | 
 | 311 |  | 
| Mingming Cao | 1c2bf37 | 2006-06-25 05:48:06 -0700 | [diff] [blame] | 312 | 	max_debt = EXT3_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, (ext3_fsblk_t)BLOCK_COST); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 313 | 	if (max_debt * INODE_COST > inodes_per_group) | 
 | 314 | 		max_debt = inodes_per_group / INODE_COST; | 
 | 315 | 	if (max_debt > 255) | 
 | 316 | 		max_debt = 255; | 
 | 317 | 	if (max_debt == 0) | 
 | 318 | 		max_debt = 1; | 
 | 319 |  | 
 | 320 | 	for (i = 0; i < ngroups; i++) { | 
 | 321 | 		group = (parent_group + i) % ngroups; | 
 | 322 | 		desc = ext3_get_group_desc (sb, group, &bh); | 
 | 323 | 		if (!desc || !desc->bg_free_inodes_count) | 
 | 324 | 			continue; | 
 | 325 | 		if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs) | 
 | 326 | 			continue; | 
 | 327 | 		if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes) | 
 | 328 | 			continue; | 
 | 329 | 		if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks) | 
 | 330 | 			continue; | 
 | 331 | 		return group; | 
 | 332 | 	} | 
 | 333 |  | 
 | 334 | fallback: | 
 | 335 | 	for (i = 0; i < ngroups; i++) { | 
 | 336 | 		group = (parent_group + i) % ngroups; | 
 | 337 | 		desc = ext3_get_group_desc (sb, group, &bh); | 
 | 338 | 		if (!desc || !desc->bg_free_inodes_count) | 
 | 339 | 			continue; | 
 | 340 | 		if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei) | 
 | 341 | 			return group; | 
 | 342 | 	} | 
 | 343 |  | 
 | 344 | 	if (avefreei) { | 
 | 345 | 		/* | 
 | 346 | 		 * The free-inodes counter is approximate, and for really small | 
 | 347 | 		 * filesystems the above test can fail to find any blockgroups | 
 | 348 | 		 */ | 
 | 349 | 		avefreei = 0; | 
 | 350 | 		goto fallback; | 
 | 351 | 	} | 
 | 352 |  | 
 | 353 | 	return -1; | 
 | 354 | } | 
 | 355 |  | 
 | 356 | static int find_group_other(struct super_block *sb, struct inode *parent) | 
 | 357 | { | 
 | 358 | 	int parent_group = EXT3_I(parent)->i_block_group; | 
 | 359 | 	int ngroups = EXT3_SB(sb)->s_groups_count; | 
 | 360 | 	struct ext3_group_desc *desc; | 
 | 361 | 	struct buffer_head *bh; | 
 | 362 | 	int group, i; | 
 | 363 |  | 
 | 364 | 	/* | 
 | 365 | 	 * Try to place the inode in its parent directory | 
 | 366 | 	 */ | 
 | 367 | 	group = parent_group; | 
 | 368 | 	desc = ext3_get_group_desc (sb, group, &bh); | 
 | 369 | 	if (desc && le16_to_cpu(desc->bg_free_inodes_count) && | 
 | 370 | 			le16_to_cpu(desc->bg_free_blocks_count)) | 
 | 371 | 		return group; | 
 | 372 |  | 
 | 373 | 	/* | 
 | 374 | 	 * We're going to place this inode in a different blockgroup from its | 
 | 375 | 	 * parent.  We want to cause files in a common directory to all land in | 
 | 376 | 	 * the same blockgroup.  But we want files which are in a different | 
 | 377 | 	 * directory which shares a blockgroup with our parent to land in a | 
 | 378 | 	 * different blockgroup. | 
 | 379 | 	 * | 
 | 380 | 	 * So add our directory's i_ino into the starting point for the hash. | 
 | 381 | 	 */ | 
 | 382 | 	group = (group + parent->i_ino) % ngroups; | 
 | 383 |  | 
 | 384 | 	/* | 
 | 385 | 	 * Use a quadratic hash to find a group with a free inode and some free | 
 | 386 | 	 * blocks. | 
 | 387 | 	 */ | 
 | 388 | 	for (i = 1; i < ngroups; i <<= 1) { | 
 | 389 | 		group += i; | 
 | 390 | 		if (group >= ngroups) | 
 | 391 | 			group -= ngroups; | 
 | 392 | 		desc = ext3_get_group_desc (sb, group, &bh); | 
 | 393 | 		if (desc && le16_to_cpu(desc->bg_free_inodes_count) && | 
 | 394 | 				le16_to_cpu(desc->bg_free_blocks_count)) | 
 | 395 | 			return group; | 
 | 396 | 	} | 
 | 397 |  | 
 | 398 | 	/* | 
 | 399 | 	 * That failed: try linear search for a free inode, even if that group | 
 | 400 | 	 * has no free blocks. | 
 | 401 | 	 */ | 
 | 402 | 	group = parent_group; | 
 | 403 | 	for (i = 0; i < ngroups; i++) { | 
 | 404 | 		if (++group >= ngroups) | 
 | 405 | 			group = 0; | 
 | 406 | 		desc = ext3_get_group_desc (sb, group, &bh); | 
 | 407 | 		if (desc && le16_to_cpu(desc->bg_free_inodes_count)) | 
 | 408 | 			return group; | 
 | 409 | 	} | 
 | 410 |  | 
 | 411 | 	return -1; | 
 | 412 | } | 
 | 413 |  | 
 | 414 | /* | 
 | 415 |  * There are two policies for allocating an inode.  If the new inode is | 
 | 416 |  * a directory, then a forward search is made for a block group with both | 
 | 417 |  * free space and a low directory-to-inode ratio; if that fails, then of | 
 | 418 |  * the groups with above-average free space, that group with the fewest | 
 | 419 |  * directories already is chosen. | 
 | 420 |  * | 
 | 421 |  * For other inodes, search forward from the parent directory's block | 
 | 422 |  * group to find a free inode. | 
 | 423 |  */ | 
 | 424 | struct inode *ext3_new_inode(handle_t *handle, struct inode * dir, int mode) | 
 | 425 | { | 
 | 426 | 	struct super_block *sb; | 
 | 427 | 	struct buffer_head *bitmap_bh = NULL; | 
 | 428 | 	struct buffer_head *bh2; | 
 | 429 | 	int group; | 
 | 430 | 	unsigned long ino = 0; | 
 | 431 | 	struct inode * inode; | 
 | 432 | 	struct ext3_group_desc * gdp = NULL; | 
 | 433 | 	struct ext3_super_block * es; | 
 | 434 | 	struct ext3_inode_info *ei; | 
 | 435 | 	struct ext3_sb_info *sbi; | 
 | 436 | 	int err = 0; | 
 | 437 | 	struct inode *ret; | 
 | 438 | 	int i; | 
 | 439 |  | 
 | 440 | 	/* Cannot create files in a deleted directory */ | 
 | 441 | 	if (!dir || !dir->i_nlink) | 
 | 442 | 		return ERR_PTR(-EPERM); | 
 | 443 |  | 
 | 444 | 	sb = dir->i_sb; | 
 | 445 | 	inode = new_inode(sb); | 
 | 446 | 	if (!inode) | 
 | 447 | 		return ERR_PTR(-ENOMEM); | 
 | 448 | 	ei = EXT3_I(inode); | 
 | 449 |  | 
 | 450 | 	sbi = EXT3_SB(sb); | 
 | 451 | 	es = sbi->s_es; | 
 | 452 | 	if (S_ISDIR(mode)) { | 
 | 453 | 		if (test_opt (sb, OLDALLOC)) | 
 | 454 | 			group = find_group_dir(sb, dir); | 
 | 455 | 		else | 
 | 456 | 			group = find_group_orlov(sb, dir); | 
 | 457 | 	} else  | 
 | 458 | 		group = find_group_other(sb, dir); | 
 | 459 |  | 
 | 460 | 	err = -ENOSPC; | 
 | 461 | 	if (group == -1) | 
 | 462 | 		goto out; | 
 | 463 |  | 
 | 464 | 	for (i = 0; i < sbi->s_groups_count; i++) { | 
 | 465 | 		err = -EIO; | 
 | 466 |  | 
 | 467 | 		gdp = ext3_get_group_desc(sb, group, &bh2); | 
 | 468 | 		if (!gdp) | 
 | 469 | 			goto fail; | 
 | 470 |  | 
 | 471 | 		brelse(bitmap_bh); | 
 | 472 | 		bitmap_bh = read_inode_bitmap(sb, group); | 
 | 473 | 		if (!bitmap_bh) | 
 | 474 | 			goto fail; | 
 | 475 |  | 
 | 476 | 		ino = 0; | 
 | 477 |  | 
 | 478 | repeat_in_this_group: | 
 | 479 | 		ino = ext3_find_next_zero_bit((unsigned long *) | 
 | 480 | 				bitmap_bh->b_data, EXT3_INODES_PER_GROUP(sb), ino); | 
 | 481 | 		if (ino < EXT3_INODES_PER_GROUP(sb)) { | 
 | 482 |  | 
 | 483 | 			BUFFER_TRACE(bitmap_bh, "get_write_access"); | 
 | 484 | 			err = ext3_journal_get_write_access(handle, bitmap_bh); | 
 | 485 | 			if (err) | 
 | 486 | 				goto fail; | 
 | 487 |  | 
 | 488 | 			if (!ext3_set_bit_atomic(sb_bgl_lock(sbi, group), | 
 | 489 | 						ino, bitmap_bh->b_data)) { | 
 | 490 | 				/* we won it */ | 
 | 491 | 				BUFFER_TRACE(bitmap_bh, | 
 | 492 | 					"call ext3_journal_dirty_metadata"); | 
 | 493 | 				err = ext3_journal_dirty_metadata(handle, | 
 | 494 | 								bitmap_bh); | 
 | 495 | 				if (err) | 
 | 496 | 					goto fail; | 
 | 497 | 				goto got; | 
 | 498 | 			} | 
 | 499 | 			/* we lost it */ | 
 | 500 | 			journal_release_buffer(handle, bitmap_bh); | 
 | 501 |  | 
 | 502 | 			if (++ino < EXT3_INODES_PER_GROUP(sb)) | 
 | 503 | 				goto repeat_in_this_group; | 
 | 504 | 		} | 
 | 505 |  | 
 | 506 | 		/* | 
 | 507 | 		 * This case is possible in concurrent environment.  It is very | 
 | 508 | 		 * rare.  We cannot repeat the find_group_xxx() call because | 
 | 509 | 		 * that will simply return the same blockgroup, because the | 
 | 510 | 		 * group descriptor metadata has not yet been updated. | 
 | 511 | 		 * So we just go onto the next blockgroup. | 
 | 512 | 		 */ | 
 | 513 | 		if (++group == sbi->s_groups_count) | 
 | 514 | 			group = 0; | 
 | 515 | 	} | 
 | 516 | 	err = -ENOSPC; | 
 | 517 | 	goto out; | 
 | 518 |  | 
 | 519 | got: | 
 | 520 | 	ino += group * EXT3_INODES_PER_GROUP(sb) + 1; | 
 | 521 | 	if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | 
 | 522 | 		ext3_error (sb, "ext3_new_inode", | 
 | 523 | 			    "reserved inode or inode > inodes count - " | 
 | 524 | 			    "block_group = %d, inode=%lu", group, ino); | 
 | 525 | 		err = -EIO; | 
 | 526 | 		goto fail; | 
 | 527 | 	} | 
 | 528 |  | 
 | 529 | 	BUFFER_TRACE(bh2, "get_write_access"); | 
 | 530 | 	err = ext3_journal_get_write_access(handle, bh2); | 
 | 531 | 	if (err) goto fail; | 
 | 532 | 	spin_lock(sb_bgl_lock(sbi, group)); | 
 | 533 | 	gdp->bg_free_inodes_count = | 
 | 534 | 		cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1); | 
 | 535 | 	if (S_ISDIR(mode)) { | 
 | 536 | 		gdp->bg_used_dirs_count = | 
 | 537 | 			cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1); | 
 | 538 | 	} | 
 | 539 | 	spin_unlock(sb_bgl_lock(sbi, group)); | 
 | 540 | 	BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata"); | 
 | 541 | 	err = ext3_journal_dirty_metadata(handle, bh2); | 
 | 542 | 	if (err) goto fail; | 
 | 543 |  | 
 | 544 | 	percpu_counter_dec(&sbi->s_freeinodes_counter); | 
 | 545 | 	if (S_ISDIR(mode)) | 
 | 546 | 		percpu_counter_inc(&sbi->s_dirs_counter); | 
 | 547 | 	sb->s_dirt = 1; | 
 | 548 |  | 
 | 549 | 	inode->i_uid = current->fsuid; | 
 | 550 | 	if (test_opt (sb, GRPID)) | 
 | 551 | 		inode->i_gid = dir->i_gid; | 
 | 552 | 	else if (dir->i_mode & S_ISGID) { | 
 | 553 | 		inode->i_gid = dir->i_gid; | 
 | 554 | 		if (S_ISDIR(mode)) | 
 | 555 | 			mode |= S_ISGID; | 
 | 556 | 	} else | 
 | 557 | 		inode->i_gid = current->fsgid; | 
 | 558 | 	inode->i_mode = mode; | 
 | 559 |  | 
 | 560 | 	inode->i_ino = ino; | 
 | 561 | 	/* This is the optimal IO size (for stat), not the fs block size */ | 
 | 562 | 	inode->i_blksize = PAGE_SIZE; | 
 | 563 | 	inode->i_blocks = 0; | 
 | 564 | 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC; | 
 | 565 |  | 
 | 566 | 	memset(ei->i_data, 0, sizeof(ei->i_data)); | 
 | 567 | 	ei->i_dir_start_lookup = 0; | 
 | 568 | 	ei->i_disksize = 0; | 
 | 569 |  | 
 | 570 | 	ei->i_flags = EXT3_I(dir)->i_flags & ~EXT3_INDEX_FL; | 
 | 571 | 	if (S_ISLNK(mode)) | 
 | 572 | 		ei->i_flags &= ~(EXT3_IMMUTABLE_FL|EXT3_APPEND_FL); | 
 | 573 | 	/* dirsync only applies to directories */ | 
 | 574 | 	if (!S_ISDIR(mode)) | 
 | 575 | 		ei->i_flags &= ~EXT3_DIRSYNC_FL; | 
 | 576 | #ifdef EXT3_FRAGMENTS | 
 | 577 | 	ei->i_faddr = 0; | 
 | 578 | 	ei->i_frag_no = 0; | 
 | 579 | 	ei->i_frag_size = 0; | 
 | 580 | #endif | 
 | 581 | 	ei->i_file_acl = 0; | 
 | 582 | 	ei->i_dir_acl = 0; | 
 | 583 | 	ei->i_dtime = 0; | 
 | 584 | 	ei->i_block_alloc_info = NULL; | 
 | 585 | 	ei->i_block_group = group; | 
 | 586 |  | 
 | 587 | 	ext3_set_inode_flags(inode); | 
 | 588 | 	if (IS_DIRSYNC(inode)) | 
 | 589 | 		handle->h_sync = 1; | 
 | 590 | 	insert_inode_hash(inode); | 
 | 591 | 	spin_lock(&sbi->s_next_gen_lock); | 
 | 592 | 	inode->i_generation = sbi->s_next_generation++; | 
 | 593 | 	spin_unlock(&sbi->s_next_gen_lock); | 
 | 594 |  | 
 | 595 | 	ei->i_state = EXT3_STATE_NEW; | 
 | 596 | 	ei->i_extra_isize = | 
 | 597 | 		(EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) ? | 
 | 598 | 		sizeof(struct ext3_inode) - EXT3_GOOD_OLD_INODE_SIZE : 0; | 
 | 599 |  | 
 | 600 | 	ret = inode; | 
 | 601 | 	if(DQUOT_ALLOC_INODE(inode)) { | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 602 | 		err = -EDQUOT; | 
| Chris Sykes | dc7b5fd | 2005-09-27 21:45:23 -0700 | [diff] [blame] | 603 | 		goto fail_drop; | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 604 | 	} | 
| Chris Sykes | dc7b5fd | 2005-09-27 21:45:23 -0700 | [diff] [blame] | 605 |  | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 606 | 	err = ext3_init_acl(handle, inode, dir); | 
| Chris Sykes | dc7b5fd | 2005-09-27 21:45:23 -0700 | [diff] [blame] | 607 | 	if (err) | 
 | 608 | 		goto fail_free_drop; | 
 | 609 |  | 
| Stephen Smalley | ac50960 | 2005-09-09 13:01:41 -0700 | [diff] [blame] | 610 | 	err = ext3_init_security(handle,inode, dir); | 
| Chris Sykes | dc7b5fd | 2005-09-27 21:45:23 -0700 | [diff] [blame] | 611 | 	if (err) | 
 | 612 | 		goto fail_free_drop; | 
 | 613 |  | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 614 | 	err = ext3_mark_inode_dirty(handle, inode); | 
 | 615 | 	if (err) { | 
 | 616 | 		ext3_std_error(sb, err); | 
| Chris Sykes | dc7b5fd | 2005-09-27 21:45:23 -0700 | [diff] [blame] | 617 | 		goto fail_free_drop; | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | 	} | 
 | 619 |  | 
 | 620 | 	ext3_debug("allocating inode %lu\n", inode->i_ino); | 
 | 621 | 	goto really_out; | 
 | 622 | fail: | 
 | 623 | 	ext3_std_error(sb, err); | 
 | 624 | out: | 
 | 625 | 	iput(inode); | 
 | 626 | 	ret = ERR_PTR(err); | 
 | 627 | really_out: | 
 | 628 | 	brelse(bitmap_bh); | 
 | 629 | 	return ret; | 
 | 630 |  | 
| Chris Sykes | dc7b5fd | 2005-09-27 21:45:23 -0700 | [diff] [blame] | 631 | fail_free_drop: | 
 | 632 | 	DQUOT_FREE_INODE(inode); | 
 | 633 |  | 
 | 634 | fail_drop: | 
 | 635 | 	DQUOT_DROP(inode); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | 	inode->i_flags |= S_NOQUOTA; | 
 | 637 | 	inode->i_nlink = 0; | 
 | 638 | 	iput(inode); | 
 | 639 | 	brelse(bitmap_bh); | 
 | 640 | 	return ERR_PTR(err); | 
 | 641 | } | 
 | 642 |  | 
 | 643 | /* Verify that we are loading a valid orphan from disk */ | 
 | 644 | struct inode *ext3_orphan_get(struct super_block *sb, unsigned long ino) | 
 | 645 | { | 
 | 646 | 	unsigned long max_ino = le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count); | 
 | 647 | 	unsigned long block_group; | 
 | 648 | 	int bit; | 
 | 649 | 	struct buffer_head *bitmap_bh = NULL; | 
 | 650 | 	struct inode *inode = NULL; | 
 | 651 |  | 
 | 652 | 	/* Error cases - e2fsck has already cleaned up for us */ | 
 | 653 | 	if (ino > max_ino) { | 
 | 654 | 		ext3_warning(sb, __FUNCTION__, | 
| Glauber de Oliveira Costa | 9f40668 | 2006-01-08 01:03:22 -0800 | [diff] [blame] | 655 | 			     "bad orphan ino %lu!  e2fsck was run?", ino); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 656 | 		goto out; | 
 | 657 | 	} | 
 | 658 |  | 
 | 659 | 	block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); | 
 | 660 | 	bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb); | 
 | 661 | 	bitmap_bh = read_inode_bitmap(sb, block_group); | 
 | 662 | 	if (!bitmap_bh) { | 
 | 663 | 		ext3_warning(sb, __FUNCTION__, | 
| Glauber de Oliveira Costa | 9f40668 | 2006-01-08 01:03:22 -0800 | [diff] [blame] | 664 | 			     "inode bitmap error for orphan %lu", ino); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 665 | 		goto out; | 
 | 666 | 	} | 
 | 667 |  | 
 | 668 | 	/* Having the inode bit set should be a 100% indicator that this | 
 | 669 | 	 * is a valid orphan (no e2fsck run on fs).  Orphans also include | 
 | 670 | 	 * inodes that were being truncated, so we can't check i_nlink==0. | 
 | 671 | 	 */ | 
 | 672 | 	if (!ext3_test_bit(bit, bitmap_bh->b_data) || | 
 | 673 | 			!(inode = iget(sb, ino)) || is_bad_inode(inode) || | 
 | 674 | 			NEXT_ORPHAN(inode) > max_ino) { | 
 | 675 | 		ext3_warning(sb, __FUNCTION__, | 
| Glauber de Oliveira Costa | 9f40668 | 2006-01-08 01:03:22 -0800 | [diff] [blame] | 676 | 			     "bad orphan inode %lu!  e2fsck was run?", ino); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | 		printk(KERN_NOTICE "ext3_test_bit(bit=%d, block=%llu) = %d\n", | 
 | 678 | 		       bit, (unsigned long long)bitmap_bh->b_blocknr, | 
 | 679 | 		       ext3_test_bit(bit, bitmap_bh->b_data)); | 
 | 680 | 		printk(KERN_NOTICE "inode=%p\n", inode); | 
 | 681 | 		if (inode) { | 
 | 682 | 			printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | 
 | 683 | 			       is_bad_inode(inode)); | 
 | 684 | 			printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | 
 | 685 | 			       NEXT_ORPHAN(inode)); | 
 | 686 | 			printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | 
 | 687 | 		} | 
 | 688 | 		/* Avoid freeing blocks if we got a bad deleted inode */ | 
 | 689 | 		if (inode && inode->i_nlink == 0) | 
 | 690 | 			inode->i_blocks = 0; | 
 | 691 | 		iput(inode); | 
 | 692 | 		inode = NULL; | 
 | 693 | 	} | 
 | 694 | out: | 
 | 695 | 	brelse(bitmap_bh); | 
 | 696 | 	return inode; | 
 | 697 | } | 
 | 698 |  | 
 | 699 | unsigned long ext3_count_free_inodes (struct super_block * sb) | 
 | 700 | { | 
 | 701 | 	unsigned long desc_count; | 
 | 702 | 	struct ext3_group_desc *gdp; | 
 | 703 | 	int i; | 
 | 704 | #ifdef EXT3FS_DEBUG | 
 | 705 | 	struct ext3_super_block *es; | 
 | 706 | 	unsigned long bitmap_count, x; | 
 | 707 | 	struct buffer_head *bitmap_bh = NULL; | 
 | 708 |  | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 709 | 	es = EXT3_SB(sb)->s_es; | 
 | 710 | 	desc_count = 0; | 
 | 711 | 	bitmap_count = 0; | 
 | 712 | 	gdp = NULL; | 
 | 713 | 	for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) { | 
 | 714 | 		gdp = ext3_get_group_desc (sb, i, NULL); | 
 | 715 | 		if (!gdp) | 
 | 716 | 			continue; | 
 | 717 | 		desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | 
 | 718 | 		brelse(bitmap_bh); | 
 | 719 | 		bitmap_bh = read_inode_bitmap(sb, i); | 
 | 720 | 		if (!bitmap_bh) | 
 | 721 | 			continue; | 
 | 722 |  | 
 | 723 | 		x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8); | 
 | 724 | 		printk("group %d: stored = %d, counted = %lu\n", | 
 | 725 | 			i, le16_to_cpu(gdp->bg_free_inodes_count), x); | 
 | 726 | 		bitmap_count += x; | 
 | 727 | 	} | 
 | 728 | 	brelse(bitmap_bh); | 
 | 729 | 	printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n", | 
 | 730 | 		le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | 	return desc_count; | 
 | 732 | #else | 
 | 733 | 	desc_count = 0; | 
 | 734 | 	for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) { | 
 | 735 | 		gdp = ext3_get_group_desc (sb, i, NULL); | 
 | 736 | 		if (!gdp) | 
 | 737 | 			continue; | 
 | 738 | 		desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | 
 | 739 | 		cond_resched(); | 
 | 740 | 	} | 
 | 741 | 	return desc_count; | 
 | 742 | #endif | 
 | 743 | } | 
 | 744 |  | 
 | 745 | /* Called at mount-time, super-block is locked */ | 
 | 746 | unsigned long ext3_count_dirs (struct super_block * sb) | 
 | 747 | { | 
 | 748 | 	unsigned long count = 0; | 
 | 749 | 	int i; | 
 | 750 |  | 
 | 751 | 	for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) { | 
 | 752 | 		struct ext3_group_desc *gdp = ext3_get_group_desc (sb, i, NULL); | 
 | 753 | 		if (!gdp) | 
 | 754 | 			continue; | 
 | 755 | 		count += le16_to_cpu(gdp->bg_used_dirs_count); | 
 | 756 | 	} | 
 | 757 | 	return count; | 
 | 758 | } | 
 | 759 |  |