blob: 25d8c9781ad94ea758781f906a34412743b3cda0 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/ialloc.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
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>
Mingming Caodab291a2006-10-11 01:21:01 -070017#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070018#include <linux/stat.h>
19#include <linux/string.h>
20#include <linux/quotaops.h>
21#include <linux/buffer_head.h>
22#include <linux/random.h>
23#include <linux/bitops.h>
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070024#include <linux/blkdev.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070025#include <asm/byteorder.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040026
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040027#include "ext4.h"
28#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070029#include "xattr.h"
30#include "acl.h"
31
Theodore Ts'o9bffad12009-06-17 11:48:11 -040032#include <trace/events/ext4.h>
33
Dave Kleikampac27a0e2006-10-11 01:20:50 -070034/*
35 * ialloc.c contains the inodes allocation and deallocation routines
36 */
37
38/*
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
42 *
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
46 */
47
Andreas Dilger717d50e2007-10-16 18:38:25 -040048/*
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
52 */
Theodore Ts'o61d08672010-10-27 21:30:15 -040053void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
Andreas Dilger717d50e2007-10-16 18:38:25 -040054{
55 int i;
56
57 if (start_bit >= end_bit)
58 return;
59
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
61 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
62 ext4_set_bit(i, bitmap);
63 if (i < end_bit)
64 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
65}
66
67/* Initializes an uninitialized inode bitmap */
Theodore Ts'o1f109d52010-10-27 21:30:14 -040068static unsigned ext4_init_inode_bitmap(struct super_block *sb,
69 struct buffer_head *bh,
70 ext4_group_t block_group,
71 struct ext4_group_desc *gdp)
Andreas Dilger717d50e2007-10-16 18:38:25 -040072{
73 struct ext4_sb_info *sbi = EXT4_SB(sb);
74
75 J_ASSERT_BH(bh, buffer_locked(bh));
76
77 /* If checksum is bad mark all blocks and inodes use to prevent
78 * allocation, essentially implementing a per-group read-only flag. */
79 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -050080 ext4_error(sb, "Checksum bad for group %u", block_group);
Theodore Ts'o021b65b2011-09-09 19:08:51 -040081 ext4_free_group_clusters_set(sb, gdp, 0);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -050082 ext4_free_inodes_set(sb, gdp, 0);
83 ext4_itable_unused_set(sb, gdp, 0);
Andreas Dilger717d50e2007-10-16 18:38:25 -040084 memset(bh->b_data, 0xff, sb->s_blocksize);
85 return 0;
86 }
87
88 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
Theodore Ts'o61d08672010-10-27 21:30:15 -040089 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
Andreas Dilger717d50e2007-10-16 18:38:25 -040090 bh->b_data);
91
92 return EXT4_INODES_PER_GROUP(sb);
93}
Dave Kleikampac27a0e2006-10-11 01:20:50 -070094
95/*
96 * Read the inode allocation bitmap for a given block_group, reading
97 * into the specified slot in the superblock's bitmap cache.
98 *
99 * Return buffer_head of bitmap on success or NULL.
100 */
101static struct buffer_head *
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400102ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700103{
Mingming Cao617ba132006-10-11 01:20:53 -0700104 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700105 struct buffer_head *bh = NULL;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400106 ext4_fsblk_t bitmap_blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700107
Mingming Cao617ba132006-10-11 01:20:53 -0700108 desc = ext4_get_group_desc(sb, block_group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700109 if (!desc)
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400110 return NULL;
Lukas Czernerbfff6872010-10-27 21:30:05 -0400111
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400112 bitmap_blk = ext4_inode_bitmap(sb, desc);
113 bh = sb_getblk(sb, bitmap_blk);
114 if (unlikely(!bh)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500115 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500116 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400117 block_group, bitmap_blk);
118 return NULL;
119 }
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500120 if (bitmap_uptodate(bh))
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400121 return bh;
122
Frederic Bohec806e682008-10-10 08:09:18 -0400123 lock_buffer(bh);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500124 if (bitmap_uptodate(bh)) {
125 unlock_buffer(bh);
126 return bh;
127 }
Lukas Czernerbfff6872010-10-27 21:30:05 -0400128
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400129 ext4_lock_group(sb, block_group);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400130 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
131 ext4_init_inode_bitmap(sb, bh, block_group, desc);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500132 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400133 set_buffer_uptodate(bh);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400134 ext4_unlock_group(sb, block_group);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500135 unlock_buffer(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400136 return bh;
137 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400138 ext4_unlock_group(sb, block_group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400139
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500140 if (buffer_uptodate(bh)) {
141 /*
142 * if not uninit if bh is uptodate,
143 * bitmap is also uptodate
144 */
145 set_bitmap_uptodate(bh);
146 unlock_buffer(bh);
147 return bh;
148 }
149 /*
150 * submit the buffer_head for read. We can
151 * safely mark the bitmap as uptodate now.
152 * We do it here so the bitmap uptodate bit
153 * get set with buffer lock held.
154 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400155 trace_ext4_load_inode_bitmap(sb, block_group);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500156 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400157 if (bh_submit_read(bh) < 0) {
158 put_bh(bh);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500159 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500160 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400161 block_group, bitmap_blk);
162 return NULL;
163 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700164 return bh;
165}
166
167/*
168 * NOTE! When we get the inode, we're the only people
169 * that have access to it, and as such there are no
170 * race conditions we have to worry about. The inode
171 * is not on the hash-lists, and it cannot be reached
172 * through the filesystem because the directory entry
173 * has been deleted earlier.
174 *
175 * HOWEVER: we must make sure that we get no aliases,
176 * which means that we have to call "clear_inode()"
177 * _before_ we mark the inode not in use in the inode
178 * bitmaps. Otherwise a newly created file might use
179 * the same inode number (not actually the same pointer
180 * though), and then we'd have two inodes sharing the
181 * same inode number and space on the harddisk.
182 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400183void ext4_free_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700184{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400185 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700186 int is_directory;
187 unsigned long ino;
188 struct buffer_head *bitmap_bh = NULL;
189 struct buffer_head *bh2;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500190 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700191 unsigned long bit;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400192 struct ext4_group_desc *gdp;
193 struct ext4_super_block *es;
Mingming Cao617ba132006-10-11 01:20:53 -0700194 struct ext4_sb_info *sbi;
Eric Sandeen7ce9d5d2009-03-04 18:38:18 -0500195 int fatal = 0, err, count, cleared;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700196
197 if (atomic_read(&inode->i_count) > 1) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400198 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
199 atomic_read(&inode->i_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700200 return;
201 }
202 if (inode->i_nlink) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400203 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
204 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700205 return;
206 }
207 if (!sb) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400208 printk(KERN_ERR "ext4_free_inode: inode on "
209 "nonexistent device\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700210 return;
211 }
Mingming Cao617ba132006-10-11 01:20:53 -0700212 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700213
214 ino = inode->i_ino;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400215 ext4_debug("freeing inode %lu\n", ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400216 trace_ext4_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700217
218 /*
219 * Note: we must free any quota before locking the superblock,
220 * as writing the quota to disk may need the lock as well.
221 */
Christoph Hellwig871a2932010-03-03 09:05:07 -0500222 dquot_initialize(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700223 ext4_xattr_delete_inode(handle, inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500224 dquot_free_inode(inode);
Christoph Hellwig9f754752010-03-03 09:05:05 -0500225 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700226
227 is_directory = S_ISDIR(inode->i_mode);
228
229 /* Do this BEFORE marking the inode not in use or returning an error */
Al Viro0930fcc2010-06-07 13:16:22 -0400230 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700231
Mingming Cao617ba132006-10-11 01:20:53 -0700232 es = EXT4_SB(sb)->s_es;
233 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500234 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700235 goto error_return;
236 }
Mingming Cao617ba132006-10-11 01:20:53 -0700237 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
238 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400239 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700240 if (!bitmap_bh)
241 goto error_return;
242
243 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700244 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700245 if (fatal)
246 goto error_return;
247
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400248 fatal = -ESRCH;
249 gdp = ext4_get_group_desc(sb, block_group, &bh2);
250 if (gdp) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700251 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700252 fatal = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700253 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400254 ext4_lock_group(sb, block_group);
Akinobu Mita597d5082011-12-28 20:32:07 -0500255 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400256 if (fatal || !cleared) {
257 ext4_unlock_group(sb, block_group);
258 goto out;
259 }
260
261 count = ext4_free_inodes_count(sb, gdp) + 1;
262 ext4_free_inodes_set(sb, gdp, count);
263 if (is_directory) {
264 count = ext4_used_dirs_count(sb, gdp) - 1;
265 ext4_used_dirs_set(sb, gdp, count);
266 percpu_counter_dec(&sbi->s_dirs_counter);
267 }
268 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
269 ext4_unlock_group(sb, block_group);
270
271 percpu_counter_inc(&sbi->s_freeinodes_counter);
272 if (sbi->s_log_groups_per_flex) {
273 ext4_group_t f = ext4_flex_group(sbi, block_group);
274
275 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
276 if (is_directory)
277 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
278 }
279 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
280 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
281out:
282 if (cleared) {
283 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
284 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
285 if (!fatal)
286 fatal = err;
Theodore Ts'oa0375152010-06-11 23:14:04 -0400287 ext4_mark_super_dirty(sb);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400288 } else
289 ext4_error(sb, "bit already cleared for inode %lu", ino);
290
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291error_return:
292 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700293 ext4_std_error(sb, fatal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700294}
295
Theodore Ts'oa4912122009-03-12 12:18:34 -0400296struct orlov_stats {
297 __u32 free_inodes;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400298 __u32 free_clusters;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400299 __u32 used_dirs;
300};
301
302/*
303 * Helper function for Orlov's allocator; returns critical information
304 * for a particular block group or flex_bg. If flex_size is 1, then g
305 * is a block group number; otherwise it is flex_bg number.
306 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400307static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
308 int flex_size, struct orlov_stats *stats)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400309{
310 struct ext4_group_desc *desc;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500311 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400312
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500313 if (flex_size > 1) {
314 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400315 stats->free_clusters = atomic_read(&flex_group[g].free_clusters);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500316 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
317 return;
318 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400319
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500320 desc = ext4_get_group_desc(sb, g, NULL);
321 if (desc) {
322 stats->free_inodes = ext4_free_inodes_count(sb, desc);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400323 stats->free_clusters = ext4_free_group_clusters(sb, desc);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500324 stats->used_dirs = ext4_used_dirs_count(sb, desc);
325 } else {
326 stats->free_inodes = 0;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400327 stats->free_clusters = 0;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500328 stats->used_dirs = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400329 }
330}
331
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700332/*
333 * Orlov's allocator for directories.
334 *
335 * We always try to spread first-level directories.
336 *
337 * If there are blockgroups with both free inodes and free blocks counts
338 * not worse than average we return one with smallest directory count.
339 * Otherwise we simply return a random group.
340 *
341 * For the rest rules look so:
342 *
343 * It's OK to put directory into a group unless
344 * it has too many directories already (max_dirs) or
345 * it has too few free inodes left (min_inodes) or
346 * it has too few free blocks left (min_blocks) or
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000347 * Parent's group is preferred, if it doesn't satisfy these
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700348 * conditions we search cyclically through the rest. If none
349 * of the groups look good we just look for a group with more
350 * free inodes than average (starting at parent's group).
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700351 */
352
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500353static int find_group_orlov(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400354 ext4_group_t *group, umode_t mode,
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400355 const struct qstr *qstr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700356{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500357 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Mingming Cao617ba132006-10-11 01:20:53 -0700358 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400359 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700360 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Theodore Ts'o14c83c92011-12-28 20:25:13 -0500361 unsigned int freei, avefreei, grp_free;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400362 ext4_fsblk_t freeb, avefreec;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700363 unsigned int ndirs;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400364 int max_dirs, min_inodes;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400365 ext4_grpblk_t min_clusters;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400366 ext4_group_t i, grp, g, ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700367 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400368 struct orlov_stats stats;
369 int flex_size = ext4_flex_bg_size(sbi);
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400370 struct dx_hash_info hinfo;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400371
Theodore Ts'o8df96752009-05-01 08:50:38 -0400372 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400373 if (flex_size > 1) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400374 ngroups = (real_ngroups + flex_size - 1) >>
Theodore Ts'oa4912122009-03-12 12:18:34 -0400375 sbi->s_log_groups_per_flex;
376 parent_group >>= sbi->s_log_groups_per_flex;
377 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700378
379 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
380 avefreei = freei / ngroups;
Theodore Ts'o57042652011-09-09 18:56:51 -0400381 freeb = EXT4_C2B(sbi,
382 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400383 avefreec = freeb;
384 do_div(avefreec, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700385 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
386
Theodore Ts'oa4912122009-03-12 12:18:34 -0400387 if (S_ISDIR(mode) &&
388 ((parent == sb->s_root->d_inode) ||
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400389 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700390 int best_ndir = inodes_per_group;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500391 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700392
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400393 if (qstr) {
394 hinfo.hash_version = DX_HASH_HALF_MD4;
395 hinfo.seed = sbi->s_hash_seed;
396 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
397 grp = hinfo.hash;
398 } else
399 get_random_bytes(&grp, sizeof(grp));
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500400 parent_group = (unsigned)grp % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700401 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400402 g = (parent_group + i) % ngroups;
403 get_orlov_stats(sb, g, flex_size, &stats);
404 if (!stats.free_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700405 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400406 if (stats.used_dirs >= best_ndir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700407 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400408 if (stats.free_inodes < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700409 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400410 if (stats.free_clusters < avefreec)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700411 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400412 grp = g;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500413 ret = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400414 best_ndir = stats.used_dirs;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700415 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400416 if (ret)
417 goto fallback;
418 found_flex_bg:
419 if (flex_size == 1) {
420 *group = grp;
421 return 0;
422 }
423
424 /*
425 * We pack inodes at the beginning of the flexgroup's
426 * inode tables. Block allocation decisions will do
427 * something similar, although regular files will
428 * start at 2nd block group of the flexgroup. See
429 * ext4_ext_find_goal() and ext4_find_near().
430 */
431 grp *= flex_size;
432 for (i = 0; i < flex_size; i++) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400433 if (grp+i >= real_ngroups)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400434 break;
435 desc = ext4_get_group_desc(sb, grp+i, NULL);
436 if (desc && ext4_free_inodes_count(sb, desc)) {
437 *group = grp+i;
438 return 0;
439 }
440 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700441 goto fallback;
442 }
443
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700444 max_dirs = ndirs / ngroups + inodes_per_group / 16;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400445 min_inodes = avefreei - inodes_per_group*flex_size / 4;
446 if (min_inodes < 1)
447 min_inodes = 1;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400448 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700449
Theodore Ts'oa4912122009-03-12 12:18:34 -0400450 /*
451 * Start looking in the flex group where we last allocated an
452 * inode for this parent directory
453 */
454 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
455 parent_group = EXT4_I(parent)->i_last_alloc_group;
456 if (flex_size > 1)
457 parent_group >>= sbi->s_log_groups_per_flex;
458 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700459
460 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400461 grp = (parent_group + i) % ngroups;
462 get_orlov_stats(sb, grp, flex_size, &stats);
463 if (stats.used_dirs >= max_dirs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700464 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400465 if (stats.free_inodes < min_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700466 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400467 if (stats.free_clusters < min_clusters)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700468 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400469 goto found_flex_bg;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700470 }
471
472fallback:
Theodore Ts'o8df96752009-05-01 08:50:38 -0400473 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400474 avefreei = freei / ngroups;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400475fallback_retry:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400476 parent_group = EXT4_I(parent)->i_block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700477 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400478 grp = (parent_group + i) % ngroups;
479 desc = ext4_get_group_desc(sb, grp, NULL);
Theodore Ts'o14c83c92011-12-28 20:25:13 -0500480 grp_free = ext4_free_inodes_count(sb, desc);
481 if (desc && grp_free && grp_free >= avefreei) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400482 *group = grp;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500483 return 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400484 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700485 }
486
487 if (avefreei) {
488 /*
489 * The free-inodes counter is approximate, and for really small
490 * filesystems the above test can fail to find any blockgroups
491 */
492 avefreei = 0;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400493 goto fallback_retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700494 }
495
496 return -1;
497}
498
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500499static int find_group_other(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400500 ext4_group_t *group, umode_t mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700501{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500502 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400503 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700504 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400505 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
506
507 /*
508 * Try to place the inode is the same flex group as its
509 * parent. If we can't find space, use the Orlov algorithm to
510 * find another flex group, and store that information in the
511 * parent directory's inode information so that use that flex
512 * group for future allocations.
513 */
514 if (flex_size > 1) {
515 int retry = 0;
516
517 try_again:
518 parent_group &= ~(flex_size-1);
519 last = parent_group + flex_size;
520 if (last > ngroups)
521 last = ngroups;
522 for (i = parent_group; i < last; i++) {
523 desc = ext4_get_group_desc(sb, i, NULL);
524 if (desc && ext4_free_inodes_count(sb, desc)) {
525 *group = i;
526 return 0;
527 }
528 }
529 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
530 retry = 1;
531 parent_group = EXT4_I(parent)->i_last_alloc_group;
532 goto try_again;
533 }
534 /*
535 * If this didn't work, use the Orlov search algorithm
536 * to find a new flex group; we pass in the mode to
537 * avoid the topdir algorithms.
538 */
539 *group = parent_group + flex_size;
540 if (*group > ngroups)
541 *group = 0;
Peter Huewe7dc57612011-02-21 21:01:42 -0500542 return find_group_orlov(sb, parent, group, mode, NULL);
Theodore Ts'oa4912122009-03-12 12:18:34 -0400543 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700544
545 /*
546 * Try to place the inode in its parent directory
547 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500548 *group = parent_group;
549 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500550 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400551 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500552 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700553
554 /*
555 * We're going to place this inode in a different blockgroup from its
556 * parent. We want to cause files in a common directory to all land in
557 * the same blockgroup. But we want files which are in a different
558 * directory which shares a blockgroup with our parent to land in a
559 * different blockgroup.
560 *
561 * So add our directory's i_ino into the starting point for the hash.
562 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500563 *group = (*group + parent->i_ino) % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700564
565 /*
566 * Use a quadratic hash to find a group with a free inode and some free
567 * blocks.
568 */
569 for (i = 1; i < ngroups; i <<= 1) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500570 *group += i;
571 if (*group >= ngroups)
572 *group -= ngroups;
573 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500574 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400575 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500576 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 }
578
579 /*
580 * That failed: try linear search for a free inode, even if that group
581 * has no free blocks.
582 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500583 *group = parent_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700584 for (i = 0; i < ngroups; i++) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500585 if (++*group >= ngroups)
586 *group = 0;
587 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500588 if (desc && ext4_free_inodes_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500589 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700590 }
591
592 return -1;
593}
594
595/*
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500596 * claim the inode from the inode bitmap. If the group
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400597 * is uninit we need to take the groups's ext4_group_lock
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500598 * and clear the uninit flag. The inode bitmap update
599 * and group desc uninit flag clear should be done
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400600 * after holding ext4_group_lock so that ext4_read_inode_bitmap
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500601 * doesn't race with the ext4_claim_inode
602 */
603static int ext4_claim_inode(struct super_block *sb,
604 struct buffer_head *inode_bitmap_bh,
Al Virodcca3fe2011-07-26 02:48:06 -0400605 unsigned long ino, ext4_group_t group, umode_t mode)
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500606{
607 int free = 0, retval = 0, count;
608 struct ext4_sb_info *sbi = EXT4_SB(sb);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400609 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500610 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
611
Lukas Czernerbfff6872010-10-27 21:30:05 -0400612 /*
613 * We have to be sure that new inode allocation does not race with
614 * inode table initialization, because otherwise we may end up
615 * allocating and writing new inode right before sb_issue_zeroout
616 * takes place and overwriting our new inode with zeroes. So we
617 * take alloc_sem to prevent it.
618 */
619 down_read(&grp->alloc_sem);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400620 ext4_lock_group(sb, group);
Akinobu Mita597d5082011-12-28 20:32:07 -0500621 if (ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data)) {
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500622 /* not a free inode */
623 retval = 1;
624 goto err_ret;
625 }
626 ino++;
627 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
628 ino > EXT4_INODES_PER_GROUP(sb)) {
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400629 ext4_unlock_group(sb, group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400630 up_read(&grp->alloc_sem);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500631 ext4_error(sb, "reserved inode or inode > inodes count - "
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500632 "block_group = %u, inode=%lu", group,
633 ino + group * EXT4_INODES_PER_GROUP(sb));
634 return 1;
635 }
636 /* If we didn't allocate from within the initialized part of the inode
637 * table then we need to initialize up to this inode. */
638 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
639
640 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
641 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
642 /* When marking the block group with
643 * ~EXT4_BG_INODE_UNINIT we don't want to depend
644 * on the value of bg_itable_unused even though
645 * mke2fs could have initialized the same for us.
646 * Instead we calculated the value below
647 */
648
649 free = 0;
650 } else {
651 free = EXT4_INODES_PER_GROUP(sb) -
652 ext4_itable_unused_count(sb, gdp);
653 }
654
655 /*
656 * Check the relative inode number against the last used
657 * relative inode number in this group. if it is greater
658 * we need to update the bg_itable_unused count
659 *
660 */
661 if (ino > free)
662 ext4_itable_unused_set(sb, gdp,
663 (EXT4_INODES_PER_GROUP(sb) - ino));
664 }
665 count = ext4_free_inodes_count(sb, gdp) - 1;
666 ext4_free_inodes_set(sb, gdp, count);
667 if (S_ISDIR(mode)) {
668 count = ext4_used_dirs_count(sb, gdp) + 1;
669 ext4_used_dirs_set(sb, gdp, count);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500670 if (sbi->s_log_groups_per_flex) {
671 ext4_group_t f = ext4_flex_group(sbi, group);
672
Eric Sandeenc4caae22010-03-23 21:32:00 -0400673 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500674 }
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500675 }
676 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
677err_ret:
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400678 ext4_unlock_group(sb, group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400679 up_read(&grp->alloc_sem);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500680 return retval;
681}
682
683/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700684 * There are two policies for allocating an inode. If the new inode is
685 * a directory, then a forward search is made for a block group with both
686 * free space and a low directory-to-inode ratio; if that fails, then of
687 * the groups with above-average free space, that group with the fewest
688 * directories already is chosen.
689 *
690 * For other inodes, search forward from the parent directory's block
691 * group to find a free inode.
692 */
Al Virodcca3fe2011-07-26 02:48:06 -0400693struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, umode_t mode,
Dmitry Monakhov5cb81da2011-10-29 09:05:00 -0400694 const struct qstr *qstr, __u32 goal, uid_t *owner)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700695{
696 struct super_block *sb;
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500697 struct buffer_head *inode_bitmap_bh = NULL;
698 struct buffer_head *group_desc_bh;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400699 ext4_group_t ngroups, group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700700 unsigned long ino = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400701 struct inode *inode;
702 struct ext4_group_desc *gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700703 struct ext4_inode_info *ei;
704 struct ext4_sb_info *sbi;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500705 int ret2, err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700706 struct inode *ret;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500707 ext4_group_t i;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400708 ext4_group_t flex_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700709
710 /* Cannot create files in a deleted directory */
711 if (!dir || !dir->i_nlink)
712 return ERR_PTR(-EPERM);
713
714 sb = dir->i_sb;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400715 ngroups = ext4_get_groups_count(sb);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400716 trace_ext4_request_inode(dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700717 inode = new_inode(sb);
718 if (!inode)
719 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700720 ei = EXT4_I(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700721 sbi = EXT4_SB(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400722
Andreas Dilger11013912009-06-13 11:45:35 -0400723 if (!goal)
724 goal = sbi->s_inode_goal;
725
Johann Lombardie6462862009-07-05 23:45:11 -0400726 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
Andreas Dilger11013912009-06-13 11:45:35 -0400727 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
728 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
729 ret2 = 0;
730 goto got_group;
731 }
732
Lukas Czerner4113c4c2011-10-08 14:34:47 -0400733 if (S_ISDIR(mode))
734 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
735 else
Theodore Ts'oa4912122009-03-12 12:18:34 -0400736 ret2 = find_group_other(sb, dir, &group, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700737
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400738got_group:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400739 EXT4_I(dir)->i_last_alloc_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700740 err = -ENOSPC;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500741 if (ret2 == -1)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700742 goto out;
743
Andreas Dilger11013912009-06-13 11:45:35 -0400744 for (i = 0; i < ngroups; i++, ino = 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700745 err = -EIO;
746
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500747 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700748 if (!gdp)
749 goto fail;
750
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500751 brelse(inode_bitmap_bh);
752 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
753 if (!inode_bitmap_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700754 goto fail;
755
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700756repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700757 ino = ext4_find_next_zero_bit((unsigned long *)
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500758 inode_bitmap_bh->b_data,
759 EXT4_INODES_PER_GROUP(sb), ino);
760
Mingming Cao617ba132006-10-11 01:20:53 -0700761 if (ino < EXT4_INODES_PER_GROUP(sb)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700762
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500763 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
764 err = ext4_journal_get_write_access(handle,
765 inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700766 if (err)
767 goto fail;
768
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500769 BUFFER_TRACE(group_desc_bh, "get_write_access");
770 err = ext4_journal_get_write_access(handle,
771 group_desc_bh);
772 if (err)
773 goto fail;
774 if (!ext4_claim_inode(sb, inode_bitmap_bh,
775 ino, group, mode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700776 /* we won it */
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500777 BUFFER_TRACE(inode_bitmap_bh,
Frank Mayhar03901312009-01-07 00:06:22 -0500778 "call ext4_handle_dirty_metadata");
779 err = ext4_handle_dirty_metadata(handle,
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -0500780 NULL,
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500781 inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700782 if (err)
783 goto fail;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500784 /* zero bit is inode number 1*/
785 ino++;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700786 goto got;
787 }
788 /* we lost it */
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500789 ext4_handle_release_buffer(handle, inode_bitmap_bh);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500790 ext4_handle_release_buffer(handle, group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700791
Mingming Cao617ba132006-10-11 01:20:53 -0700792 if (++ino < EXT4_INODES_PER_GROUP(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700793 goto repeat_in_this_group;
794 }
795
796 /*
797 * This case is possible in concurrent environment. It is very
798 * rare. We cannot repeat the find_group_xxx() call because
799 * that will simply return the same blockgroup, because the
800 * group descriptor metadata has not yet been updated.
801 * So we just go onto the next blockgroup.
802 */
Theodore Ts'o8df96752009-05-01 08:50:38 -0400803 if (++group == ngroups)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700804 group = 0;
805 }
806 err = -ENOSPC;
807 goto out;
808
809got:
Andreas Dilger717d50e2007-10-16 18:38:25 -0400810 /* We may have to initialize the block bitmap if it isn't already */
811 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
812 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500813 struct buffer_head *block_bitmap_bh;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400814
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500815 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
816 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
817 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400818 if (err) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500819 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400820 goto fail;
821 }
822
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400823 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
824 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
825 brelse(block_bitmap_bh);
826
Andreas Dilger717d50e2007-10-16 18:38:25 -0400827 /* recheck and clear flag under lock if we still need to */
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400828 ext4_lock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400829 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500830 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400831 ext4_free_group_clusters_set(sb, gdp,
Theodore Ts'ocff1dfd72011-09-09 19:12:51 -0400832 ext4_free_clusters_after_init(sb, group, gdp));
Frederic Bohe23712a92008-11-07 09:21:01 -0500833 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
834 gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400835 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400836 ext4_unlock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400837
Andreas Dilger717d50e2007-10-16 18:38:25 -0400838 if (err)
839 goto fail;
840 }
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500841 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
842 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500843 if (err)
844 goto fail;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700845
846 percpu_counter_dec(&sbi->s_freeinodes_counter);
847 if (S_ISDIR(mode))
848 percpu_counter_inc(&sbi->s_dirs_counter);
Theodore Ts'oa0375152010-06-11 23:14:04 -0400849 ext4_mark_super_dirty(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700850
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400851 if (sbi->s_log_groups_per_flex) {
852 flex_group = ext4_flex_group(sbi, group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500853 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400854 }
Dmitry Monakhov5cb81da2011-10-29 09:05:00 -0400855 if (owner) {
856 inode->i_mode = mode;
857 inode->i_uid = owner[0];
858 inode->i_gid = owner[1];
859 } else if (test_opt(sb, GRPID)) {
Dmitry Monakhovb10b8522010-03-04 17:31:51 +0300860 inode->i_mode = mode;
861 inode->i_uid = current_fsuid();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700862 inode->i_gid = dir->i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700863 } else
Dmitry Monakhovb10b8522010-03-04 17:31:51 +0300864 inode_init_owner(inode, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700865
Andreas Dilger717d50e2007-10-16 18:38:25 -0400866 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700867 /* This is the optimal IO size (for stat), not the fs block size */
868 inode->i_blocks = 0;
Kalpak Shahef7f3832007-07-18 09:15:20 -0400869 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
870 ext4_current_time(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700871
872 memset(ei->i_data, 0, sizeof(ei->i_data));
873 ei->i_dir_start_lookup = 0;
874 ei->i_disksize = 0;
875
Eryu Guan4af83502011-10-31 18:21:29 -0400876 /* Don't inherit extent flag from directory, amongst others. */
Duane Griffin2dc6b0d2009-02-15 18:09:20 -0500877 ei->i_flags =
878 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700879 ei->i_file_acl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700880 ei->i_dtime = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700881 ei->i_block_group = group;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400882 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700883
Mingming Cao617ba132006-10-11 01:20:53 -0700884 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700885 if (IS_DIRSYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -0500886 ext4_handle_sync(handle);
Al Viro6b38e842008-12-30 02:03:31 -0500887 if (insert_inode_locked(inode) < 0) {
Jan Karaacd6ad82011-12-18 17:37:02 -0500888 /*
889 * Likely a bitmap corruption causing inode to be allocated
890 * twice.
891 */
892 err = -EIO;
893 goto fail;
Al Viro6b38e842008-12-30 02:03:31 -0500894 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700895 spin_lock(&sbi->s_next_gen_lock);
896 inode->i_generation = sbi->s_next_generation++;
897 spin_unlock(&sbi->s_next_gen_lock);
898
Theodore Ts'o353eb832011-01-10 12:18:25 -0500899 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -0500900 ext4_set_inode_state(inode, EXT4_STATE_NEW);
Kalpak Shahef7f3832007-07-18 09:15:20 -0400901
902 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700903
904 ret = inode;
Christoph Hellwig871a2932010-03-03 09:05:07 -0500905 dquot_initialize(inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500906 err = dquot_alloc_inode(inode);
907 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700908 goto fail_drop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700909
Mingming Cao617ba132006-10-11 01:20:53 -0700910 err = ext4_init_acl(handle, inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700911 if (err)
912 goto fail_free_drop;
913
Eric Paris2a7dba32011-02-01 11:05:39 -0500914 err = ext4_init_security(handle, inode, dir, qstr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700915 if (err)
916 goto fail_free_drop;
917
Theodore Ts'o83982b62009-01-06 14:53:16 -0500918 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Eric Sandeene4079a12008-07-11 19:27:31 -0400919 /* set extent flag only for directory, file and normal symlink*/
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -0400920 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400921 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -0500922 ext4_ext_tree_init(handle, inode);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -0500923 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700924 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700925
Theodore Ts'o688f8692011-03-16 17:16:31 -0400926 if (ext4_handle_valid(handle)) {
927 ei->i_sync_tid = handle->h_transaction->t_tid;
928 ei->i_datasync_tid = handle->h_transaction->t_tid;
929 }
930
Aneesh Kumar K.V8753e882008-04-29 22:00:36 -0400931 err = ext4_mark_inode_dirty(handle, inode);
932 if (err) {
933 ext4_std_error(sb, err);
934 goto fail_free_drop;
935 }
936
Mingming Cao617ba132006-10-11 01:20:53 -0700937 ext4_debug("allocating inode %lu\n", inode->i_ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400938 trace_ext4_allocate_inode(inode, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700939 goto really_out;
940fail:
Mingming Cao617ba132006-10-11 01:20:53 -0700941 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700942out:
943 iput(inode);
944 ret = ERR_PTR(err);
945really_out:
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500946 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700947 return ret;
948
949fail_free_drop:
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500950 dquot_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700951
952fail_drop:
Christoph Hellwig9f754752010-03-03 09:05:05 -0500953 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700954 inode->i_flags |= S_NOQUOTA;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +0200955 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -0500956 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700957 iput(inode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500958 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700959 return ERR_PTR(err);
960}
961
962/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -0700963struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700964{
Mingming Cao617ba132006-10-11 01:20:53 -0700965 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500966 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700967 int bit;
David Howells1d1fe1e2008-02-07 00:15:37 -0800968 struct buffer_head *bitmap_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700969 struct inode *inode = NULL;
David Howells1d1fe1e2008-02-07 00:15:37 -0800970 long err = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700971
972 /* Error cases - e2fsck has already cleaned up for us */
973 if (ino > max_ino) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500974 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -0800975 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700976 }
977
Mingming Cao617ba132006-10-11 01:20:53 -0700978 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
979 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400980 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700981 if (!bitmap_bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500982 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -0800983 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700984 }
985
986 /* Having the inode bit set should be a 100% indicator that this
987 * is a valid orphan (no e2fsck run on fs). Orphans also include
988 * inodes that were being truncated, so we can't check i_nlink==0.
989 */
David Howells1d1fe1e2008-02-07 00:15:37 -0800990 if (!ext4_test_bit(bit, bitmap_bh->b_data))
991 goto bad_orphan;
992
993 inode = ext4_iget(sb, ino);
994 if (IS_ERR(inode))
995 goto iget_failed;
996
Duane Griffin91ef4ca2008-07-11 19:27:31 -0400997 /*
998 * If the orphans has i_nlinks > 0 then it should be able to be
999 * truncated, otherwise it won't be removed from the orphan list
1000 * during processing and an infinite loop will result.
1001 */
1002 if (inode->i_nlink && !ext4_can_truncate(inode))
1003 goto bad_orphan;
1004
David Howells1d1fe1e2008-02-07 00:15:37 -08001005 if (NEXT_ORPHAN(inode) > max_ino)
1006 goto bad_orphan;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001007 brelse(bitmap_bh);
1008 return inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08001009
1010iget_failed:
1011 err = PTR_ERR(inode);
1012 inode = NULL;
1013bad_orphan:
Eric Sandeen12062dd2010-02-15 14:19:27 -05001014 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001015 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1016 bit, (unsigned long long)bitmap_bh->b_blocknr,
1017 ext4_test_bit(bit, bitmap_bh->b_data));
1018 printk(KERN_NOTICE "inode=%p\n", inode);
1019 if (inode) {
1020 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1021 is_bad_inode(inode));
1022 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1023 NEXT_ORPHAN(inode));
1024 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001025 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
David Howells1d1fe1e2008-02-07 00:15:37 -08001026 /* Avoid freeing blocks if we got a bad deleted inode */
1027 if (inode->i_nlink == 0)
1028 inode->i_blocks = 0;
1029 iput(inode);
1030 }
1031 brelse(bitmap_bh);
1032error:
1033 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001034}
1035
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001036unsigned long ext4_count_free_inodes(struct super_block *sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001037{
1038 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001039 struct ext4_group_desc *gdp;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001040 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001041#ifdef EXT4FS_DEBUG
1042 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001043 unsigned long bitmap_count, x;
1044 struct buffer_head *bitmap_bh = NULL;
1045
Mingming Cao617ba132006-10-11 01:20:53 -07001046 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001047 desc_count = 0;
1048 bitmap_count = 0;
1049 gdp = NULL;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001050 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001051 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001052 if (!gdp)
1053 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001054 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001055 brelse(bitmap_bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001056 bitmap_bh = ext4_read_inode_bitmap(sb, i);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001057 if (!bitmap_bh)
1058 continue;
1059
Mingming Cao617ba132006-10-11 01:20:53 -07001060 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
Eric Sandeenc549a952008-01-28 23:58:27 -05001061 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
Peng Tao785b4b32009-07-27 21:44:40 -04001062 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001063 bitmap_count += x;
1064 }
1065 brelse(bitmap_bh);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001066 printk(KERN_DEBUG "ext4_count_free_inodes: "
1067 "stored = %u, computed = %lu, %lu\n",
1068 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001069 return desc_count;
1070#else
1071 desc_count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001072 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001073 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001074 if (!gdp)
1075 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001076 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001077 cond_resched();
1078 }
1079 return desc_count;
1080#endif
1081}
1082
1083/* Called at mount-time, super-block is locked */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001084unsigned long ext4_count_dirs(struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001085{
1086 unsigned long count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001087 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001088
Theodore Ts'o8df96752009-05-01 08:50:38 -04001089 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001090 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001091 if (!gdp)
1092 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001093 count += ext4_used_dirs_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001094 }
1095 return count;
1096}
Lukas Czernerbfff6872010-10-27 21:30:05 -04001097
1098/*
1099 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1100 * inode table. Must be called without any spinlock held. The only place
1101 * where it is called from on active part of filesystem is ext4lazyinit
1102 * thread, so we do not need any special locks, however we have to prevent
1103 * inode allocation from the current group, so we take alloc_sem lock, to
1104 * block ext4_claim_inode until we are finished.
1105 */
H Hartley Sweetene0cbee32011-10-18 10:57:51 -04001106int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
Lukas Czernerbfff6872010-10-27 21:30:05 -04001107 int barrier)
1108{
1109 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1110 struct ext4_sb_info *sbi = EXT4_SB(sb);
1111 struct ext4_group_desc *gdp = NULL;
1112 struct buffer_head *group_desc_bh;
1113 handle_t *handle;
1114 ext4_fsblk_t blk;
1115 int num, ret = 0, used_blks = 0;
Lukas Czernerbfff6872010-10-27 21:30:05 -04001116
1117 /* This should not happen, but just to be sure check this */
1118 if (sb->s_flags & MS_RDONLY) {
1119 ret = 1;
1120 goto out;
1121 }
1122
1123 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1124 if (!gdp)
1125 goto out;
1126
1127 /*
1128 * We do not need to lock this, because we are the only one
1129 * handling this flag.
1130 */
1131 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1132 goto out;
1133
1134 handle = ext4_journal_start_sb(sb, 1);
1135 if (IS_ERR(handle)) {
1136 ret = PTR_ERR(handle);
1137 goto out;
1138 }
1139
1140 down_write(&grp->alloc_sem);
1141 /*
1142 * If inode bitmap was already initialized there may be some
1143 * used inodes so we need to skip blocks with used inodes in
1144 * inode table.
1145 */
1146 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1147 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1148 ext4_itable_unused_count(sb, gdp)),
1149 sbi->s_inodes_per_block);
1150
Lukas Czerner857ac882010-10-27 21:30:05 -04001151 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1152 ext4_error(sb, "Something is wrong with group %u\n"
1153 "Used itable blocks: %d"
1154 "itable unused count: %u\n",
1155 group, used_blks,
1156 ext4_itable_unused_count(sb, gdp));
1157 ret = 1;
Yongqiang Yang33853a02011-08-01 06:32:19 -04001158 goto err_out;
Lukas Czerner857ac882010-10-27 21:30:05 -04001159 }
1160
Lukas Czernerbfff6872010-10-27 21:30:05 -04001161 blk = ext4_inode_table(sb, gdp) + used_blks;
1162 num = sbi->s_itb_per_group - used_blks;
1163
1164 BUFFER_TRACE(group_desc_bh, "get_write_access");
1165 ret = ext4_journal_get_write_access(handle,
1166 group_desc_bh);
1167 if (ret)
1168 goto err_out;
1169
Lukas Czernerbfff6872010-10-27 21:30:05 -04001170 /*
1171 * Skip zeroout if the inode table is full. But we set the ZEROED
1172 * flag anyway, because obviously, when it is full it does not need
1173 * further zeroing.
1174 */
1175 if (unlikely(num == 0))
1176 goto skip_zeroout;
1177
1178 ext4_debug("going to zero out inode table in group %d\n",
1179 group);
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001180 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001181 if (ret < 0)
1182 goto err_out;
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001183 if (barrier)
1184 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001185
1186skip_zeroout:
1187 ext4_lock_group(sb, group);
1188 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1189 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1190 ext4_unlock_group(sb, group);
1191
1192 BUFFER_TRACE(group_desc_bh,
1193 "call ext4_handle_dirty_metadata");
1194 ret = ext4_handle_dirty_metadata(handle, NULL,
1195 group_desc_bh);
1196
1197err_out:
1198 up_write(&grp->alloc_sem);
1199 ext4_journal_stop(handle);
1200out:
1201 return ret;
1202}