blob: 32fd2b9075dd4029925d11adb0b84df1a399a8b1 [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{
Andreas Dilger717d50e2007-10-16 18:38:25 -040073 J_ASSERT_BH(bh, buffer_locked(bh));
74
75 /* If checksum is bad mark all blocks and inodes use to prevent
76 * allocation, essentially implementing a per-group read-only flag. */
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -040077 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -050078 ext4_error(sb, "Checksum bad for group %u", block_group);
Theodore Ts'o021b65b2011-09-09 19:08:51 -040079 ext4_free_group_clusters_set(sb, gdp, 0);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -050080 ext4_free_inodes_set(sb, gdp, 0);
81 ext4_itable_unused_set(sb, gdp, 0);
Andreas Dilger717d50e2007-10-16 18:38:25 -040082 memset(bh->b_data, 0xff, sb->s_blocksize);
Darrick J. Wong41a246d2012-04-29 18:33:10 -040083 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
84 EXT4_INODES_PER_GROUP(sb) / 8);
Andreas Dilger717d50e2007-10-16 18:38:25 -040085 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);
Darrick J. Wong41a246d2012-04-29 18:33:10 -040091 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
92 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -040093 ext4_group_desc_csum_set(sb, block_group, gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -040094
95 return EXT4_INODES_PER_GROUP(sb);
96}
Dave Kleikampac27a0e2006-10-11 01:20:50 -070097
Theodore Ts'o813e5722012-02-20 17:52:46 -050098void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
99{
100 if (uptodate) {
101 set_buffer_uptodate(bh);
102 set_bitmap_uptodate(bh);
103 }
104 unlock_buffer(bh);
105 put_bh(bh);
106}
107
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700108/*
109 * Read the inode allocation bitmap for a given block_group, reading
110 * into the specified slot in the superblock's bitmap cache.
111 *
112 * Return buffer_head of bitmap on success or NULL.
113 */
114static struct buffer_head *
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400115ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700116{
Mingming Cao617ba132006-10-11 01:20:53 -0700117 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700118 struct buffer_head *bh = NULL;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400119 ext4_fsblk_t bitmap_blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700120
Mingming Cao617ba132006-10-11 01:20:53 -0700121 desc = ext4_get_group_desc(sb, block_group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700122 if (!desc)
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400123 return NULL;
Lukas Czernerbfff6872010-10-27 21:30:05 -0400124
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400125 bitmap_blk = ext4_inode_bitmap(sb, desc);
126 bh = sb_getblk(sb, bitmap_blk);
127 if (unlikely(!bh)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500128 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500129 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400130 block_group, bitmap_blk);
131 return NULL;
132 }
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500133 if (bitmap_uptodate(bh))
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400134 goto verify;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400135
Frederic Bohec806e682008-10-10 08:09:18 -0400136 lock_buffer(bh);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500137 if (bitmap_uptodate(bh)) {
138 unlock_buffer(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400139 goto verify;
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500140 }
Lukas Czernerbfff6872010-10-27 21:30:05 -0400141
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400142 ext4_lock_group(sb, block_group);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400143 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
144 ext4_init_inode_bitmap(sb, bh, block_group, desc);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500145 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400146 set_buffer_uptodate(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400147 set_buffer_verified(bh);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400148 ext4_unlock_group(sb, block_group);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500149 unlock_buffer(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400150 return bh;
151 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400152 ext4_unlock_group(sb, block_group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400153
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500154 if (buffer_uptodate(bh)) {
155 /*
156 * if not uninit if bh is uptodate,
157 * bitmap is also uptodate
158 */
159 set_bitmap_uptodate(bh);
160 unlock_buffer(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400161 goto verify;
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500162 }
163 /*
Theodore Ts'o813e5722012-02-20 17:52:46 -0500164 * submit the buffer_head for reading
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500165 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400166 trace_ext4_load_inode_bitmap(sb, block_group);
Theodore Ts'o813e5722012-02-20 17:52:46 -0500167 bh->b_end_io = ext4_end_bitmap_read;
168 get_bh(bh);
169 submit_bh(READ, bh);
170 wait_on_buffer(bh);
171 if (!buffer_uptodate(bh)) {
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400172 put_bh(bh);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500173 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'o813e5722012-02-20 17:52:46 -0500174 "block_group = %u, inode_bitmap = %llu",
175 block_group, bitmap_blk);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400176 return NULL;
177 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400178
179verify:
180 ext4_lock_group(sb, block_group);
181 if (!buffer_verified(bh) &&
182 !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
183 EXT4_INODES_PER_GROUP(sb) / 8)) {
184 ext4_unlock_group(sb, block_group);
185 put_bh(bh);
186 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
187 "inode_bitmap = %llu", block_group, bitmap_blk);
188 return NULL;
189 }
190 ext4_unlock_group(sb, block_group);
191 set_buffer_verified(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700192 return bh;
193}
194
195/*
196 * NOTE! When we get the inode, we're the only people
197 * that have access to it, and as such there are no
198 * race conditions we have to worry about. The inode
199 * is not on the hash-lists, and it cannot be reached
200 * through the filesystem because the directory entry
201 * has been deleted earlier.
202 *
203 * HOWEVER: we must make sure that we get no aliases,
204 * which means that we have to call "clear_inode()"
205 * _before_ we mark the inode not in use in the inode
206 * bitmaps. Otherwise a newly created file might use
207 * the same inode number (not actually the same pointer
208 * though), and then we'd have two inodes sharing the
209 * same inode number and space on the harddisk.
210 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400211void ext4_free_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700212{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400213 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700214 int is_directory;
215 unsigned long ino;
216 struct buffer_head *bitmap_bh = NULL;
217 struct buffer_head *bh2;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500218 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700219 unsigned long bit;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400220 struct ext4_group_desc *gdp;
221 struct ext4_super_block *es;
Mingming Cao617ba132006-10-11 01:20:53 -0700222 struct ext4_sb_info *sbi;
Eric Sandeen7ce9d5d2009-03-04 18:38:18 -0500223 int fatal = 0, err, count, cleared;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700224
Theodore Ts'o92b97812012-03-19 23:41:49 -0400225 if (!sb) {
226 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
227 "nonexistent device\n", __func__, __LINE__);
228 return;
229 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700230 if (atomic_read(&inode->i_count) > 1) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400231 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
232 __func__, __LINE__, inode->i_ino,
233 atomic_read(&inode->i_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234 return;
235 }
236 if (inode->i_nlink) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400237 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
238 __func__, __LINE__, inode->i_ino, inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700239 return;
240 }
Mingming Cao617ba132006-10-11 01:20:53 -0700241 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700242
243 ino = inode->i_ino;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400244 ext4_debug("freeing inode %lu\n", ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400245 trace_ext4_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700246
247 /*
248 * Note: we must free any quota before locking the superblock,
249 * as writing the quota to disk may need the lock as well.
250 */
Christoph Hellwig871a2932010-03-03 09:05:07 -0500251 dquot_initialize(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700252 ext4_xattr_delete_inode(handle, inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500253 dquot_free_inode(inode);
Christoph Hellwig9f754752010-03-03 09:05:05 -0500254 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700255
256 is_directory = S_ISDIR(inode->i_mode);
257
258 /* Do this BEFORE marking the inode not in use or returning an error */
Al Viro0930fcc2010-06-07 13:16:22 -0400259 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700260
Mingming Cao617ba132006-10-11 01:20:53 -0700261 es = EXT4_SB(sb)->s_es;
262 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500263 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700264 goto error_return;
265 }
Mingming Cao617ba132006-10-11 01:20:53 -0700266 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
267 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400268 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700269 if (!bitmap_bh)
270 goto error_return;
271
272 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700273 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700274 if (fatal)
275 goto error_return;
276
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400277 fatal = -ESRCH;
278 gdp = ext4_get_group_desc(sb, block_group, &bh2);
279 if (gdp) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700280 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700281 fatal = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700282 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400283 ext4_lock_group(sb, block_group);
Akinobu Mita597d5082011-12-28 20:32:07 -0500284 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400285 if (fatal || !cleared) {
286 ext4_unlock_group(sb, block_group);
287 goto out;
288 }
289
290 count = ext4_free_inodes_count(sb, gdp) + 1;
291 ext4_free_inodes_set(sb, gdp, count);
292 if (is_directory) {
293 count = ext4_used_dirs_count(sb, gdp) - 1;
294 ext4_used_dirs_set(sb, gdp, count);
295 percpu_counter_dec(&sbi->s_dirs_counter);
296 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400297 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
298 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400299 ext4_group_desc_csum_set(sb, block_group, gdp);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400300 ext4_unlock_group(sb, block_group);
301
302 percpu_counter_inc(&sbi->s_freeinodes_counter);
303 if (sbi->s_log_groups_per_flex) {
304 ext4_group_t f = ext4_flex_group(sbi, block_group);
305
306 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
307 if (is_directory)
308 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
309 }
310 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
311 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
312out:
313 if (cleared) {
314 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
315 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
316 if (!fatal)
317 fatal = err;
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400318 } else
319 ext4_error(sb, "bit already cleared for inode %lu", ino);
320
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700321error_return:
322 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700323 ext4_std_error(sb, fatal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700324}
325
Theodore Ts'oa4912122009-03-12 12:18:34 -0400326struct orlov_stats {
327 __u32 free_inodes;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400328 __u32 free_clusters;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400329 __u32 used_dirs;
330};
331
332/*
333 * Helper function for Orlov's allocator; returns critical information
334 * for a particular block group or flex_bg. If flex_size is 1, then g
335 * is a block group number; otherwise it is flex_bg number.
336 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400337static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
338 int flex_size, struct orlov_stats *stats)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400339{
340 struct ext4_group_desc *desc;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500341 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400342
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500343 if (flex_size > 1) {
344 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400345 stats->free_clusters = atomic_read(&flex_group[g].free_clusters);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500346 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
347 return;
348 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400349
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500350 desc = ext4_get_group_desc(sb, g, NULL);
351 if (desc) {
352 stats->free_inodes = ext4_free_inodes_count(sb, desc);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400353 stats->free_clusters = ext4_free_group_clusters(sb, desc);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500354 stats->used_dirs = ext4_used_dirs_count(sb, desc);
355 } else {
356 stats->free_inodes = 0;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400357 stats->free_clusters = 0;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500358 stats->used_dirs = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400359 }
360}
361
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700362/*
363 * Orlov's allocator for directories.
364 *
365 * We always try to spread first-level directories.
366 *
367 * If there are blockgroups with both free inodes and free blocks counts
368 * not worse than average we return one with smallest directory count.
369 * Otherwise we simply return a random group.
370 *
371 * For the rest rules look so:
372 *
373 * It's OK to put directory into a group unless
374 * it has too many directories already (max_dirs) or
375 * it has too few free inodes left (min_inodes) or
376 * it has too few free blocks left (min_blocks) or
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000377 * Parent's group is preferred, if it doesn't satisfy these
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700378 * conditions we search cyclically through the rest. If none
379 * of the groups look good we just look for a group with more
380 * free inodes than average (starting at parent's group).
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700381 */
382
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500383static int find_group_orlov(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400384 ext4_group_t *group, umode_t mode,
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400385 const struct qstr *qstr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700386{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500387 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Mingming Cao617ba132006-10-11 01:20:53 -0700388 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400389 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700390 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Theodore Ts'o14c83c92011-12-28 20:25:13 -0500391 unsigned int freei, avefreei, grp_free;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400392 ext4_fsblk_t freeb, avefreec;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700393 unsigned int ndirs;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400394 int max_dirs, min_inodes;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400395 ext4_grpblk_t min_clusters;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400396 ext4_group_t i, grp, g, ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700397 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400398 struct orlov_stats stats;
399 int flex_size = ext4_flex_bg_size(sbi);
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400400 struct dx_hash_info hinfo;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400401
Theodore Ts'o8df96752009-05-01 08:50:38 -0400402 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400403 if (flex_size > 1) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400404 ngroups = (real_ngroups + flex_size - 1) >>
Theodore Ts'oa4912122009-03-12 12:18:34 -0400405 sbi->s_log_groups_per_flex;
406 parent_group >>= sbi->s_log_groups_per_flex;
407 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700408
409 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
410 avefreei = freei / ngroups;
Theodore Ts'o57042652011-09-09 18:56:51 -0400411 freeb = EXT4_C2B(sbi,
412 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400413 avefreec = freeb;
414 do_div(avefreec, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700415 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
416
Theodore Ts'oa4912122009-03-12 12:18:34 -0400417 if (S_ISDIR(mode) &&
418 ((parent == sb->s_root->d_inode) ||
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400419 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700420 int best_ndir = inodes_per_group;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500421 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700422
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400423 if (qstr) {
424 hinfo.hash_version = DX_HASH_HALF_MD4;
425 hinfo.seed = sbi->s_hash_seed;
426 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
427 grp = hinfo.hash;
428 } else
429 get_random_bytes(&grp, sizeof(grp));
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500430 parent_group = (unsigned)grp % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700431 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400432 g = (parent_group + i) % ngroups;
433 get_orlov_stats(sb, g, flex_size, &stats);
434 if (!stats.free_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700435 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400436 if (stats.used_dirs >= best_ndir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700437 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400438 if (stats.free_inodes < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700439 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400440 if (stats.free_clusters < avefreec)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700441 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400442 grp = g;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500443 ret = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400444 best_ndir = stats.used_dirs;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700445 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400446 if (ret)
447 goto fallback;
448 found_flex_bg:
449 if (flex_size == 1) {
450 *group = grp;
451 return 0;
452 }
453
454 /*
455 * We pack inodes at the beginning of the flexgroup's
456 * inode tables. Block allocation decisions will do
457 * something similar, although regular files will
458 * start at 2nd block group of the flexgroup. See
459 * ext4_ext_find_goal() and ext4_find_near().
460 */
461 grp *= flex_size;
462 for (i = 0; i < flex_size; i++) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400463 if (grp+i >= real_ngroups)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400464 break;
465 desc = ext4_get_group_desc(sb, grp+i, NULL);
466 if (desc && ext4_free_inodes_count(sb, desc)) {
467 *group = grp+i;
468 return 0;
469 }
470 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700471 goto fallback;
472 }
473
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700474 max_dirs = ndirs / ngroups + inodes_per_group / 16;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400475 min_inodes = avefreei - inodes_per_group*flex_size / 4;
476 if (min_inodes < 1)
477 min_inodes = 1;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400478 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700479
Theodore Ts'oa4912122009-03-12 12:18:34 -0400480 /*
481 * Start looking in the flex group where we last allocated an
482 * inode for this parent directory
483 */
484 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
485 parent_group = EXT4_I(parent)->i_last_alloc_group;
486 if (flex_size > 1)
487 parent_group >>= sbi->s_log_groups_per_flex;
488 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700489
490 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400491 grp = (parent_group + i) % ngroups;
492 get_orlov_stats(sb, grp, flex_size, &stats);
493 if (stats.used_dirs >= max_dirs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700494 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400495 if (stats.free_inodes < min_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700496 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400497 if (stats.free_clusters < min_clusters)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700498 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400499 goto found_flex_bg;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700500 }
501
502fallback:
Theodore Ts'o8df96752009-05-01 08:50:38 -0400503 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400504 avefreei = freei / ngroups;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400505fallback_retry:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400506 parent_group = EXT4_I(parent)->i_block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700507 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400508 grp = (parent_group + i) % ngroups;
509 desc = ext4_get_group_desc(sb, grp, NULL);
Dan Carpenterbb3d1322012-05-28 14:16:57 -0400510 if (desc) {
511 grp_free = ext4_free_inodes_count(sb, desc);
512 if (grp_free && grp_free >= avefreei) {
513 *group = grp;
514 return 0;
515 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400516 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700517 }
518
519 if (avefreei) {
520 /*
521 * The free-inodes counter is approximate, and for really small
522 * filesystems the above test can fail to find any blockgroups
523 */
524 avefreei = 0;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400525 goto fallback_retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700526 }
527
528 return -1;
529}
530
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500531static int find_group_other(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400532 ext4_group_t *group, umode_t mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700533{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500534 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400535 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700536 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400537 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
538
539 /*
540 * Try to place the inode is the same flex group as its
541 * parent. If we can't find space, use the Orlov algorithm to
542 * find another flex group, and store that information in the
543 * parent directory's inode information so that use that flex
544 * group for future allocations.
545 */
546 if (flex_size > 1) {
547 int retry = 0;
548
549 try_again:
550 parent_group &= ~(flex_size-1);
551 last = parent_group + flex_size;
552 if (last > ngroups)
553 last = ngroups;
554 for (i = parent_group; i < last; i++) {
555 desc = ext4_get_group_desc(sb, i, NULL);
556 if (desc && ext4_free_inodes_count(sb, desc)) {
557 *group = i;
558 return 0;
559 }
560 }
561 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
562 retry = 1;
563 parent_group = EXT4_I(parent)->i_last_alloc_group;
564 goto try_again;
565 }
566 /*
567 * If this didn't work, use the Orlov search algorithm
568 * to find a new flex group; we pass in the mode to
569 * avoid the topdir algorithms.
570 */
571 *group = parent_group + flex_size;
572 if (*group > ngroups)
573 *group = 0;
Peter Huewe7dc57612011-02-21 21:01:42 -0500574 return find_group_orlov(sb, parent, group, mode, NULL);
Theodore Ts'oa4912122009-03-12 12:18:34 -0400575 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700576
577 /*
578 * Try to place the inode in its parent directory
579 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500580 *group = parent_group;
581 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500582 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400583 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500584 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700585
586 /*
587 * We're going to place this inode in a different blockgroup from its
588 * parent. We want to cause files in a common directory to all land in
589 * the same blockgroup. But we want files which are in a different
590 * directory which shares a blockgroup with our parent to land in a
591 * different blockgroup.
592 *
593 * So add our directory's i_ino into the starting point for the hash.
594 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500595 *group = (*group + parent->i_ino) % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700596
597 /*
598 * Use a quadratic hash to find a group with a free inode and some free
599 * blocks.
600 */
601 for (i = 1; i < ngroups; i <<= 1) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500602 *group += i;
603 if (*group >= ngroups)
604 *group -= ngroups;
605 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500606 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400607 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500608 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700609 }
610
611 /*
612 * That failed: try linear search for a free inode, even if that group
613 * has no free blocks.
614 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500615 *group = parent_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700616 for (i = 0; i < ngroups; i++) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500617 if (++*group >= ngroups)
618 *group = 0;
619 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500620 if (desc && ext4_free_inodes_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500621 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700622 }
623
624 return -1;
625}
626
627/*
628 * There are two policies for allocating an inode. If the new inode is
629 * a directory, then a forward search is made for a block group with both
630 * free space and a low directory-to-inode ratio; if that fails, then of
631 * the groups with above-average free space, that group with the fewest
632 * directories already is chosen.
633 *
634 * For other inodes, search forward from the parent directory's block
635 * group to find a free inode.
636 */
Theodore Ts'o11395752013-02-09 16:27:09 -0500637struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
638 umode_t mode, const struct qstr *qstr,
639 __u32 goal, uid_t *owner, int handle_type,
640 unsigned int line_no, int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700641{
642 struct super_block *sb;
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500643 struct buffer_head *inode_bitmap_bh = NULL;
644 struct buffer_head *group_desc_bh;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400645 ext4_group_t ngroups, group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700646 unsigned long ino = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400647 struct inode *inode;
648 struct ext4_group_desc *gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700649 struct ext4_inode_info *ei;
650 struct ext4_sb_info *sbi;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500651 int ret2, err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700652 struct inode *ret;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500653 ext4_group_t i;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400654 ext4_group_t flex_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700655
656 /* Cannot create files in a deleted directory */
657 if (!dir || !dir->i_nlink)
658 return ERR_PTR(-EPERM);
659
660 sb = dir->i_sb;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400661 ngroups = ext4_get_groups_count(sb);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400662 trace_ext4_request_inode(dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700663 inode = new_inode(sb);
664 if (!inode)
665 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700666 ei = EXT4_I(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700667 sbi = EXT4_SB(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400668
Andreas Dilger11013912009-06-13 11:45:35 -0400669 if (!goal)
670 goal = sbi->s_inode_goal;
671
Johann Lombardie6462862009-07-05 23:45:11 -0400672 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
Andreas Dilger11013912009-06-13 11:45:35 -0400673 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
674 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
675 ret2 = 0;
676 goto got_group;
677 }
678
Lukas Czerner4113c4c2011-10-08 14:34:47 -0400679 if (S_ISDIR(mode))
680 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
681 else
Theodore Ts'oa4912122009-03-12 12:18:34 -0400682 ret2 = find_group_other(sb, dir, &group, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700683
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400684got_group:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400685 EXT4_I(dir)->i_last_alloc_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700686 err = -ENOSPC;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500687 if (ret2 == -1)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700688 goto out;
689
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500690 /*
691 * Normally we will only go through one pass of this loop,
692 * unless we get unlucky and it turns out the group we selected
693 * had its last inode grabbed by someone else.
694 */
Andreas Dilger11013912009-06-13 11:45:35 -0400695 for (i = 0; i < ngroups; i++, ino = 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700696 err = -EIO;
697
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500698 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700699 if (!gdp)
700 goto fail;
701
Yongqiang Yangf2a09af2012-09-23 23:16:03 -0400702 /*
703 * Check free inodes count before loading bitmap.
704 */
705 if (ext4_free_inodes_count(sb, gdp) == 0) {
706 if (++group == ngroups)
707 group = 0;
708 continue;
709 }
710
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500711 brelse(inode_bitmap_bh);
712 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
713 if (!inode_bitmap_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700714 goto fail;
715
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700716repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700717 ino = ext4_find_next_zero_bit((unsigned long *)
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500718 inode_bitmap_bh->b_data,
719 EXT4_INODES_PER_GROUP(sb), ino);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500720 if (ino >= EXT4_INODES_PER_GROUP(sb)) {
721 if (++group == ngroups)
722 group = 0;
723 continue;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700724 }
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500725 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
726 ext4_error(sb, "reserved inode found cleared - "
727 "inode=%lu", ino + 1);
728 continue;
729 }
Theodore Ts'o11395752013-02-09 16:27:09 -0500730 if (!handle) {
731 BUG_ON(nblocks <= 0);
732 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
733 handle_type, nblocks);
734 if (IS_ERR(handle)) {
735 err = PTR_ERR(handle);
736 goto fail;
737 }
738 }
Eric Sandeenffb53872012-10-28 22:24:57 -0400739 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
740 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
741 if (err)
742 goto fail;
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500743 ext4_lock_group(sb, group);
744 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
745 ext4_unlock_group(sb, group);
746 ino++; /* the inode bitmap is zero-based */
747 if (!ret2)
748 goto got; /* we grabbed the inode! */
749 if (ino < EXT4_INODES_PER_GROUP(sb))
750 goto repeat_in_this_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700751 }
752 err = -ENOSPC;
753 goto out;
754
755got:
Eric Sandeenffb53872012-10-28 22:24:57 -0400756 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
757 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
758 if (err)
759 goto fail;
760
Andreas Dilger717d50e2007-10-16 18:38:25 -0400761 /* We may have to initialize the block bitmap if it isn't already */
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400762 if (ext4_has_group_desc_csum(sb) &&
Andreas Dilger717d50e2007-10-16 18:38:25 -0400763 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500764 struct buffer_head *block_bitmap_bh;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400765
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500766 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
767 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
768 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400769 if (err) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500770 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400771 goto fail;
772 }
773
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400774 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
775 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400776
Andreas Dilger717d50e2007-10-16 18:38:25 -0400777 /* recheck and clear flag under lock if we still need to */
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400778 ext4_lock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400779 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500780 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400781 ext4_free_group_clusters_set(sb, gdp,
Theodore Ts'ocff1dfd72011-09-09 19:12:51 -0400782 ext4_free_clusters_after_init(sb, group, gdp));
Darrick J. Wongfa77dcf2012-04-29 18:35:10 -0400783 ext4_block_bitmap_csum_set(sb, group, gdp,
Tao Ma79f1ba42012-10-22 00:34:32 -0400784 block_bitmap_bh);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400785 ext4_group_desc_csum_set(sb, group, gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400786 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400787 ext4_unlock_group(sb, group);
Theodore Ts'oaeb1e5d62012-11-29 21:21:22 -0500788 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400789
Andreas Dilger717d50e2007-10-16 18:38:25 -0400790 if (err)
791 goto fail;
792 }
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500793
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500794 BUFFER_TRACE(group_desc_bh, "get_write_access");
795 err = ext4_journal_get_write_access(handle, group_desc_bh);
796 if (err)
797 goto fail;
798
799 /* Update the relevant bg descriptor fields */
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400800 if (ext4_has_group_desc_csum(sb)) {
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500801 int free;
802 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
803
804 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
805 ext4_lock_group(sb, group); /* while we modify the bg desc */
806 free = EXT4_INODES_PER_GROUP(sb) -
807 ext4_itable_unused_count(sb, gdp);
808 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
809 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
810 free = 0;
811 }
812 /*
813 * Check the relative inode number against the last used
814 * relative inode number in this group. if it is greater
815 * we need to update the bg_itable_unused count
816 */
817 if (ino > free)
818 ext4_itable_unused_set(sb, gdp,
819 (EXT4_INODES_PER_GROUP(sb) - ino));
820 up_read(&grp->alloc_sem);
Tao Ma6f2e9f02012-05-28 18:20:59 -0400821 } else {
822 ext4_lock_group(sb, group);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500823 }
Tao Ma6f2e9f02012-05-28 18:20:59 -0400824
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500825 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
826 if (S_ISDIR(mode)) {
827 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
828 if (sbi->s_log_groups_per_flex) {
829 ext4_group_t f = ext4_flex_group(sbi, group);
830
831 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
832 }
833 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400834 if (ext4_has_group_desc_csum(sb)) {
835 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
836 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400837 ext4_group_desc_csum_set(sb, group, gdp);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500838 }
Tao Ma6f2e9f02012-05-28 18:20:59 -0400839 ext4_unlock_group(sb, group);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500840
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);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700849
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400850 if (sbi->s_log_groups_per_flex) {
851 flex_group = ext4_flex_group(sbi, group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500852 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400853 }
Dmitry Monakhov5cb81da2011-10-29 09:05:00 -0400854 if (owner) {
855 inode->i_mode = mode;
Eric W. Biederman08cefc72012-02-07 15:41:49 -0800856 i_uid_write(inode, owner[0]);
857 i_gid_write(inode, owner[1]);
Dmitry Monakhov5cb81da2011-10-29 09:05:00 -0400858 } else if (test_opt(sb, GRPID)) {
Dmitry Monakhovb10b8522010-03-04 17:31:51 +0300859 inode->i_mode = mode;
860 inode->i_uid = current_fsuid();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700861 inode->i_gid = dir->i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700862 } else
Dmitry Monakhovb10b8522010-03-04 17:31:51 +0300863 inode_init_owner(inode, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700864
Andreas Dilger717d50e2007-10-16 18:38:25 -0400865 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700866 /* This is the optimal IO size (for stat), not the fs block size */
867 inode->i_blocks = 0;
Kalpak Shahef7f3832007-07-18 09:15:20 -0400868 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
869 ext4_current_time(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700870
871 memset(ei->i_data, 0, sizeof(ei->i_data));
872 ei->i_dir_start_lookup = 0;
873 ei->i_disksize = 0;
874
Eryu Guan4af83502011-10-31 18:21:29 -0400875 /* Don't inherit extent flag from directory, amongst others. */
Duane Griffin2dc6b0d2009-02-15 18:09:20 -0500876 ei->i_flags =
877 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700878 ei->i_file_acl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700879 ei->i_dtime = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700880 ei->i_block_group = group;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400881 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700882
Mingming Cao617ba132006-10-11 01:20:53 -0700883 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700884 if (IS_DIRSYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -0500885 ext4_handle_sync(handle);
Al Viro6b38e842008-12-30 02:03:31 -0500886 if (insert_inode_locked(inode) < 0) {
Jan Karaacd6ad82011-12-18 17:37:02 -0500887 /*
888 * Likely a bitmap corruption causing inode to be allocated
889 * twice.
890 */
891 err = -EIO;
892 goto fail;
Al Viro6b38e842008-12-30 02:03:31 -0500893 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700894 spin_lock(&sbi->s_next_gen_lock);
895 inode->i_generation = sbi->s_next_generation++;
896 spin_unlock(&sbi->s_next_gen_lock);
897
Darrick J. Wong814525f2012-04-29 18:31:10 -0400898 /* Precompute checksum seed for inode metadata */
899 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
900 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
901 __u32 csum;
902 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
903 __le32 inum = cpu_to_le32(inode->i_ino);
904 __le32 gen = cpu_to_le32(inode->i_generation);
905 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
906 sizeof(inum));
907 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
908 sizeof(gen));
909 }
910
Theodore Ts'o353eb832011-01-10 12:18:25 -0500911 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -0500912 ext4_set_inode_state(inode, EXT4_STATE_NEW);
Kalpak Shahef7f3832007-07-18 09:15:20 -0400913
914 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700915
Tao Maf08225d2012-12-10 14:06:03 -0500916 ei->i_inline_off = 0;
917 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA))
918 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
919
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700920 ret = inode;
Christoph Hellwig871a2932010-03-03 09:05:07 -0500921 dquot_initialize(inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500922 err = dquot_alloc_inode(inode);
923 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700924 goto fail_drop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700925
Mingming Cao617ba132006-10-11 01:20:53 -0700926 err = ext4_init_acl(handle, inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700927 if (err)
928 goto fail_free_drop;
929
Eric Paris2a7dba32011-02-01 11:05:39 -0500930 err = ext4_init_security(handle, inode, dir, qstr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700931 if (err)
932 goto fail_free_drop;
933
Theodore Ts'o83982b62009-01-06 14:53:16 -0500934 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Eric Sandeene4079a12008-07-11 19:27:31 -0400935 /* set extent flag only for directory, file and normal symlink*/
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -0400936 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400937 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -0500938 ext4_ext_tree_init(handle, inode);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -0500939 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700940 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700941
Theodore Ts'o688f8692011-03-16 17:16:31 -0400942 if (ext4_handle_valid(handle)) {
943 ei->i_sync_tid = handle->h_transaction->t_tid;
944 ei->i_datasync_tid = handle->h_transaction->t_tid;
945 }
946
Aneesh Kumar K.V8753e882008-04-29 22:00:36 -0400947 err = ext4_mark_inode_dirty(handle, inode);
948 if (err) {
949 ext4_std_error(sb, err);
950 goto fail_free_drop;
951 }
952
Mingming Cao617ba132006-10-11 01:20:53 -0700953 ext4_debug("allocating inode %lu\n", inode->i_ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400954 trace_ext4_allocate_inode(inode, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700955 goto really_out;
956fail:
Mingming Cao617ba132006-10-11 01:20:53 -0700957 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700958out:
959 iput(inode);
960 ret = ERR_PTR(err);
961really_out:
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500962 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700963 return ret;
964
965fail_free_drop:
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500966 dquot_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700967
968fail_drop:
Christoph Hellwig9f754752010-03-03 09:05:05 -0500969 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700970 inode->i_flags |= S_NOQUOTA;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +0200971 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -0500972 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700973 iput(inode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500974 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700975 return ERR_PTR(err);
976}
977
978/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -0700979struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700980{
Mingming Cao617ba132006-10-11 01:20:53 -0700981 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500982 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700983 int bit;
David Howells1d1fe1e2008-02-07 00:15:37 -0800984 struct buffer_head *bitmap_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700985 struct inode *inode = NULL;
David Howells1d1fe1e2008-02-07 00:15:37 -0800986 long err = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700987
988 /* Error cases - e2fsck has already cleaned up for us */
989 if (ino > max_ino) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500990 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -0800991 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700992 }
993
Mingming Cao617ba132006-10-11 01:20:53 -0700994 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
995 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400996 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700997 if (!bitmap_bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500998 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -0800999 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001000 }
1001
1002 /* Having the inode bit set should be a 100% indicator that this
1003 * is a valid orphan (no e2fsck run on fs). Orphans also include
1004 * inodes that were being truncated, so we can't check i_nlink==0.
1005 */
David Howells1d1fe1e2008-02-07 00:15:37 -08001006 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1007 goto bad_orphan;
1008
1009 inode = ext4_iget(sb, ino);
1010 if (IS_ERR(inode))
1011 goto iget_failed;
1012
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001013 /*
1014 * If the orphans has i_nlinks > 0 then it should be able to be
1015 * truncated, otherwise it won't be removed from the orphan list
1016 * during processing and an infinite loop will result.
1017 */
1018 if (inode->i_nlink && !ext4_can_truncate(inode))
1019 goto bad_orphan;
1020
David Howells1d1fe1e2008-02-07 00:15:37 -08001021 if (NEXT_ORPHAN(inode) > max_ino)
1022 goto bad_orphan;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001023 brelse(bitmap_bh);
1024 return inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08001025
1026iget_failed:
1027 err = PTR_ERR(inode);
1028 inode = NULL;
1029bad_orphan:
Eric Sandeen12062dd2010-02-15 14:19:27 -05001030 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001031 printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001032 bit, (unsigned long long)bitmap_bh->b_blocknr,
1033 ext4_test_bit(bit, bitmap_bh->b_data));
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001034 printk(KERN_WARNING "inode=%p\n", inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08001035 if (inode) {
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001036 printk(KERN_WARNING "is_bad_inode(inode)=%d\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001037 is_bad_inode(inode));
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001038 printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001039 NEXT_ORPHAN(inode));
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001040 printk(KERN_WARNING "max_ino=%lu\n", max_ino);
1041 printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink);
David Howells1d1fe1e2008-02-07 00:15:37 -08001042 /* Avoid freeing blocks if we got a bad deleted inode */
1043 if (inode->i_nlink == 0)
1044 inode->i_blocks = 0;
1045 iput(inode);
1046 }
1047 brelse(bitmap_bh);
1048error:
1049 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001050}
1051
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001052unsigned long ext4_count_free_inodes(struct super_block *sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001053{
1054 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001055 struct ext4_group_desc *gdp;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001056 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001057#ifdef EXT4FS_DEBUG
1058 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001059 unsigned long bitmap_count, x;
1060 struct buffer_head *bitmap_bh = NULL;
1061
Mingming Cao617ba132006-10-11 01:20:53 -07001062 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001063 desc_count = 0;
1064 bitmap_count = 0;
1065 gdp = NULL;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001066 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001067 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001068 if (!gdp)
1069 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001070 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001071 brelse(bitmap_bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001072 bitmap_bh = ext4_read_inode_bitmap(sb, i);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001073 if (!bitmap_bh)
1074 continue;
1075
Theodore Ts'of6fb99c2012-06-30 19:14:57 -04001076 x = ext4_count_free(bitmap_bh->b_data,
1077 EXT4_INODES_PER_GROUP(sb) / 8);
Eric Sandeenc549a952008-01-28 23:58:27 -05001078 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
Peng Tao785b4b32009-07-27 21:44:40 -04001079 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001080 bitmap_count += x;
1081 }
1082 brelse(bitmap_bh);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001083 printk(KERN_DEBUG "ext4_count_free_inodes: "
1084 "stored = %u, computed = %lu, %lu\n",
1085 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001086 return desc_count;
1087#else
1088 desc_count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001089 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001090 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 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001094 cond_resched();
1095 }
1096 return desc_count;
1097#endif
1098}
1099
1100/* Called at mount-time, super-block is locked */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001101unsigned long ext4_count_dirs(struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001102{
1103 unsigned long count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001104 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001105
Theodore Ts'o8df96752009-05-01 08:50:38 -04001106 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001107 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001108 if (!gdp)
1109 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001110 count += ext4_used_dirs_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001111 }
1112 return count;
1113}
Lukas Czernerbfff6872010-10-27 21:30:05 -04001114
1115/*
1116 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1117 * inode table. Must be called without any spinlock held. The only place
1118 * where it is called from on active part of filesystem is ext4lazyinit
1119 * thread, so we do not need any special locks, however we have to prevent
1120 * inode allocation from the current group, so we take alloc_sem lock, to
Theodore Ts'o119c0d42012-02-06 20:12:03 -05001121 * block ext4_new_inode() until we are finished.
Lukas Czernerbfff6872010-10-27 21:30:05 -04001122 */
H Hartley Sweetene0cbee32011-10-18 10:57:51 -04001123int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
Lukas Czernerbfff6872010-10-27 21:30:05 -04001124 int barrier)
1125{
1126 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1127 struct ext4_sb_info *sbi = EXT4_SB(sb);
1128 struct ext4_group_desc *gdp = NULL;
1129 struct buffer_head *group_desc_bh;
1130 handle_t *handle;
1131 ext4_fsblk_t blk;
1132 int num, ret = 0, used_blks = 0;
Lukas Czernerbfff6872010-10-27 21:30:05 -04001133
1134 /* This should not happen, but just to be sure check this */
1135 if (sb->s_flags & MS_RDONLY) {
1136 ret = 1;
1137 goto out;
1138 }
1139
1140 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1141 if (!gdp)
1142 goto out;
1143
1144 /*
1145 * We do not need to lock this, because we are the only one
1146 * handling this flag.
1147 */
1148 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1149 goto out;
1150
Theodore Ts'o9924a922013-02-08 21:59:22 -05001151 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001152 if (IS_ERR(handle)) {
1153 ret = PTR_ERR(handle);
1154 goto out;
1155 }
1156
1157 down_write(&grp->alloc_sem);
1158 /*
1159 * If inode bitmap was already initialized there may be some
1160 * used inodes so we need to skip blocks with used inodes in
1161 * inode table.
1162 */
1163 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1164 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1165 ext4_itable_unused_count(sb, gdp)),
1166 sbi->s_inodes_per_block);
1167
Lukas Czerner857ac882010-10-27 21:30:05 -04001168 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
Theodore Ts'o1084f252012-03-19 23:13:43 -04001169 ext4_error(sb, "Something is wrong with group %u: "
1170 "used itable blocks: %d; "
1171 "itable unused count: %u",
Lukas Czerner857ac882010-10-27 21:30:05 -04001172 group, used_blks,
1173 ext4_itable_unused_count(sb, gdp));
1174 ret = 1;
Yongqiang Yang33853a02011-08-01 06:32:19 -04001175 goto err_out;
Lukas Czerner857ac882010-10-27 21:30:05 -04001176 }
1177
Lukas Czernerbfff6872010-10-27 21:30:05 -04001178 blk = ext4_inode_table(sb, gdp) + used_blks;
1179 num = sbi->s_itb_per_group - used_blks;
1180
1181 BUFFER_TRACE(group_desc_bh, "get_write_access");
1182 ret = ext4_journal_get_write_access(handle,
1183 group_desc_bh);
1184 if (ret)
1185 goto err_out;
1186
Lukas Czernerbfff6872010-10-27 21:30:05 -04001187 /*
1188 * Skip zeroout if the inode table is full. But we set the ZEROED
1189 * flag anyway, because obviously, when it is full it does not need
1190 * further zeroing.
1191 */
1192 if (unlikely(num == 0))
1193 goto skip_zeroout;
1194
1195 ext4_debug("going to zero out inode table in group %d\n",
1196 group);
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001197 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001198 if (ret < 0)
1199 goto err_out;
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001200 if (barrier)
1201 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001202
1203skip_zeroout:
1204 ext4_lock_group(sb, group);
1205 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001206 ext4_group_desc_csum_set(sb, group, gdp);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001207 ext4_unlock_group(sb, group);
1208
1209 BUFFER_TRACE(group_desc_bh,
1210 "call ext4_handle_dirty_metadata");
1211 ret = ext4_handle_dirty_metadata(handle, NULL,
1212 group_desc_bh);
1213
1214err_out:
1215 up_write(&grp->alloc_sem);
1216 ext4_journal_stop(handle);
1217out:
1218 return ret;
1219}