blob: 137193ff389b88510f414499116c8a9ee858abe5 [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{
Darrick J. Wongbdfb6ff2013-08-28 18:46:56 -040073 struct ext4_group_info *grp;
Andreas Dilger717d50e2007-10-16 18:38:25 -040074 J_ASSERT_BH(bh, buffer_locked(bh));
75
76 /* If checksum is bad mark all blocks and inodes use to prevent
77 * allocation, essentially implementing a per-group read-only flag. */
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -040078 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -050079 ext4_error(sb, "Checksum bad for group %u", block_group);
Darrick J. Wongbdfb6ff2013-08-28 18:46:56 -040080 grp = ext4_get_group_info(sb, block_group);
81 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
82 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
Andreas Dilger717d50e2007-10-16 18:38:25 -040083 return 0;
84 }
85
86 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
Theodore Ts'o61d08672010-10-27 21:30:15 -040087 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
Andreas Dilger717d50e2007-10-16 18:38:25 -040088 bh->b_data);
Darrick J. Wong41a246d2012-04-29 18:33:10 -040089 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
90 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -040091 ext4_group_desc_csum_set(sb, block_group, gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -040092
93 return EXT4_INODES_PER_GROUP(sb);
94}
Dave Kleikampac27a0e2006-10-11 01:20:50 -070095
Theodore Ts'o813e5722012-02-20 17:52:46 -050096void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
97{
98 if (uptodate) {
99 set_buffer_uptodate(bh);
100 set_bitmap_uptodate(bh);
101 }
102 unlock_buffer(bh);
103 put_bh(bh);
104}
105
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106/*
107 * Read the inode allocation bitmap for a given block_group, reading
108 * into the specified slot in the superblock's bitmap cache.
109 *
110 * Return buffer_head of bitmap on success or NULL.
111 */
112static struct buffer_head *
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400113ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700114{
Mingming Cao617ba132006-10-11 01:20:53 -0700115 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700116 struct buffer_head *bh = NULL;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400117 ext4_fsblk_t bitmap_blk;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400118 struct ext4_group_info *grp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700119
Mingming Cao617ba132006-10-11 01:20:53 -0700120 desc = ext4_get_group_desc(sb, block_group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700121 if (!desc)
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400122 return NULL;
Lukas Czernerbfff6872010-10-27 21:30:05 -0400123
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400124 bitmap_blk = ext4_inode_bitmap(sb, desc);
125 bh = sb_getblk(sb, bitmap_blk);
126 if (unlikely(!bh)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500127 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500128 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400129 block_group, bitmap_blk);
130 return NULL;
131 }
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500132 if (bitmap_uptodate(bh))
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400133 goto verify;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400134
Frederic Bohec806e682008-10-10 08:09:18 -0400135 lock_buffer(bh);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500136 if (bitmap_uptodate(bh)) {
137 unlock_buffer(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400138 goto verify;
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500139 }
Lukas Czernerbfff6872010-10-27 21:30:05 -0400140
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400141 ext4_lock_group(sb, block_group);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400142 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
143 ext4_init_inode_bitmap(sb, bh, block_group, desc);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500144 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400145 set_buffer_uptodate(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400146 set_buffer_verified(bh);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400147 ext4_unlock_group(sb, block_group);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500148 unlock_buffer(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400149 return bh;
150 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400151 ext4_unlock_group(sb, block_group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400152
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500153 if (buffer_uptodate(bh)) {
154 /*
155 * if not uninit if bh is uptodate,
156 * bitmap is also uptodate
157 */
158 set_bitmap_uptodate(bh);
159 unlock_buffer(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400160 goto verify;
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500161 }
162 /*
Theodore Ts'o813e5722012-02-20 17:52:46 -0500163 * submit the buffer_head for reading
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500164 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400165 trace_ext4_load_inode_bitmap(sb, block_group);
Theodore Ts'o813e5722012-02-20 17:52:46 -0500166 bh->b_end_io = ext4_end_bitmap_read;
167 get_bh(bh);
Theodore Ts'o9f203502013-04-20 15:46:17 -0400168 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Theodore Ts'o813e5722012-02-20 17:52:46 -0500169 wait_on_buffer(bh);
170 if (!buffer_uptodate(bh)) {
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400171 put_bh(bh);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500172 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'o813e5722012-02-20 17:52:46 -0500173 "block_group = %u, inode_bitmap = %llu",
174 block_group, bitmap_blk);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400175 return NULL;
176 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400177
178verify:
179 ext4_lock_group(sb, block_group);
180 if (!buffer_verified(bh) &&
181 !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
182 EXT4_INODES_PER_GROUP(sb) / 8)) {
183 ext4_unlock_group(sb, block_group);
184 put_bh(bh);
185 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
186 "inode_bitmap = %llu", block_group, bitmap_blk);
Darrick J. Wong87a39382013-08-28 18:32:58 -0400187 grp = ext4_get_group_info(sb, block_group);
188 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400189 return NULL;
190 }
191 ext4_unlock_group(sb, block_group);
192 set_buffer_verified(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700193 return bh;
194}
195
196/*
197 * NOTE! When we get the inode, we're the only people
198 * that have access to it, and as such there are no
199 * race conditions we have to worry about. The inode
200 * is not on the hash-lists, and it cannot be reached
201 * through the filesystem because the directory entry
202 * has been deleted earlier.
203 *
204 * HOWEVER: we must make sure that we get no aliases,
205 * which means that we have to call "clear_inode()"
206 * _before_ we mark the inode not in use in the inode
207 * bitmaps. Otherwise a newly created file might use
208 * the same inode number (not actually the same pointer
209 * though), and then we'd have two inodes sharing the
210 * same inode number and space on the harddisk.
211 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400212void ext4_free_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700213{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400214 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700215 int is_directory;
216 unsigned long ino;
217 struct buffer_head *bitmap_bh = NULL;
218 struct buffer_head *bh2;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500219 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700220 unsigned long bit;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400221 struct ext4_group_desc *gdp;
222 struct ext4_super_block *es;
Mingming Cao617ba132006-10-11 01:20:53 -0700223 struct ext4_sb_info *sbi;
Eric Sandeen7ce9d5d2009-03-04 18:38:18 -0500224 int fatal = 0, err, count, cleared;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400225 struct ext4_group_info *grp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700226
Theodore Ts'o92b97812012-03-19 23:41:49 -0400227 if (!sb) {
228 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
229 "nonexistent device\n", __func__, __LINE__);
230 return;
231 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700232 if (atomic_read(&inode->i_count) > 1) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400233 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
234 __func__, __LINE__, inode->i_ino,
235 atomic_read(&inode->i_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700236 return;
237 }
238 if (inode->i_nlink) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400239 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
240 __func__, __LINE__, inode->i_ino, inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700241 return;
242 }
Mingming Cao617ba132006-10-11 01:20:53 -0700243 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700244
245 ino = inode->i_ino;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400246 ext4_debug("freeing inode %lu\n", ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400247 trace_ext4_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700248
249 /*
250 * Note: we must free any quota before locking the superblock,
251 * as writing the quota to disk may need the lock as well.
252 */
Christoph Hellwig871a2932010-03-03 09:05:07 -0500253 dquot_initialize(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700254 ext4_xattr_delete_inode(handle, inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500255 dquot_free_inode(inode);
Christoph Hellwig9f754752010-03-03 09:05:05 -0500256 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700257
258 is_directory = S_ISDIR(inode->i_mode);
259
260 /* Do this BEFORE marking the inode not in use or returning an error */
Al Viro0930fcc2010-06-07 13:16:22 -0400261 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700262
Mingming Cao617ba132006-10-11 01:20:53 -0700263 es = EXT4_SB(sb)->s_es;
264 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500265 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700266 goto error_return;
267 }
Mingming Cao617ba132006-10-11 01:20:53 -0700268 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
269 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400270 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Darrick J. Wong87a39382013-08-28 18:32:58 -0400271 /* Don't bother if the inode bitmap is corrupt. */
272 grp = ext4_get_group_info(sb, block_group);
273 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) || !bitmap_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700274 goto error_return;
275
276 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700277 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700278 if (fatal)
279 goto error_return;
280
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400281 fatal = -ESRCH;
282 gdp = ext4_get_group_desc(sb, block_group, &bh2);
283 if (gdp) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700284 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700285 fatal = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700286 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400287 ext4_lock_group(sb, block_group);
Akinobu Mita597d5082011-12-28 20:32:07 -0500288 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400289 if (fatal || !cleared) {
290 ext4_unlock_group(sb, block_group);
291 goto out;
292 }
293
294 count = ext4_free_inodes_count(sb, gdp) + 1;
295 ext4_free_inodes_set(sb, gdp, count);
296 if (is_directory) {
297 count = ext4_used_dirs_count(sb, gdp) - 1;
298 ext4_used_dirs_set(sb, gdp, count);
299 percpu_counter_dec(&sbi->s_dirs_counter);
300 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400301 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
302 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400303 ext4_group_desc_csum_set(sb, block_group, gdp);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400304 ext4_unlock_group(sb, block_group);
305
306 percpu_counter_inc(&sbi->s_freeinodes_counter);
307 if (sbi->s_log_groups_per_flex) {
308 ext4_group_t f = ext4_flex_group(sbi, block_group);
309
310 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
311 if (is_directory)
312 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
313 }
314 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
315 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
316out:
317 if (cleared) {
318 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
319 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
320 if (!fatal)
321 fatal = err;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400322 } else {
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400323 ext4_error(sb, "bit already cleared for inode %lu", ino);
Darrick J. Wong87a39382013-08-28 18:32:58 -0400324 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
325 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400326
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700327error_return:
328 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700329 ext4_std_error(sb, fatal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700330}
331
Theodore Ts'oa4912122009-03-12 12:18:34 -0400332struct orlov_stats {
Theodore Ts'o90ba9832013-03-11 23:39:59 -0400333 __u64 free_clusters;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400334 __u32 free_inodes;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400335 __u32 used_dirs;
336};
337
338/*
339 * Helper function for Orlov's allocator; returns critical information
340 * for a particular block group or flex_bg. If flex_size is 1, then g
341 * is a block group number; otherwise it is flex_bg number.
342 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400343static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
344 int flex_size, struct orlov_stats *stats)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400345{
346 struct ext4_group_desc *desc;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500347 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400348
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500349 if (flex_size > 1) {
350 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
Theodore Ts'o90ba9832013-03-11 23:39:59 -0400351 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500352 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
353 return;
354 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400355
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500356 desc = ext4_get_group_desc(sb, g, NULL);
357 if (desc) {
358 stats->free_inodes = ext4_free_inodes_count(sb, desc);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400359 stats->free_clusters = ext4_free_group_clusters(sb, desc);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500360 stats->used_dirs = ext4_used_dirs_count(sb, desc);
361 } else {
362 stats->free_inodes = 0;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400363 stats->free_clusters = 0;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500364 stats->used_dirs = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400365 }
366}
367
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700368/*
369 * Orlov's allocator for directories.
370 *
371 * We always try to spread first-level directories.
372 *
373 * If there are blockgroups with both free inodes and free blocks counts
374 * not worse than average we return one with smallest directory count.
375 * Otherwise we simply return a random group.
376 *
377 * For the rest rules look so:
378 *
379 * It's OK to put directory into a group unless
380 * it has too many directories already (max_dirs) or
381 * it has too few free inodes left (min_inodes) or
382 * it has too few free blocks left (min_blocks) or
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000383 * Parent's group is preferred, if it doesn't satisfy these
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700384 * conditions we search cyclically through the rest. If none
385 * of the groups look good we just look for a group with more
386 * free inodes than average (starting at parent's group).
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700387 */
388
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500389static int find_group_orlov(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400390 ext4_group_t *group, umode_t mode,
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400391 const struct qstr *qstr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700392{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500393 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Mingming Cao617ba132006-10-11 01:20:53 -0700394 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400395 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700396 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Theodore Ts'o14c83c92011-12-28 20:25:13 -0500397 unsigned int freei, avefreei, grp_free;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400398 ext4_fsblk_t freeb, avefreec;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700399 unsigned int ndirs;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400400 int max_dirs, min_inodes;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400401 ext4_grpblk_t min_clusters;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400402 ext4_group_t i, grp, g, ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700403 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400404 struct orlov_stats stats;
405 int flex_size = ext4_flex_bg_size(sbi);
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400406 struct dx_hash_info hinfo;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400407
Theodore Ts'o8df96752009-05-01 08:50:38 -0400408 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400409 if (flex_size > 1) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400410 ngroups = (real_ngroups + flex_size - 1) >>
Theodore Ts'oa4912122009-03-12 12:18:34 -0400411 sbi->s_log_groups_per_flex;
412 parent_group >>= sbi->s_log_groups_per_flex;
413 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700414
415 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
416 avefreei = freei / ngroups;
Theodore Ts'o57042652011-09-09 18:56:51 -0400417 freeb = EXT4_C2B(sbi,
418 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400419 avefreec = freeb;
420 do_div(avefreec, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700421 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
422
Theodore Ts'oa4912122009-03-12 12:18:34 -0400423 if (S_ISDIR(mode) &&
424 ((parent == sb->s_root->d_inode) ||
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400425 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700426 int best_ndir = inodes_per_group;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500427 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700428
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400429 if (qstr) {
430 hinfo.hash_version = DX_HASH_HALF_MD4;
431 hinfo.seed = sbi->s_hash_seed;
432 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
433 grp = hinfo.hash;
434 } else
435 get_random_bytes(&grp, sizeof(grp));
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500436 parent_group = (unsigned)grp % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700437 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400438 g = (parent_group + i) % ngroups;
439 get_orlov_stats(sb, g, flex_size, &stats);
440 if (!stats.free_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700441 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400442 if (stats.used_dirs >= best_ndir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700443 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400444 if (stats.free_inodes < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700445 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400446 if (stats.free_clusters < avefreec)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700447 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400448 grp = g;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500449 ret = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400450 best_ndir = stats.used_dirs;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700451 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400452 if (ret)
453 goto fallback;
454 found_flex_bg:
455 if (flex_size == 1) {
456 *group = grp;
457 return 0;
458 }
459
460 /*
461 * We pack inodes at the beginning of the flexgroup's
462 * inode tables. Block allocation decisions will do
463 * something similar, although regular files will
464 * start at 2nd block group of the flexgroup. See
465 * ext4_ext_find_goal() and ext4_find_near().
466 */
467 grp *= flex_size;
468 for (i = 0; i < flex_size; i++) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400469 if (grp+i >= real_ngroups)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400470 break;
471 desc = ext4_get_group_desc(sb, grp+i, NULL);
472 if (desc && ext4_free_inodes_count(sb, desc)) {
473 *group = grp+i;
474 return 0;
475 }
476 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700477 goto fallback;
478 }
479
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700480 max_dirs = ndirs / ngroups + inodes_per_group / 16;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400481 min_inodes = avefreei - inodes_per_group*flex_size / 4;
482 if (min_inodes < 1)
483 min_inodes = 1;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400484 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700485
Theodore Ts'oa4912122009-03-12 12:18:34 -0400486 /*
487 * Start looking in the flex group where we last allocated an
488 * inode for this parent directory
489 */
490 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
491 parent_group = EXT4_I(parent)->i_last_alloc_group;
492 if (flex_size > 1)
493 parent_group >>= sbi->s_log_groups_per_flex;
494 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700495
496 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400497 grp = (parent_group + i) % ngroups;
498 get_orlov_stats(sb, grp, flex_size, &stats);
499 if (stats.used_dirs >= max_dirs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700500 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400501 if (stats.free_inodes < min_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700502 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400503 if (stats.free_clusters < min_clusters)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700504 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400505 goto found_flex_bg;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700506 }
507
508fallback:
Theodore Ts'o8df96752009-05-01 08:50:38 -0400509 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400510 avefreei = freei / ngroups;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400511fallback_retry:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400512 parent_group = EXT4_I(parent)->i_block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700513 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400514 grp = (parent_group + i) % ngroups;
515 desc = ext4_get_group_desc(sb, grp, NULL);
Dan Carpenterbb3d1322012-05-28 14:16:57 -0400516 if (desc) {
517 grp_free = ext4_free_inodes_count(sb, desc);
518 if (grp_free && grp_free >= avefreei) {
519 *group = grp;
520 return 0;
521 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400522 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700523 }
524
525 if (avefreei) {
526 /*
527 * The free-inodes counter is approximate, and for really small
528 * filesystems the above test can fail to find any blockgroups
529 */
530 avefreei = 0;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400531 goto fallback_retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700532 }
533
534 return -1;
535}
536
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500537static int find_group_other(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400538 ext4_group_t *group, umode_t mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700539{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500540 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400541 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700542 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400543 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
544
545 /*
546 * Try to place the inode is the same flex group as its
547 * parent. If we can't find space, use the Orlov algorithm to
548 * find another flex group, and store that information in the
549 * parent directory's inode information so that use that flex
550 * group for future allocations.
551 */
552 if (flex_size > 1) {
553 int retry = 0;
554
555 try_again:
556 parent_group &= ~(flex_size-1);
557 last = parent_group + flex_size;
558 if (last > ngroups)
559 last = ngroups;
560 for (i = parent_group; i < last; i++) {
561 desc = ext4_get_group_desc(sb, i, NULL);
562 if (desc && ext4_free_inodes_count(sb, desc)) {
563 *group = i;
564 return 0;
565 }
566 }
567 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
568 retry = 1;
569 parent_group = EXT4_I(parent)->i_last_alloc_group;
570 goto try_again;
571 }
572 /*
573 * If this didn't work, use the Orlov search algorithm
574 * to find a new flex group; we pass in the mode to
575 * avoid the topdir algorithms.
576 */
577 *group = parent_group + flex_size;
578 if (*group > ngroups)
579 *group = 0;
Peter Huewe7dc57612011-02-21 21:01:42 -0500580 return find_group_orlov(sb, parent, group, mode, NULL);
Theodore Ts'oa4912122009-03-12 12:18:34 -0400581 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700582
583 /*
584 * Try to place the inode in its parent directory
585 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500586 *group = parent_group;
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) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400589 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500590 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700591
592 /*
593 * We're going to place this inode in a different blockgroup from its
594 * parent. We want to cause files in a common directory to all land in
595 * the same blockgroup. But we want files which are in a different
596 * directory which shares a blockgroup with our parent to land in a
597 * different blockgroup.
598 *
599 * So add our directory's i_ino into the starting point for the hash.
600 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500601 *group = (*group + parent->i_ino) % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700602
603 /*
604 * Use a quadratic hash to find a group with a free inode and some free
605 * blocks.
606 */
607 for (i = 1; i < ngroups; i <<= 1) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500608 *group += i;
609 if (*group >= ngroups)
610 *group -= ngroups;
611 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500612 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400613 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500614 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700615 }
616
617 /*
618 * That failed: try linear search for a free inode, even if that group
619 * has no free blocks.
620 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500621 *group = parent_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700622 for (i = 0; i < ngroups; i++) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500623 if (++*group >= ngroups)
624 *group = 0;
625 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500626 if (desc && ext4_free_inodes_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500627 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700628 }
629
630 return -1;
631}
632
633/*
Theodore Ts'o19883bd2013-08-16 22:06:55 -0400634 * In no journal mode, if an inode has recently been deleted, we want
635 * to avoid reusing it until we're reasonably sure the inode table
636 * block has been written back to disk. (Yes, these values are
637 * somewhat arbitrary...)
638 */
639#define RECENTCY_MIN 5
640#define RECENTCY_DIRTY 30
641
642static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
643{
644 struct ext4_group_desc *gdp;
645 struct ext4_inode *raw_inode;
646 struct buffer_head *bh;
647 unsigned long dtime, now;
648 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
649 int offset, ret = 0, recentcy = RECENTCY_MIN;
650
651 gdp = ext4_get_group_desc(sb, group, NULL);
652 if (unlikely(!gdp))
653 return 0;
654
655 bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
656 (ino / inodes_per_block));
657 if (unlikely(!bh) || !buffer_uptodate(bh))
658 /*
659 * If the block is not in the buffer cache, then it
660 * must have been written out.
661 */
662 goto out;
663
664 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
665 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
666 dtime = le32_to_cpu(raw_inode->i_dtime);
667 now = get_seconds();
668 if (buffer_dirty(bh))
669 recentcy += RECENTCY_DIRTY;
670
671 if (dtime && (dtime < now) && (now < dtime + recentcy))
672 ret = 1;
673out:
674 brelse(bh);
675 return ret;
676}
677
678/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700679 * There are two policies for allocating an inode. If the new inode is
680 * a directory, then a forward search is made for a block group with both
681 * free space and a low directory-to-inode ratio; if that fails, then of
682 * the groups with above-average free space, that group with the fewest
683 * directories already is chosen.
684 *
685 * For other inodes, search forward from the parent directory's block
686 * group to find a free inode.
687 */
Theodore Ts'o11395752013-02-09 16:27:09 -0500688struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
689 umode_t mode, const struct qstr *qstr,
690 __u32 goal, uid_t *owner, int handle_type,
691 unsigned int line_no, int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700692{
693 struct super_block *sb;
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500694 struct buffer_head *inode_bitmap_bh = NULL;
695 struct buffer_head *group_desc_bh;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400696 ext4_group_t ngroups, group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700697 unsigned long ino = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400698 struct inode *inode;
699 struct ext4_group_desc *gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700700 struct ext4_inode_info *ei;
701 struct ext4_sb_info *sbi;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500702 int ret2, err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700703 struct inode *ret;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500704 ext4_group_t i;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400705 ext4_group_t flex_group;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400706 struct ext4_group_info *grp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700707
708 /* Cannot create files in a deleted directory */
709 if (!dir || !dir->i_nlink)
710 return ERR_PTR(-EPERM);
711
712 sb = dir->i_sb;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400713 ngroups = ext4_get_groups_count(sb);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400714 trace_ext4_request_inode(dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700715 inode = new_inode(sb);
716 if (!inode)
717 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700718 ei = EXT4_I(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700719 sbi = EXT4_SB(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400720
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400721 /*
722 * Initalize owners and quota early so that we don't have to account
723 * for quota initialization worst case in standard inode creating
724 * transaction
725 */
726 if (owner) {
727 inode->i_mode = mode;
728 i_uid_write(inode, owner[0]);
729 i_gid_write(inode, owner[1]);
730 } else if (test_opt(sb, GRPID)) {
731 inode->i_mode = mode;
732 inode->i_uid = current_fsuid();
733 inode->i_gid = dir->i_gid;
734 } else
735 inode_init_owner(inode, dir, mode);
736 dquot_initialize(inode);
737
Andreas Dilger11013912009-06-13 11:45:35 -0400738 if (!goal)
739 goal = sbi->s_inode_goal;
740
Johann Lombardie6462862009-07-05 23:45:11 -0400741 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
Andreas Dilger11013912009-06-13 11:45:35 -0400742 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
743 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
744 ret2 = 0;
745 goto got_group;
746 }
747
Lukas Czerner4113c4c2011-10-08 14:34:47 -0400748 if (S_ISDIR(mode))
749 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
750 else
Theodore Ts'oa4912122009-03-12 12:18:34 -0400751 ret2 = find_group_other(sb, dir, &group, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700752
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400753got_group:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400754 EXT4_I(dir)->i_last_alloc_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700755 err = -ENOSPC;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500756 if (ret2 == -1)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700757 goto out;
758
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500759 /*
760 * Normally we will only go through one pass of this loop,
761 * unless we get unlucky and it turns out the group we selected
762 * had its last inode grabbed by someone else.
763 */
Andreas Dilger11013912009-06-13 11:45:35 -0400764 for (i = 0; i < ngroups; i++, ino = 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700765 err = -EIO;
766
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500767 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700768 if (!gdp)
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400769 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700770
Yongqiang Yangf2a09af2012-09-23 23:16:03 -0400771 /*
772 * Check free inodes count before loading bitmap.
773 */
774 if (ext4_free_inodes_count(sb, gdp) == 0) {
775 if (++group == ngroups)
776 group = 0;
777 continue;
778 }
779
Darrick J. Wong87a39382013-08-28 18:32:58 -0400780 grp = ext4_get_group_info(sb, group);
781 /* Skip groups with already-known suspicious inode tables */
782 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
783 if (++group == ngroups)
784 group = 0;
785 continue;
786 }
787
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500788 brelse(inode_bitmap_bh);
789 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
Darrick J. Wong87a39382013-08-28 18:32:58 -0400790 /* Skip groups with suspicious inode tables */
791 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || !inode_bitmap_bh) {
792 if (++group == ngroups)
793 group = 0;
794 continue;
795 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700796
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700797repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700798 ino = ext4_find_next_zero_bit((unsigned long *)
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500799 inode_bitmap_bh->b_data,
800 EXT4_INODES_PER_GROUP(sb), ino);
Theodore Ts'oa34eb502013-07-26 15:15:46 -0400801 if (ino >= EXT4_INODES_PER_GROUP(sb))
802 goto next_group;
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500803 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
804 ext4_error(sb, "reserved inode found cleared - "
805 "inode=%lu", ino + 1);
806 continue;
807 }
Theodore Ts'o19883bd2013-08-16 22:06:55 -0400808 if ((EXT4_SB(sb)->s_journal == NULL) &&
809 recently_deleted(sb, group, ino)) {
810 ino++;
811 goto next_inode;
812 }
Theodore Ts'o11395752013-02-09 16:27:09 -0500813 if (!handle) {
814 BUG_ON(nblocks <= 0);
815 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
Jan Kara5fe2fe82013-06-04 12:37:50 -0400816 handle_type, nblocks,
817 0);
Theodore Ts'o11395752013-02-09 16:27:09 -0500818 if (IS_ERR(handle)) {
819 err = PTR_ERR(handle);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400820 ext4_std_error(sb, err);
821 goto out;
Theodore Ts'o11395752013-02-09 16:27:09 -0500822 }
823 }
Eric Sandeenffb53872012-10-28 22:24:57 -0400824 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
825 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400826 if (err) {
827 ext4_std_error(sb, err);
828 goto out;
829 }
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500830 ext4_lock_group(sb, group);
831 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
832 ext4_unlock_group(sb, group);
833 ino++; /* the inode bitmap is zero-based */
834 if (!ret2)
835 goto got; /* we grabbed the inode! */
Theodore Ts'o19883bd2013-08-16 22:06:55 -0400836next_inode:
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500837 if (ino < EXT4_INODES_PER_GROUP(sb))
838 goto repeat_in_this_group;
Theodore Ts'oa34eb502013-07-26 15:15:46 -0400839next_group:
840 if (++group == ngroups)
841 group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700842 }
843 err = -ENOSPC;
844 goto out;
845
846got:
Eric Sandeenffb53872012-10-28 22:24:57 -0400847 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
848 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400849 if (err) {
850 ext4_std_error(sb, err);
851 goto out;
852 }
Eric Sandeenffb53872012-10-28 22:24:57 -0400853
Andreas Dilger717d50e2007-10-16 18:38:25 -0400854 /* We may have to initialize the block bitmap if it isn't already */
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400855 if (ext4_has_group_desc_csum(sb) &&
Andreas Dilger717d50e2007-10-16 18:38:25 -0400856 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500857 struct buffer_head *block_bitmap_bh;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400858
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500859 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
860 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
861 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400862 if (err) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500863 brelse(block_bitmap_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400864 ext4_std_error(sb, err);
865 goto out;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400866 }
867
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400868 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
869 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400870
Andreas Dilger717d50e2007-10-16 18:38:25 -0400871 /* recheck and clear flag under lock if we still need to */
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400872 ext4_lock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400873 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500874 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400875 ext4_free_group_clusters_set(sb, gdp,
Theodore Ts'ocff1dfd72011-09-09 19:12:51 -0400876 ext4_free_clusters_after_init(sb, group, gdp));
Darrick J. Wongfa77dcf2012-04-29 18:35:10 -0400877 ext4_block_bitmap_csum_set(sb, group, gdp,
Tao Ma79f1ba42012-10-22 00:34:32 -0400878 block_bitmap_bh);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400879 ext4_group_desc_csum_set(sb, group, gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400880 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400881 ext4_unlock_group(sb, group);
Theodore Ts'oaeb1e5d62012-11-29 21:21:22 -0500882 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400883
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400884 if (err) {
885 ext4_std_error(sb, err);
886 goto out;
887 }
Andreas Dilger717d50e2007-10-16 18:38:25 -0400888 }
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500889
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500890 BUFFER_TRACE(group_desc_bh, "get_write_access");
891 err = ext4_journal_get_write_access(handle, group_desc_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400892 if (err) {
893 ext4_std_error(sb, err);
894 goto out;
895 }
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500896
897 /* Update the relevant bg descriptor fields */
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400898 if (ext4_has_group_desc_csum(sb)) {
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500899 int free;
900 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
901
902 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
903 ext4_lock_group(sb, group); /* while we modify the bg desc */
904 free = EXT4_INODES_PER_GROUP(sb) -
905 ext4_itable_unused_count(sb, gdp);
906 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
907 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
908 free = 0;
909 }
910 /*
911 * Check the relative inode number against the last used
912 * relative inode number in this group. if it is greater
913 * we need to update the bg_itable_unused count
914 */
915 if (ino > free)
916 ext4_itable_unused_set(sb, gdp,
917 (EXT4_INODES_PER_GROUP(sb) - ino));
918 up_read(&grp->alloc_sem);
Tao Ma6f2e9f02012-05-28 18:20:59 -0400919 } else {
920 ext4_lock_group(sb, group);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500921 }
Tao Ma6f2e9f02012-05-28 18:20:59 -0400922
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500923 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
924 if (S_ISDIR(mode)) {
925 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
926 if (sbi->s_log_groups_per_flex) {
927 ext4_group_t f = ext4_flex_group(sbi, group);
928
929 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
930 }
931 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400932 if (ext4_has_group_desc_csum(sb)) {
933 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
934 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400935 ext4_group_desc_csum_set(sb, group, gdp);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500936 }
Tao Ma6f2e9f02012-05-28 18:20:59 -0400937 ext4_unlock_group(sb, group);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500938
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500939 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
940 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400941 if (err) {
942 ext4_std_error(sb, err);
943 goto out;
944 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700945
946 percpu_counter_dec(&sbi->s_freeinodes_counter);
947 if (S_ISDIR(mode))
948 percpu_counter_inc(&sbi->s_dirs_counter);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700949
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400950 if (sbi->s_log_groups_per_flex) {
951 flex_group = ext4_flex_group(sbi, group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500952 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400953 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700954
Andreas Dilger717d50e2007-10-16 18:38:25 -0400955 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700956 /* This is the optimal IO size (for stat), not the fs block size */
957 inode->i_blocks = 0;
Kalpak Shahef7f3832007-07-18 09:15:20 -0400958 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
959 ext4_current_time(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700960
961 memset(ei->i_data, 0, sizeof(ei->i_data));
962 ei->i_dir_start_lookup = 0;
963 ei->i_disksize = 0;
964
Eryu Guan4af83502011-10-31 18:21:29 -0400965 /* Don't inherit extent flag from directory, amongst others. */
Duane Griffin2dc6b0d2009-02-15 18:09:20 -0500966 ei->i_flags =
967 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700968 ei->i_file_acl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700969 ei->i_dtime = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700970 ei->i_block_group = group;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400971 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700972
Mingming Cao617ba132006-10-11 01:20:53 -0700973 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700974 if (IS_DIRSYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -0500975 ext4_handle_sync(handle);
Al Viro6b38e842008-12-30 02:03:31 -0500976 if (insert_inode_locked(inode) < 0) {
Jan Karaacd6ad82011-12-18 17:37:02 -0500977 /*
978 * Likely a bitmap corruption causing inode to be allocated
979 * twice.
980 */
981 err = -EIO;
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400982 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
983 inode->i_ino);
984 goto out;
Al Viro6b38e842008-12-30 02:03:31 -0500985 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700986 spin_lock(&sbi->s_next_gen_lock);
987 inode->i_generation = sbi->s_next_generation++;
988 spin_unlock(&sbi->s_next_gen_lock);
989
Darrick J. Wong814525f2012-04-29 18:31:10 -0400990 /* Precompute checksum seed for inode metadata */
991 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
992 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
993 __u32 csum;
Darrick J. Wong814525f2012-04-29 18:31:10 -0400994 __le32 inum = cpu_to_le32(inode->i_ino);
995 __le32 gen = cpu_to_le32(inode->i_generation);
996 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
997 sizeof(inum));
998 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
999 sizeof(gen));
1000 }
1001
Theodore Ts'o353eb832011-01-10 12:18:25 -05001002 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001003 ext4_set_inode_state(inode, EXT4_STATE_NEW);
Kalpak Shahef7f3832007-07-18 09:15:20 -04001004
1005 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001006
Tao Maf08225d2012-12-10 14:06:03 -05001007 ei->i_inline_off = 0;
1008 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA))
1009 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1010
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001011 ret = inode;
Christoph Hellwig63936dd2010-03-03 09:05:01 -05001012 err = dquot_alloc_inode(inode);
1013 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001014 goto fail_drop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001015
Mingming Cao617ba132006-10-11 01:20:53 -07001016 err = ext4_init_acl(handle, inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001017 if (err)
1018 goto fail_free_drop;
1019
Eric Paris2a7dba32011-02-01 11:05:39 -05001020 err = ext4_init_security(handle, inode, dir, qstr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001021 if (err)
1022 goto fail_free_drop;
1023
Theodore Ts'o83982b62009-01-06 14:53:16 -05001024 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Eric Sandeene4079a12008-07-11 19:27:31 -04001025 /* set extent flag only for directory, file and normal symlink*/
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04001026 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001027 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001028 ext4_ext_tree_init(handle, inode);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001029 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001030 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001031
Theodore Ts'o688f8692011-03-16 17:16:31 -04001032 if (ext4_handle_valid(handle)) {
1033 ei->i_sync_tid = handle->h_transaction->t_tid;
1034 ei->i_datasync_tid = handle->h_transaction->t_tid;
1035 }
1036
Aneesh Kumar K.V8753e882008-04-29 22:00:36 -04001037 err = ext4_mark_inode_dirty(handle, inode);
1038 if (err) {
1039 ext4_std_error(sb, err);
1040 goto fail_free_drop;
1041 }
1042
Mingming Cao617ba132006-10-11 01:20:53 -07001043 ext4_debug("allocating inode %lu\n", inode->i_ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001044 trace_ext4_allocate_inode(inode, dir, mode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001045 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001046 return ret;
1047
1048fail_free_drop:
Christoph Hellwig63936dd2010-03-03 09:05:01 -05001049 dquot_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001050fail_drop:
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02001051 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001052 unlock_new_inode(inode);
Jan Karaeb9cc7e2013-04-19 13:38:14 -04001053out:
1054 dquot_drop(inode);
1055 inode->i_flags |= S_NOQUOTA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001056 iput(inode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001057 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001058 return ERR_PTR(err);
1059}
1060
1061/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -07001062struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001063{
Mingming Cao617ba132006-10-11 01:20:53 -07001064 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Avantika Mathurfd2d4292008-01-28 23:58:27 -05001065 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001066 int bit;
David Howells1d1fe1e2008-02-07 00:15:37 -08001067 struct buffer_head *bitmap_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001068 struct inode *inode = NULL;
David Howells1d1fe1e2008-02-07 00:15:37 -08001069 long err = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001070
1071 /* Error cases - e2fsck has already cleaned up for us */
1072 if (ino > max_ino) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001073 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001074 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001075 }
1076
Mingming Cao617ba132006-10-11 01:20:53 -07001077 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1078 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001079 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001080 if (!bitmap_bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001081 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001082 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001083 }
1084
1085 /* Having the inode bit set should be a 100% indicator that this
1086 * is a valid orphan (no e2fsck run on fs). Orphans also include
1087 * inodes that were being truncated, so we can't check i_nlink==0.
1088 */
David Howells1d1fe1e2008-02-07 00:15:37 -08001089 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1090 goto bad_orphan;
1091
1092 inode = ext4_iget(sb, ino);
1093 if (IS_ERR(inode))
1094 goto iget_failed;
1095
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001096 /*
1097 * If the orphans has i_nlinks > 0 then it should be able to be
1098 * truncated, otherwise it won't be removed from the orphan list
1099 * during processing and an infinite loop will result.
1100 */
1101 if (inode->i_nlink && !ext4_can_truncate(inode))
1102 goto bad_orphan;
1103
David Howells1d1fe1e2008-02-07 00:15:37 -08001104 if (NEXT_ORPHAN(inode) > max_ino)
1105 goto bad_orphan;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001106 brelse(bitmap_bh);
1107 return inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08001108
1109iget_failed:
1110 err = PTR_ERR(inode);
1111 inode = NULL;
1112bad_orphan:
Eric Sandeen12062dd2010-02-15 14:19:27 -05001113 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001114 printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001115 bit, (unsigned long long)bitmap_bh->b_blocknr,
1116 ext4_test_bit(bit, bitmap_bh->b_data));
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001117 printk(KERN_WARNING "inode=%p\n", inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08001118 if (inode) {
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001119 printk(KERN_WARNING "is_bad_inode(inode)=%d\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001120 is_bad_inode(inode));
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001121 printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001122 NEXT_ORPHAN(inode));
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001123 printk(KERN_WARNING "max_ino=%lu\n", max_ino);
1124 printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink);
David Howells1d1fe1e2008-02-07 00:15:37 -08001125 /* Avoid freeing blocks if we got a bad deleted inode */
1126 if (inode->i_nlink == 0)
1127 inode->i_blocks = 0;
1128 iput(inode);
1129 }
1130 brelse(bitmap_bh);
1131error:
1132 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001133}
1134
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001135unsigned long ext4_count_free_inodes(struct super_block *sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001136{
1137 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001138 struct ext4_group_desc *gdp;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001139 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001140#ifdef EXT4FS_DEBUG
1141 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001142 unsigned long bitmap_count, x;
1143 struct buffer_head *bitmap_bh = NULL;
1144
Mingming Cao617ba132006-10-11 01:20:53 -07001145 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001146 desc_count = 0;
1147 bitmap_count = 0;
1148 gdp = NULL;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001149 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001150 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001151 if (!gdp)
1152 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001153 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001154 brelse(bitmap_bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001155 bitmap_bh = ext4_read_inode_bitmap(sb, i);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001156 if (!bitmap_bh)
1157 continue;
1158
Theodore Ts'of6fb99c2012-06-30 19:14:57 -04001159 x = ext4_count_free(bitmap_bh->b_data,
1160 EXT4_INODES_PER_GROUP(sb) / 8);
Eric Sandeenc549a952008-01-28 23:58:27 -05001161 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
Peng Tao785b4b32009-07-27 21:44:40 -04001162 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001163 bitmap_count += x;
1164 }
1165 brelse(bitmap_bh);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001166 printk(KERN_DEBUG "ext4_count_free_inodes: "
1167 "stored = %u, computed = %lu, %lu\n",
1168 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001169 return desc_count;
1170#else
1171 desc_count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001172 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001173 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001174 if (!gdp)
1175 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001176 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001177 cond_resched();
1178 }
1179 return desc_count;
1180#endif
1181}
1182
1183/* Called at mount-time, super-block is locked */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001184unsigned long ext4_count_dirs(struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001185{
1186 unsigned long count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001187 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001188
Theodore Ts'o8df96752009-05-01 08:50:38 -04001189 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001190 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001191 if (!gdp)
1192 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001193 count += ext4_used_dirs_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001194 }
1195 return count;
1196}
Lukas Czernerbfff6872010-10-27 21:30:05 -04001197
1198/*
1199 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1200 * inode table. Must be called without any spinlock held. The only place
1201 * where it is called from on active part of filesystem is ext4lazyinit
1202 * thread, so we do not need any special locks, however we have to prevent
1203 * inode allocation from the current group, so we take alloc_sem lock, to
Theodore Ts'o119c0d42012-02-06 20:12:03 -05001204 * block ext4_new_inode() until we are finished.
Lukas Czernerbfff6872010-10-27 21:30:05 -04001205 */
H Hartley Sweetene0cbee32011-10-18 10:57:51 -04001206int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
Lukas Czernerbfff6872010-10-27 21:30:05 -04001207 int barrier)
1208{
1209 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1210 struct ext4_sb_info *sbi = EXT4_SB(sb);
1211 struct ext4_group_desc *gdp = NULL;
1212 struct buffer_head *group_desc_bh;
1213 handle_t *handle;
1214 ext4_fsblk_t blk;
1215 int num, ret = 0, used_blks = 0;
Lukas Czernerbfff6872010-10-27 21:30:05 -04001216
1217 /* This should not happen, but just to be sure check this */
1218 if (sb->s_flags & MS_RDONLY) {
1219 ret = 1;
1220 goto out;
1221 }
1222
1223 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1224 if (!gdp)
1225 goto out;
1226
1227 /*
1228 * We do not need to lock this, because we are the only one
1229 * handling this flag.
1230 */
1231 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1232 goto out;
1233
Theodore Ts'o9924a922013-02-08 21:59:22 -05001234 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001235 if (IS_ERR(handle)) {
1236 ret = PTR_ERR(handle);
1237 goto out;
1238 }
1239
1240 down_write(&grp->alloc_sem);
1241 /*
1242 * If inode bitmap was already initialized there may be some
1243 * used inodes so we need to skip blocks with used inodes in
1244 * inode table.
1245 */
1246 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1247 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1248 ext4_itable_unused_count(sb, gdp)),
1249 sbi->s_inodes_per_block);
1250
Lukas Czerner857ac882010-10-27 21:30:05 -04001251 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
Theodore Ts'o1084f252012-03-19 23:13:43 -04001252 ext4_error(sb, "Something is wrong with group %u: "
1253 "used itable blocks: %d; "
1254 "itable unused count: %u",
Lukas Czerner857ac882010-10-27 21:30:05 -04001255 group, used_blks,
1256 ext4_itable_unused_count(sb, gdp));
1257 ret = 1;
Yongqiang Yang33853a02011-08-01 06:32:19 -04001258 goto err_out;
Lukas Czerner857ac882010-10-27 21:30:05 -04001259 }
1260
Lukas Czernerbfff6872010-10-27 21:30:05 -04001261 blk = ext4_inode_table(sb, gdp) + used_blks;
1262 num = sbi->s_itb_per_group - used_blks;
1263
1264 BUFFER_TRACE(group_desc_bh, "get_write_access");
1265 ret = ext4_journal_get_write_access(handle,
1266 group_desc_bh);
1267 if (ret)
1268 goto err_out;
1269
Lukas Czernerbfff6872010-10-27 21:30:05 -04001270 /*
1271 * Skip zeroout if the inode table is full. But we set the ZEROED
1272 * flag anyway, because obviously, when it is full it does not need
1273 * further zeroing.
1274 */
1275 if (unlikely(num == 0))
1276 goto skip_zeroout;
1277
1278 ext4_debug("going to zero out inode table in group %d\n",
1279 group);
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001280 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001281 if (ret < 0)
1282 goto err_out;
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001283 if (barrier)
1284 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001285
1286skip_zeroout:
1287 ext4_lock_group(sb, group);
1288 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001289 ext4_group_desc_csum_set(sb, group, gdp);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001290 ext4_unlock_group(sb, group);
1291
1292 BUFFER_TRACE(group_desc_bh,
1293 "call ext4_handle_dirty_metadata");
1294 ret = ext4_handle_dirty_metadata(handle, NULL,
1295 group_desc_bh);
1296
1297err_out:
1298 up_write(&grp->alloc_sem);
1299 ext4_journal_stop(handle);
1300out:
1301 return ret;
1302}