blob: ac644c31ca67472f3a16f91c6107e3d06772f66f [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;
Namjae Jeone43bb4e2014-06-26 10:11:53 -040074 struct ext4_sb_info *sbi = EXT4_SB(sb);
Andreas Dilger717d50e2007-10-16 18:38:25 -040075 J_ASSERT_BH(bh, buffer_locked(bh));
76
77 /* If checksum is bad mark all blocks and inodes use to prevent
78 * allocation, essentially implementing a per-group read-only flag. */
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -040079 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -050080 ext4_error(sb, "Checksum bad for group %u", block_group);
Darrick J. Wongbdfb6ff2013-08-28 18:46:56 -040081 grp = ext4_get_group_info(sb, block_group);
Namjae Jeone43bb4e2014-06-26 10:11:53 -040082 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
83 percpu_counter_sub(&sbi->s_freeclusters_counter,
84 grp->bb_free);
Darrick J. Wongbdfb6ff2013-08-28 18:46:56 -040085 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
Namjae Jeone43bb4e2014-06-26 10:11:53 -040086 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
87 int count;
88 count = ext4_free_inodes_count(sb, gdp);
89 percpu_counter_sub(&sbi->s_freeinodes_counter,
90 count);
91 }
Darrick J. Wongbdfb6ff2013-08-28 18:46:56 -040092 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
Andreas Dilger717d50e2007-10-16 18:38:25 -040093 return 0;
94 }
95
96 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
Theodore Ts'o61d08672010-10-27 21:30:15 -040097 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
Andreas Dilger717d50e2007-10-16 18:38:25 -040098 bh->b_data);
Darrick J. Wong41a246d2012-04-29 18:33:10 -040099 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
100 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400101 ext4_group_desc_csum_set(sb, block_group, gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400102
103 return EXT4_INODES_PER_GROUP(sb);
104}
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700105
Theodore Ts'o813e5722012-02-20 17:52:46 -0500106void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
107{
108 if (uptodate) {
109 set_buffer_uptodate(bh);
110 set_bitmap_uptodate(bh);
111 }
112 unlock_buffer(bh);
113 put_bh(bh);
114}
115
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700116/*
117 * Read the inode allocation bitmap for a given block_group, reading
118 * into the specified slot in the superblock's bitmap cache.
119 *
120 * Return buffer_head of bitmap on success or NULL.
121 */
122static struct buffer_head *
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400123ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700124{
Mingming Cao617ba132006-10-11 01:20:53 -0700125 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700126 struct buffer_head *bh = NULL;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400127 ext4_fsblk_t bitmap_blk;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400128 struct ext4_group_info *grp;
Namjae Jeone43bb4e2014-06-26 10:11:53 -0400129 struct ext4_sb_info *sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700130
Mingming Cao617ba132006-10-11 01:20:53 -0700131 desc = ext4_get_group_desc(sb, block_group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700132 if (!desc)
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400133 return NULL;
Lukas Czernerbfff6872010-10-27 21:30:05 -0400134
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400135 bitmap_blk = ext4_inode_bitmap(sb, desc);
136 bh = sb_getblk(sb, bitmap_blk);
137 if (unlikely(!bh)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500138 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500139 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400140 block_group, bitmap_blk);
141 return NULL;
142 }
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500143 if (bitmap_uptodate(bh))
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400144 goto verify;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400145
Frederic Bohec806e682008-10-10 08:09:18 -0400146 lock_buffer(bh);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500147 if (bitmap_uptodate(bh)) {
148 unlock_buffer(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400149 goto verify;
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500150 }
Lukas Czernerbfff6872010-10-27 21:30:05 -0400151
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400152 ext4_lock_group(sb, block_group);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400153 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
154 ext4_init_inode_bitmap(sb, bh, block_group, desc);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500155 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400156 set_buffer_uptodate(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400157 set_buffer_verified(bh);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400158 ext4_unlock_group(sb, block_group);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500159 unlock_buffer(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400160 return bh;
161 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400162 ext4_unlock_group(sb, block_group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400163
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500164 if (buffer_uptodate(bh)) {
165 /*
166 * if not uninit if bh is uptodate,
167 * bitmap is also uptodate
168 */
169 set_bitmap_uptodate(bh);
170 unlock_buffer(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400171 goto verify;
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500172 }
173 /*
Theodore Ts'o813e5722012-02-20 17:52:46 -0500174 * submit the buffer_head for reading
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500175 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400176 trace_ext4_load_inode_bitmap(sb, block_group);
Theodore Ts'o813e5722012-02-20 17:52:46 -0500177 bh->b_end_io = ext4_end_bitmap_read;
178 get_bh(bh);
Theodore Ts'o9f203502013-04-20 15:46:17 -0400179 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Theodore Ts'o813e5722012-02-20 17:52:46 -0500180 wait_on_buffer(bh);
181 if (!buffer_uptodate(bh)) {
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400182 put_bh(bh);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500183 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'o813e5722012-02-20 17:52:46 -0500184 "block_group = %u, inode_bitmap = %llu",
185 block_group, bitmap_blk);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400186 return NULL;
187 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400188
189verify:
190 ext4_lock_group(sb, block_group);
191 if (!buffer_verified(bh) &&
192 !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
193 EXT4_INODES_PER_GROUP(sb) / 8)) {
194 ext4_unlock_group(sb, block_group);
195 put_bh(bh);
196 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
197 "inode_bitmap = %llu", block_group, bitmap_blk);
Darrick J. Wong87a39382013-08-28 18:32:58 -0400198 grp = ext4_get_group_info(sb, block_group);
Namjae Jeone43bb4e2014-06-26 10:11:53 -0400199 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
200 int count;
201 count = ext4_free_inodes_count(sb, desc);
202 percpu_counter_sub(&sbi->s_freeinodes_counter,
203 count);
204 }
Darrick J. Wong87a39382013-08-28 18:32:58 -0400205 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400206 return NULL;
207 }
208 ext4_unlock_group(sb, block_group);
209 set_buffer_verified(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700210 return bh;
211}
212
213/*
214 * NOTE! When we get the inode, we're the only people
215 * that have access to it, and as such there are no
216 * race conditions we have to worry about. The inode
217 * is not on the hash-lists, and it cannot be reached
218 * through the filesystem because the directory entry
219 * has been deleted earlier.
220 *
221 * HOWEVER: we must make sure that we get no aliases,
222 * which means that we have to call "clear_inode()"
223 * _before_ we mark the inode not in use in the inode
224 * bitmaps. Otherwise a newly created file might use
225 * the same inode number (not actually the same pointer
226 * though), and then we'd have two inodes sharing the
227 * same inode number and space on the harddisk.
228 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400229void ext4_free_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700230{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400231 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700232 int is_directory;
233 unsigned long ino;
234 struct buffer_head *bitmap_bh = NULL;
235 struct buffer_head *bh2;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500236 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700237 unsigned long bit;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400238 struct ext4_group_desc *gdp;
239 struct ext4_super_block *es;
Mingming Cao617ba132006-10-11 01:20:53 -0700240 struct ext4_sb_info *sbi;
Eric Sandeen7ce9d5d2009-03-04 18:38:18 -0500241 int fatal = 0, err, count, cleared;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400242 struct ext4_group_info *grp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700243
Theodore Ts'o92b97812012-03-19 23:41:49 -0400244 if (!sb) {
245 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
246 "nonexistent device\n", __func__, __LINE__);
247 return;
248 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700249 if (atomic_read(&inode->i_count) > 1) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400250 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
251 __func__, __LINE__, inode->i_ino,
252 atomic_read(&inode->i_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700253 return;
254 }
255 if (inode->i_nlink) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400256 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
257 __func__, __LINE__, inode->i_ino, inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700258 return;
259 }
Mingming Cao617ba132006-10-11 01:20:53 -0700260 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700261
262 ino = inode->i_ino;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400263 ext4_debug("freeing inode %lu\n", ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400264 trace_ext4_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700265
266 /*
267 * Note: we must free any quota before locking the superblock,
268 * as writing the quota to disk may need the lock as well.
269 */
Christoph Hellwig871a2932010-03-03 09:05:07 -0500270 dquot_initialize(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700271 ext4_xattr_delete_inode(handle, inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500272 dquot_free_inode(inode);
Christoph Hellwig9f754752010-03-03 09:05:05 -0500273 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700274
275 is_directory = S_ISDIR(inode->i_mode);
276
277 /* Do this BEFORE marking the inode not in use or returning an error */
Al Viro0930fcc2010-06-07 13:16:22 -0400278 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700279
Mingming Cao617ba132006-10-11 01:20:53 -0700280 es = EXT4_SB(sb)->s_es;
281 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500282 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700283 goto error_return;
284 }
Mingming Cao617ba132006-10-11 01:20:53 -0700285 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
286 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400287 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Darrick J. Wong87a39382013-08-28 18:32:58 -0400288 /* Don't bother if the inode bitmap is corrupt. */
289 grp = ext4_get_group_info(sb, block_group);
290 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) || !bitmap_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291 goto error_return;
292
293 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700294 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700295 if (fatal)
296 goto error_return;
297
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400298 fatal = -ESRCH;
299 gdp = ext4_get_group_desc(sb, block_group, &bh2);
300 if (gdp) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700301 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700302 fatal = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700303 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400304 ext4_lock_group(sb, block_group);
Akinobu Mita597d5082011-12-28 20:32:07 -0500305 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400306 if (fatal || !cleared) {
307 ext4_unlock_group(sb, block_group);
308 goto out;
309 }
310
311 count = ext4_free_inodes_count(sb, gdp) + 1;
312 ext4_free_inodes_set(sb, gdp, count);
313 if (is_directory) {
314 count = ext4_used_dirs_count(sb, gdp) - 1;
315 ext4_used_dirs_set(sb, gdp, count);
316 percpu_counter_dec(&sbi->s_dirs_counter);
317 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400318 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
319 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400320 ext4_group_desc_csum_set(sb, block_group, gdp);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400321 ext4_unlock_group(sb, block_group);
322
323 percpu_counter_inc(&sbi->s_freeinodes_counter);
324 if (sbi->s_log_groups_per_flex) {
325 ext4_group_t f = ext4_flex_group(sbi, block_group);
326
327 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
328 if (is_directory)
329 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
330 }
331 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
332 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
333out:
334 if (cleared) {
335 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
336 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
337 if (!fatal)
338 fatal = err;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400339 } else {
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400340 ext4_error(sb, "bit already cleared for inode %lu", ino);
Namjae Jeonbf40c922014-07-12 16:11:42 -0400341 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
Namjae Jeone43bb4e2014-06-26 10:11:53 -0400342 int count;
343 count = ext4_free_inodes_count(sb, gdp);
344 percpu_counter_sub(&sbi->s_freeinodes_counter,
345 count);
346 }
Darrick J. Wong87a39382013-08-28 18:32:58 -0400347 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
348 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400349
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700350error_return:
351 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700352 ext4_std_error(sb, fatal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700353}
354
Theodore Ts'oa4912122009-03-12 12:18:34 -0400355struct orlov_stats {
Theodore Ts'o90ba9832013-03-11 23:39:59 -0400356 __u64 free_clusters;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400357 __u32 free_inodes;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400358 __u32 used_dirs;
359};
360
361/*
362 * Helper function for Orlov's allocator; returns critical information
363 * for a particular block group or flex_bg. If flex_size is 1, then g
364 * is a block group number; otherwise it is flex_bg number.
365 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400366static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
367 int flex_size, struct orlov_stats *stats)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400368{
369 struct ext4_group_desc *desc;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500370 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400371
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500372 if (flex_size > 1) {
373 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
Theodore Ts'o90ba9832013-03-11 23:39:59 -0400374 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500375 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
376 return;
377 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400378
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500379 desc = ext4_get_group_desc(sb, g, NULL);
380 if (desc) {
381 stats->free_inodes = ext4_free_inodes_count(sb, desc);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400382 stats->free_clusters = ext4_free_group_clusters(sb, desc);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500383 stats->used_dirs = ext4_used_dirs_count(sb, desc);
384 } else {
385 stats->free_inodes = 0;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400386 stats->free_clusters = 0;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500387 stats->used_dirs = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400388 }
389}
390
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700391/*
392 * Orlov's allocator for directories.
393 *
394 * We always try to spread first-level directories.
395 *
396 * If there are blockgroups with both free inodes and free blocks counts
397 * not worse than average we return one with smallest directory count.
398 * Otherwise we simply return a random group.
399 *
400 * For the rest rules look so:
401 *
402 * It's OK to put directory into a group unless
403 * it has too many directories already (max_dirs) or
404 * it has too few free inodes left (min_inodes) or
405 * it has too few free blocks left (min_blocks) or
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000406 * Parent's group is preferred, if it doesn't satisfy these
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700407 * conditions we search cyclically through the rest. If none
408 * of the groups look good we just look for a group with more
409 * free inodes than average (starting at parent's group).
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700410 */
411
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500412static int find_group_orlov(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400413 ext4_group_t *group, umode_t mode,
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400414 const struct qstr *qstr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700415{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500416 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Mingming Cao617ba132006-10-11 01:20:53 -0700417 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400418 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700419 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Theodore Ts'o14c83c92011-12-28 20:25:13 -0500420 unsigned int freei, avefreei, grp_free;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400421 ext4_fsblk_t freeb, avefreec;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700422 unsigned int ndirs;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400423 int max_dirs, min_inodes;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400424 ext4_grpblk_t min_clusters;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400425 ext4_group_t i, grp, g, ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700426 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400427 struct orlov_stats stats;
428 int flex_size = ext4_flex_bg_size(sbi);
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400429 struct dx_hash_info hinfo;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400430
Theodore Ts'o8df96752009-05-01 08:50:38 -0400431 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400432 if (flex_size > 1) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400433 ngroups = (real_ngroups + flex_size - 1) >>
Theodore Ts'oa4912122009-03-12 12:18:34 -0400434 sbi->s_log_groups_per_flex;
435 parent_group >>= sbi->s_log_groups_per_flex;
436 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700437
438 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
439 avefreei = freei / ngroups;
Theodore Ts'o57042652011-09-09 18:56:51 -0400440 freeb = EXT4_C2B(sbi,
441 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400442 avefreec = freeb;
443 do_div(avefreec, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700444 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
445
Theodore Ts'oa4912122009-03-12 12:18:34 -0400446 if (S_ISDIR(mode) &&
447 ((parent == sb->s_root->d_inode) ||
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400448 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700449 int best_ndir = inodes_per_group;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500450 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700451
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400452 if (qstr) {
453 hinfo.hash_version = DX_HASH_HALF_MD4;
454 hinfo.seed = sbi->s_hash_seed;
455 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
456 grp = hinfo.hash;
457 } else
Theodore Ts'odd1f7232013-11-08 00:14:53 -0500458 grp = prandom_u32();
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500459 parent_group = (unsigned)grp % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700460 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400461 g = (parent_group + i) % ngroups;
462 get_orlov_stats(sb, g, flex_size, &stats);
463 if (!stats.free_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700464 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400465 if (stats.used_dirs >= best_ndir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700466 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400467 if (stats.free_inodes < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700468 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400469 if (stats.free_clusters < avefreec)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700470 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400471 grp = g;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500472 ret = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400473 best_ndir = stats.used_dirs;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700474 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400475 if (ret)
476 goto fallback;
477 found_flex_bg:
478 if (flex_size == 1) {
479 *group = grp;
480 return 0;
481 }
482
483 /*
484 * We pack inodes at the beginning of the flexgroup's
485 * inode tables. Block allocation decisions will do
486 * something similar, although regular files will
487 * start at 2nd block group of the flexgroup. See
488 * ext4_ext_find_goal() and ext4_find_near().
489 */
490 grp *= flex_size;
491 for (i = 0; i < flex_size; i++) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400492 if (grp+i >= real_ngroups)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400493 break;
494 desc = ext4_get_group_desc(sb, grp+i, NULL);
495 if (desc && ext4_free_inodes_count(sb, desc)) {
496 *group = grp+i;
497 return 0;
498 }
499 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700500 goto fallback;
501 }
502
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700503 max_dirs = ndirs / ngroups + inodes_per_group / 16;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400504 min_inodes = avefreei - inodes_per_group*flex_size / 4;
505 if (min_inodes < 1)
506 min_inodes = 1;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400507 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700508
Theodore Ts'oa4912122009-03-12 12:18:34 -0400509 /*
510 * Start looking in the flex group where we last allocated an
511 * inode for this parent directory
512 */
513 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
514 parent_group = EXT4_I(parent)->i_last_alloc_group;
515 if (flex_size > 1)
516 parent_group >>= sbi->s_log_groups_per_flex;
517 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700518
519 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400520 grp = (parent_group + i) % ngroups;
521 get_orlov_stats(sb, grp, flex_size, &stats);
522 if (stats.used_dirs >= max_dirs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700523 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400524 if (stats.free_inodes < min_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700525 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400526 if (stats.free_clusters < min_clusters)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700527 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400528 goto found_flex_bg;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700529 }
530
531fallback:
Theodore Ts'o8df96752009-05-01 08:50:38 -0400532 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400533 avefreei = freei / ngroups;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400534fallback_retry:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400535 parent_group = EXT4_I(parent)->i_block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700536 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400537 grp = (parent_group + i) % ngroups;
538 desc = ext4_get_group_desc(sb, grp, NULL);
Dan Carpenterbb3d1322012-05-28 14:16:57 -0400539 if (desc) {
540 grp_free = ext4_free_inodes_count(sb, desc);
541 if (grp_free && grp_free >= avefreei) {
542 *group = grp;
543 return 0;
544 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400545 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700546 }
547
548 if (avefreei) {
549 /*
550 * The free-inodes counter is approximate, and for really small
551 * filesystems the above test can fail to find any blockgroups
552 */
553 avefreei = 0;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400554 goto fallback_retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700555 }
556
557 return -1;
558}
559
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500560static int find_group_other(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400561 ext4_group_t *group, umode_t mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700562{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500563 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400564 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700565 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400566 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
567
568 /*
569 * Try to place the inode is the same flex group as its
570 * parent. If we can't find space, use the Orlov algorithm to
571 * find another flex group, and store that information in the
572 * parent directory's inode information so that use that flex
573 * group for future allocations.
574 */
575 if (flex_size > 1) {
576 int retry = 0;
577
578 try_again:
579 parent_group &= ~(flex_size-1);
580 last = parent_group + flex_size;
581 if (last > ngroups)
582 last = ngroups;
583 for (i = parent_group; i < last; i++) {
584 desc = ext4_get_group_desc(sb, i, NULL);
585 if (desc && ext4_free_inodes_count(sb, desc)) {
586 *group = i;
587 return 0;
588 }
589 }
590 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
591 retry = 1;
592 parent_group = EXT4_I(parent)->i_last_alloc_group;
593 goto try_again;
594 }
595 /*
596 * If this didn't work, use the Orlov search algorithm
597 * to find a new flex group; we pass in the mode to
598 * avoid the topdir algorithms.
599 */
600 *group = parent_group + flex_size;
601 if (*group > ngroups)
602 *group = 0;
Peter Huewe7dc57612011-02-21 21:01:42 -0500603 return find_group_orlov(sb, parent, group, mode, NULL);
Theodore Ts'oa4912122009-03-12 12:18:34 -0400604 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700605
606 /*
607 * Try to place the inode in its parent directory
608 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500609 *group = parent_group;
610 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500611 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400612 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500613 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700614
615 /*
616 * We're going to place this inode in a different blockgroup from its
617 * parent. We want to cause files in a common directory to all land in
618 * the same blockgroup. But we want files which are in a different
619 * directory which shares a blockgroup with our parent to land in a
620 * different blockgroup.
621 *
622 * So add our directory's i_ino into the starting point for the hash.
623 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500624 *group = (*group + parent->i_ino) % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700625
626 /*
627 * Use a quadratic hash to find a group with a free inode and some free
628 * blocks.
629 */
630 for (i = 1; i < ngroups; i <<= 1) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500631 *group += i;
632 if (*group >= ngroups)
633 *group -= ngroups;
634 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500635 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400636 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500637 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700638 }
639
640 /*
641 * That failed: try linear search for a free inode, even if that group
642 * has no free blocks.
643 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500644 *group = parent_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700645 for (i = 0; i < ngroups; i++) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500646 if (++*group >= ngroups)
647 *group = 0;
648 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500649 if (desc && ext4_free_inodes_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500650 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700651 }
652
653 return -1;
654}
655
656/*
Theodore Ts'o19883bd2013-08-16 22:06:55 -0400657 * In no journal mode, if an inode has recently been deleted, we want
658 * to avoid reusing it until we're reasonably sure the inode table
659 * block has been written back to disk. (Yes, these values are
660 * somewhat arbitrary...)
661 */
662#define RECENTCY_MIN 5
663#define RECENTCY_DIRTY 30
664
665static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
666{
667 struct ext4_group_desc *gdp;
668 struct ext4_inode *raw_inode;
669 struct buffer_head *bh;
670 unsigned long dtime, now;
671 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
672 int offset, ret = 0, recentcy = RECENTCY_MIN;
673
674 gdp = ext4_get_group_desc(sb, group, NULL);
675 if (unlikely(!gdp))
676 return 0;
677
678 bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
679 (ino / inodes_per_block));
680 if (unlikely(!bh) || !buffer_uptodate(bh))
681 /*
682 * If the block is not in the buffer cache, then it
683 * must have been written out.
684 */
685 goto out;
686
687 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
688 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
689 dtime = le32_to_cpu(raw_inode->i_dtime);
690 now = get_seconds();
691 if (buffer_dirty(bh))
692 recentcy += RECENTCY_DIRTY;
693
694 if (dtime && (dtime < now) && (now < dtime + recentcy))
695 ret = 1;
696out:
697 brelse(bh);
698 return ret;
699}
700
701/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700702 * There are two policies for allocating an inode. If the new inode is
703 * a directory, then a forward search is made for a block group with both
704 * free space and a low directory-to-inode ratio; if that fails, then of
705 * the groups with above-average free space, that group with the fewest
706 * directories already is chosen.
707 *
708 * For other inodes, search forward from the parent directory's block
709 * group to find a free inode.
710 */
Theodore Ts'o11395752013-02-09 16:27:09 -0500711struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
712 umode_t mode, const struct qstr *qstr,
713 __u32 goal, uid_t *owner, int handle_type,
714 unsigned int line_no, int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700715{
716 struct super_block *sb;
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500717 struct buffer_head *inode_bitmap_bh = NULL;
718 struct buffer_head *group_desc_bh;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400719 ext4_group_t ngroups, group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700720 unsigned long ino = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400721 struct inode *inode;
722 struct ext4_group_desc *gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700723 struct ext4_inode_info *ei;
724 struct ext4_sb_info *sbi;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500725 int ret2, err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700726 struct inode *ret;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500727 ext4_group_t i;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400728 ext4_group_t flex_group;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400729 struct ext4_group_info *grp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700730
731 /* Cannot create files in a deleted directory */
732 if (!dir || !dir->i_nlink)
733 return ERR_PTR(-EPERM);
734
735 sb = dir->i_sb;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400736 ngroups = ext4_get_groups_count(sb);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400737 trace_ext4_request_inode(dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700738 inode = new_inode(sb);
739 if (!inode)
740 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700741 ei = EXT4_I(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700742 sbi = EXT4_SB(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400743
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400744 /*
745 * Initalize owners and quota early so that we don't have to account
746 * for quota initialization worst case in standard inode creating
747 * transaction
748 */
749 if (owner) {
750 inode->i_mode = mode;
751 i_uid_write(inode, owner[0]);
752 i_gid_write(inode, owner[1]);
753 } else if (test_opt(sb, GRPID)) {
754 inode->i_mode = mode;
755 inode->i_uid = current_fsuid();
756 inode->i_gid = dir->i_gid;
757 } else
758 inode_init_owner(inode, dir, mode);
759 dquot_initialize(inode);
760
Andreas Dilger11013912009-06-13 11:45:35 -0400761 if (!goal)
762 goal = sbi->s_inode_goal;
763
Johann Lombardie6462862009-07-05 23:45:11 -0400764 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
Andreas Dilger11013912009-06-13 11:45:35 -0400765 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
766 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
767 ret2 = 0;
768 goto got_group;
769 }
770
Lukas Czerner4113c4c2011-10-08 14:34:47 -0400771 if (S_ISDIR(mode))
772 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
773 else
Theodore Ts'oa4912122009-03-12 12:18:34 -0400774 ret2 = find_group_other(sb, dir, &group, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700775
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400776got_group:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400777 EXT4_I(dir)->i_last_alloc_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700778 err = -ENOSPC;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500779 if (ret2 == -1)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700780 goto out;
781
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500782 /*
783 * Normally we will only go through one pass of this loop,
784 * unless we get unlucky and it turns out the group we selected
785 * had its last inode grabbed by someone else.
786 */
Andreas Dilger11013912009-06-13 11:45:35 -0400787 for (i = 0; i < ngroups; i++, ino = 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700788 err = -EIO;
789
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500790 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700791 if (!gdp)
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400792 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700793
Yongqiang Yangf2a09af2012-09-23 23:16:03 -0400794 /*
795 * Check free inodes count before loading bitmap.
796 */
797 if (ext4_free_inodes_count(sb, gdp) == 0) {
798 if (++group == ngroups)
799 group = 0;
800 continue;
801 }
802
Darrick J. Wong87a39382013-08-28 18:32:58 -0400803 grp = ext4_get_group_info(sb, group);
804 /* Skip groups with already-known suspicious inode tables */
805 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
806 if (++group == ngroups)
807 group = 0;
808 continue;
809 }
810
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500811 brelse(inode_bitmap_bh);
812 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
Darrick J. Wong87a39382013-08-28 18:32:58 -0400813 /* Skip groups with suspicious inode tables */
814 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || !inode_bitmap_bh) {
815 if (++group == ngroups)
816 group = 0;
817 continue;
818 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700819
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700820repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700821 ino = ext4_find_next_zero_bit((unsigned long *)
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500822 inode_bitmap_bh->b_data,
823 EXT4_INODES_PER_GROUP(sb), ino);
Theodore Ts'oa34eb502013-07-26 15:15:46 -0400824 if (ino >= EXT4_INODES_PER_GROUP(sb))
825 goto next_group;
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500826 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
827 ext4_error(sb, "reserved inode found cleared - "
828 "inode=%lu", ino + 1);
829 continue;
830 }
Theodore Ts'o19883bd2013-08-16 22:06:55 -0400831 if ((EXT4_SB(sb)->s_journal == NULL) &&
832 recently_deleted(sb, group, ino)) {
833 ino++;
834 goto next_inode;
835 }
Theodore Ts'o11395752013-02-09 16:27:09 -0500836 if (!handle) {
837 BUG_ON(nblocks <= 0);
838 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
Jan Kara5fe2fe82013-06-04 12:37:50 -0400839 handle_type, nblocks,
840 0);
Theodore Ts'o11395752013-02-09 16:27:09 -0500841 if (IS_ERR(handle)) {
842 err = PTR_ERR(handle);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400843 ext4_std_error(sb, err);
844 goto out;
Theodore Ts'o11395752013-02-09 16:27:09 -0500845 }
846 }
Eric Sandeenffb53872012-10-28 22:24:57 -0400847 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
848 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400849 if (err) {
850 ext4_std_error(sb, err);
851 goto out;
852 }
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500853 ext4_lock_group(sb, group);
854 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
855 ext4_unlock_group(sb, group);
856 ino++; /* the inode bitmap is zero-based */
857 if (!ret2)
858 goto got; /* we grabbed the inode! */
Theodore Ts'o19883bd2013-08-16 22:06:55 -0400859next_inode:
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500860 if (ino < EXT4_INODES_PER_GROUP(sb))
861 goto repeat_in_this_group;
Theodore Ts'oa34eb502013-07-26 15:15:46 -0400862next_group:
863 if (++group == ngroups)
864 group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700865 }
866 err = -ENOSPC;
867 goto out;
868
869got:
Eric Sandeenffb53872012-10-28 22:24:57 -0400870 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
871 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400872 if (err) {
873 ext4_std_error(sb, err);
874 goto out;
875 }
Eric Sandeenffb53872012-10-28 22:24:57 -0400876
Theodore Ts'o61c219f2014-07-05 16:28:35 -0400877 BUFFER_TRACE(group_desc_bh, "get_write_access");
878 err = ext4_journal_get_write_access(handle, group_desc_bh);
879 if (err) {
880 ext4_std_error(sb, err);
881 goto out;
882 }
883
Andreas Dilger717d50e2007-10-16 18:38:25 -0400884 /* We may have to initialize the block bitmap if it isn't already */
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400885 if (ext4_has_group_desc_csum(sb) &&
Andreas Dilger717d50e2007-10-16 18:38:25 -0400886 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500887 struct buffer_head *block_bitmap_bh;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400888
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500889 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
Jan Kara599a9b72014-10-30 10:53:16 -0400890 if (!block_bitmap_bh) {
891 err = -EIO;
892 goto out;
893 }
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500894 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
895 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400896 if (err) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500897 brelse(block_bitmap_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400898 ext4_std_error(sb, err);
899 goto out;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400900 }
901
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400902 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
903 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400904
Andreas Dilger717d50e2007-10-16 18:38:25 -0400905 /* recheck and clear flag under lock if we still need to */
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400906 ext4_lock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400907 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500908 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400909 ext4_free_group_clusters_set(sb, gdp,
Theodore Ts'ocff1dfd72011-09-09 19:12:51 -0400910 ext4_free_clusters_after_init(sb, group, gdp));
Darrick J. Wongfa77dcf2012-04-29 18:35:10 -0400911 ext4_block_bitmap_csum_set(sb, group, gdp,
Tao Ma79f1ba42012-10-22 00:34:32 -0400912 block_bitmap_bh);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400913 ext4_group_desc_csum_set(sb, group, gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400914 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400915 ext4_unlock_group(sb, group);
Theodore Ts'oaeb1e5d62012-11-29 21:21:22 -0500916 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400917
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400918 if (err) {
919 ext4_std_error(sb, err);
920 goto out;
921 }
Andreas Dilger717d50e2007-10-16 18:38:25 -0400922 }
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500923
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500924 /* Update the relevant bg descriptor fields */
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400925 if (ext4_has_group_desc_csum(sb)) {
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500926 int free;
927 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
928
929 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
930 ext4_lock_group(sb, group); /* while we modify the bg desc */
931 free = EXT4_INODES_PER_GROUP(sb) -
932 ext4_itable_unused_count(sb, gdp);
933 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
934 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
935 free = 0;
936 }
937 /*
938 * Check the relative inode number against the last used
939 * relative inode number in this group. if it is greater
940 * we need to update the bg_itable_unused count
941 */
942 if (ino > free)
943 ext4_itable_unused_set(sb, gdp,
944 (EXT4_INODES_PER_GROUP(sb) - ino));
945 up_read(&grp->alloc_sem);
Tao Ma6f2e9f02012-05-28 18:20:59 -0400946 } else {
947 ext4_lock_group(sb, group);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500948 }
Tao Ma6f2e9f02012-05-28 18:20:59 -0400949
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500950 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
951 if (S_ISDIR(mode)) {
952 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
953 if (sbi->s_log_groups_per_flex) {
954 ext4_group_t f = ext4_flex_group(sbi, group);
955
956 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
957 }
958 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400959 if (ext4_has_group_desc_csum(sb)) {
960 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
961 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400962 ext4_group_desc_csum_set(sb, group, gdp);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500963 }
Tao Ma6f2e9f02012-05-28 18:20:59 -0400964 ext4_unlock_group(sb, group);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500965
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500966 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
967 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400968 if (err) {
969 ext4_std_error(sb, err);
970 goto out;
971 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700972
973 percpu_counter_dec(&sbi->s_freeinodes_counter);
974 if (S_ISDIR(mode))
975 percpu_counter_inc(&sbi->s_dirs_counter);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700976
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400977 if (sbi->s_log_groups_per_flex) {
978 flex_group = ext4_flex_group(sbi, group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500979 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400980 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700981
Andreas Dilger717d50e2007-10-16 18:38:25 -0400982 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700983 /* This is the optimal IO size (for stat), not the fs block size */
984 inode->i_blocks = 0;
Kalpak Shahef7f3832007-07-18 09:15:20 -0400985 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
986 ext4_current_time(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700987
988 memset(ei->i_data, 0, sizeof(ei->i_data));
989 ei->i_dir_start_lookup = 0;
990 ei->i_disksize = 0;
991
Eryu Guan4af83502011-10-31 18:21:29 -0400992 /* Don't inherit extent flag from directory, amongst others. */
Duane Griffin2dc6b0d2009-02-15 18:09:20 -0500993 ei->i_flags =
994 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700995 ei->i_file_acl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700996 ei->i_dtime = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700997 ei->i_block_group = group;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400998 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700999
Mingming Cao617ba132006-10-11 01:20:53 -07001000 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001001 if (IS_DIRSYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05001002 ext4_handle_sync(handle);
Al Viro6b38e842008-12-30 02:03:31 -05001003 if (insert_inode_locked(inode) < 0) {
Jan Karaacd6ad82011-12-18 17:37:02 -05001004 /*
1005 * Likely a bitmap corruption causing inode to be allocated
1006 * twice.
1007 */
1008 err = -EIO;
Jan Karaeb9cc7e2013-04-19 13:38:14 -04001009 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1010 inode->i_ino);
1011 goto out;
Al Viro6b38e842008-12-30 02:03:31 -05001012 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001013 spin_lock(&sbi->s_next_gen_lock);
1014 inode->i_generation = sbi->s_next_generation++;
1015 spin_unlock(&sbi->s_next_gen_lock);
1016
Darrick J. Wong814525f2012-04-29 18:31:10 -04001017 /* Precompute checksum seed for inode metadata */
Dmitry Monakhov9aa5d322014-10-13 03:36:16 -04001018 if (ext4_has_metadata_csum(sb)) {
Darrick J. Wong814525f2012-04-29 18:31:10 -04001019 __u32 csum;
Darrick J. Wong814525f2012-04-29 18:31:10 -04001020 __le32 inum = cpu_to_le32(inode->i_ino);
1021 __le32 gen = cpu_to_le32(inode->i_generation);
1022 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1023 sizeof(inum));
1024 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1025 sizeof(gen));
1026 }
1027
Theodore Ts'o353eb832011-01-10 12:18:25 -05001028 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001029 ext4_set_inode_state(inode, EXT4_STATE_NEW);
Kalpak Shahef7f3832007-07-18 09:15:20 -04001030
1031 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001032
Tao Maf08225d2012-12-10 14:06:03 -05001033 ei->i_inline_off = 0;
1034 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA))
1035 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1036
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001037 ret = inode;
Christoph Hellwig63936dd2010-03-03 09:05:01 -05001038 err = dquot_alloc_inode(inode);
1039 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001040 goto fail_drop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001041
Mingming Cao617ba132006-10-11 01:20:53 -07001042 err = ext4_init_acl(handle, inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001043 if (err)
1044 goto fail_free_drop;
1045
Eric Paris2a7dba32011-02-01 11:05:39 -05001046 err = ext4_init_security(handle, inode, dir, qstr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001047 if (err)
1048 goto fail_free_drop;
1049
Theodore Ts'o83982b62009-01-06 14:53:16 -05001050 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Eric Sandeene4079a12008-07-11 19:27:31 -04001051 /* set extent flag only for directory, file and normal symlink*/
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04001052 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001053 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001054 ext4_ext_tree_init(handle, inode);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001055 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001056 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001057
Theodore Ts'o688f8692011-03-16 17:16:31 -04001058 if (ext4_handle_valid(handle)) {
1059 ei->i_sync_tid = handle->h_transaction->t_tid;
1060 ei->i_datasync_tid = handle->h_transaction->t_tid;
1061 }
1062
Aneesh Kumar K.V8753e882008-04-29 22:00:36 -04001063 err = ext4_mark_inode_dirty(handle, inode);
1064 if (err) {
1065 ext4_std_error(sb, err);
1066 goto fail_free_drop;
1067 }
1068
Mingming Cao617ba132006-10-11 01:20:53 -07001069 ext4_debug("allocating inode %lu\n", inode->i_ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001070 trace_ext4_allocate_inode(inode, dir, mode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001071 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001072 return ret;
1073
1074fail_free_drop:
Christoph Hellwig63936dd2010-03-03 09:05:01 -05001075 dquot_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001076fail_drop:
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02001077 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001078 unlock_new_inode(inode);
Jan Karaeb9cc7e2013-04-19 13:38:14 -04001079out:
1080 dquot_drop(inode);
1081 inode->i_flags |= S_NOQUOTA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001082 iput(inode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001083 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001084 return ERR_PTR(err);
1085}
1086
1087/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -07001088struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001089{
Mingming Cao617ba132006-10-11 01:20:53 -07001090 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Avantika Mathurfd2d4292008-01-28 23:58:27 -05001091 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001092 int bit;
David Howells1d1fe1e2008-02-07 00:15:37 -08001093 struct buffer_head *bitmap_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001094 struct inode *inode = NULL;
David Howells1d1fe1e2008-02-07 00:15:37 -08001095 long err = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001096
1097 /* Error cases - e2fsck has already cleaned up for us */
1098 if (ino > max_ino) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001099 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001100 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001101 }
1102
Mingming Cao617ba132006-10-11 01:20:53 -07001103 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1104 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001105 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001106 if (!bitmap_bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001107 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001108 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001109 }
1110
1111 /* Having the inode bit set should be a 100% indicator that this
1112 * is a valid orphan (no e2fsck run on fs). Orphans also include
1113 * inodes that were being truncated, so we can't check i_nlink==0.
1114 */
David Howells1d1fe1e2008-02-07 00:15:37 -08001115 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1116 goto bad_orphan;
1117
1118 inode = ext4_iget(sb, ino);
1119 if (IS_ERR(inode))
1120 goto iget_failed;
1121
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001122 /*
1123 * If the orphans has i_nlinks > 0 then it should be able to be
1124 * truncated, otherwise it won't be removed from the orphan list
1125 * during processing and an infinite loop will result.
1126 */
1127 if (inode->i_nlink && !ext4_can_truncate(inode))
1128 goto bad_orphan;
1129
David Howells1d1fe1e2008-02-07 00:15:37 -08001130 if (NEXT_ORPHAN(inode) > max_ino)
1131 goto bad_orphan;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001132 brelse(bitmap_bh);
1133 return inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08001134
1135iget_failed:
1136 err = PTR_ERR(inode);
1137 inode = NULL;
1138bad_orphan:
Eric Sandeen12062dd2010-02-15 14:19:27 -05001139 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001140 printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001141 bit, (unsigned long long)bitmap_bh->b_blocknr,
1142 ext4_test_bit(bit, bitmap_bh->b_data));
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001143 printk(KERN_WARNING "inode=%p\n", inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08001144 if (inode) {
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001145 printk(KERN_WARNING "is_bad_inode(inode)=%d\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001146 is_bad_inode(inode));
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001147 printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001148 NEXT_ORPHAN(inode));
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001149 printk(KERN_WARNING "max_ino=%lu\n", max_ino);
1150 printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink);
David Howells1d1fe1e2008-02-07 00:15:37 -08001151 /* Avoid freeing blocks if we got a bad deleted inode */
1152 if (inode->i_nlink == 0)
1153 inode->i_blocks = 0;
1154 iput(inode);
1155 }
1156 brelse(bitmap_bh);
1157error:
1158 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001159}
1160
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001161unsigned long ext4_count_free_inodes(struct super_block *sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001162{
1163 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001164 struct ext4_group_desc *gdp;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001165 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001166#ifdef EXT4FS_DEBUG
1167 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001168 unsigned long bitmap_count, x;
1169 struct buffer_head *bitmap_bh = NULL;
1170
Mingming Cao617ba132006-10-11 01:20:53 -07001171 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001172 desc_count = 0;
1173 bitmap_count = 0;
1174 gdp = NULL;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001175 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001176 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001177 if (!gdp)
1178 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001179 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001180 brelse(bitmap_bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001181 bitmap_bh = ext4_read_inode_bitmap(sb, i);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001182 if (!bitmap_bh)
1183 continue;
1184
Theodore Ts'of6fb99c2012-06-30 19:14:57 -04001185 x = ext4_count_free(bitmap_bh->b_data,
1186 EXT4_INODES_PER_GROUP(sb) / 8);
Eric Sandeenc549a952008-01-28 23:58:27 -05001187 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
Peng Tao785b4b32009-07-27 21:44:40 -04001188 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001189 bitmap_count += x;
1190 }
1191 brelse(bitmap_bh);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001192 printk(KERN_DEBUG "ext4_count_free_inodes: "
1193 "stored = %u, computed = %lu, %lu\n",
1194 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001195 return desc_count;
1196#else
1197 desc_count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001198 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001199 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001200 if (!gdp)
1201 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001202 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001203 cond_resched();
1204 }
1205 return desc_count;
1206#endif
1207}
1208
1209/* Called at mount-time, super-block is locked */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001210unsigned long ext4_count_dirs(struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001211{
1212 unsigned long count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001213 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001214
Theodore Ts'o8df96752009-05-01 08:50:38 -04001215 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001216 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001217 if (!gdp)
1218 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001219 count += ext4_used_dirs_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001220 }
1221 return count;
1222}
Lukas Czernerbfff6872010-10-27 21:30:05 -04001223
1224/*
1225 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1226 * inode table. Must be called without any spinlock held. The only place
1227 * where it is called from on active part of filesystem is ext4lazyinit
1228 * thread, so we do not need any special locks, however we have to prevent
1229 * inode allocation from the current group, so we take alloc_sem lock, to
Theodore Ts'o119c0d42012-02-06 20:12:03 -05001230 * block ext4_new_inode() until we are finished.
Lukas Czernerbfff6872010-10-27 21:30:05 -04001231 */
H Hartley Sweetene0cbee32011-10-18 10:57:51 -04001232int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
Lukas Czernerbfff6872010-10-27 21:30:05 -04001233 int barrier)
1234{
1235 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1236 struct ext4_sb_info *sbi = EXT4_SB(sb);
1237 struct ext4_group_desc *gdp = NULL;
1238 struct buffer_head *group_desc_bh;
1239 handle_t *handle;
1240 ext4_fsblk_t blk;
1241 int num, ret = 0, used_blks = 0;
Lukas Czernerbfff6872010-10-27 21:30:05 -04001242
1243 /* This should not happen, but just to be sure check this */
1244 if (sb->s_flags & MS_RDONLY) {
1245 ret = 1;
1246 goto out;
1247 }
1248
1249 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1250 if (!gdp)
1251 goto out;
1252
1253 /*
1254 * We do not need to lock this, because we are the only one
1255 * handling this flag.
1256 */
1257 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1258 goto out;
1259
Theodore Ts'o9924a922013-02-08 21:59:22 -05001260 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001261 if (IS_ERR(handle)) {
1262 ret = PTR_ERR(handle);
1263 goto out;
1264 }
1265
1266 down_write(&grp->alloc_sem);
1267 /*
1268 * If inode bitmap was already initialized there may be some
1269 * used inodes so we need to skip blocks with used inodes in
1270 * inode table.
1271 */
1272 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1273 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1274 ext4_itable_unused_count(sb, gdp)),
1275 sbi->s_inodes_per_block);
1276
Lukas Czerner857ac882010-10-27 21:30:05 -04001277 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
Theodore Ts'o1084f252012-03-19 23:13:43 -04001278 ext4_error(sb, "Something is wrong with group %u: "
1279 "used itable blocks: %d; "
1280 "itable unused count: %u",
Lukas Czerner857ac882010-10-27 21:30:05 -04001281 group, used_blks,
1282 ext4_itable_unused_count(sb, gdp));
1283 ret = 1;
Yongqiang Yang33853a02011-08-01 06:32:19 -04001284 goto err_out;
Lukas Czerner857ac882010-10-27 21:30:05 -04001285 }
1286
Lukas Czernerbfff6872010-10-27 21:30:05 -04001287 blk = ext4_inode_table(sb, gdp) + used_blks;
1288 num = sbi->s_itb_per_group - used_blks;
1289
1290 BUFFER_TRACE(group_desc_bh, "get_write_access");
1291 ret = ext4_journal_get_write_access(handle,
1292 group_desc_bh);
1293 if (ret)
1294 goto err_out;
1295
Lukas Czernerbfff6872010-10-27 21:30:05 -04001296 /*
1297 * Skip zeroout if the inode table is full. But we set the ZEROED
1298 * flag anyway, because obviously, when it is full it does not need
1299 * further zeroing.
1300 */
1301 if (unlikely(num == 0))
1302 goto skip_zeroout;
1303
1304 ext4_debug("going to zero out inode table in group %d\n",
1305 group);
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001306 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001307 if (ret < 0)
1308 goto err_out;
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001309 if (barrier)
1310 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001311
1312skip_zeroout:
1313 ext4_lock_group(sb, group);
1314 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001315 ext4_group_desc_csum_set(sb, group, gdp);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001316 ext4_unlock_group(sb, group);
1317
1318 BUFFER_TRACE(group_desc_bh,
1319 "call ext4_handle_dirty_metadata");
1320 ret = ext4_handle_dirty_metadata(handle, NULL,
1321 group_desc_bh);
1322
1323err_out:
1324 up_write(&grp->alloc_sem);
1325 ext4_journal_stop(handle);
1326out:
1327 return ret;
1328}