blob: 9c63f273b550497759103ad0f6b4aaafd8e2049d [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/ialloc.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * BSD ufs-inspired inode and directory allocation by
10 * Stephen Tweedie (sct@redhat.com), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
13 */
14
15#include <linux/time.h>
16#include <linux/fs.h>
Mingming Caodab291a2006-10-11 01:21:01 -070017#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070018#include <linux/stat.h>
19#include <linux/string.h>
20#include <linux/quotaops.h>
21#include <linux/buffer_head.h>
22#include <linux/random.h>
23#include <linux/bitops.h>
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070024#include <linux/blkdev.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070025#include <asm/byteorder.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040026
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040027#include "ext4.h"
28#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070029#include "xattr.h"
30#include "acl.h"
31
Theodore Ts'o9bffad12009-06-17 11:48:11 -040032#include <trace/events/ext4.h>
33
Dave Kleikampac27a0e2006-10-11 01:20:50 -070034/*
35 * ialloc.c contains the inodes allocation and deallocation routines
36 */
37
38/*
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
42 *
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
46 */
47
Andreas Dilger717d50e2007-10-16 18:38:25 -040048/*
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
52 */
Theodore Ts'o61d08672010-10-27 21:30:15 -040053void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
Andreas Dilger717d50e2007-10-16 18:38:25 -040054{
55 int i;
56
57 if (start_bit >= end_bit)
58 return;
59
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
61 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
62 ext4_set_bit(i, bitmap);
63 if (i < end_bit)
64 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
65}
66
67/* Initializes an uninitialized inode bitmap */
Theodore Ts'o1f109d52010-10-27 21:30:14 -040068static unsigned ext4_init_inode_bitmap(struct super_block *sb,
69 struct buffer_head *bh,
70 ext4_group_t block_group,
71 struct ext4_group_desc *gdp)
Andreas Dilger717d50e2007-10-16 18:38:25 -040072{
73 struct ext4_sb_info *sbi = EXT4_SB(sb);
74
75 J_ASSERT_BH(bh, buffer_locked(bh));
76
77 /* If checksum is bad mark all blocks and inodes use to prevent
78 * allocation, essentially implementing a per-group read-only flag. */
79 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -050080 ext4_error(sb, "Checksum bad for group %u", block_group);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -050081 ext4_free_blks_set(sb, gdp, 0);
82 ext4_free_inodes_set(sb, gdp, 0);
83 ext4_itable_unused_set(sb, gdp, 0);
Andreas Dilger717d50e2007-10-16 18:38:25 -040084 memset(bh->b_data, 0xff, sb->s_blocksize);
85 return 0;
86 }
87
88 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
Theodore Ts'o61d08672010-10-27 21:30:15 -040089 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
Andreas Dilger717d50e2007-10-16 18:38:25 -040090 bh->b_data);
91
92 return EXT4_INODES_PER_GROUP(sb);
93}
Dave Kleikampac27a0e2006-10-11 01:20:50 -070094
95/*
96 * Read the inode allocation bitmap for a given block_group, reading
97 * into the specified slot in the superblock's bitmap cache.
98 *
99 * Return buffer_head of bitmap on success or NULL.
100 */
101static struct buffer_head *
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400102ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700103{
Mingming Cao617ba132006-10-11 01:20:53 -0700104 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700105 struct buffer_head *bh = NULL;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400106 ext4_fsblk_t bitmap_blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700107
Mingming Cao617ba132006-10-11 01:20:53 -0700108 desc = ext4_get_group_desc(sb, block_group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700109 if (!desc)
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400110 return NULL;
Lukas Czernerbfff6872010-10-27 21:30:05 -0400111
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400112 bitmap_blk = ext4_inode_bitmap(sb, desc);
113 bh = sb_getblk(sb, bitmap_blk);
114 if (unlikely(!bh)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500115 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500116 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400117 block_group, bitmap_blk);
118 return NULL;
119 }
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500120 if (bitmap_uptodate(bh))
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400121 return bh;
122
Frederic Bohec806e682008-10-10 08:09:18 -0400123 lock_buffer(bh);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500124 if (bitmap_uptodate(bh)) {
125 unlock_buffer(bh);
126 return bh;
127 }
Lukas Czernerbfff6872010-10-27 21:30:05 -0400128
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400129 ext4_lock_group(sb, block_group);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400130 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
131 ext4_init_inode_bitmap(sb, bh, block_group, desc);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500132 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400133 set_buffer_uptodate(bh);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400134 ext4_unlock_group(sb, block_group);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500135 unlock_buffer(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400136 return bh;
137 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400138 ext4_unlock_group(sb, block_group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400139
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500140 if (buffer_uptodate(bh)) {
141 /*
142 * if not uninit if bh is uptodate,
143 * bitmap is also uptodate
144 */
145 set_bitmap_uptodate(bh);
146 unlock_buffer(bh);
147 return bh;
148 }
149 /*
150 * submit the buffer_head for read. We can
151 * safely mark the bitmap as uptodate now.
152 * We do it here so the bitmap uptodate bit
153 * get set with buffer lock held.
154 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400155 trace_ext4_load_inode_bitmap(sb, block_group);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500156 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400157 if (bh_submit_read(bh) < 0) {
158 put_bh(bh);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500159 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500160 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400161 block_group, bitmap_blk);
162 return NULL;
163 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700164 return bh;
165}
166
167/*
168 * NOTE! When we get the inode, we're the only people
169 * that have access to it, and as such there are no
170 * race conditions we have to worry about. The inode
171 * is not on the hash-lists, and it cannot be reached
172 * through the filesystem because the directory entry
173 * has been deleted earlier.
174 *
175 * HOWEVER: we must make sure that we get no aliases,
176 * which means that we have to call "clear_inode()"
177 * _before_ we mark the inode not in use in the inode
178 * bitmaps. Otherwise a newly created file might use
179 * the same inode number (not actually the same pointer
180 * though), and then we'd have two inodes sharing the
181 * same inode number and space on the harddisk.
182 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400183void ext4_free_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700184{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400185 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700186 int is_directory;
187 unsigned long ino;
188 struct buffer_head *bitmap_bh = NULL;
189 struct buffer_head *bh2;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500190 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700191 unsigned long bit;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400192 struct ext4_group_desc *gdp;
193 struct ext4_super_block *es;
Mingming Cao617ba132006-10-11 01:20:53 -0700194 struct ext4_sb_info *sbi;
Eric Sandeen7ce9d5d2009-03-04 18:38:18 -0500195 int fatal = 0, err, count, cleared;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700196
197 if (atomic_read(&inode->i_count) > 1) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400198 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
199 atomic_read(&inode->i_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700200 return;
201 }
202 if (inode->i_nlink) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400203 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
204 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700205 return;
206 }
207 if (!sb) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400208 printk(KERN_ERR "ext4_free_inode: inode on "
209 "nonexistent device\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700210 return;
211 }
Mingming Cao617ba132006-10-11 01:20:53 -0700212 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700213
214 ino = inode->i_ino;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400215 ext4_debug("freeing inode %lu\n", ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400216 trace_ext4_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700217
218 /*
219 * Note: we must free any quota before locking the superblock,
220 * as writing the quota to disk may need the lock as well.
221 */
Christoph Hellwig871a2932010-03-03 09:05:07 -0500222 dquot_initialize(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700223 ext4_xattr_delete_inode(handle, inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500224 dquot_free_inode(inode);
Christoph Hellwig9f754752010-03-03 09:05:05 -0500225 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700226
227 is_directory = S_ISDIR(inode->i_mode);
228
229 /* Do this BEFORE marking the inode not in use or returning an error */
Al Viro0930fcc2010-06-07 13:16:22 -0400230 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700231
Mingming Cao617ba132006-10-11 01:20:53 -0700232 es = EXT4_SB(sb)->s_es;
233 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500234 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700235 goto error_return;
236 }
Mingming Cao617ba132006-10-11 01:20:53 -0700237 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
238 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400239 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700240 if (!bitmap_bh)
241 goto error_return;
242
243 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700244 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700245 if (fatal)
246 goto error_return;
247
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400248 fatal = -ESRCH;
249 gdp = ext4_get_group_desc(sb, block_group, &bh2);
250 if (gdp) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700251 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700252 fatal = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700253 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400254 ext4_lock_group(sb, block_group);
255 cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
256 if (fatal || !cleared) {
257 ext4_unlock_group(sb, block_group);
258 goto out;
259 }
260
261 count = ext4_free_inodes_count(sb, gdp) + 1;
262 ext4_free_inodes_set(sb, gdp, count);
263 if (is_directory) {
264 count = ext4_used_dirs_count(sb, gdp) - 1;
265 ext4_used_dirs_set(sb, gdp, count);
266 percpu_counter_dec(&sbi->s_dirs_counter);
267 }
268 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
269 ext4_unlock_group(sb, block_group);
270
271 percpu_counter_inc(&sbi->s_freeinodes_counter);
272 if (sbi->s_log_groups_per_flex) {
273 ext4_group_t f = ext4_flex_group(sbi, block_group);
274
275 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
276 if (is_directory)
277 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
278 }
279 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
280 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
281out:
282 if (cleared) {
283 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
284 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
285 if (!fatal)
286 fatal = err;
Theodore Ts'oa0375152010-06-11 23:14:04 -0400287 ext4_mark_super_dirty(sb);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400288 } else
289 ext4_error(sb, "bit already cleared for inode %lu", ino);
290
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291error_return:
292 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700293 ext4_std_error(sb, fatal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700294}
295
296/*
297 * There are two policies for allocating an inode. If the new inode is
298 * a directory, then a forward search is made for a block group with both
299 * free space and a low directory-to-inode ratio; if that fails, then of
300 * the groups with above-average free space, that group with the fewest
301 * directories already is chosen.
302 *
303 * For other inodes, search forward from the parent directory\'s block
304 * group to find a free inode.
305 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500306static int find_group_dir(struct super_block *sb, struct inode *parent,
307 ext4_group_t *best_group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700308{
Theodore Ts'o8df96752009-05-01 08:50:38 -0400309 ext4_group_t ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700310 unsigned int freei, avefreei;
Mingming Cao617ba132006-10-11 01:20:53 -0700311 struct ext4_group_desc *desc, *best_desc = NULL;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500312 ext4_group_t group;
313 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700314
Mingming Cao617ba132006-10-11 01:20:53 -0700315 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700316 avefreei = freei / ngroups;
317
318 for (group = 0; group < ngroups; group++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400319 desc = ext4_get_group_desc(sb, group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500320 if (!desc || !ext4_free_inodes_count(sb, desc))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700321 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500322 if (ext4_free_inodes_count(sb, desc) < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700323 continue;
324 if (!best_desc ||
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500325 (ext4_free_blks_count(sb, desc) >
326 ext4_free_blks_count(sb, best_desc))) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500327 *best_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700328 best_desc = desc;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500329 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700330 }
331 }
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500332 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700333}
334
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400335#define free_block_ratio 10
336
337static int find_group_flex(struct super_block *sb, struct inode *parent,
338 ext4_group_t *best_group)
339{
340 struct ext4_sb_info *sbi = EXT4_SB(sb);
341 struct ext4_group_desc *desc;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400342 struct flex_groups *flex_group = sbi->s_flex_groups;
343 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
344 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400345 ext4_group_t ngroups = ext4_get_groups_count(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400346 int flex_size = ext4_flex_bg_size(sbi);
347 ext4_group_t best_flex = parent_fbg_group;
348 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
349 int flexbg_free_blocks;
350 int flex_freeb_ratio;
351 ext4_group_t n_fbg_groups;
352 ext4_group_t i;
353
Theodore Ts'o8df96752009-05-01 08:50:38 -0400354 n_fbg_groups = (ngroups + flex_size - 1) >>
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400355 sbi->s_log_groups_per_flex;
356
357find_close_to_parent:
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500358 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400359 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500360 if (atomic_read(&flex_group[best_flex].free_inodes) &&
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400361 flex_freeb_ratio > free_block_ratio)
362 goto found_flexbg;
363
364 if (best_flex && best_flex == parent_fbg_group) {
365 best_flex--;
366 goto find_close_to_parent;
367 }
368
369 for (i = 0; i < n_fbg_groups; i++) {
370 if (i == parent_fbg_group || i == parent_fbg_group - 1)
371 continue;
372
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500373 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400374 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
375
376 if (flex_freeb_ratio > free_block_ratio &&
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500377 (atomic_read(&flex_group[i].free_inodes))) {
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400378 best_flex = i;
379 goto found_flexbg;
380 }
381
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500382 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
383 ((atomic_read(&flex_group[i].free_blocks) >
384 atomic_read(&flex_group[best_flex].free_blocks)) &&
385 atomic_read(&flex_group[i].free_inodes)))
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400386 best_flex = i;
387 }
388
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500389 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
390 !atomic_read(&flex_group[best_flex].free_blocks))
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400391 return -1;
392
393found_flexbg:
394 for (i = best_flex * flex_size; i < ngroups &&
395 i < (best_flex + 1) * flex_size; i++) {
Theodore Ts'o88b6edd2009-05-25 11:50:39 -0400396 desc = ext4_get_group_desc(sb, i, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500397 if (ext4_free_inodes_count(sb, desc)) {
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400398 *best_group = i;
399 goto out;
400 }
401 }
402
403 return -1;
404out:
405 return 0;
406}
407
Theodore Ts'oa4912122009-03-12 12:18:34 -0400408struct orlov_stats {
409 __u32 free_inodes;
410 __u32 free_blocks;
411 __u32 used_dirs;
412};
413
414/*
415 * Helper function for Orlov's allocator; returns critical information
416 * for a particular block group or flex_bg. If flex_size is 1, then g
417 * is a block group number; otherwise it is flex_bg number.
418 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400419static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
420 int flex_size, struct orlov_stats *stats)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400421{
422 struct ext4_group_desc *desc;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500423 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400424
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500425 if (flex_size > 1) {
426 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
427 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
428 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
429 return;
430 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400431
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500432 desc = ext4_get_group_desc(sb, g, NULL);
433 if (desc) {
434 stats->free_inodes = ext4_free_inodes_count(sb, desc);
435 stats->free_blocks = ext4_free_blks_count(sb, desc);
436 stats->used_dirs = ext4_used_dirs_count(sb, desc);
437 } else {
438 stats->free_inodes = 0;
439 stats->free_blocks = 0;
440 stats->used_dirs = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400441 }
442}
443
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700444/*
445 * Orlov's allocator for directories.
446 *
447 * We always try to spread first-level directories.
448 *
449 * If there are blockgroups with both free inodes and free blocks counts
450 * not worse than average we return one with smallest directory count.
451 * Otherwise we simply return a random group.
452 *
453 * For the rest rules look so:
454 *
455 * It's OK to put directory into a group unless
456 * it has too many directories already (max_dirs) or
457 * it has too few free inodes left (min_inodes) or
458 * it has too few free blocks left (min_blocks) or
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000459 * Parent's group is preferred, if it doesn't satisfy these
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700460 * conditions we search cyclically through the rest. If none
461 * of the groups look good we just look for a group with more
462 * free inodes than average (starting at parent's group).
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700463 */
464
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500465static int find_group_orlov(struct super_block *sb, struct inode *parent,
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400466 ext4_group_t *group, int mode,
467 const struct qstr *qstr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700468{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500469 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Mingming Cao617ba132006-10-11 01:20:53 -0700470 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400471 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700472 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700473 unsigned int freei, avefreei;
Mingming Cao617ba132006-10-11 01:20:53 -0700474 ext4_fsblk_t freeb, avefreeb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700475 unsigned int ndirs;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400476 int max_dirs, min_inodes;
Mingming Cao617ba132006-10-11 01:20:53 -0700477 ext4_grpblk_t min_blocks;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400478 ext4_group_t i, grp, g, ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700479 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400480 struct orlov_stats stats;
481 int flex_size = ext4_flex_bg_size(sbi);
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400482 struct dx_hash_info hinfo;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400483
Theodore Ts'o8df96752009-05-01 08:50:38 -0400484 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400485 if (flex_size > 1) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400486 ngroups = (real_ngroups + flex_size - 1) >>
Theodore Ts'oa4912122009-03-12 12:18:34 -0400487 sbi->s_log_groups_per_flex;
488 parent_group >>= sbi->s_log_groups_per_flex;
489 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700490
491 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
492 avefreei = freei / ngroups;
493 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700494 avefreeb = freeb;
Andrew Mortonf4e5bc22006-10-11 01:21:19 -0700495 do_div(avefreeb, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700496 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
497
Theodore Ts'oa4912122009-03-12 12:18:34 -0400498 if (S_ISDIR(mode) &&
499 ((parent == sb->s_root->d_inode) ||
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400500 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700501 int best_ndir = inodes_per_group;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500502 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700503
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400504 if (qstr) {
505 hinfo.hash_version = DX_HASH_HALF_MD4;
506 hinfo.seed = sbi->s_hash_seed;
507 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
508 grp = hinfo.hash;
509 } else
510 get_random_bytes(&grp, sizeof(grp));
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500511 parent_group = (unsigned)grp % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700512 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400513 g = (parent_group + i) % ngroups;
514 get_orlov_stats(sb, g, flex_size, &stats);
515 if (!stats.free_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700516 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400517 if (stats.used_dirs >= best_ndir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700518 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400519 if (stats.free_inodes < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700520 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400521 if (stats.free_blocks < avefreeb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700522 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400523 grp = g;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500524 ret = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400525 best_ndir = stats.used_dirs;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700526 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400527 if (ret)
528 goto fallback;
529 found_flex_bg:
530 if (flex_size == 1) {
531 *group = grp;
532 return 0;
533 }
534
535 /*
536 * We pack inodes at the beginning of the flexgroup's
537 * inode tables. Block allocation decisions will do
538 * something similar, although regular files will
539 * start at 2nd block group of the flexgroup. See
540 * ext4_ext_find_goal() and ext4_find_near().
541 */
542 grp *= flex_size;
543 for (i = 0; i < flex_size; i++) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400544 if (grp+i >= real_ngroups)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400545 break;
546 desc = ext4_get_group_desc(sb, grp+i, NULL);
547 if (desc && ext4_free_inodes_count(sb, desc)) {
548 *group = grp+i;
549 return 0;
550 }
551 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700552 goto fallback;
553 }
554
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700555 max_dirs = ndirs / ngroups + inodes_per_group / 16;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400556 min_inodes = avefreei - inodes_per_group*flex_size / 4;
557 if (min_inodes < 1)
558 min_inodes = 1;
559 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700560
Theodore Ts'oa4912122009-03-12 12:18:34 -0400561 /*
562 * Start looking in the flex group where we last allocated an
563 * inode for this parent directory
564 */
565 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
566 parent_group = EXT4_I(parent)->i_last_alloc_group;
567 if (flex_size > 1)
568 parent_group >>= sbi->s_log_groups_per_flex;
569 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700570
571 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400572 grp = (parent_group + i) % ngroups;
573 get_orlov_stats(sb, grp, flex_size, &stats);
574 if (stats.used_dirs >= max_dirs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700575 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400576 if (stats.free_inodes < min_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400578 if (stats.free_blocks < min_blocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700579 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400580 goto found_flex_bg;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700581 }
582
583fallback:
Theodore Ts'o8df96752009-05-01 08:50:38 -0400584 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400585 avefreei = freei / ngroups;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400586fallback_retry:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400587 parent_group = EXT4_I(parent)->i_block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700588 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400589 grp = (parent_group + i) % ngroups;
590 desc = ext4_get_group_desc(sb, grp, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500591 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'oa4912122009-03-12 12:18:34 -0400592 ext4_free_inodes_count(sb, desc) >= avefreei) {
593 *group = grp;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500594 return 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400595 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700596 }
597
598 if (avefreei) {
599 /*
600 * The free-inodes counter is approximate, and for really small
601 * filesystems the above test can fail to find any blockgroups
602 */
603 avefreei = 0;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400604 goto fallback_retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700605 }
606
607 return -1;
608}
609
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500610static int find_group_other(struct super_block *sb, struct inode *parent,
Theodore Ts'oa4912122009-03-12 12:18:34 -0400611 ext4_group_t *group, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700612{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500613 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400614 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700615 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400616 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
617
618 /*
619 * Try to place the inode is the same flex group as its
620 * parent. If we can't find space, use the Orlov algorithm to
621 * find another flex group, and store that information in the
622 * parent directory's inode information so that use that flex
623 * group for future allocations.
624 */
625 if (flex_size > 1) {
626 int retry = 0;
627
628 try_again:
629 parent_group &= ~(flex_size-1);
630 last = parent_group + flex_size;
631 if (last > ngroups)
632 last = ngroups;
633 for (i = parent_group; i < last; i++) {
634 desc = ext4_get_group_desc(sb, i, NULL);
635 if (desc && ext4_free_inodes_count(sb, desc)) {
636 *group = i;
637 return 0;
638 }
639 }
640 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
641 retry = 1;
642 parent_group = EXT4_I(parent)->i_last_alloc_group;
643 goto try_again;
644 }
645 /*
646 * If this didn't work, use the Orlov search algorithm
647 * to find a new flex group; we pass in the mode to
648 * avoid the topdir algorithms.
649 */
650 *group = parent_group + flex_size;
651 if (*group > ngroups)
652 *group = 0;
Peter Huewe7dc57612011-02-21 21:01:42 -0500653 return find_group_orlov(sb, parent, group, mode, NULL);
Theodore Ts'oa4912122009-03-12 12:18:34 -0400654 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700655
656 /*
657 * Try to place the inode in its parent directory
658 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500659 *group = parent_group;
660 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500661 if (desc && ext4_free_inodes_count(sb, desc) &&
662 ext4_free_blks_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500663 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700664
665 /*
666 * We're going to place this inode in a different blockgroup from its
667 * parent. We want to cause files in a common directory to all land in
668 * the same blockgroup. But we want files which are in a different
669 * directory which shares a blockgroup with our parent to land in a
670 * different blockgroup.
671 *
672 * So add our directory's i_ino into the starting point for the hash.
673 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500674 *group = (*group + parent->i_ino) % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700675
676 /*
677 * Use a quadratic hash to find a group with a free inode and some free
678 * blocks.
679 */
680 for (i = 1; i < ngroups; i <<= 1) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500681 *group += i;
682 if (*group >= ngroups)
683 *group -= ngroups;
684 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500685 if (desc && ext4_free_inodes_count(sb, desc) &&
686 ext4_free_blks_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500687 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700688 }
689
690 /*
691 * That failed: try linear search for a free inode, even if that group
692 * has no free blocks.
693 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500694 *group = parent_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700695 for (i = 0; i < ngroups; i++) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500696 if (++*group >= ngroups)
697 *group = 0;
698 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500699 if (desc && ext4_free_inodes_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500700 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700701 }
702
703 return -1;
704}
705
706/*
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500707 * claim the inode from the inode bitmap. If the group
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400708 * is uninit we need to take the groups's ext4_group_lock
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500709 * and clear the uninit flag. The inode bitmap update
710 * and group desc uninit flag clear should be done
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400711 * after holding ext4_group_lock so that ext4_read_inode_bitmap
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500712 * doesn't race with the ext4_claim_inode
713 */
714static int ext4_claim_inode(struct super_block *sb,
715 struct buffer_head *inode_bitmap_bh,
716 unsigned long ino, ext4_group_t group, int mode)
717{
718 int free = 0, retval = 0, count;
719 struct ext4_sb_info *sbi = EXT4_SB(sb);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400720 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500721 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
722
Lukas Czernerbfff6872010-10-27 21:30:05 -0400723 /*
724 * We have to be sure that new inode allocation does not race with
725 * inode table initialization, because otherwise we may end up
726 * allocating and writing new inode right before sb_issue_zeroout
727 * takes place and overwriting our new inode with zeroes. So we
728 * take alloc_sem to prevent it.
729 */
730 down_read(&grp->alloc_sem);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400731 ext4_lock_group(sb, group);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500732 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
733 /* not a free inode */
734 retval = 1;
735 goto err_ret;
736 }
737 ino++;
738 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
739 ino > EXT4_INODES_PER_GROUP(sb)) {
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400740 ext4_unlock_group(sb, group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400741 up_read(&grp->alloc_sem);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500742 ext4_error(sb, "reserved inode or inode > inodes count - "
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500743 "block_group = %u, inode=%lu", group,
744 ino + group * EXT4_INODES_PER_GROUP(sb));
745 return 1;
746 }
747 /* If we didn't allocate from within the initialized part of the inode
748 * table then we need to initialize up to this inode. */
749 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
750
751 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
752 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
753 /* When marking the block group with
754 * ~EXT4_BG_INODE_UNINIT we don't want to depend
755 * on the value of bg_itable_unused even though
756 * mke2fs could have initialized the same for us.
757 * Instead we calculated the value below
758 */
759
760 free = 0;
761 } else {
762 free = EXT4_INODES_PER_GROUP(sb) -
763 ext4_itable_unused_count(sb, gdp);
764 }
765
766 /*
767 * Check the relative inode number against the last used
768 * relative inode number in this group. if it is greater
769 * we need to update the bg_itable_unused count
770 *
771 */
772 if (ino > free)
773 ext4_itable_unused_set(sb, gdp,
774 (EXT4_INODES_PER_GROUP(sb) - ino));
775 }
776 count = ext4_free_inodes_count(sb, gdp) - 1;
777 ext4_free_inodes_set(sb, gdp, count);
778 if (S_ISDIR(mode)) {
779 count = ext4_used_dirs_count(sb, gdp) + 1;
780 ext4_used_dirs_set(sb, gdp, count);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500781 if (sbi->s_log_groups_per_flex) {
782 ext4_group_t f = ext4_flex_group(sbi, group);
783
Eric Sandeenc4caae22010-03-23 21:32:00 -0400784 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500785 }
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500786 }
787 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
788err_ret:
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400789 ext4_unlock_group(sb, group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400790 up_read(&grp->alloc_sem);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500791 return retval;
792}
793
794/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700795 * There are two policies for allocating an inode. If the new inode is
796 * a directory, then a forward search is made for a block group with both
797 * free space and a low directory-to-inode ratio; if that fails, then of
798 * the groups with above-average free space, that group with the fewest
799 * directories already is chosen.
800 *
801 * For other inodes, search forward from the parent directory's block
802 * group to find a free inode.
803 */
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400804struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
Andreas Dilger11013912009-06-13 11:45:35 -0400805 const struct qstr *qstr, __u32 goal)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700806{
807 struct super_block *sb;
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500808 struct buffer_head *inode_bitmap_bh = NULL;
809 struct buffer_head *group_desc_bh;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400810 ext4_group_t ngroups, group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700811 unsigned long ino = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400812 struct inode *inode;
813 struct ext4_group_desc *gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700814 struct ext4_inode_info *ei;
815 struct ext4_sb_info *sbi;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500816 int ret2, err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700817 struct inode *ret;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500818 ext4_group_t i;
819 int free = 0;
Theodore Ts'o2842c3b2009-03-12 12:20:01 -0400820 static int once = 1;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400821 ext4_group_t flex_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700822
823 /* Cannot create files in a deleted directory */
824 if (!dir || !dir->i_nlink)
825 return ERR_PTR(-EPERM);
826
827 sb = dir->i_sb;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400828 ngroups = ext4_get_groups_count(sb);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400829 trace_ext4_request_inode(dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700830 inode = new_inode(sb);
831 if (!inode)
832 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700833 ei = EXT4_I(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700834 sbi = EXT4_SB(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400835
Andreas Dilger11013912009-06-13 11:45:35 -0400836 if (!goal)
837 goal = sbi->s_inode_goal;
838
Johann Lombardie6462862009-07-05 23:45:11 -0400839 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
Andreas Dilger11013912009-06-13 11:45:35 -0400840 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
841 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
842 ret2 = 0;
843 goto got_group;
844 }
845
Theodore Ts'oa4912122009-03-12 12:18:34 -0400846 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400847 ret2 = find_group_flex(sb, dir, &group);
Theodore Ts'o05bf9e82009-02-21 12:13:24 -0500848 if (ret2 == -1) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400849 ret2 = find_group_other(sb, dir, &group, mode);
Chuck Ebbert6b82f3c2009-04-14 07:37:40 -0400850 if (ret2 == 0 && once) {
Theodore Ts'o2842c3b2009-03-12 12:20:01 -0400851 once = 0;
Theodore Ts'o05bf9e82009-02-21 12:13:24 -0500852 printk(KERN_NOTICE "ext4: find_group_flex "
853 "failed, fallback succeeded dir %lu\n",
854 dir->i_ino);
Chuck Ebbert6b82f3c2009-04-14 07:37:40 -0400855 }
Theodore Ts'o05bf9e82009-02-21 12:13:24 -0500856 }
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400857 goto got_group;
858 }
859
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700860 if (S_ISDIR(mode)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400861 if (test_opt(sb, OLDALLOC))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500862 ret2 = find_group_dir(sb, dir, &group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700863 else
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400864 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700865 } else
Theodore Ts'oa4912122009-03-12 12:18:34 -0400866 ret2 = find_group_other(sb, dir, &group, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700867
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400868got_group:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400869 EXT4_I(dir)->i_last_alloc_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700870 err = -ENOSPC;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500871 if (ret2 == -1)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700872 goto out;
873
Andreas Dilger11013912009-06-13 11:45:35 -0400874 for (i = 0; i < ngroups; i++, ino = 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700875 err = -EIO;
876
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500877 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700878 if (!gdp)
879 goto fail;
880
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500881 brelse(inode_bitmap_bh);
882 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
883 if (!inode_bitmap_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700884 goto fail;
885
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700886repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700887 ino = ext4_find_next_zero_bit((unsigned long *)
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500888 inode_bitmap_bh->b_data,
889 EXT4_INODES_PER_GROUP(sb), ino);
890
Mingming Cao617ba132006-10-11 01:20:53 -0700891 if (ino < EXT4_INODES_PER_GROUP(sb)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700892
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500893 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
894 err = ext4_journal_get_write_access(handle,
895 inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700896 if (err)
897 goto fail;
898
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500899 BUFFER_TRACE(group_desc_bh, "get_write_access");
900 err = ext4_journal_get_write_access(handle,
901 group_desc_bh);
902 if (err)
903 goto fail;
904 if (!ext4_claim_inode(sb, inode_bitmap_bh,
905 ino, group, mode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700906 /* we won it */
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500907 BUFFER_TRACE(inode_bitmap_bh,
Frank Mayhar03901312009-01-07 00:06:22 -0500908 "call ext4_handle_dirty_metadata");
909 err = ext4_handle_dirty_metadata(handle,
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -0500910 NULL,
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500911 inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700912 if (err)
913 goto fail;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500914 /* zero bit is inode number 1*/
915 ino++;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700916 goto got;
917 }
918 /* we lost it */
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500919 ext4_handle_release_buffer(handle, inode_bitmap_bh);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500920 ext4_handle_release_buffer(handle, group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700921
Mingming Cao617ba132006-10-11 01:20:53 -0700922 if (++ino < EXT4_INODES_PER_GROUP(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700923 goto repeat_in_this_group;
924 }
925
926 /*
927 * This case is possible in concurrent environment. It is very
928 * rare. We cannot repeat the find_group_xxx() call because
929 * that will simply return the same blockgroup, because the
930 * group descriptor metadata has not yet been updated.
931 * So we just go onto the next blockgroup.
932 */
Theodore Ts'o8df96752009-05-01 08:50:38 -0400933 if (++group == ngroups)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700934 group = 0;
935 }
936 err = -ENOSPC;
937 goto out;
938
939got:
Andreas Dilger717d50e2007-10-16 18:38:25 -0400940 /* We may have to initialize the block bitmap if it isn't already */
941 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
942 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500943 struct buffer_head *block_bitmap_bh;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400944
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500945 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
946 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
947 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400948 if (err) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500949 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400950 goto fail;
951 }
952
953 free = 0;
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400954 ext4_lock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400955 /* recheck and clear flag under lock if we still need to */
956 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Andreas Dilger717d50e2007-10-16 18:38:25 -0400957 free = ext4_free_blocks_after_init(sb, group, gdp);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500958 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500959 ext4_free_blks_set(sb, gdp, free);
Frederic Bohe23712a92008-11-07 09:21:01 -0500960 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
961 gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400962 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400963 ext4_unlock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400964
965 /* Don't need to dirty bitmap block if we didn't change it */
966 if (free) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500967 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
Frank Mayhar03901312009-01-07 00:06:22 -0500968 err = ext4_handle_dirty_metadata(handle,
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500969 NULL, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400970 }
971
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500972 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400973 if (err)
974 goto fail;
975 }
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500976 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
977 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500978 if (err)
979 goto fail;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700980
981 percpu_counter_dec(&sbi->s_freeinodes_counter);
982 if (S_ISDIR(mode))
983 percpu_counter_inc(&sbi->s_dirs_counter);
Theodore Ts'oa0375152010-06-11 23:14:04 -0400984 ext4_mark_super_dirty(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700985
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400986 if (sbi->s_log_groups_per_flex) {
987 flex_group = ext4_flex_group(sbi, group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500988 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400989 }
990
Dmitry Monakhovb10b8522010-03-04 17:31:51 +0300991 if (test_opt(sb, GRPID)) {
992 inode->i_mode = mode;
993 inode->i_uid = current_fsuid();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700994 inode->i_gid = dir->i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700995 } else
Dmitry Monakhovb10b8522010-03-04 17:31:51 +0300996 inode_init_owner(inode, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700997
Andreas Dilger717d50e2007-10-16 18:38:25 -0400998 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700999 /* This is the optimal IO size (for stat), not the fs block size */
1000 inode->i_blocks = 0;
Kalpak Shahef7f3832007-07-18 09:15:20 -04001001 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1002 ext4_current_time(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001003
1004 memset(ei->i_data, 0, sizeof(ei->i_data));
1005 ei->i_dir_start_lookup = 0;
1006 ei->i_disksize = 0;
1007
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001008 /*
Duane Griffin2dc6b0d2009-02-15 18:09:20 -05001009 * Don't inherit extent flag from directory, amongst others. We set
1010 * extent flag on newly created directory and file only if -o extent
1011 * mount option is specified
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001012 */
Duane Griffin2dc6b0d2009-02-15 18:09:20 -05001013 ei->i_flags =
1014 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001015 ei->i_file_acl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001016 ei->i_dtime = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001017 ei->i_block_group = group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04001018 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001019
Mingming Cao617ba132006-10-11 01:20:53 -07001020 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001021 if (IS_DIRSYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05001022 ext4_handle_sync(handle);
Al Viro6b38e842008-12-30 02:03:31 -05001023 if (insert_inode_locked(inode) < 0) {
1024 err = -EINVAL;
1025 goto fail_drop;
1026 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001027 spin_lock(&sbi->s_next_gen_lock);
1028 inode->i_generation = sbi->s_next_generation++;
1029 spin_unlock(&sbi->s_next_gen_lock);
1030
Theodore Ts'o353eb832011-01-10 12:18:25 -05001031 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001032 ext4_set_inode_state(inode, EXT4_STATE_NEW);
Kalpak Shahef7f3832007-07-18 09:15:20 -04001033
1034 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001035
1036 ret = inode;
Christoph Hellwig871a2932010-03-03 09:05:07 -05001037 dquot_initialize(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);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001071 goto really_out;
1072fail:
Mingming Cao617ba132006-10-11 01:20:53 -07001073 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001074out:
1075 iput(inode);
1076 ret = ERR_PTR(err);
1077really_out:
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001078 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001079 return ret;
1080
1081fail_free_drop:
Christoph Hellwig63936dd2010-03-03 09:05:01 -05001082 dquot_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001083
1084fail_drop:
Christoph Hellwig9f754752010-03-03 09:05:05 -05001085 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001086 inode->i_flags |= S_NOQUOTA;
1087 inode->i_nlink = 0;
Al Viro6b38e842008-12-30 02:03:31 -05001088 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001089 iput(inode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001090 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001091 return ERR_PTR(err);
1092}
1093
1094/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -07001095struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001096{
Mingming Cao617ba132006-10-11 01:20:53 -07001097 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Avantika Mathurfd2d4292008-01-28 23:58:27 -05001098 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001099 int bit;
David Howells1d1fe1e2008-02-07 00:15:37 -08001100 struct buffer_head *bitmap_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001101 struct inode *inode = NULL;
David Howells1d1fe1e2008-02-07 00:15:37 -08001102 long err = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001103
1104 /* Error cases - e2fsck has already cleaned up for us */
1105 if (ino > max_ino) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001106 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001107 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001108 }
1109
Mingming Cao617ba132006-10-11 01:20:53 -07001110 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1111 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001112 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001113 if (!bitmap_bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001114 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001115 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001116 }
1117
1118 /* Having the inode bit set should be a 100% indicator that this
1119 * is a valid orphan (no e2fsck run on fs). Orphans also include
1120 * inodes that were being truncated, so we can't check i_nlink==0.
1121 */
David Howells1d1fe1e2008-02-07 00:15:37 -08001122 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1123 goto bad_orphan;
1124
1125 inode = ext4_iget(sb, ino);
1126 if (IS_ERR(inode))
1127 goto iget_failed;
1128
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001129 /*
1130 * If the orphans has i_nlinks > 0 then it should be able to be
1131 * truncated, otherwise it won't be removed from the orphan list
1132 * during processing and an infinite loop will result.
1133 */
1134 if (inode->i_nlink && !ext4_can_truncate(inode))
1135 goto bad_orphan;
1136
David Howells1d1fe1e2008-02-07 00:15:37 -08001137 if (NEXT_ORPHAN(inode) > max_ino)
1138 goto bad_orphan;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001139 brelse(bitmap_bh);
1140 return inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08001141
1142iget_failed:
1143 err = PTR_ERR(inode);
1144 inode = NULL;
1145bad_orphan:
Eric Sandeen12062dd2010-02-15 14:19:27 -05001146 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001147 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1148 bit, (unsigned long long)bitmap_bh->b_blocknr,
1149 ext4_test_bit(bit, bitmap_bh->b_data));
1150 printk(KERN_NOTICE "inode=%p\n", inode);
1151 if (inode) {
1152 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1153 is_bad_inode(inode));
1154 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1155 NEXT_ORPHAN(inode));
1156 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001157 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
David Howells1d1fe1e2008-02-07 00:15:37 -08001158 /* Avoid freeing blocks if we got a bad deleted inode */
1159 if (inode->i_nlink == 0)
1160 inode->i_blocks = 0;
1161 iput(inode);
1162 }
1163 brelse(bitmap_bh);
1164error:
1165 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001166}
1167
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001168unsigned long ext4_count_free_inodes(struct super_block *sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001169{
1170 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001171 struct ext4_group_desc *gdp;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001172 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001173#ifdef EXT4FS_DEBUG
1174 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001175 unsigned long bitmap_count, x;
1176 struct buffer_head *bitmap_bh = NULL;
1177
Mingming Cao617ba132006-10-11 01:20:53 -07001178 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001179 desc_count = 0;
1180 bitmap_count = 0;
1181 gdp = NULL;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001182 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001183 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001184 if (!gdp)
1185 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001186 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001187 brelse(bitmap_bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001188 bitmap_bh = ext4_read_inode_bitmap(sb, i);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001189 if (!bitmap_bh)
1190 continue;
1191
Mingming Cao617ba132006-10-11 01:20:53 -07001192 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
Eric Sandeenc549a952008-01-28 23:58:27 -05001193 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
Peng Tao785b4b32009-07-27 21:44:40 -04001194 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001195 bitmap_count += x;
1196 }
1197 brelse(bitmap_bh);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001198 printk(KERN_DEBUG "ext4_count_free_inodes: "
1199 "stored = %u, computed = %lu, %lu\n",
1200 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001201 return desc_count;
1202#else
1203 desc_count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001204 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001205 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001206 if (!gdp)
1207 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001208 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001209 cond_resched();
1210 }
1211 return desc_count;
1212#endif
1213}
1214
1215/* Called at mount-time, super-block is locked */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001216unsigned long ext4_count_dirs(struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001217{
1218 unsigned long count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001219 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001220
Theodore Ts'o8df96752009-05-01 08:50:38 -04001221 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001222 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001223 if (!gdp)
1224 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001225 count += ext4_used_dirs_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001226 }
1227 return count;
1228}
Lukas Czernerbfff6872010-10-27 21:30:05 -04001229
1230/*
1231 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1232 * inode table. Must be called without any spinlock held. The only place
1233 * where it is called from on active part of filesystem is ext4lazyinit
1234 * thread, so we do not need any special locks, however we have to prevent
1235 * inode allocation from the current group, so we take alloc_sem lock, to
1236 * block ext4_claim_inode until we are finished.
1237 */
1238extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1239 int barrier)
1240{
1241 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1242 struct ext4_sb_info *sbi = EXT4_SB(sb);
1243 struct ext4_group_desc *gdp = NULL;
1244 struct buffer_head *group_desc_bh;
1245 handle_t *handle;
1246 ext4_fsblk_t blk;
1247 int num, ret = 0, used_blks = 0;
Lukas Czernerbfff6872010-10-27 21:30:05 -04001248
1249 /* This should not happen, but just to be sure check this */
1250 if (sb->s_flags & MS_RDONLY) {
1251 ret = 1;
1252 goto out;
1253 }
1254
1255 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1256 if (!gdp)
1257 goto out;
1258
1259 /*
1260 * We do not need to lock this, because we are the only one
1261 * handling this flag.
1262 */
1263 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1264 goto out;
1265
1266 handle = ext4_journal_start_sb(sb, 1);
1267 if (IS_ERR(handle)) {
1268 ret = PTR_ERR(handle);
1269 goto out;
1270 }
1271
1272 down_write(&grp->alloc_sem);
1273 /*
1274 * If inode bitmap was already initialized there may be some
1275 * used inodes so we need to skip blocks with used inodes in
1276 * inode table.
1277 */
1278 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1279 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1280 ext4_itable_unused_count(sb, gdp)),
1281 sbi->s_inodes_per_block);
1282
Lukas Czerner857ac882010-10-27 21:30:05 -04001283 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1284 ext4_error(sb, "Something is wrong with group %u\n"
1285 "Used itable blocks: %d"
1286 "itable unused count: %u\n",
1287 group, used_blks,
1288 ext4_itable_unused_count(sb, gdp));
1289 ret = 1;
Yongqiang Yang33853a02011-08-01 06:32:19 -04001290 goto err_out;
Lukas Czerner857ac882010-10-27 21:30:05 -04001291 }
1292
Lukas Czernerbfff6872010-10-27 21:30:05 -04001293 blk = ext4_inode_table(sb, gdp) + used_blks;
1294 num = sbi->s_itb_per_group - used_blks;
1295
1296 BUFFER_TRACE(group_desc_bh, "get_write_access");
1297 ret = ext4_journal_get_write_access(handle,
1298 group_desc_bh);
1299 if (ret)
1300 goto err_out;
1301
Lukas Czernerbfff6872010-10-27 21:30:05 -04001302 /*
1303 * Skip zeroout if the inode table is full. But we set the ZEROED
1304 * flag anyway, because obviously, when it is full it does not need
1305 * further zeroing.
1306 */
1307 if (unlikely(num == 0))
1308 goto skip_zeroout;
1309
1310 ext4_debug("going to zero out inode table in group %d\n",
1311 group);
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001312 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001313 if (ret < 0)
1314 goto err_out;
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001315 if (barrier)
1316 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001317
1318skip_zeroout:
1319 ext4_lock_group(sb, group);
1320 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1321 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1322 ext4_unlock_group(sb, group);
1323
1324 BUFFER_TRACE(group_desc_bh,
1325 "call ext4_handle_dirty_metadata");
1326 ret = ext4_handle_dirty_metadata(handle, NULL,
1327 group_desc_bh);
1328
1329err_out:
1330 up_write(&grp->alloc_sem);
1331 ext4_journal_stop(handle);
1332out:
1333 return ret;
1334}