blob: 1ce240a23ebb84963ce2126b78f126657fd7d6df [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 */
155 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400156 if (bh_submit_read(bh) < 0) {
157 put_bh(bh);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500158 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500159 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400160 block_group, bitmap_blk);
161 return NULL;
162 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700163 return bh;
164}
165
166/*
167 * NOTE! When we get the inode, we're the only people
168 * that have access to it, and as such there are no
169 * race conditions we have to worry about. The inode
170 * is not on the hash-lists, and it cannot be reached
171 * through the filesystem because the directory entry
172 * has been deleted earlier.
173 *
174 * HOWEVER: we must make sure that we get no aliases,
175 * which means that we have to call "clear_inode()"
176 * _before_ we mark the inode not in use in the inode
177 * bitmaps. Otherwise a newly created file might use
178 * the same inode number (not actually the same pointer
179 * though), and then we'd have two inodes sharing the
180 * same inode number and space on the harddisk.
181 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400182void ext4_free_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700183{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400184 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700185 int is_directory;
186 unsigned long ino;
187 struct buffer_head *bitmap_bh = NULL;
188 struct buffer_head *bh2;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500189 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700190 unsigned long bit;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400191 struct ext4_group_desc *gdp;
192 struct ext4_super_block *es;
Mingming Cao617ba132006-10-11 01:20:53 -0700193 struct ext4_sb_info *sbi;
Eric Sandeen7ce9d5d2009-03-04 18:38:18 -0500194 int fatal = 0, err, count, cleared;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700195
196 if (atomic_read(&inode->i_count) > 1) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400197 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
198 atomic_read(&inode->i_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700199 return;
200 }
201 if (inode->i_nlink) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400202 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
203 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700204 return;
205 }
206 if (!sb) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400207 printk(KERN_ERR "ext4_free_inode: inode on "
208 "nonexistent device\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700209 return;
210 }
Mingming Cao617ba132006-10-11 01:20:53 -0700211 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700212
213 ino = inode->i_ino;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400214 ext4_debug("freeing inode %lu\n", ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400215 trace_ext4_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700216
217 /*
218 * Note: we must free any quota before locking the superblock,
219 * as writing the quota to disk may need the lock as well.
220 */
Christoph Hellwig871a2932010-03-03 09:05:07 -0500221 dquot_initialize(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700222 ext4_xattr_delete_inode(handle, inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500223 dquot_free_inode(inode);
Christoph Hellwig9f754752010-03-03 09:05:05 -0500224 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700225
226 is_directory = S_ISDIR(inode->i_mode);
227
228 /* Do this BEFORE marking the inode not in use or returning an error */
Al Viro0930fcc2010-06-07 13:16:22 -0400229 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700230
Mingming Cao617ba132006-10-11 01:20:53 -0700231 es = EXT4_SB(sb)->s_es;
232 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500233 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234 goto error_return;
235 }
Mingming Cao617ba132006-10-11 01:20:53 -0700236 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
237 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400238 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700239 if (!bitmap_bh)
240 goto error_return;
241
242 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700243 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700244 if (fatal)
245 goto error_return;
246
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400247 fatal = -ESRCH;
248 gdp = ext4_get_group_desc(sb, block_group, &bh2);
249 if (gdp) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700250 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700251 fatal = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700252 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400253 ext4_lock_group(sb, block_group);
254 cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
255 if (fatal || !cleared) {
256 ext4_unlock_group(sb, block_group);
257 goto out;
258 }
259
260 count = ext4_free_inodes_count(sb, gdp) + 1;
261 ext4_free_inodes_set(sb, gdp, count);
262 if (is_directory) {
263 count = ext4_used_dirs_count(sb, gdp) - 1;
264 ext4_used_dirs_set(sb, gdp, count);
265 percpu_counter_dec(&sbi->s_dirs_counter);
266 }
267 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
268 ext4_unlock_group(sb, block_group);
269
270 percpu_counter_inc(&sbi->s_freeinodes_counter);
271 if (sbi->s_log_groups_per_flex) {
272 ext4_group_t f = ext4_flex_group(sbi, block_group);
273
274 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
275 if (is_directory)
276 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
277 }
278 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
279 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
280out:
281 if (cleared) {
282 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
283 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
284 if (!fatal)
285 fatal = err;
Theodore Ts'oa0375152010-06-11 23:14:04 -0400286 ext4_mark_super_dirty(sb);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400287 } else
288 ext4_error(sb, "bit already cleared for inode %lu", ino);
289
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700290error_return:
291 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700292 ext4_std_error(sb, fatal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700293}
294
295/*
296 * There are two policies for allocating an inode. If the new inode is
297 * a directory, then a forward search is made for a block group with both
298 * free space and a low directory-to-inode ratio; if that fails, then of
299 * the groups with above-average free space, that group with the fewest
300 * directories already is chosen.
301 *
302 * For other inodes, search forward from the parent directory\'s block
303 * group to find a free inode.
304 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500305static int find_group_dir(struct super_block *sb, struct inode *parent,
306 ext4_group_t *best_group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700307{
Theodore Ts'o8df96752009-05-01 08:50:38 -0400308 ext4_group_t ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700309 unsigned int freei, avefreei;
Mingming Cao617ba132006-10-11 01:20:53 -0700310 struct ext4_group_desc *desc, *best_desc = NULL;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500311 ext4_group_t group;
312 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700313
Mingming Cao617ba132006-10-11 01:20:53 -0700314 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700315 avefreei = freei / ngroups;
316
317 for (group = 0; group < ngroups; group++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400318 desc = ext4_get_group_desc(sb, group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500319 if (!desc || !ext4_free_inodes_count(sb, desc))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700320 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500321 if (ext4_free_inodes_count(sb, desc) < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700322 continue;
323 if (!best_desc ||
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500324 (ext4_free_blks_count(sb, desc) >
325 ext4_free_blks_count(sb, best_desc))) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500326 *best_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700327 best_desc = desc;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500328 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700329 }
330 }
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500331 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700332}
333
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400334#define free_block_ratio 10
335
336static int find_group_flex(struct super_block *sb, struct inode *parent,
337 ext4_group_t *best_group)
338{
339 struct ext4_sb_info *sbi = EXT4_SB(sb);
340 struct ext4_group_desc *desc;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400341 struct flex_groups *flex_group = sbi->s_flex_groups;
342 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
343 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400344 ext4_group_t ngroups = ext4_get_groups_count(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400345 int flex_size = ext4_flex_bg_size(sbi);
346 ext4_group_t best_flex = parent_fbg_group;
347 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
348 int flexbg_free_blocks;
349 int flex_freeb_ratio;
350 ext4_group_t n_fbg_groups;
351 ext4_group_t i;
352
Theodore Ts'o8df96752009-05-01 08:50:38 -0400353 n_fbg_groups = (ngroups + flex_size - 1) >>
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400354 sbi->s_log_groups_per_flex;
355
356find_close_to_parent:
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500357 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400358 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500359 if (atomic_read(&flex_group[best_flex].free_inodes) &&
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400360 flex_freeb_ratio > free_block_ratio)
361 goto found_flexbg;
362
363 if (best_flex && best_flex == parent_fbg_group) {
364 best_flex--;
365 goto find_close_to_parent;
366 }
367
368 for (i = 0; i < n_fbg_groups; i++) {
369 if (i == parent_fbg_group || i == parent_fbg_group - 1)
370 continue;
371
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500372 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400373 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
374
375 if (flex_freeb_ratio > free_block_ratio &&
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500376 (atomic_read(&flex_group[i].free_inodes))) {
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400377 best_flex = i;
378 goto found_flexbg;
379 }
380
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500381 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
382 ((atomic_read(&flex_group[i].free_blocks) >
383 atomic_read(&flex_group[best_flex].free_blocks)) &&
384 atomic_read(&flex_group[i].free_inodes)))
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400385 best_flex = i;
386 }
387
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500388 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
389 !atomic_read(&flex_group[best_flex].free_blocks))
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400390 return -1;
391
392found_flexbg:
393 for (i = best_flex * flex_size; i < ngroups &&
394 i < (best_flex + 1) * flex_size; i++) {
Theodore Ts'o88b6edd2009-05-25 11:50:39 -0400395 desc = ext4_get_group_desc(sb, i, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500396 if (ext4_free_inodes_count(sb, desc)) {
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400397 *best_group = i;
398 goto out;
399 }
400 }
401
402 return -1;
403out:
404 return 0;
405}
406
Theodore Ts'oa4912122009-03-12 12:18:34 -0400407struct orlov_stats {
408 __u32 free_inodes;
409 __u32 free_blocks;
410 __u32 used_dirs;
411};
412
413/*
414 * Helper function for Orlov's allocator; returns critical information
415 * for a particular block group or flex_bg. If flex_size is 1, then g
416 * is a block group number; otherwise it is flex_bg number.
417 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400418static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
419 int flex_size, struct orlov_stats *stats)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400420{
421 struct ext4_group_desc *desc;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500422 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400423
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500424 if (flex_size > 1) {
425 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
426 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
427 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
428 return;
429 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400430
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500431 desc = ext4_get_group_desc(sb, g, NULL);
432 if (desc) {
433 stats->free_inodes = ext4_free_inodes_count(sb, desc);
434 stats->free_blocks = ext4_free_blks_count(sb, desc);
435 stats->used_dirs = ext4_used_dirs_count(sb, desc);
436 } else {
437 stats->free_inodes = 0;
438 stats->free_blocks = 0;
439 stats->used_dirs = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400440 }
441}
442
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700443/*
444 * Orlov's allocator for directories.
445 *
446 * We always try to spread first-level directories.
447 *
448 * If there are blockgroups with both free inodes and free blocks counts
449 * not worse than average we return one with smallest directory count.
450 * Otherwise we simply return a random group.
451 *
452 * For the rest rules look so:
453 *
454 * It's OK to put directory into a group unless
455 * it has too many directories already (max_dirs) or
456 * it has too few free inodes left (min_inodes) or
457 * it has too few free blocks left (min_blocks) or
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000458 * Parent's group is preferred, if it doesn't satisfy these
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700459 * conditions we search cyclically through the rest. If none
460 * of the groups look good we just look for a group with more
461 * free inodes than average (starting at parent's group).
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700462 */
463
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500464static int find_group_orlov(struct super_block *sb, struct inode *parent,
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400465 ext4_group_t *group, int mode,
466 const struct qstr *qstr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700467{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500468 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Mingming Cao617ba132006-10-11 01:20:53 -0700469 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400470 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700471 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700472 unsigned int freei, avefreei;
Mingming Cao617ba132006-10-11 01:20:53 -0700473 ext4_fsblk_t freeb, avefreeb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700474 unsigned int ndirs;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400475 int max_dirs, min_inodes;
Mingming Cao617ba132006-10-11 01:20:53 -0700476 ext4_grpblk_t min_blocks;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400477 ext4_group_t i, grp, g, ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700478 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400479 struct orlov_stats stats;
480 int flex_size = ext4_flex_bg_size(sbi);
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400481 struct dx_hash_info hinfo;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400482
Theodore Ts'o8df96752009-05-01 08:50:38 -0400483 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400484 if (flex_size > 1) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400485 ngroups = (real_ngroups + flex_size - 1) >>
Theodore Ts'oa4912122009-03-12 12:18:34 -0400486 sbi->s_log_groups_per_flex;
487 parent_group >>= sbi->s_log_groups_per_flex;
488 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700489
490 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
491 avefreei = freei / ngroups;
492 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700493 avefreeb = freeb;
Andrew Mortonf4e5bc22006-10-11 01:21:19 -0700494 do_div(avefreeb, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700495 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
496
Theodore Ts'oa4912122009-03-12 12:18:34 -0400497 if (S_ISDIR(mode) &&
498 ((parent == sb->s_root->d_inode) ||
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400499 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700500 int best_ndir = inodes_per_group;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500501 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700502
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400503 if (qstr) {
504 hinfo.hash_version = DX_HASH_HALF_MD4;
505 hinfo.seed = sbi->s_hash_seed;
506 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
507 grp = hinfo.hash;
508 } else
509 get_random_bytes(&grp, sizeof(grp));
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500510 parent_group = (unsigned)grp % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700511 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400512 g = (parent_group + i) % ngroups;
513 get_orlov_stats(sb, g, flex_size, &stats);
514 if (!stats.free_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700515 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400516 if (stats.used_dirs >= best_ndir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700517 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400518 if (stats.free_inodes < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700519 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400520 if (stats.free_blocks < avefreeb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700521 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400522 grp = g;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500523 ret = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400524 best_ndir = stats.used_dirs;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700525 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400526 if (ret)
527 goto fallback;
528 found_flex_bg:
529 if (flex_size == 1) {
530 *group = grp;
531 return 0;
532 }
533
534 /*
535 * We pack inodes at the beginning of the flexgroup's
536 * inode tables. Block allocation decisions will do
537 * something similar, although regular files will
538 * start at 2nd block group of the flexgroup. See
539 * ext4_ext_find_goal() and ext4_find_near().
540 */
541 grp *= flex_size;
542 for (i = 0; i < flex_size; i++) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400543 if (grp+i >= real_ngroups)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400544 break;
545 desc = ext4_get_group_desc(sb, grp+i, NULL);
546 if (desc && ext4_free_inodes_count(sb, desc)) {
547 *group = grp+i;
548 return 0;
549 }
550 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700551 goto fallback;
552 }
553
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700554 max_dirs = ndirs / ngroups + inodes_per_group / 16;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400555 min_inodes = avefreei - inodes_per_group*flex_size / 4;
556 if (min_inodes < 1)
557 min_inodes = 1;
558 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700559
Theodore Ts'oa4912122009-03-12 12:18:34 -0400560 /*
561 * Start looking in the flex group where we last allocated an
562 * inode for this parent directory
563 */
564 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
565 parent_group = EXT4_I(parent)->i_last_alloc_group;
566 if (flex_size > 1)
567 parent_group >>= sbi->s_log_groups_per_flex;
568 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700569
570 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400571 grp = (parent_group + i) % ngroups;
572 get_orlov_stats(sb, grp, flex_size, &stats);
573 if (stats.used_dirs >= max_dirs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700574 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400575 if (stats.free_inodes < min_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700576 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400577 if (stats.free_blocks < min_blocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700578 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400579 goto found_flex_bg;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700580 }
581
582fallback:
Theodore Ts'o8df96752009-05-01 08:50:38 -0400583 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400584 avefreei = freei / ngroups;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400585fallback_retry:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400586 parent_group = EXT4_I(parent)->i_block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700587 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400588 grp = (parent_group + i) % ngroups;
589 desc = ext4_get_group_desc(sb, grp, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500590 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'oa4912122009-03-12 12:18:34 -0400591 ext4_free_inodes_count(sb, desc) >= avefreei) {
592 *group = grp;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500593 return 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400594 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700595 }
596
597 if (avefreei) {
598 /*
599 * The free-inodes counter is approximate, and for really small
600 * filesystems the above test can fail to find any blockgroups
601 */
602 avefreei = 0;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400603 goto fallback_retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700604 }
605
606 return -1;
607}
608
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500609static int find_group_other(struct super_block *sb, struct inode *parent,
Theodore Ts'oa4912122009-03-12 12:18:34 -0400610 ext4_group_t *group, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700611{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500612 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400613 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700614 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400615 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
616
617 /*
618 * Try to place the inode is the same flex group as its
619 * parent. If we can't find space, use the Orlov algorithm to
620 * find another flex group, and store that information in the
621 * parent directory's inode information so that use that flex
622 * group for future allocations.
623 */
624 if (flex_size > 1) {
625 int retry = 0;
626
627 try_again:
628 parent_group &= ~(flex_size-1);
629 last = parent_group + flex_size;
630 if (last > ngroups)
631 last = ngroups;
632 for (i = parent_group; i < last; i++) {
633 desc = ext4_get_group_desc(sb, i, NULL);
634 if (desc && ext4_free_inodes_count(sb, desc)) {
635 *group = i;
636 return 0;
637 }
638 }
639 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
640 retry = 1;
641 parent_group = EXT4_I(parent)->i_last_alloc_group;
642 goto try_again;
643 }
644 /*
645 * If this didn't work, use the Orlov search algorithm
646 * to find a new flex group; we pass in the mode to
647 * avoid the topdir algorithms.
648 */
649 *group = parent_group + flex_size;
650 if (*group > ngroups)
651 *group = 0;
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400652 return find_group_orlov(sb, parent, group, mode, 0);
Theodore Ts'oa4912122009-03-12 12:18:34 -0400653 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700654
655 /*
656 * Try to place the inode in its parent directory
657 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500658 *group = parent_group;
659 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500660 if (desc && ext4_free_inodes_count(sb, desc) &&
661 ext4_free_blks_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500662 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700663
664 /*
665 * We're going to place this inode in a different blockgroup from its
666 * parent. We want to cause files in a common directory to all land in
667 * the same blockgroup. But we want files which are in a different
668 * directory which shares a blockgroup with our parent to land in a
669 * different blockgroup.
670 *
671 * So add our directory's i_ino into the starting point for the hash.
672 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500673 *group = (*group + parent->i_ino) % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700674
675 /*
676 * Use a quadratic hash to find a group with a free inode and some free
677 * blocks.
678 */
679 for (i = 1; i < ngroups; i <<= 1) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500680 *group += i;
681 if (*group >= ngroups)
682 *group -= ngroups;
683 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500684 if (desc && ext4_free_inodes_count(sb, desc) &&
685 ext4_free_blks_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500686 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700687 }
688
689 /*
690 * That failed: try linear search for a free inode, even if that group
691 * has no free blocks.
692 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500693 *group = parent_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700694 for (i = 0; i < ngroups; i++) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500695 if (++*group >= ngroups)
696 *group = 0;
697 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500698 if (desc && ext4_free_inodes_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500699 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700700 }
701
702 return -1;
703}
704
705/*
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500706 * claim the inode from the inode bitmap. If the group
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400707 * is uninit we need to take the groups's ext4_group_lock
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500708 * and clear the uninit flag. The inode bitmap update
709 * and group desc uninit flag clear should be done
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400710 * after holding ext4_group_lock so that ext4_read_inode_bitmap
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500711 * doesn't race with the ext4_claim_inode
712 */
713static int ext4_claim_inode(struct super_block *sb,
714 struct buffer_head *inode_bitmap_bh,
715 unsigned long ino, ext4_group_t group, int mode)
716{
717 int free = 0, retval = 0, count;
718 struct ext4_sb_info *sbi = EXT4_SB(sb);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400719 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500720 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
721
Lukas Czernerbfff6872010-10-27 21:30:05 -0400722 /*
723 * We have to be sure that new inode allocation does not race with
724 * inode table initialization, because otherwise we may end up
725 * allocating and writing new inode right before sb_issue_zeroout
726 * takes place and overwriting our new inode with zeroes. So we
727 * take alloc_sem to prevent it.
728 */
729 down_read(&grp->alloc_sem);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400730 ext4_lock_group(sb, group);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500731 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
732 /* not a free inode */
733 retval = 1;
734 goto err_ret;
735 }
736 ino++;
737 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
738 ino > EXT4_INODES_PER_GROUP(sb)) {
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400739 ext4_unlock_group(sb, group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400740 up_read(&grp->alloc_sem);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500741 ext4_error(sb, "reserved inode or inode > inodes count - "
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500742 "block_group = %u, inode=%lu", group,
743 ino + group * EXT4_INODES_PER_GROUP(sb));
744 return 1;
745 }
746 /* If we didn't allocate from within the initialized part of the inode
747 * table then we need to initialize up to this inode. */
748 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
749
750 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
751 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
752 /* When marking the block group with
753 * ~EXT4_BG_INODE_UNINIT we don't want to depend
754 * on the value of bg_itable_unused even though
755 * mke2fs could have initialized the same for us.
756 * Instead we calculated the value below
757 */
758
759 free = 0;
760 } else {
761 free = EXT4_INODES_PER_GROUP(sb) -
762 ext4_itable_unused_count(sb, gdp);
763 }
764
765 /*
766 * Check the relative inode number against the last used
767 * relative inode number in this group. if it is greater
768 * we need to update the bg_itable_unused count
769 *
770 */
771 if (ino > free)
772 ext4_itable_unused_set(sb, gdp,
773 (EXT4_INODES_PER_GROUP(sb) - ino));
774 }
775 count = ext4_free_inodes_count(sb, gdp) - 1;
776 ext4_free_inodes_set(sb, gdp, count);
777 if (S_ISDIR(mode)) {
778 count = ext4_used_dirs_count(sb, gdp) + 1;
779 ext4_used_dirs_set(sb, gdp, count);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500780 if (sbi->s_log_groups_per_flex) {
781 ext4_group_t f = ext4_flex_group(sbi, group);
782
Eric Sandeenc4caae22010-03-23 21:32:00 -0400783 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500784 }
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500785 }
786 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
787err_ret:
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400788 ext4_unlock_group(sb, group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400789 up_read(&grp->alloc_sem);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500790 return retval;
791}
792
793/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700794 * There are two policies for allocating an inode. If the new inode is
795 * a directory, then a forward search is made for a block group with both
796 * free space and a low directory-to-inode ratio; if that fails, then of
797 * the groups with above-average free space, that group with the fewest
798 * directories already is chosen.
799 *
800 * For other inodes, search forward from the parent directory's block
801 * group to find a free inode.
802 */
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400803struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
Andreas Dilger11013912009-06-13 11:45:35 -0400804 const struct qstr *qstr, __u32 goal)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700805{
806 struct super_block *sb;
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500807 struct buffer_head *inode_bitmap_bh = NULL;
808 struct buffer_head *group_desc_bh;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400809 ext4_group_t ngroups, group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700810 unsigned long ino = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400811 struct inode *inode;
812 struct ext4_group_desc *gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700813 struct ext4_inode_info *ei;
814 struct ext4_sb_info *sbi;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500815 int ret2, err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700816 struct inode *ret;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500817 ext4_group_t i;
818 int free = 0;
Theodore Ts'o2842c3b2009-03-12 12:20:01 -0400819 static int once = 1;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400820 ext4_group_t flex_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700821
822 /* Cannot create files in a deleted directory */
823 if (!dir || !dir->i_nlink)
824 return ERR_PTR(-EPERM);
825
826 sb = dir->i_sb;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400827 ngroups = ext4_get_groups_count(sb);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400828 trace_ext4_request_inode(dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700829 inode = new_inode(sb);
830 if (!inode)
831 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700832 ei = EXT4_I(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700833 sbi = EXT4_SB(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400834
Andreas Dilger11013912009-06-13 11:45:35 -0400835 if (!goal)
836 goal = sbi->s_inode_goal;
837
Johann Lombardie6462862009-07-05 23:45:11 -0400838 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
Andreas Dilger11013912009-06-13 11:45:35 -0400839 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
840 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
841 ret2 = 0;
842 goto got_group;
843 }
844
Theodore Ts'oa4912122009-03-12 12:18:34 -0400845 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400846 ret2 = find_group_flex(sb, dir, &group);
Theodore Ts'o05bf9e82009-02-21 12:13:24 -0500847 if (ret2 == -1) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400848 ret2 = find_group_other(sb, dir, &group, mode);
Chuck Ebbert6b82f3c2009-04-14 07:37:40 -0400849 if (ret2 == 0 && once) {
Theodore Ts'o2842c3b2009-03-12 12:20:01 -0400850 once = 0;
Theodore Ts'o05bf9e82009-02-21 12:13:24 -0500851 printk(KERN_NOTICE "ext4: find_group_flex "
852 "failed, fallback succeeded dir %lu\n",
853 dir->i_ino);
Chuck Ebbert6b82f3c2009-04-14 07:37:40 -0400854 }
Theodore Ts'o05bf9e82009-02-21 12:13:24 -0500855 }
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400856 goto got_group;
857 }
858
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700859 if (S_ISDIR(mode)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400860 if (test_opt(sb, OLDALLOC))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500861 ret2 = find_group_dir(sb, dir, &group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700862 else
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400863 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700864 } else
Theodore Ts'oa4912122009-03-12 12:18:34 -0400865 ret2 = find_group_other(sb, dir, &group, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700866
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400867got_group:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400868 EXT4_I(dir)->i_last_alloc_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700869 err = -ENOSPC;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500870 if (ret2 == -1)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700871 goto out;
872
Andreas Dilger11013912009-06-13 11:45:35 -0400873 for (i = 0; i < ngroups; i++, ino = 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700874 err = -EIO;
875
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500876 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700877 if (!gdp)
878 goto fail;
879
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500880 brelse(inode_bitmap_bh);
881 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
882 if (!inode_bitmap_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700883 goto fail;
884
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700885repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700886 ino = ext4_find_next_zero_bit((unsigned long *)
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500887 inode_bitmap_bh->b_data,
888 EXT4_INODES_PER_GROUP(sb), ino);
889
Mingming Cao617ba132006-10-11 01:20:53 -0700890 if (ino < EXT4_INODES_PER_GROUP(sb)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700891
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500892 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
893 err = ext4_journal_get_write_access(handle,
894 inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700895 if (err)
896 goto fail;
897
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500898 BUFFER_TRACE(group_desc_bh, "get_write_access");
899 err = ext4_journal_get_write_access(handle,
900 group_desc_bh);
901 if (err)
902 goto fail;
903 if (!ext4_claim_inode(sb, inode_bitmap_bh,
904 ino, group, mode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700905 /* we won it */
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500906 BUFFER_TRACE(inode_bitmap_bh,
Frank Mayhar03901312009-01-07 00:06:22 -0500907 "call ext4_handle_dirty_metadata");
908 err = ext4_handle_dirty_metadata(handle,
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -0500909 NULL,
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500910 inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700911 if (err)
912 goto fail;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500913 /* zero bit is inode number 1*/
914 ino++;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700915 goto got;
916 }
917 /* we lost it */
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500918 ext4_handle_release_buffer(handle, inode_bitmap_bh);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500919 ext4_handle_release_buffer(handle, group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700920
Mingming Cao617ba132006-10-11 01:20:53 -0700921 if (++ino < EXT4_INODES_PER_GROUP(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700922 goto repeat_in_this_group;
923 }
924
925 /*
926 * This case is possible in concurrent environment. It is very
927 * rare. We cannot repeat the find_group_xxx() call because
928 * that will simply return the same blockgroup, because the
929 * group descriptor metadata has not yet been updated.
930 * So we just go onto the next blockgroup.
931 */
Theodore Ts'o8df96752009-05-01 08:50:38 -0400932 if (++group == ngroups)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700933 group = 0;
934 }
935 err = -ENOSPC;
936 goto out;
937
938got:
Andreas Dilger717d50e2007-10-16 18:38:25 -0400939 /* We may have to initialize the block bitmap if it isn't already */
940 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
941 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500942 struct buffer_head *block_bitmap_bh;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400943
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500944 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
945 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
946 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400947 if (err) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500948 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400949 goto fail;
950 }
951
952 free = 0;
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400953 ext4_lock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400954 /* recheck and clear flag under lock if we still need to */
955 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Andreas Dilger717d50e2007-10-16 18:38:25 -0400956 free = ext4_free_blocks_after_init(sb, group, gdp);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500957 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500958 ext4_free_blks_set(sb, gdp, free);
Frederic Bohe23712a92008-11-07 09:21:01 -0500959 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
960 gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400961 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400962 ext4_unlock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400963
964 /* Don't need to dirty bitmap block if we didn't change it */
965 if (free) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500966 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
Frank Mayhar03901312009-01-07 00:06:22 -0500967 err = ext4_handle_dirty_metadata(handle,
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500968 NULL, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400969 }
970
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500971 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400972 if (err)
973 goto fail;
974 }
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500975 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
976 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500977 if (err)
978 goto fail;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700979
980 percpu_counter_dec(&sbi->s_freeinodes_counter);
981 if (S_ISDIR(mode))
982 percpu_counter_inc(&sbi->s_dirs_counter);
Theodore Ts'oa0375152010-06-11 23:14:04 -0400983 ext4_mark_super_dirty(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700984
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400985 if (sbi->s_log_groups_per_flex) {
986 flex_group = ext4_flex_group(sbi, group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500987 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400988 }
989
Dmitry Monakhovb10b8522010-03-04 17:31:51 +0300990 if (test_opt(sb, GRPID)) {
991 inode->i_mode = mode;
992 inode->i_uid = current_fsuid();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700993 inode->i_gid = dir->i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700994 } else
Dmitry Monakhovb10b8522010-03-04 17:31:51 +0300995 inode_init_owner(inode, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700996
Andreas Dilger717d50e2007-10-16 18:38:25 -0400997 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700998 /* This is the optimal IO size (for stat), not the fs block size */
999 inode->i_blocks = 0;
Kalpak Shahef7f3832007-07-18 09:15:20 -04001000 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1001 ext4_current_time(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001002
1003 memset(ei->i_data, 0, sizeof(ei->i_data));
1004 ei->i_dir_start_lookup = 0;
1005 ei->i_disksize = 0;
1006
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001007 /*
Duane Griffin2dc6b0d2009-02-15 18:09:20 -05001008 * Don't inherit extent flag from directory, amongst others. We set
1009 * extent flag on newly created directory and file only if -o extent
1010 * mount option is specified
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001011 */
Duane Griffin2dc6b0d2009-02-15 18:09:20 -05001012 ei->i_flags =
1013 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001014 ei->i_file_acl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001015 ei->i_dtime = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001016 ei->i_block_group = group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04001017 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001018
Mingming Cao617ba132006-10-11 01:20:53 -07001019 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001020 if (IS_DIRSYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05001021 ext4_handle_sync(handle);
Al Viro6b38e842008-12-30 02:03:31 -05001022 if (insert_inode_locked(inode) < 0) {
1023 err = -EINVAL;
1024 goto fail_drop;
1025 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001026 spin_lock(&sbi->s_next_gen_lock);
1027 inode->i_generation = sbi->s_next_generation++;
1028 spin_unlock(&sbi->s_next_gen_lock);
1029
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001030 ei->i_state_flags = 0;
1031 ext4_set_inode_state(inode, EXT4_STATE_NEW);
Kalpak Shahef7f3832007-07-18 09:15:20 -04001032
1033 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001034
1035 ret = inode;
Christoph Hellwig871a2932010-03-03 09:05:07 -05001036 dquot_initialize(inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -05001037 err = dquot_alloc_inode(inode);
1038 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001039 goto fail_drop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001040
Mingming Cao617ba132006-10-11 01:20:53 -07001041 err = ext4_init_acl(handle, inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001042 if (err)
1043 goto fail_free_drop;
1044
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001045 err = ext4_init_security(handle, inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001046 if (err)
1047 goto fail_free_drop;
1048
Theodore Ts'o83982b62009-01-06 14:53:16 -05001049 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Eric Sandeene4079a12008-07-11 19:27:31 -04001050 /* set extent flag only for directory, file and normal symlink*/
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04001051 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001052 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001053 ext4_ext_tree_init(handle, inode);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001054 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001055 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001056
Aneesh Kumar K.V8753e882008-04-29 22:00:36 -04001057 err = ext4_mark_inode_dirty(handle, inode);
1058 if (err) {
1059 ext4_std_error(sb, err);
1060 goto fail_free_drop;
1061 }
1062
Mingming Cao617ba132006-10-11 01:20:53 -07001063 ext4_debug("allocating inode %lu\n", inode->i_ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001064 trace_ext4_allocate_inode(inode, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001065 goto really_out;
1066fail:
Mingming Cao617ba132006-10-11 01:20:53 -07001067 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001068out:
1069 iput(inode);
1070 ret = ERR_PTR(err);
1071really_out:
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001072 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001073 return ret;
1074
1075fail_free_drop:
Christoph Hellwig63936dd2010-03-03 09:05:01 -05001076 dquot_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001077
1078fail_drop:
Christoph Hellwig9f754752010-03-03 09:05:05 -05001079 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001080 inode->i_flags |= S_NOQUOTA;
1081 inode->i_nlink = 0;
Al Viro6b38e842008-12-30 02:03:31 -05001082 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001083 iput(inode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001084 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001085 return ERR_PTR(err);
1086}
1087
1088/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -07001089struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001090{
Mingming Cao617ba132006-10-11 01:20:53 -07001091 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Avantika Mathurfd2d4292008-01-28 23:58:27 -05001092 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001093 int bit;
David Howells1d1fe1e2008-02-07 00:15:37 -08001094 struct buffer_head *bitmap_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001095 struct inode *inode = NULL;
David Howells1d1fe1e2008-02-07 00:15:37 -08001096 long err = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001097
1098 /* Error cases - e2fsck has already cleaned up for us */
1099 if (ino > max_ino) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001100 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001101 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001102 }
1103
Mingming Cao617ba132006-10-11 01:20:53 -07001104 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1105 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001106 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001107 if (!bitmap_bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001108 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001109 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001110 }
1111
1112 /* Having the inode bit set should be a 100% indicator that this
1113 * is a valid orphan (no e2fsck run on fs). Orphans also include
1114 * inodes that were being truncated, so we can't check i_nlink==0.
1115 */
David Howells1d1fe1e2008-02-07 00:15:37 -08001116 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1117 goto bad_orphan;
1118
1119 inode = ext4_iget(sb, ino);
1120 if (IS_ERR(inode))
1121 goto iget_failed;
1122
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001123 /*
1124 * If the orphans has i_nlinks > 0 then it should be able to be
1125 * truncated, otherwise it won't be removed from the orphan list
1126 * during processing and an infinite loop will result.
1127 */
1128 if (inode->i_nlink && !ext4_can_truncate(inode))
1129 goto bad_orphan;
1130
David Howells1d1fe1e2008-02-07 00:15:37 -08001131 if (NEXT_ORPHAN(inode) > max_ino)
1132 goto bad_orphan;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001133 brelse(bitmap_bh);
1134 return inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08001135
1136iget_failed:
1137 err = PTR_ERR(inode);
1138 inode = NULL;
1139bad_orphan:
Eric Sandeen12062dd2010-02-15 14:19:27 -05001140 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001141 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1142 bit, (unsigned long long)bitmap_bh->b_blocknr,
1143 ext4_test_bit(bit, bitmap_bh->b_data));
1144 printk(KERN_NOTICE "inode=%p\n", inode);
1145 if (inode) {
1146 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1147 is_bad_inode(inode));
1148 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1149 NEXT_ORPHAN(inode));
1150 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001151 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
David Howells1d1fe1e2008-02-07 00:15:37 -08001152 /* Avoid freeing blocks if we got a bad deleted inode */
1153 if (inode->i_nlink == 0)
1154 inode->i_blocks = 0;
1155 iput(inode);
1156 }
1157 brelse(bitmap_bh);
1158error:
1159 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001160}
1161
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001162unsigned long ext4_count_free_inodes(struct super_block *sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001163{
1164 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001165 struct ext4_group_desc *gdp;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001166 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001167#ifdef EXT4FS_DEBUG
1168 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001169 unsigned long bitmap_count, x;
1170 struct buffer_head *bitmap_bh = NULL;
1171
Mingming Cao617ba132006-10-11 01:20:53 -07001172 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001173 desc_count = 0;
1174 bitmap_count = 0;
1175 gdp = NULL;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001176 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001177 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001178 if (!gdp)
1179 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001180 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001181 brelse(bitmap_bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001182 bitmap_bh = ext4_read_inode_bitmap(sb, i);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001183 if (!bitmap_bh)
1184 continue;
1185
Mingming Cao617ba132006-10-11 01:20:53 -07001186 x = ext4_count_free(bitmap_bh, 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
1230 * block ext4_claim_inode until we are finished.
1231 */
1232extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1233 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
1260 handle = ext4_journal_start_sb(sb, 1);
1261 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)) {
1278 ext4_error(sb, "Something is wrong with group %u\n"
1279 "Used itable blocks: %d"
1280 "itable unused count: %u\n",
1281 group, used_blks,
1282 ext4_itable_unused_count(sb, gdp));
1283 ret = 1;
1284 goto out;
1285 }
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);
1315 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1316 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}