blob: 29e6dc7299b8b6186cf46a39deac0a88594b6028 [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 */
53void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
54{
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 */
Avantika Mathurfd2d4292008-01-28 23:58:27 -050068unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
69 ext4_group_t block_group,
Andreas Dilger717d50e2007-10-16 18:38:25 -040070 struct ext4_group_desc *gdp)
71{
72 struct ext4_sb_info *sbi = EXT4_SB(sb);
73
74 J_ASSERT_BH(bh, buffer_locked(bh));
75
76 /* If checksum is bad mark all blocks and inodes use to prevent
77 * allocation, essentially implementing a per-group read-only flag. */
78 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
Theodore Ts'oa9df9a42009-01-05 22:18:16 -050079 ext4_error(sb, __func__, "Checksum bad for group %u",
Andreas Dilger717d50e2007-10-16 18:38:25 -040080 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);
Aneesh Kumar K.V648f5872009-01-05 21:46:04 -050089 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;
111 bitmap_blk = ext4_inode_bitmap(sb, desc);
112 bh = sb_getblk(sb, bitmap_blk);
113 if (unlikely(!bh)) {
114 ext4_error(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700115 "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 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400128 ext4_lock_group(sb, block_group);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400129 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
130 ext4_init_inode_bitmap(sb, bh, block_group, desc);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500131 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400132 set_buffer_uptodate(bh);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400133 ext4_unlock_group(sb, block_group);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500134 unlock_buffer(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400135 return bh;
136 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400137 ext4_unlock_group(sb, block_group);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500138 if (buffer_uptodate(bh)) {
139 /*
140 * if not uninit if bh is uptodate,
141 * bitmap is also uptodate
142 */
143 set_bitmap_uptodate(bh);
144 unlock_buffer(bh);
145 return bh;
146 }
147 /*
148 * submit the buffer_head for read. We can
149 * safely mark the bitmap as uptodate now.
150 * We do it here so the bitmap uptodate bit
151 * get set with buffer lock held.
152 */
153 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400154 if (bh_submit_read(bh) < 0) {
155 put_bh(bh);
156 ext4_error(sb, __func__,
157 "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500158 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400159 block_group, bitmap_blk);
160 return NULL;
161 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700162 return bh;
163}
164
165/*
166 * NOTE! When we get the inode, we're the only people
167 * that have access to it, and as such there are no
168 * race conditions we have to worry about. The inode
169 * is not on the hash-lists, and it cannot be reached
170 * through the filesystem because the directory entry
171 * has been deleted earlier.
172 *
173 * HOWEVER: we must make sure that we get no aliases,
174 * which means that we have to call "clear_inode()"
175 * _before_ we mark the inode not in use in the inode
176 * bitmaps. Otherwise a newly created file might use
177 * the same inode number (not actually the same pointer
178 * though), and then we'd have two inodes sharing the
179 * same inode number and space on the harddisk.
180 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400181void ext4_free_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700182{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400183 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700184 int is_directory;
185 unsigned long ino;
186 struct buffer_head *bitmap_bh = NULL;
187 struct buffer_head *bh2;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500188 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700189 unsigned long bit;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400190 struct ext4_group_desc *gdp;
191 struct ext4_super_block *es;
Mingming Cao617ba132006-10-11 01:20:53 -0700192 struct ext4_sb_info *sbi;
Eric Sandeen7ce9d5d2009-03-04 18:38:18 -0500193 int fatal = 0, err, count, cleared;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700194
195 if (atomic_read(&inode->i_count) > 1) {
Theodore Ts'o47760042008-09-08 23:00:52 -0400196 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
197 atomic_read(&inode->i_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700198 return;
199 }
200 if (inode->i_nlink) {
Theodore Ts'o47760042008-09-08 23:00:52 -0400201 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
202 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700203 return;
204 }
205 if (!sb) {
Theodore Ts'o47760042008-09-08 23:00:52 -0400206 printk(KERN_ERR "ext4_free_inode: inode on "
207 "nonexistent device\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700208 return;
209 }
Mingming Cao617ba132006-10-11 01:20:53 -0700210 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700211
212 ino = inode->i_ino;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400213 ext4_debug("freeing inode %lu\n", ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400214 trace_ext4_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700215
216 /*
217 * Note: we must free any quota before locking the superblock,
218 * as writing the quota to disk may need the lock as well.
219 */
Jan Karaa269eb12009-01-26 17:04:39 +0100220 vfs_dq_init(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700221 ext4_xattr_delete_inode(handle, inode);
Jan Karaa269eb12009-01-26 17:04:39 +0100222 vfs_dq_free_inode(inode);
223 vfs_dq_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700224
225 is_directory = S_ISDIR(inode->i_mode);
226
227 /* Do this BEFORE marking the inode not in use or returning an error */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400228 clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700229
Mingming Cao617ba132006-10-11 01:20:53 -0700230 es = EXT4_SB(sb)->s_es;
231 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400232 ext4_error(sb, "ext4_free_inode",
233 "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
247 /* Ok, now we can actually update the inode bitmaps.. */
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400248 cleared = ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group),
249 bit, bitmap_bh->b_data);
Eric Sandeen7ce9d5d2009-03-04 18:38:18 -0500250 if (!cleared)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400251 ext4_error(sb, "ext4_free_inode",
252 "bit already cleared for inode %lu", ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700253 else {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400254 gdp = ext4_get_group_desc(sb, block_group, &bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700255
256 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700257 fatal = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700258 if (fatal) goto error_return;
259
260 if (gdp) {
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400261 ext4_lock_group(sb, block_group);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500262 count = ext4_free_inodes_count(sb, gdp) + 1;
263 ext4_free_inodes_set(sb, gdp, count);
264 if (is_directory) {
265 count = ext4_used_dirs_count(sb, gdp) - 1;
266 ext4_used_dirs_set(sb, gdp, count);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500267 if (sbi->s_log_groups_per_flex) {
268 ext4_group_t f;
269
270 f = ext4_flex_group(sbi, block_group);
271 atomic_dec(&sbi->s_flex_groups[f].free_inodes);
272 }
273
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500274 }
Andreas Dilger717d50e2007-10-16 18:38:25 -0400275 gdp->bg_checksum = ext4_group_desc_csum(sbi,
276 block_group, gdp);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400277 ext4_unlock_group(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700278 percpu_counter_inc(&sbi->s_freeinodes_counter);
279 if (is_directory)
280 percpu_counter_dec(&sbi->s_dirs_counter);
281
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400282 if (sbi->s_log_groups_per_flex) {
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500283 ext4_group_t f;
284
285 f = ext4_flex_group(sbi, block_group);
286 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400287 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700288 }
Frank Mayhar03901312009-01-07 00:06:22 -0500289 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
290 err = ext4_handle_dirty_metadata(handle, NULL, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291 if (!fatal) fatal = err;
292 }
Frank Mayhar03901312009-01-07 00:06:22 -0500293 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
294 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700295 if (!fatal)
296 fatal = err;
297 sb->s_dirt = 1;
298error_return:
299 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700300 ext4_std_error(sb, fatal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700301}
302
303/*
304 * There are two policies for allocating an inode. If the new inode is
305 * a directory, then a forward search is made for a block group with both
306 * free space and a low directory-to-inode ratio; if that fails, then of
307 * the groups with above-average free space, that group with the fewest
308 * directories already is chosen.
309 *
310 * For other inodes, search forward from the parent directory\'s block
311 * group to find a free inode.
312 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500313static int find_group_dir(struct super_block *sb, struct inode *parent,
314 ext4_group_t *best_group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700315{
Theodore Ts'o8df96752009-05-01 08:50:38 -0400316 ext4_group_t ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700317 unsigned int freei, avefreei;
Mingming Cao617ba132006-10-11 01:20:53 -0700318 struct ext4_group_desc *desc, *best_desc = NULL;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500319 ext4_group_t group;
320 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700321
Mingming Cao617ba132006-10-11 01:20:53 -0700322 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700323 avefreei = freei / ngroups;
324
325 for (group = 0; group < ngroups; group++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400326 desc = ext4_get_group_desc(sb, group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500327 if (!desc || !ext4_free_inodes_count(sb, desc))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700328 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500329 if (ext4_free_inodes_count(sb, desc) < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700330 continue;
331 if (!best_desc ||
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500332 (ext4_free_blks_count(sb, desc) >
333 ext4_free_blks_count(sb, best_desc))) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500334 *best_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700335 best_desc = desc;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500336 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700337 }
338 }
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500339 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700340}
341
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400342#define free_block_ratio 10
343
344static int find_group_flex(struct super_block *sb, struct inode *parent,
345 ext4_group_t *best_group)
346{
347 struct ext4_sb_info *sbi = EXT4_SB(sb);
348 struct ext4_group_desc *desc;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400349 struct flex_groups *flex_group = sbi->s_flex_groups;
350 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
351 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400352 ext4_group_t ngroups = ext4_get_groups_count(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400353 int flex_size = ext4_flex_bg_size(sbi);
354 ext4_group_t best_flex = parent_fbg_group;
355 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
356 int flexbg_free_blocks;
357 int flex_freeb_ratio;
358 ext4_group_t n_fbg_groups;
359 ext4_group_t i;
360
Theodore Ts'o8df96752009-05-01 08:50:38 -0400361 n_fbg_groups = (ngroups + flex_size - 1) >>
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400362 sbi->s_log_groups_per_flex;
363
364find_close_to_parent:
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500365 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400366 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500367 if (atomic_read(&flex_group[best_flex].free_inodes) &&
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400368 flex_freeb_ratio > free_block_ratio)
369 goto found_flexbg;
370
371 if (best_flex && best_flex == parent_fbg_group) {
372 best_flex--;
373 goto find_close_to_parent;
374 }
375
376 for (i = 0; i < n_fbg_groups; i++) {
377 if (i == parent_fbg_group || i == parent_fbg_group - 1)
378 continue;
379
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500380 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400381 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
382
383 if (flex_freeb_ratio > free_block_ratio &&
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500384 (atomic_read(&flex_group[i].free_inodes))) {
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400385 best_flex = i;
386 goto found_flexbg;
387 }
388
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500389 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
390 ((atomic_read(&flex_group[i].free_blocks) >
391 atomic_read(&flex_group[best_flex].free_blocks)) &&
392 atomic_read(&flex_group[i].free_inodes)))
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400393 best_flex = i;
394 }
395
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500396 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
397 !atomic_read(&flex_group[best_flex].free_blocks))
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400398 return -1;
399
400found_flexbg:
401 for (i = best_flex * flex_size; i < ngroups &&
402 i < (best_flex + 1) * flex_size; i++) {
Theodore Ts'o88b6edd2009-05-25 11:50:39 -0400403 desc = ext4_get_group_desc(sb, i, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500404 if (ext4_free_inodes_count(sb, desc)) {
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400405 *best_group = i;
406 goto out;
407 }
408 }
409
410 return -1;
411out:
412 return 0;
413}
414
Theodore Ts'oa4912122009-03-12 12:18:34 -0400415struct orlov_stats {
416 __u32 free_inodes;
417 __u32 free_blocks;
418 __u32 used_dirs;
419};
420
421/*
422 * Helper function for Orlov's allocator; returns critical information
423 * for a particular block group or flex_bg. If flex_size is 1, then g
424 * is a block group number; otherwise it is flex_bg number.
425 */
426void get_orlov_stats(struct super_block *sb, ext4_group_t g,
427 int flex_size, struct orlov_stats *stats)
428{
429 struct ext4_group_desc *desc;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500430 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400431
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500432 if (flex_size > 1) {
433 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
434 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
435 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
436 return;
437 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400438
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500439 desc = ext4_get_group_desc(sb, g, NULL);
440 if (desc) {
441 stats->free_inodes = ext4_free_inodes_count(sb, desc);
442 stats->free_blocks = ext4_free_blks_count(sb, desc);
443 stats->used_dirs = ext4_used_dirs_count(sb, desc);
444 } else {
445 stats->free_inodes = 0;
446 stats->free_blocks = 0;
447 stats->used_dirs = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400448 }
449}
450
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700451/*
452 * Orlov's allocator for directories.
453 *
454 * We always try to spread first-level directories.
455 *
456 * If there are blockgroups with both free inodes and free blocks counts
457 * not worse than average we return one with smallest directory count.
458 * Otherwise we simply return a random group.
459 *
460 * For the rest rules look so:
461 *
462 * It's OK to put directory into a group unless
463 * it has too many directories already (max_dirs) or
464 * it has too few free inodes left (min_inodes) or
465 * it has too few free blocks left (min_blocks) or
Benoit Boissinot1cc8dcf2008-04-21 22:45:55 +0000466 * Parent's group is preferred, if it doesn't satisfy these
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700467 * conditions we search cyclically through the rest. If none
468 * of the groups look good we just look for a group with more
469 * free inodes than average (starting at parent's group).
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700470 */
471
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500472static int find_group_orlov(struct super_block *sb, struct inode *parent,
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400473 ext4_group_t *group, int mode,
474 const struct qstr *qstr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700475{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500476 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Mingming Cao617ba132006-10-11 01:20:53 -0700477 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400478 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700479 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700480 unsigned int freei, avefreei;
Mingming Cao617ba132006-10-11 01:20:53 -0700481 ext4_fsblk_t freeb, avefreeb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700482 unsigned int ndirs;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400483 int max_dirs, min_inodes;
Mingming Cao617ba132006-10-11 01:20:53 -0700484 ext4_grpblk_t min_blocks;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400485 ext4_group_t i, grp, g, ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700486 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400487 struct orlov_stats stats;
488 int flex_size = ext4_flex_bg_size(sbi);
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400489 struct dx_hash_info hinfo;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400490
Theodore Ts'o8df96752009-05-01 08:50:38 -0400491 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400492 if (flex_size > 1) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400493 ngroups = (real_ngroups + flex_size - 1) >>
Theodore Ts'oa4912122009-03-12 12:18:34 -0400494 sbi->s_log_groups_per_flex;
495 parent_group >>= sbi->s_log_groups_per_flex;
496 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700497
498 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
499 avefreei = freei / ngroups;
500 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700501 avefreeb = freeb;
Andrew Mortonf4e5bc22006-10-11 01:21:19 -0700502 do_div(avefreeb, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700503 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
504
Theodore Ts'oa4912122009-03-12 12:18:34 -0400505 if (S_ISDIR(mode) &&
506 ((parent == sb->s_root->d_inode) ||
507 (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700508 int best_ndir = inodes_per_group;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500509 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700510
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400511 if (qstr) {
512 hinfo.hash_version = DX_HASH_HALF_MD4;
513 hinfo.seed = sbi->s_hash_seed;
514 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
515 grp = hinfo.hash;
516 } else
517 get_random_bytes(&grp, sizeof(grp));
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500518 parent_group = (unsigned)grp % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700519 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400520 g = (parent_group + i) % ngroups;
521 get_orlov_stats(sb, g, flex_size, &stats);
522 if (!stats.free_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700523 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400524 if (stats.used_dirs >= best_ndir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700525 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400526 if (stats.free_inodes < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700527 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400528 if (stats.free_blocks < avefreeb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700529 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400530 grp = g;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500531 ret = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400532 best_ndir = stats.used_dirs;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700533 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400534 if (ret)
535 goto fallback;
536 found_flex_bg:
537 if (flex_size == 1) {
538 *group = grp;
539 return 0;
540 }
541
542 /*
543 * We pack inodes at the beginning of the flexgroup's
544 * inode tables. Block allocation decisions will do
545 * something similar, although regular files will
546 * start at 2nd block group of the flexgroup. See
547 * ext4_ext_find_goal() and ext4_find_near().
548 */
549 grp *= flex_size;
550 for (i = 0; i < flex_size; i++) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400551 if (grp+i >= real_ngroups)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400552 break;
553 desc = ext4_get_group_desc(sb, grp+i, NULL);
554 if (desc && ext4_free_inodes_count(sb, desc)) {
555 *group = grp+i;
556 return 0;
557 }
558 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700559 goto fallback;
560 }
561
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700562 max_dirs = ndirs / ngroups + inodes_per_group / 16;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400563 min_inodes = avefreei - inodes_per_group*flex_size / 4;
564 if (min_inodes < 1)
565 min_inodes = 1;
566 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700567
Theodore Ts'oa4912122009-03-12 12:18:34 -0400568 /*
569 * Start looking in the flex group where we last allocated an
570 * inode for this parent directory
571 */
572 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
573 parent_group = EXT4_I(parent)->i_last_alloc_group;
574 if (flex_size > 1)
575 parent_group >>= sbi->s_log_groups_per_flex;
576 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577
578 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400579 grp = (parent_group + i) % ngroups;
580 get_orlov_stats(sb, grp, flex_size, &stats);
581 if (stats.used_dirs >= max_dirs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700582 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400583 if (stats.free_inodes < min_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700584 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400585 if (stats.free_blocks < min_blocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700586 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400587 goto found_flex_bg;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700588 }
589
590fallback:
Theodore Ts'o8df96752009-05-01 08:50:38 -0400591 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400592 avefreei = freei / ngroups;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400593fallback_retry:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400594 parent_group = EXT4_I(parent)->i_block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700595 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400596 grp = (parent_group + i) % ngroups;
597 desc = ext4_get_group_desc(sb, grp, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500598 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'oa4912122009-03-12 12:18:34 -0400599 ext4_free_inodes_count(sb, desc) >= avefreei) {
600 *group = grp;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500601 return 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400602 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700603 }
604
605 if (avefreei) {
606 /*
607 * The free-inodes counter is approximate, and for really small
608 * filesystems the above test can fail to find any blockgroups
609 */
610 avefreei = 0;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400611 goto fallback_retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700612 }
613
614 return -1;
615}
616
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500617static int find_group_other(struct super_block *sb, struct inode *parent,
Theodore Ts'oa4912122009-03-12 12:18:34 -0400618 ext4_group_t *group, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700619{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500620 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400621 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700622 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400623 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
624
625 /*
626 * Try to place the inode is the same flex group as its
627 * parent. If we can't find space, use the Orlov algorithm to
628 * find another flex group, and store that information in the
629 * parent directory's inode information so that use that flex
630 * group for future allocations.
631 */
632 if (flex_size > 1) {
633 int retry = 0;
634
635 try_again:
636 parent_group &= ~(flex_size-1);
637 last = parent_group + flex_size;
638 if (last > ngroups)
639 last = ngroups;
640 for (i = parent_group; i < last; i++) {
641 desc = ext4_get_group_desc(sb, i, NULL);
642 if (desc && ext4_free_inodes_count(sb, desc)) {
643 *group = i;
644 return 0;
645 }
646 }
647 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
648 retry = 1;
649 parent_group = EXT4_I(parent)->i_last_alloc_group;
650 goto try_again;
651 }
652 /*
653 * If this didn't work, use the Orlov search algorithm
654 * to find a new flex group; we pass in the mode to
655 * avoid the topdir algorithms.
656 */
657 *group = parent_group + flex_size;
658 if (*group > ngroups)
659 *group = 0;
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400660 return find_group_orlov(sb, parent, group, mode, 0);
Theodore Ts'oa4912122009-03-12 12:18:34 -0400661 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700662
663 /*
664 * Try to place the inode in its parent directory
665 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500666 *group = parent_group;
667 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500668 if (desc && ext4_free_inodes_count(sb, desc) &&
669 ext4_free_blks_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500670 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700671
672 /*
673 * We're going to place this inode in a different blockgroup from its
674 * parent. We want to cause files in a common directory to all land in
675 * the same blockgroup. But we want files which are in a different
676 * directory which shares a blockgroup with our parent to land in a
677 * different blockgroup.
678 *
679 * So add our directory's i_ino into the starting point for the hash.
680 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500681 *group = (*group + parent->i_ino) % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700682
683 /*
684 * Use a quadratic hash to find a group with a free inode and some free
685 * blocks.
686 */
687 for (i = 1; i < ngroups; i <<= 1) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500688 *group += i;
689 if (*group >= ngroups)
690 *group -= ngroups;
691 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500692 if (desc && ext4_free_inodes_count(sb, desc) &&
693 ext4_free_blks_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500694 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700695 }
696
697 /*
698 * That failed: try linear search for a free inode, even if that group
699 * has no free blocks.
700 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500701 *group = parent_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700702 for (i = 0; i < ngroups; i++) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500703 if (++*group >= ngroups)
704 *group = 0;
705 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500706 if (desc && ext4_free_inodes_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500707 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700708 }
709
710 return -1;
711}
712
713/*
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500714 * claim the inode from the inode bitmap. If the group
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400715 * is uninit we need to take the groups's ext4_group_lock
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500716 * and clear the uninit flag. The inode bitmap update
717 * and group desc uninit flag clear should be done
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400718 * after holding ext4_group_lock so that ext4_read_inode_bitmap
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500719 * doesn't race with the ext4_claim_inode
720 */
721static int ext4_claim_inode(struct super_block *sb,
722 struct buffer_head *inode_bitmap_bh,
723 unsigned long ino, ext4_group_t group, int mode)
724{
725 int free = 0, retval = 0, count;
726 struct ext4_sb_info *sbi = EXT4_SB(sb);
727 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
728
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400729 ext4_lock_group(sb, group);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500730 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
731 /* not a free inode */
732 retval = 1;
733 goto err_ret;
734 }
735 ino++;
736 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
737 ino > EXT4_INODES_PER_GROUP(sb)) {
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400738 ext4_unlock_group(sb, group);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500739 ext4_error(sb, __func__,
740 "reserved inode or inode > inodes count - "
741 "block_group = %u, inode=%lu", group,
742 ino + group * EXT4_INODES_PER_GROUP(sb));
743 return 1;
744 }
745 /* If we didn't allocate from within the initialized part of the inode
746 * table then we need to initialize up to this inode. */
747 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
748
749 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
750 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
751 /* When marking the block group with
752 * ~EXT4_BG_INODE_UNINIT we don't want to depend
753 * on the value of bg_itable_unused even though
754 * mke2fs could have initialized the same for us.
755 * Instead we calculated the value below
756 */
757
758 free = 0;
759 } else {
760 free = EXT4_INODES_PER_GROUP(sb) -
761 ext4_itable_unused_count(sb, gdp);
762 }
763
764 /*
765 * Check the relative inode number against the last used
766 * relative inode number in this group. if it is greater
767 * we need to update the bg_itable_unused count
768 *
769 */
770 if (ino > free)
771 ext4_itable_unused_set(sb, gdp,
772 (EXT4_INODES_PER_GROUP(sb) - ino));
773 }
774 count = ext4_free_inodes_count(sb, gdp) - 1;
775 ext4_free_inodes_set(sb, gdp, count);
776 if (S_ISDIR(mode)) {
777 count = ext4_used_dirs_count(sb, gdp) + 1;
778 ext4_used_dirs_set(sb, gdp, count);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500779 if (sbi->s_log_groups_per_flex) {
780 ext4_group_t f = ext4_flex_group(sbi, group);
781
782 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
783 }
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500784 }
785 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
786err_ret:
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400787 ext4_unlock_group(sb, group);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500788 return retval;
789}
790
791/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700792 * There are two policies for allocating an inode. If the new inode is
793 * a directory, then a forward search is made for a block group with both
794 * free space and a low directory-to-inode ratio; if that fails, then of
795 * the groups with above-average free space, that group with the fewest
796 * directories already is chosen.
797 *
798 * For other inodes, search forward from the parent directory's block
799 * group to find a free inode.
800 */
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400801struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
Andreas Dilger11013912009-06-13 11:45:35 -0400802 const struct qstr *qstr, __u32 goal)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700803{
804 struct super_block *sb;
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500805 struct buffer_head *inode_bitmap_bh = NULL;
806 struct buffer_head *group_desc_bh;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400807 ext4_group_t ngroups, group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700808 unsigned long ino = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400809 struct inode *inode;
810 struct ext4_group_desc *gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700811 struct ext4_inode_info *ei;
812 struct ext4_sb_info *sbi;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500813 int ret2, err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700814 struct inode *ret;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500815 ext4_group_t i;
816 int free = 0;
Theodore Ts'o2842c3b2009-03-12 12:20:01 -0400817 static int once = 1;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400818 ext4_group_t flex_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700819
820 /* Cannot create files in a deleted directory */
821 if (!dir || !dir->i_nlink)
822 return ERR_PTR(-EPERM);
823
824 sb = dir->i_sb;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400825 ngroups = ext4_get_groups_count(sb);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400826 trace_ext4_request_inode(dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700827 inode = new_inode(sb);
828 if (!inode)
829 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700830 ei = EXT4_I(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700831 sbi = EXT4_SB(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400832
Andreas Dilger11013912009-06-13 11:45:35 -0400833 if (!goal)
834 goal = sbi->s_inode_goal;
835
Johann Lombardie6462862009-07-05 23:45:11 -0400836 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
Andreas Dilger11013912009-06-13 11:45:35 -0400837 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
838 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
839 ret2 = 0;
840 goto got_group;
841 }
842
Theodore Ts'oa4912122009-03-12 12:18:34 -0400843 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400844 ret2 = find_group_flex(sb, dir, &group);
Theodore Ts'o05bf9e82009-02-21 12:13:24 -0500845 if (ret2 == -1) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400846 ret2 = find_group_other(sb, dir, &group, mode);
Chuck Ebbert6b82f3c2009-04-14 07:37:40 -0400847 if (ret2 == 0 && once) {
Theodore Ts'o2842c3b2009-03-12 12:20:01 -0400848 once = 0;
Theodore Ts'o05bf9e82009-02-21 12:13:24 -0500849 printk(KERN_NOTICE "ext4: find_group_flex "
850 "failed, fallback succeeded dir %lu\n",
851 dir->i_ino);
Chuck Ebbert6b82f3c2009-04-14 07:37:40 -0400852 }
Theodore Ts'o05bf9e82009-02-21 12:13:24 -0500853 }
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400854 goto got_group;
855 }
856
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700857 if (S_ISDIR(mode)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400858 if (test_opt(sb, OLDALLOC))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500859 ret2 = find_group_dir(sb, dir, &group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700860 else
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400861 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700862 } else
Theodore Ts'oa4912122009-03-12 12:18:34 -0400863 ret2 = find_group_other(sb, dir, &group, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700864
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400865got_group:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400866 EXT4_I(dir)->i_last_alloc_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700867 err = -ENOSPC;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500868 if (ret2 == -1)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700869 goto out;
870
Andreas Dilger11013912009-06-13 11:45:35 -0400871 for (i = 0; i < ngroups; i++, ino = 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700872 err = -EIO;
873
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500874 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700875 if (!gdp)
876 goto fail;
877
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500878 brelse(inode_bitmap_bh);
879 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
880 if (!inode_bitmap_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700881 goto fail;
882
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700883repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700884 ino = ext4_find_next_zero_bit((unsigned long *)
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500885 inode_bitmap_bh->b_data,
886 EXT4_INODES_PER_GROUP(sb), ino);
887
Mingming Cao617ba132006-10-11 01:20:53 -0700888 if (ino < EXT4_INODES_PER_GROUP(sb)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700889
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500890 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
891 err = ext4_journal_get_write_access(handle,
892 inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700893 if (err)
894 goto fail;
895
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500896 BUFFER_TRACE(group_desc_bh, "get_write_access");
897 err = ext4_journal_get_write_access(handle,
898 group_desc_bh);
899 if (err)
900 goto fail;
901 if (!ext4_claim_inode(sb, inode_bitmap_bh,
902 ino, group, mode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700903 /* we won it */
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500904 BUFFER_TRACE(inode_bitmap_bh,
Frank Mayhar03901312009-01-07 00:06:22 -0500905 "call ext4_handle_dirty_metadata");
906 err = ext4_handle_dirty_metadata(handle,
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500907 inode,
908 inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700909 if (err)
910 goto fail;
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500911 /* zero bit is inode number 1*/
912 ino++;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700913 goto got;
914 }
915 /* we lost it */
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500916 ext4_handle_release_buffer(handle, inode_bitmap_bh);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500917 ext4_handle_release_buffer(handle, group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700918
Mingming Cao617ba132006-10-11 01:20:53 -0700919 if (++ino < EXT4_INODES_PER_GROUP(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700920 goto repeat_in_this_group;
921 }
922
923 /*
924 * This case is possible in concurrent environment. It is very
925 * rare. We cannot repeat the find_group_xxx() call because
926 * that will simply return the same blockgroup, because the
927 * group descriptor metadata has not yet been updated.
928 * So we just go onto the next blockgroup.
929 */
Theodore Ts'o8df96752009-05-01 08:50:38 -0400930 if (++group == ngroups)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700931 group = 0;
932 }
933 err = -ENOSPC;
934 goto out;
935
936got:
Andreas Dilger717d50e2007-10-16 18:38:25 -0400937 /* We may have to initialize the block bitmap if it isn't already */
938 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
939 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500940 struct buffer_head *block_bitmap_bh;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400941
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500942 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
943 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
944 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400945 if (err) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500946 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400947 goto fail;
948 }
949
950 free = 0;
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400951 ext4_lock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400952 /* recheck and clear flag under lock if we still need to */
953 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Andreas Dilger717d50e2007-10-16 18:38:25 -0400954 free = ext4_free_blocks_after_init(sb, group, gdp);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500955 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500956 ext4_free_blks_set(sb, gdp, free);
Frederic Bohe23712a92008-11-07 09:21:01 -0500957 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
958 gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400959 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400960 ext4_unlock_group(sb, group);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400961
962 /* Don't need to dirty bitmap block if we didn't change it */
963 if (free) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500964 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
Frank Mayhar03901312009-01-07 00:06:22 -0500965 err = ext4_handle_dirty_metadata(handle,
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500966 NULL, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400967 }
968
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500969 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400970 if (err)
971 goto fail;
972 }
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500973 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
974 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
Aneesh Kumar K.V39341862009-01-05 21:38:14 -0500975 if (err)
976 goto fail;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700977
978 percpu_counter_dec(&sbi->s_freeinodes_counter);
979 if (S_ISDIR(mode))
980 percpu_counter_inc(&sbi->s_dirs_counter);
981 sb->s_dirt = 1;
982
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400983 if (sbi->s_log_groups_per_flex) {
984 flex_group = ext4_flex_group(sbi, group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500985 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400986 }
987
David Howells4c9c5442008-11-14 10:38:51 +1100988 inode->i_uid = current_fsuid();
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400989 if (test_opt(sb, GRPID))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700990 inode->i_gid = dir->i_gid;
991 else if (dir->i_mode & S_ISGID) {
992 inode->i_gid = dir->i_gid;
993 if (S_ISDIR(mode))
994 mode |= S_ISGID;
995 } else
David Howells4c9c5442008-11-14 10:38:51 +1100996 inode->i_gid = current_fsgid();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700997 inode->i_mode = mode;
998
Andreas Dilger717d50e2007-10-16 18:38:25 -0400999 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001000 /* This is the optimal IO size (for stat), not the fs block size */
1001 inode->i_blocks = 0;
Kalpak Shahef7f3832007-07-18 09:15:20 -04001002 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1003 ext4_current_time(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001004
1005 memset(ei->i_data, 0, sizeof(ei->i_data));
1006 ei->i_dir_start_lookup = 0;
1007 ei->i_disksize = 0;
1008
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001009 /*
Duane Griffin2dc6b0d2009-02-15 18:09:20 -05001010 * Don't inherit extent flag from directory, amongst others. We set
1011 * extent flag on newly created directory and file only if -o extent
1012 * mount option is specified
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001013 */
Duane Griffin2dc6b0d2009-02-15 18:09:20 -05001014 ei->i_flags =
1015 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001016 ei->i_file_acl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001017 ei->i_dtime = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001018 ei->i_block_group = group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04001019 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001020
Mingming Cao617ba132006-10-11 01:20:53 -07001021 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001022 if (IS_DIRSYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05001023 ext4_handle_sync(handle);
Al Viro6b38e842008-12-30 02:03:31 -05001024 if (insert_inode_locked(inode) < 0) {
1025 err = -EINVAL;
1026 goto fail_drop;
1027 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001028 spin_lock(&sbi->s_next_gen_lock);
1029 inode->i_generation = sbi->s_next_generation++;
1030 spin_unlock(&sbi->s_next_gen_lock);
1031
Mingming Cao617ba132006-10-11 01:20:53 -07001032 ei->i_state = 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;
Jan Karaa269eb12009-01-26 17:04:39 +01001037 if (vfs_dq_alloc_inode(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001038 err = -EDQUOT;
1039 goto fail_drop;
1040 }
1041
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
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001046 err = ext4_init_security(handle, inode, dir);
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)) {
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001053 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
1054 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
Aneesh Kumar K.V8753e882008-04-29 22:00:36 -04001058 err = ext4_mark_inode_dirty(handle, inode);
1059 if (err) {
1060 ext4_std_error(sb, err);
1061 goto fail_free_drop;
1062 }
1063
Mingming Cao617ba132006-10-11 01:20:53 -07001064 ext4_debug("allocating inode %lu\n", inode->i_ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001065 trace_ext4_allocate_inode(inode, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001066 goto really_out;
1067fail:
Mingming Cao617ba132006-10-11 01:20:53 -07001068 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001069out:
1070 iput(inode);
1071 ret = ERR_PTR(err);
1072really_out:
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001073 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001074 return ret;
1075
1076fail_free_drop:
Jan Karaa269eb12009-01-26 17:04:39 +01001077 vfs_dq_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001078
1079fail_drop:
Jan Karaa269eb12009-01-26 17:04:39 +01001080 vfs_dq_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001081 inode->i_flags |= S_NOQUOTA;
1082 inode->i_nlink = 0;
Al Viro6b38e842008-12-30 02:03:31 -05001083 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001084 iput(inode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001085 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001086 return ERR_PTR(err);
1087}
1088
1089/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -07001090struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001091{
Mingming Cao617ba132006-10-11 01:20:53 -07001092 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Avantika Mathurfd2d4292008-01-28 23:58:27 -05001093 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001094 int bit;
David Howells1d1fe1e2008-02-07 00:15:37 -08001095 struct buffer_head *bitmap_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001096 struct inode *inode = NULL;
David Howells1d1fe1e2008-02-07 00:15:37 -08001097 long err = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001098
1099 /* Error cases - e2fsck has already cleaned up for us */
1100 if (ino > max_ino) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001101 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001102 "bad orphan ino %lu! e2fsck was run?", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001103 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001104 }
1105
Mingming Cao617ba132006-10-11 01:20:53 -07001106 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1107 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001108 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001109 if (!bitmap_bh) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001110 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001111 "inode bitmap error for orphan %lu", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001112 goto error;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001113 }
1114
1115 /* Having the inode bit set should be a 100% indicator that this
1116 * is a valid orphan (no e2fsck run on fs). Orphans also include
1117 * inodes that were being truncated, so we can't check i_nlink==0.
1118 */
David Howells1d1fe1e2008-02-07 00:15:37 -08001119 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1120 goto bad_orphan;
1121
1122 inode = ext4_iget(sb, ino);
1123 if (IS_ERR(inode))
1124 goto iget_failed;
1125
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001126 /*
1127 * If the orphans has i_nlinks > 0 then it should be able to be
1128 * truncated, otherwise it won't be removed from the orphan list
1129 * during processing and an infinite loop will result.
1130 */
1131 if (inode->i_nlink && !ext4_can_truncate(inode))
1132 goto bad_orphan;
1133
David Howells1d1fe1e2008-02-07 00:15:37 -08001134 if (NEXT_ORPHAN(inode) > max_ino)
1135 goto bad_orphan;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001136 brelse(bitmap_bh);
1137 return inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08001138
1139iget_failed:
1140 err = PTR_ERR(inode);
1141 inode = NULL;
1142bad_orphan:
Harvey Harrison46e665e2008-04-17 10:38:59 -04001143 ext4_warning(sb, __func__,
David Howells1d1fe1e2008-02-07 00:15:37 -08001144 "bad orphan inode %lu! e2fsck was run?", ino);
1145 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1146 bit, (unsigned long long)bitmap_bh->b_blocknr,
1147 ext4_test_bit(bit, bitmap_bh->b_data));
1148 printk(KERN_NOTICE "inode=%p\n", inode);
1149 if (inode) {
1150 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1151 is_bad_inode(inode));
1152 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1153 NEXT_ORPHAN(inode));
1154 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001155 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
David Howells1d1fe1e2008-02-07 00:15:37 -08001156 /* Avoid freeing blocks if we got a bad deleted inode */
1157 if (inode->i_nlink == 0)
1158 inode->i_blocks = 0;
1159 iput(inode);
1160 }
1161 brelse(bitmap_bh);
1162error:
1163 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001164}
1165
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001166unsigned long ext4_count_free_inodes(struct super_block *sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001167{
1168 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001169 struct ext4_group_desc *gdp;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001170 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001171#ifdef EXT4FS_DEBUG
1172 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001173 unsigned long bitmap_count, x;
1174 struct buffer_head *bitmap_bh = NULL;
1175
Mingming Cao617ba132006-10-11 01:20:53 -07001176 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001177 desc_count = 0;
1178 bitmap_count = 0;
1179 gdp = NULL;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001180 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001181 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001182 if (!gdp)
1183 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001184 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001185 brelse(bitmap_bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001186 bitmap_bh = ext4_read_inode_bitmap(sb, i);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001187 if (!bitmap_bh)
1188 continue;
1189
Mingming Cao617ba132006-10-11 01:20:53 -07001190 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
Eric Sandeenc549a952008-01-28 23:58:27 -05001191 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001192 i, ext4_free_inodes_count(sb, gdp), x);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001193 bitmap_count += x;
1194 }
1195 brelse(bitmap_bh);
Theodore Ts'o47760042008-09-08 23:00:52 -04001196 printk(KERN_DEBUG "ext4_count_free_inodes: "
1197 "stored = %u, computed = %lu, %lu\n",
1198 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001199 return desc_count;
1200#else
1201 desc_count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001202 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001203 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001204 if (!gdp)
1205 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001206 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001207 cond_resched();
1208 }
1209 return desc_count;
1210#endif
1211}
1212
1213/* Called at mount-time, super-block is locked */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001214unsigned long ext4_count_dirs(struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001215{
1216 unsigned long count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001217 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001218
Theodore Ts'o8df96752009-05-01 08:50:38 -04001219 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001220 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001221 if (!gdp)
1222 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001223 count += ext4_used_dirs_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001224 }
1225 return count;
1226}