blob: 4f78e099de1d9438b8edc27fc72c4b2ccc6ac55b [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>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070017#include <linux/stat.h>
18#include <linux/string.h>
19#include <linux/quotaops.h>
20#include <linux/buffer_head.h>
21#include <linux/random.h>
22#include <linux/bitops.h>
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070023#include <linux/blkdev.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070024#include <asm/byteorder.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040025
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040026#include "ext4.h"
27#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070028#include "xattr.h"
29#include "acl.h"
30
Theodore Ts'o9bffad12009-06-17 11:48:11 -040031#include <trace/events/ext4.h>
32
Dave Kleikampac27a0e2006-10-11 01:20:50 -070033/*
34 * ialloc.c contains the inodes allocation and deallocation routines
35 */
36
37/*
38 * The free inodes are managed by bitmaps. A file system contains several
39 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
40 * block for inodes, N blocks for the inode table and data blocks.
41 *
42 * The file system contains group descriptors which are located after the
43 * super block. Each descriptor contains the number of the bitmap block and
44 * the free blocks count in the block.
45 */
46
Andreas Dilger717d50e2007-10-16 18:38:25 -040047/*
48 * To avoid calling the atomic setbit hundreds or thousands of times, we only
49 * need to use it within a single byte (to ensure we get endianness right).
50 * We can use memset for the rest of the bitmap as there are no other users.
51 */
Theodore Ts'o61d08672010-10-27 21:30:15 -040052void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
Andreas Dilger717d50e2007-10-16 18:38:25 -040053{
54 int i;
55
56 if (start_bit >= end_bit)
57 return;
58
59 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
60 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
61 ext4_set_bit(i, bitmap);
62 if (i < end_bit)
63 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
64}
65
Theodore Ts'o813e5722012-02-20 17:52:46 -050066void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
67{
68 if (uptodate) {
69 set_buffer_uptodate(bh);
70 set_bitmap_uptodate(bh);
71 }
72 unlock_buffer(bh);
73 put_bh(bh);
74}
75
Darrick J. Wong9008a582015-10-17 21:33:24 -040076static int ext4_validate_inode_bitmap(struct super_block *sb,
77 struct ext4_group_desc *desc,
78 ext4_group_t block_group,
79 struct buffer_head *bh)
80{
81 ext4_fsblk_t blk;
82 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
83 struct ext4_sb_info *sbi = EXT4_SB(sb);
84
85 if (buffer_verified(bh))
86 return 0;
87 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
88 return -EFSCORRUPTED;
89
90 ext4_lock_group(sb, block_group);
Theodore Ts'o262a62c2018-07-12 19:08:05 -040091 if (buffer_verified(bh))
92 goto verified;
Darrick J. Wong9008a582015-10-17 21:33:24 -040093 blk = ext4_inode_bitmap(sb, desc);
94 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
95 EXT4_INODES_PER_GROUP(sb) / 8)) {
96 ext4_unlock_group(sb, block_group);
97 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
98 "inode_bitmap = %llu", block_group, blk);
99 grp = ext4_get_group_info(sb, block_group);
100 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
101 int count;
102 count = ext4_free_inodes_count(sb, desc);
103 percpu_counter_sub(&sbi->s_freeinodes_counter,
104 count);
105 }
106 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
107 return -EFSBADCRC;
108 }
109 set_buffer_verified(bh);
Theodore Ts'o262a62c2018-07-12 19:08:05 -0400110verified:
Darrick J. Wong9008a582015-10-17 21:33:24 -0400111 ext4_unlock_group(sb, block_group);
112 return 0;
113}
114
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700115/*
116 * Read the inode allocation bitmap for a given block_group, reading
117 * into the specified slot in the superblock's bitmap cache.
118 *
119 * Return buffer_head of bitmap on success or NULL.
120 */
121static struct buffer_head *
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400122ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700123{
Mingming Cao617ba132006-10-11 01:20:53 -0700124 struct ext4_group_desc *desc;
Theodore Ts'o76964812018-03-26 23:54:10 -0400125 struct ext4_sb_info *sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700126 struct buffer_head *bh = NULL;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400127 ext4_fsblk_t bitmap_blk;
Darrick J. Wong9008a582015-10-17 21:33:24 -0400128 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700129
Mingming Cao617ba132006-10-11 01:20:53 -0700130 desc = ext4_get_group_desc(sb, block_group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700131 if (!desc)
Darrick J. Wong9008a582015-10-17 21:33:24 -0400132 return ERR_PTR(-EFSCORRUPTED);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400133
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400134 bitmap_blk = ext4_inode_bitmap(sb, desc);
Theodore Ts'o76964812018-03-26 23:54:10 -0400135 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
136 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
137 ext4_error(sb, "Invalid inode bitmap blk %llu in "
138 "block_group %u", bitmap_blk, block_group);
139 return ERR_PTR(-EFSCORRUPTED);
140 }
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400141 bh = sb_getblk(sb, bitmap_blk);
142 if (unlikely(!bh)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500143 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'oa9df9a42009-01-05 22:18:16 -0500144 "block_group = %u, inode_bitmap = %llu",
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400145 block_group, bitmap_blk);
Darrick J. Wong9008a582015-10-17 21:33:24 -0400146 return ERR_PTR(-EIO);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400147 }
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500148 if (bitmap_uptodate(bh))
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400149 goto verify;
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400150
Frederic Bohec806e682008-10-10 08:09:18 -0400151 lock_buffer(bh);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500152 if (bitmap_uptodate(bh)) {
153 unlock_buffer(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400154 goto verify;
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500155 }
Lukas Czernerbfff6872010-10-27 21:30:05 -0400156
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400157 ext4_lock_group(sb, block_group);
Theodore Ts'o5ae57322018-06-14 00:58:00 -0400158 if (ext4_has_group_desc_csum(sb) &&
159 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
160 if (block_group == 0) {
161 ext4_unlock_group(sb, block_group);
162 unlock_buffer(bh);
163 ext4_error(sb, "Inode bitmap for bg 0 marked "
164 "uninitialized");
165 err = -EFSCORRUPTED;
166 goto out;
167 }
Theodore Ts'oa06b7982018-02-19 14:16:47 -0500168 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
169 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
170 sb->s_blocksize * 8, bh->b_data);
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500171 set_bitmap_uptodate(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400172 set_buffer_uptodate(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400173 set_buffer_verified(bh);
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400174 ext4_unlock_group(sb, block_group);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500175 unlock_buffer(bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400176 return bh;
177 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400178 ext4_unlock_group(sb, block_group);
Lukas Czernerbfff6872010-10-27 21:30:05 -0400179
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500180 if (buffer_uptodate(bh)) {
181 /*
182 * if not uninit if bh is uptodate,
183 * bitmap is also uptodate
184 */
185 set_bitmap_uptodate(bh);
186 unlock_buffer(bh);
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400187 goto verify;
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500188 }
189 /*
Theodore Ts'o813e5722012-02-20 17:52:46 -0500190 * submit the buffer_head for reading
Aneesh Kumar K.V2ccb5fb2009-01-05 21:49:55 -0500191 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400192 trace_ext4_load_inode_bitmap(sb, block_group);
Theodore Ts'o813e5722012-02-20 17:52:46 -0500193 bh->b_end_io = ext4_end_bitmap_read;
194 get_bh(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -0500195 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
Theodore Ts'o813e5722012-02-20 17:52:46 -0500196 wait_on_buffer(bh);
197 if (!buffer_uptodate(bh)) {
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400198 put_bh(bh);
Eric Sandeen12062dd2010-02-15 14:19:27 -0500199 ext4_error(sb, "Cannot read inode bitmap - "
Theodore Ts'o813e5722012-02-20 17:52:46 -0500200 "block_group = %u, inode_bitmap = %llu",
201 block_group, bitmap_blk);
Darrick J. Wong9008a582015-10-17 21:33:24 -0400202 return ERR_PTR(-EIO);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400203 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400204
205verify:
Darrick J. Wong9008a582015-10-17 21:33:24 -0400206 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
207 if (err)
208 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700209 return bh;
Darrick J. Wong9008a582015-10-17 21:33:24 -0400210out:
211 put_bh(bh);
212 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700213}
214
215/*
216 * NOTE! When we get the inode, we're the only people
217 * that have access to it, and as such there are no
218 * race conditions we have to worry about. The inode
219 * is not on the hash-lists, and it cannot be reached
220 * through the filesystem because the directory entry
221 * has been deleted earlier.
222 *
223 * HOWEVER: we must make sure that we get no aliases,
224 * which means that we have to call "clear_inode()"
225 * _before_ we mark the inode not in use in the inode
226 * bitmaps. Otherwise a newly created file might use
227 * the same inode number (not actually the same pointer
228 * though), and then we'd have two inodes sharing the
229 * same inode number and space on the harddisk.
230 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400231void ext4_free_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700232{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400233 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234 int is_directory;
235 unsigned long ino;
236 struct buffer_head *bitmap_bh = NULL;
237 struct buffer_head *bh2;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500238 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700239 unsigned long bit;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400240 struct ext4_group_desc *gdp;
241 struct ext4_super_block *es;
Mingming Cao617ba132006-10-11 01:20:53 -0700242 struct ext4_sb_info *sbi;
Eric Sandeen7ce9d5d2009-03-04 18:38:18 -0500243 int fatal = 0, err, count, cleared;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400244 struct ext4_group_info *grp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700245
Theodore Ts'o92b97812012-03-19 23:41:49 -0400246 if (!sb) {
247 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
248 "nonexistent device\n", __func__, __LINE__);
249 return;
250 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700251 if (atomic_read(&inode->i_count) > 1) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400252 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
253 __func__, __LINE__, inode->i_ino,
254 atomic_read(&inode->i_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700255 return;
256 }
257 if (inode->i_nlink) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400258 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
259 __func__, __LINE__, inode->i_ino, inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700260 return;
261 }
Mingming Cao617ba132006-10-11 01:20:53 -0700262 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700263
264 ino = inode->i_ino;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400265 ext4_debug("freeing inode %lu\n", ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400266 trace_ext4_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700267
268 /*
269 * Note: we must free any quota before locking the superblock,
270 * as writing the quota to disk may need the lock as well.
271 */
Christoph Hellwig871a2932010-03-03 09:05:07 -0500272 dquot_initialize(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700273 ext4_xattr_delete_inode(handle, inode);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500274 dquot_free_inode(inode);
Christoph Hellwig9f754752010-03-03 09:05:05 -0500275 dquot_drop(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700276
277 is_directory = S_ISDIR(inode->i_mode);
278
279 /* Do this BEFORE marking the inode not in use or returning an error */
Al Viro0930fcc2010-06-07 13:16:22 -0400280 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700281
Mingming Cao617ba132006-10-11 01:20:53 -0700282 es = EXT4_SB(sb)->s_es;
283 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500284 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700285 goto error_return;
286 }
Mingming Cao617ba132006-10-11 01:20:53 -0700287 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
288 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -0400289 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Darrick J. Wong87a39382013-08-28 18:32:58 -0400290 /* Don't bother if the inode bitmap is corrupt. */
291 grp = ext4_get_group_info(sb, block_group);
Darrick J. Wong9008a582015-10-17 21:33:24 -0400292 if (IS_ERR(bitmap_bh)) {
293 fatal = PTR_ERR(bitmap_bh);
294 bitmap_bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700295 goto error_return;
Darrick J. Wong9008a582015-10-17 21:33:24 -0400296 }
297 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
298 fatal = -EFSCORRUPTED;
299 goto error_return;
300 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700301
302 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700303 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700304 if (fatal)
305 goto error_return;
306
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400307 fatal = -ESRCH;
308 gdp = ext4_get_group_desc(sb, block_group, &bh2);
309 if (gdp) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700310 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700311 fatal = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700312 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400313 ext4_lock_group(sb, block_group);
Akinobu Mita597d5082011-12-28 20:32:07 -0500314 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400315 if (fatal || !cleared) {
316 ext4_unlock_group(sb, block_group);
317 goto out;
318 }
319
320 count = ext4_free_inodes_count(sb, gdp) + 1;
321 ext4_free_inodes_set(sb, gdp, count);
322 if (is_directory) {
323 count = ext4_used_dirs_count(sb, gdp) - 1;
324 ext4_used_dirs_set(sb, gdp, count);
325 percpu_counter_dec(&sbi->s_dirs_counter);
326 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400327 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
328 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400329 ext4_group_desc_csum_set(sb, block_group, gdp);
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400330 ext4_unlock_group(sb, block_group);
331
332 percpu_counter_inc(&sbi->s_freeinodes_counter);
333 if (sbi->s_log_groups_per_flex) {
334 ext4_group_t f = ext4_flex_group(sbi, block_group);
335
336 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
337 if (is_directory)
338 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
339 }
340 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
341 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
342out:
343 if (cleared) {
344 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
345 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
346 if (!fatal)
347 fatal = err;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400348 } else {
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400349 ext4_error(sb, "bit already cleared for inode %lu", ino);
Namjae Jeonbf40c922014-07-12 16:11:42 -0400350 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
Namjae Jeone43bb4e2014-06-26 10:11:53 -0400351 int count;
352 count = ext4_free_inodes_count(sb, gdp);
353 percpu_counter_sub(&sbi->s_freeinodes_counter,
354 count);
355 }
Darrick J. Wong87a39382013-08-28 18:32:58 -0400356 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
357 }
Dmitry Monakhovd17413c2010-05-16 07:00:00 -0400358
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700359error_return:
360 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700361 ext4_std_error(sb, fatal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700362}
363
Theodore Ts'oa4912122009-03-12 12:18:34 -0400364struct orlov_stats {
Theodore Ts'o90ba9832013-03-11 23:39:59 -0400365 __u64 free_clusters;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400366 __u32 free_inodes;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400367 __u32 used_dirs;
368};
369
370/*
371 * Helper function for Orlov's allocator; returns critical information
372 * for a particular block group or flex_bg. If flex_size is 1, then g
373 * is a block group number; otherwise it is flex_bg number.
374 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400375static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
376 int flex_size, struct orlov_stats *stats)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400377{
378 struct ext4_group_desc *desc;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500379 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400380
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500381 if (flex_size > 1) {
382 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
Theodore Ts'o90ba9832013-03-11 23:39:59 -0400383 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500384 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
385 return;
386 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400387
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500388 desc = ext4_get_group_desc(sb, g, NULL);
389 if (desc) {
390 stats->free_inodes = ext4_free_inodes_count(sb, desc);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400391 stats->free_clusters = ext4_free_group_clusters(sb, desc);
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500392 stats->used_dirs = ext4_used_dirs_count(sb, desc);
393 } else {
394 stats->free_inodes = 0;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400395 stats->free_clusters = 0;
Theodore Ts'o7d39db12009-03-04 19:31:53 -0500396 stats->used_dirs = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400397 }
398}
399
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700400/*
401 * Orlov's allocator for directories.
402 *
403 * We always try to spread first-level directories.
404 *
405 * If there are blockgroups with both free inodes and free blocks counts
406 * not worse than average we return one with smallest directory count.
407 * Otherwise we simply return a random group.
408 *
409 * For the rest rules look so:
410 *
411 * It's OK to put directory into a group unless
412 * it has too many directories already (max_dirs) or
413 * it has too few free inodes left (min_inodes) or
414 * it has too few free blocks left (min_blocks) or
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000415 * Parent's group is preferred, if it doesn't satisfy these
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700416 * conditions we search cyclically through the rest. If none
417 * of the groups look good we just look for a group with more
418 * free inodes than average (starting at parent's group).
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700419 */
420
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500421static int find_group_orlov(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400422 ext4_group_t *group, umode_t mode,
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400423 const struct qstr *qstr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700424{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500425 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Mingming Cao617ba132006-10-11 01:20:53 -0700426 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o8df96752009-05-01 08:50:38 -0400427 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700428 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Theodore Ts'o14c83c92011-12-28 20:25:13 -0500429 unsigned int freei, avefreei, grp_free;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400430 ext4_fsblk_t freeb, avefreec;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700431 unsigned int ndirs;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400432 int max_dirs, min_inodes;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400433 ext4_grpblk_t min_clusters;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400434 ext4_group_t i, grp, g, ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700435 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400436 struct orlov_stats stats;
437 int flex_size = ext4_flex_bg_size(sbi);
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400438 struct dx_hash_info hinfo;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400439
Theodore Ts'o8df96752009-05-01 08:50:38 -0400440 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400441 if (flex_size > 1) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400442 ngroups = (real_ngroups + flex_size - 1) >>
Theodore Ts'oa4912122009-03-12 12:18:34 -0400443 sbi->s_log_groups_per_flex;
444 parent_group >>= sbi->s_log_groups_per_flex;
445 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700446
447 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
448 avefreei = freei / ngroups;
Theodore Ts'o57042652011-09-09 18:56:51 -0400449 freeb = EXT4_C2B(sbi,
450 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400451 avefreec = freeb;
452 do_div(avefreec, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700453 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
454
Theodore Ts'oa4912122009-03-12 12:18:34 -0400455 if (S_ISDIR(mode) &&
David Howells2b0143b2015-03-17 22:25:59 +0000456 ((parent == d_inode(sb->s_root)) ||
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400457 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700458 int best_ndir = inodes_per_group;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500459 int ret = -1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700460
Theodore Ts'of157a4a2009-06-13 11:09:42 -0400461 if (qstr) {
462 hinfo.hash_version = DX_HASH_HALF_MD4;
463 hinfo.seed = sbi->s_hash_seed;
464 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
465 grp = hinfo.hash;
466 } else
Theodore Ts'odd1f7232013-11-08 00:14:53 -0500467 grp = prandom_u32();
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500468 parent_group = (unsigned)grp % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700469 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400470 g = (parent_group + i) % ngroups;
471 get_orlov_stats(sb, g, flex_size, &stats);
472 if (!stats.free_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700473 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400474 if (stats.used_dirs >= best_ndir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700475 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400476 if (stats.free_inodes < avefreei)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700477 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400478 if (stats.free_clusters < avefreec)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700479 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400480 grp = g;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500481 ret = 0;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400482 best_ndir = stats.used_dirs;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700483 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400484 if (ret)
485 goto fallback;
486 found_flex_bg:
487 if (flex_size == 1) {
488 *group = grp;
489 return 0;
490 }
491
492 /*
493 * We pack inodes at the beginning of the flexgroup's
494 * inode tables. Block allocation decisions will do
495 * something similar, although regular files will
496 * start at 2nd block group of the flexgroup. See
497 * ext4_ext_find_goal() and ext4_find_near().
498 */
499 grp *= flex_size;
500 for (i = 0; i < flex_size; i++) {
Theodore Ts'o8df96752009-05-01 08:50:38 -0400501 if (grp+i >= real_ngroups)
Theodore Ts'oa4912122009-03-12 12:18:34 -0400502 break;
503 desc = ext4_get_group_desc(sb, grp+i, NULL);
504 if (desc && ext4_free_inodes_count(sb, desc)) {
505 *group = grp+i;
506 return 0;
507 }
508 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700509 goto fallback;
510 }
511
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700512 max_dirs = ndirs / ngroups + inodes_per_group / 16;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400513 min_inodes = avefreei - inodes_per_group*flex_size / 4;
514 if (min_inodes < 1)
515 min_inodes = 1;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400516 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700517
Theodore Ts'oa4912122009-03-12 12:18:34 -0400518 /*
519 * Start looking in the flex group where we last allocated an
520 * inode for this parent directory
521 */
522 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
523 parent_group = EXT4_I(parent)->i_last_alloc_group;
524 if (flex_size > 1)
525 parent_group >>= sbi->s_log_groups_per_flex;
526 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700527
528 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400529 grp = (parent_group + i) % ngroups;
530 get_orlov_stats(sb, grp, flex_size, &stats);
531 if (stats.used_dirs >= max_dirs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700532 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400533 if (stats.free_inodes < min_inodes)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700534 continue;
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -0400535 if (stats.free_clusters < min_clusters)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700536 continue;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400537 goto found_flex_bg;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700538 }
539
540fallback:
Theodore Ts'o8df96752009-05-01 08:50:38 -0400541 ngroups = real_ngroups;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400542 avefreei = freei / ngroups;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400543fallback_retry:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400544 parent_group = EXT4_I(parent)->i_block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700545 for (i = 0; i < ngroups; i++) {
Theodore Ts'oa4912122009-03-12 12:18:34 -0400546 grp = (parent_group + i) % ngroups;
547 desc = ext4_get_group_desc(sb, grp, NULL);
Dan Carpenterbb3d1322012-05-28 14:16:57 -0400548 if (desc) {
549 grp_free = ext4_free_inodes_count(sb, desc);
550 if (grp_free && grp_free >= avefreei) {
551 *group = grp;
552 return 0;
553 }
Theodore Ts'oa4912122009-03-12 12:18:34 -0400554 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700555 }
556
557 if (avefreei) {
558 /*
559 * The free-inodes counter is approximate, and for really small
560 * filesystems the above test can fail to find any blockgroups
561 */
562 avefreei = 0;
Theodore Ts'ob5451f72009-04-22 21:00:36 -0400563 goto fallback_retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700564 }
565
566 return -1;
567}
568
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500569static int find_group_other(struct super_block *sb, struct inode *parent,
Al Virodcca3fe2011-07-26 02:48:06 -0400570 ext4_group_t *group, umode_t mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700571{
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500572 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400573 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700574 struct ext4_group_desc *desc;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400575 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
576
577 /*
578 * Try to place the inode is the same flex group as its
579 * parent. If we can't find space, use the Orlov algorithm to
580 * find another flex group, and store that information in the
581 * parent directory's inode information so that use that flex
582 * group for future allocations.
583 */
584 if (flex_size > 1) {
585 int retry = 0;
586
587 try_again:
588 parent_group &= ~(flex_size-1);
589 last = parent_group + flex_size;
590 if (last > ngroups)
591 last = ngroups;
592 for (i = parent_group; i < last; i++) {
593 desc = ext4_get_group_desc(sb, i, NULL);
594 if (desc && ext4_free_inodes_count(sb, desc)) {
595 *group = i;
596 return 0;
597 }
598 }
599 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
600 retry = 1;
601 parent_group = EXT4_I(parent)->i_last_alloc_group;
602 goto try_again;
603 }
604 /*
605 * If this didn't work, use the Orlov search algorithm
606 * to find a new flex group; we pass in the mode to
607 * avoid the topdir algorithms.
608 */
609 *group = parent_group + flex_size;
610 if (*group > ngroups)
611 *group = 0;
Peter Huewe7dc57612011-02-21 21:01:42 -0500612 return find_group_orlov(sb, parent, group, mode, NULL);
Theodore Ts'oa4912122009-03-12 12:18:34 -0400613 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700614
615 /*
616 * Try to place the inode in its parent directory
617 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500618 *group = parent_group;
619 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500620 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400621 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500622 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700623
624 /*
625 * We're going to place this inode in a different blockgroup from its
626 * parent. We want to cause files in a common directory to all land in
627 * the same blockgroup. But we want files which are in a different
628 * directory which shares a blockgroup with our parent to land in a
629 * different blockgroup.
630 *
631 * So add our directory's i_ino into the starting point for the hash.
632 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500633 *group = (*group + parent->i_ino) % ngroups;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700634
635 /*
636 * Use a quadratic hash to find a group with a free inode and some free
637 * blocks.
638 */
639 for (i = 1; i < ngroups; i <<= 1) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500640 *group += i;
641 if (*group >= ngroups)
642 *group -= ngroups;
643 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500644 if (desc && ext4_free_inodes_count(sb, desc) &&
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400645 ext4_free_group_clusters(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500646 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700647 }
648
649 /*
650 * That failed: try linear search for a free inode, even if that group
651 * has no free blocks.
652 */
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500653 *group = parent_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700654 for (i = 0; i < ngroups; i++) {
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500655 if (++*group >= ngroups)
656 *group = 0;
657 desc = ext4_get_group_desc(sb, *group, NULL);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500658 if (desc && ext4_free_inodes_count(sb, desc))
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500659 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700660 }
661
662 return -1;
663}
664
665/*
Theodore Ts'o19883bd2013-08-16 22:06:55 -0400666 * In no journal mode, if an inode has recently been deleted, we want
667 * to avoid reusing it until we're reasonably sure the inode table
668 * block has been written back to disk. (Yes, these values are
669 * somewhat arbitrary...)
670 */
671#define RECENTCY_MIN 5
672#define RECENTCY_DIRTY 30
673
674static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
675{
676 struct ext4_group_desc *gdp;
677 struct ext4_inode *raw_inode;
678 struct buffer_head *bh;
679 unsigned long dtime, now;
680 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
681 int offset, ret = 0, recentcy = RECENTCY_MIN;
682
683 gdp = ext4_get_group_desc(sb, group, NULL);
684 if (unlikely(!gdp))
685 return 0;
686
687 bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
688 (ino / inodes_per_block));
689 if (unlikely(!bh) || !buffer_uptodate(bh))
690 /*
691 * If the block is not in the buffer cache, then it
692 * must have been written out.
693 */
694 goto out;
695
696 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
697 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
698 dtime = le32_to_cpu(raw_inode->i_dtime);
699 now = get_seconds();
700 if (buffer_dirty(bh))
701 recentcy += RECENTCY_DIRTY;
702
703 if (dtime && (dtime < now) && (now < dtime + recentcy))
704 ret = 1;
705out:
706 brelse(bh);
707 return ret;
708}
709
710/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700711 * There are two policies for allocating an inode. If the new inode is
712 * a directory, then a forward search is made for a block group with both
713 * free space and a low directory-to-inode ratio; if that fails, then of
714 * the groups with above-average free space, that group with the fewest
715 * directories already is chosen.
716 *
717 * For other inodes, search forward from the parent directory's block
718 * group to find a free inode.
719 */
Theodore Ts'o11395752013-02-09 16:27:09 -0500720struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
721 umode_t mode, const struct qstr *qstr,
722 __u32 goal, uid_t *owner, int handle_type,
723 unsigned int line_no, int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700724{
725 struct super_block *sb;
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500726 struct buffer_head *inode_bitmap_bh = NULL;
727 struct buffer_head *group_desc_bh;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400728 ext4_group_t ngroups, group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700729 unsigned long ino = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400730 struct inode *inode;
731 struct ext4_group_desc *gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700732 struct ext4_inode_info *ei;
733 struct ext4_sb_info *sbi;
Jan Karaa7cdade2015-06-29 16:22:54 +0200734 int ret2, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700735 struct inode *ret;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500736 ext4_group_t i;
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400737 ext4_group_t flex_group;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400738 struct ext4_group_info *grp;
Theodore Ts'oe709e9d2015-05-31 13:35:02 -0400739 int encrypt = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700740
741 /* Cannot create files in a deleted directory */
742 if (!dir || !dir->i_nlink)
743 return ERR_PTR(-EPERM);
744
Theodore Ts'oe709e9d2015-05-31 13:35:02 -0400745 if ((ext4_encrypted_inode(dir) ||
746 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
747 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400748 err = fscrypt_get_encryption_info(dir);
Theodore Ts'oe709e9d2015-05-31 13:35:02 -0400749 if (err)
750 return ERR_PTR(err);
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400751 if (!fscrypt_has_encryption_key(dir))
Eric Biggersd612bee2016-12-05 11:12:44 -0800752 return ERR_PTR(-ENOKEY);
Theodore Ts'oe709e9d2015-05-31 13:35:02 -0400753 if (!handle)
754 nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
755 encrypt = 1;
756 }
757
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700758 sb = dir->i_sb;
Theodore Ts'o8df96752009-05-01 08:50:38 -0400759 ngroups = ext4_get_groups_count(sb);
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400760 trace_ext4_request_inode(dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700761 inode = new_inode(sb);
762 if (!inode)
763 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700764 ei = EXT4_I(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700765 sbi = EXT4_SB(sb);
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400766
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400767 /*
Adam Buchbinderb8a074632016-03-09 23:49:05 -0500768 * Initialize owners and quota early so that we don't have to account
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400769 * for quota initialization worst case in standard inode creating
770 * transaction
771 */
772 if (owner) {
773 inode->i_mode = mode;
774 i_uid_write(inode, owner[0]);
775 i_gid_write(inode, owner[1]);
776 } else if (test_opt(sb, GRPID)) {
777 inode->i_mode = mode;
778 inode->i_uid = current_fsuid();
779 inode->i_gid = dir->i_gid;
780 } else
781 inode_init_owner(inode, dir, mode);
Li Xi040cb372016-01-08 16:01:21 -0500782
Kaho Ng0b7b7772016-09-05 23:11:58 -0400783 if (ext4_has_feature_project(sb) &&
Li Xi040cb372016-01-08 16:01:21 -0500784 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
785 ei->i_projid = EXT4_I(dir)->i_projid;
786 else
787 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
788
Jan Karaa7cdade2015-06-29 16:22:54 +0200789 err = dquot_initialize(inode);
790 if (err)
791 goto out;
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400792
Andreas Dilger11013912009-06-13 11:45:35 -0400793 if (!goal)
794 goal = sbi->s_inode_goal;
795
Johann Lombardie6462862009-07-05 23:45:11 -0400796 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
Andreas Dilger11013912009-06-13 11:45:35 -0400797 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
798 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
799 ret2 = 0;
800 goto got_group;
801 }
802
Lukas Czerner4113c4c2011-10-08 14:34:47 -0400803 if (S_ISDIR(mode))
804 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
805 else
Theodore Ts'oa4912122009-03-12 12:18:34 -0400806 ret2 = find_group_other(sb, dir, &group, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700807
Jose R. Santos772cb7c2008-07-11 19:27:31 -0400808got_group:
Theodore Ts'oa4912122009-03-12 12:18:34 -0400809 EXT4_I(dir)->i_last_alloc_group = group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700810 err = -ENOSPC;
Avantika Mathur2aa9fc42008-01-28 23:58:27 -0500811 if (ret2 == -1)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700812 goto out;
813
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500814 /*
815 * Normally we will only go through one pass of this loop,
816 * unless we get unlucky and it turns out the group we selected
817 * had its last inode grabbed by someone else.
818 */
Andreas Dilger11013912009-06-13 11:45:35 -0400819 for (i = 0; i < ngroups; i++, ino = 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700820 err = -EIO;
821
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500822 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700823 if (!gdp)
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400824 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700825
Yongqiang Yangf2a09af2012-09-23 23:16:03 -0400826 /*
827 * Check free inodes count before loading bitmap.
828 */
829 if (ext4_free_inodes_count(sb, gdp) == 0) {
830 if (++group == ngroups)
831 group = 0;
832 continue;
833 }
834
Darrick J. Wong87a39382013-08-28 18:32:58 -0400835 grp = ext4_get_group_info(sb, group);
836 /* Skip groups with already-known suspicious inode tables */
837 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
838 if (++group == ngroups)
839 group = 0;
840 continue;
841 }
842
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500843 brelse(inode_bitmap_bh);
844 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
Darrick J. Wong87a39382013-08-28 18:32:58 -0400845 /* Skip groups with suspicious inode tables */
Darrick J. Wong9008a582015-10-17 21:33:24 -0400846 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
847 IS_ERR(inode_bitmap_bh)) {
848 inode_bitmap_bh = NULL;
Darrick J. Wong87a39382013-08-28 18:32:58 -0400849 if (++group == ngroups)
850 group = 0;
851 continue;
852 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700853
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700854repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700855 ino = ext4_find_next_zero_bit((unsigned long *)
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500856 inode_bitmap_bh->b_data,
857 EXT4_INODES_PER_GROUP(sb), ino);
Theodore Ts'oa34eb502013-07-26 15:15:46 -0400858 if (ino >= EXT4_INODES_PER_GROUP(sb))
859 goto next_group;
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500860 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
861 ext4_error(sb, "reserved inode found cleared - "
862 "inode=%lu", ino + 1);
863 continue;
864 }
Theodore Ts'o19883bd2013-08-16 22:06:55 -0400865 if ((EXT4_SB(sb)->s_journal == NULL) &&
866 recently_deleted(sb, group, ino)) {
867 ino++;
868 goto next_inode;
869 }
Theodore Ts'o11395752013-02-09 16:27:09 -0500870 if (!handle) {
871 BUG_ON(nblocks <= 0);
872 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
Jan Kara5fe2fe82013-06-04 12:37:50 -0400873 handle_type, nblocks,
874 0);
Theodore Ts'o11395752013-02-09 16:27:09 -0500875 if (IS_ERR(handle)) {
876 err = PTR_ERR(handle);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400877 ext4_std_error(sb, err);
878 goto out;
Theodore Ts'o11395752013-02-09 16:27:09 -0500879 }
880 }
Eric Sandeenffb53872012-10-28 22:24:57 -0400881 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
882 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400883 if (err) {
884 ext4_std_error(sb, err);
885 goto out;
886 }
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500887 ext4_lock_group(sb, group);
888 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
889 ext4_unlock_group(sb, group);
890 ino++; /* the inode bitmap is zero-based */
891 if (!ret2)
892 goto got; /* we grabbed the inode! */
Theodore Ts'o19883bd2013-08-16 22:06:55 -0400893next_inode:
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500894 if (ino < EXT4_INODES_PER_GROUP(sb))
895 goto repeat_in_this_group;
Theodore Ts'oa34eb502013-07-26 15:15:46 -0400896next_group:
897 if (++group == ngroups)
898 group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700899 }
900 err = -ENOSPC;
901 goto out;
902
903got:
Eric Sandeenffb53872012-10-28 22:24:57 -0400904 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
905 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400906 if (err) {
907 ext4_std_error(sb, err);
908 goto out;
909 }
Eric Sandeenffb53872012-10-28 22:24:57 -0400910
Theodore Ts'o61c219f2014-07-05 16:28:35 -0400911 BUFFER_TRACE(group_desc_bh, "get_write_access");
912 err = ext4_journal_get_write_access(handle, group_desc_bh);
913 if (err) {
914 ext4_std_error(sb, err);
915 goto out;
916 }
917
Andreas Dilger717d50e2007-10-16 18:38:25 -0400918 /* We may have to initialize the block bitmap if it isn't already */
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400919 if (ext4_has_group_desc_csum(sb) &&
Andreas Dilger717d50e2007-10-16 18:38:25 -0400920 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500921 struct buffer_head *block_bitmap_bh;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400922
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500923 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
Darrick J. Wong9008a582015-10-17 21:33:24 -0400924 if (IS_ERR(block_bitmap_bh)) {
925 err = PTR_ERR(block_bitmap_bh);
Jan Kara599a9b72014-10-30 10:53:16 -0400926 goto out;
927 }
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500928 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
929 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400930 if (err) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500931 brelse(block_bitmap_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400932 ext4_std_error(sb, err);
933 goto out;
Andreas Dilger717d50e2007-10-16 18:38:25 -0400934 }
935
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400936 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
937 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400938
Andreas Dilger717d50e2007-10-16 18:38:25 -0400939 /* recheck and clear flag under lock if we still need to */
Theodore Ts'ofd034a82011-09-09 18:42:51 -0400940 ext4_lock_group(sb, group);
Theodore Ts'o5ae57322018-06-14 00:58:00 -0400941 if (ext4_has_group_desc_csum(sb) &&
942 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -0500943 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
Theodore Ts'o021b65b2011-09-09 19:08:51 -0400944 ext4_free_group_clusters_set(sb, gdp,
Theodore Ts'ocff1dfd72011-09-09 19:12:51 -0400945 ext4_free_clusters_after_init(sb, group, gdp));
Darrick J. Wongfa77dcf2012-04-29 18:35:10 -0400946 ext4_block_bitmap_csum_set(sb, group, gdp,
Tao Ma79f1ba42012-10-22 00:34:32 -0400947 block_bitmap_bh);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400948 ext4_group_desc_csum_set(sb, group, gdp);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400949 }
Aneesh Kumar K.V955ce5f2009-05-02 20:35:09 -0400950 ext4_unlock_group(sb, group);
Theodore Ts'oaeb1e5d62012-11-29 21:21:22 -0500951 brelse(block_bitmap_bh);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400952
Jan Karaeb9cc7e2013-04-19 13:38:14 -0400953 if (err) {
954 ext4_std_error(sb, err);
955 goto out;
956 }
Andreas Dilger717d50e2007-10-16 18:38:25 -0400957 }
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500958
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500959 /* Update the relevant bg descriptor fields */
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400960 if (ext4_has_group_desc_csum(sb)) {
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500961 int free;
962 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
963
964 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
965 ext4_lock_group(sb, group); /* while we modify the bg desc */
966 free = EXT4_INODES_PER_GROUP(sb) -
967 ext4_itable_unused_count(sb, gdp);
968 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
969 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
970 free = 0;
971 }
972 /*
973 * Check the relative inode number against the last used
974 * relative inode number in this group. if it is greater
975 * we need to update the bg_itable_unused count
976 */
977 if (ino > free)
978 ext4_itable_unused_set(sb, gdp,
979 (EXT4_INODES_PER_GROUP(sb) - ino));
980 up_read(&grp->alloc_sem);
Tao Ma6f2e9f02012-05-28 18:20:59 -0400981 } else {
982 ext4_lock_group(sb, group);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500983 }
Tao Ma6f2e9f02012-05-28 18:20:59 -0400984
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500985 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
986 if (S_ISDIR(mode)) {
987 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
988 if (sbi->s_log_groups_per_flex) {
989 ext4_group_t f = ext4_flex_group(sbi, group);
990
991 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
992 }
993 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -0400994 if (ext4_has_group_desc_csum(sb)) {
995 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
996 EXT4_INODES_PER_GROUP(sb) / 8);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -0400997 ext4_group_desc_csum_set(sb, group, gdp);
Theodore Ts'o119c0d42012-02-06 20:12:03 -0500998 }
Tao Ma6f2e9f02012-05-28 18:20:59 -0400999 ext4_unlock_group(sb, group);
Theodore Ts'o119c0d42012-02-06 20:12:03 -05001000
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001001 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1002 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
Jan Karaeb9cc7e2013-04-19 13:38:14 -04001003 if (err) {
1004 ext4_std_error(sb, err);
1005 goto out;
1006 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001007
1008 percpu_counter_dec(&sbi->s_freeinodes_counter);
1009 if (S_ISDIR(mode))
1010 percpu_counter_inc(&sbi->s_dirs_counter);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001011
Jose R. Santos772cb7c2008-07-11 19:27:31 -04001012 if (sbi->s_log_groups_per_flex) {
1013 flex_group = ext4_flex_group(sbi, group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -05001014 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
Jose R. Santos772cb7c2008-07-11 19:27:31 -04001015 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001016
Andreas Dilger717d50e2007-10-16 18:38:25 -04001017 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001018 /* This is the optimal IO size (for stat), not the fs block size */
1019 inode->i_blocks = 0;
Kalpak Shahef7f3832007-07-18 09:15:20 -04001020 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1021 ext4_current_time(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001022
1023 memset(ei->i_data, 0, sizeof(ei->i_data));
1024 ei->i_dir_start_lookup = 0;
1025 ei->i_disksize = 0;
1026
Eryu Guan4af83502011-10-31 18:21:29 -04001027 /* Don't inherit extent flag from directory, amongst others. */
Duane Griffin2dc6b0d2009-02-15 18:09:20 -05001028 ei->i_flags =
1029 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001030 ei->i_file_acl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001031 ei->i_dtime = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001032 ei->i_block_group = group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04001033 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001034
Mingming Cao617ba132006-10-11 01:20:53 -07001035 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001036 if (IS_DIRSYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05001037 ext4_handle_sync(handle);
Al Viro6b38e842008-12-30 02:03:31 -05001038 if (insert_inode_locked(inode) < 0) {
Jan Karaacd6ad82011-12-18 17:37:02 -05001039 /*
1040 * Likely a bitmap corruption causing inode to be allocated
1041 * twice.
1042 */
1043 err = -EIO;
Jan Karaeb9cc7e2013-04-19 13:38:14 -04001044 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1045 inode->i_ino);
1046 goto out;
Al Viro6b38e842008-12-30 02:03:31 -05001047 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001048 spin_lock(&sbi->s_next_gen_lock);
1049 inode->i_generation = sbi->s_next_generation++;
1050 spin_unlock(&sbi->s_next_gen_lock);
1051
Darrick J. Wong814525f2012-04-29 18:31:10 -04001052 /* Precompute checksum seed for inode metadata */
Dmitry Monakhov9aa5d322014-10-13 03:36:16 -04001053 if (ext4_has_metadata_csum(sb)) {
Darrick J. Wong814525f2012-04-29 18:31:10 -04001054 __u32 csum;
Darrick J. Wong814525f2012-04-29 18:31:10 -04001055 __le32 inum = cpu_to_le32(inode->i_ino);
1056 __le32 gen = cpu_to_le32(inode->i_generation);
1057 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1058 sizeof(inum));
1059 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1060 sizeof(gen));
1061 }
1062
Theodore Ts'o353eb832011-01-10 12:18:25 -05001063 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001064 ext4_set_inode_state(inode, EXT4_STATE_NEW);
Kalpak Shahef7f3832007-07-18 09:15:20 -04001065
1066 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
Tao Maf08225d2012-12-10 14:06:03 -05001067 ei->i_inline_off = 0;
Darrick J. Wonge2b911c2015-10-17 16:18:43 -04001068 if (ext4_has_feature_inline_data(sb))
Tao Maf08225d2012-12-10 14:06:03 -05001069 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001070 ret = inode;
Christoph Hellwig63936dd2010-03-03 09:05:01 -05001071 err = dquot_alloc_inode(inode);
1072 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001073 goto fail_drop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001074
Mingming Cao617ba132006-10-11 01:20:53 -07001075 err = ext4_init_acl(handle, inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001076 if (err)
1077 goto fail_free_drop;
1078
Eric Paris2a7dba32011-02-01 11:05:39 -05001079 err = ext4_init_security(handle, inode, dir, qstr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001080 if (err)
1081 goto fail_free_drop;
1082
Darrick J. Wonge2b911c2015-10-17 16:18:43 -04001083 if (ext4_has_feature_extents(sb)) {
Eric Sandeene4079a12008-07-11 19:27:31 -04001084 /* set extent flag only for directory, file and normal symlink*/
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04001085 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001086 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001087 ext4_ext_tree_init(handle, inode);
Aneesh Kumar K.V42bf0382008-02-25 16:38:03 -05001088 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001089 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001090
Theodore Ts'o688f8692011-03-16 17:16:31 -04001091 if (ext4_handle_valid(handle)) {
1092 ei->i_sync_tid = handle->h_transaction->t_tid;
1093 ei->i_datasync_tid = handle->h_transaction->t_tid;
1094 }
1095
Theodore Ts'oe709e9d2015-05-31 13:35:02 -04001096 if (encrypt) {
Jaegeuk Kima7550b32016-07-10 14:01:03 -04001097 /* give pointer to avoid set_context with journal ops. */
1098 err = fscrypt_inherit_context(dir, inode, &encrypt, true);
Theodore Ts'oe709e9d2015-05-31 13:35:02 -04001099 if (err)
1100 goto fail_free_drop;
1101 }
1102
Aneesh Kumar K.V8753e882008-04-29 22:00:36 -04001103 err = ext4_mark_inode_dirty(handle, inode);
1104 if (err) {
1105 ext4_std_error(sb, err);
1106 goto fail_free_drop;
1107 }
1108
Mingming Cao617ba132006-10-11 01:20:53 -07001109 ext4_debug("allocating inode %lu\n", inode->i_ino);
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001110 trace_ext4_allocate_inode(inode, dir, mode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001111 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001112 return ret;
1113
1114fail_free_drop:
Christoph Hellwig63936dd2010-03-03 09:05:01 -05001115 dquot_free_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001116fail_drop:
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02001117 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001118 unlock_new_inode(inode);
Jan Karaeb9cc7e2013-04-19 13:38:14 -04001119out:
1120 dquot_drop(inode);
1121 inode->i_flags |= S_NOQUOTA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001122 iput(inode);
Aneesh Kumar K.V3300bed2009-01-03 22:33:39 -05001123 brelse(inode_bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001124 return ERR_PTR(err);
1125}
1126
1127/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -07001128struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001129{
Mingming Cao617ba132006-10-11 01:20:53 -07001130 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Avantika Mathurfd2d4292008-01-28 23:58:27 -05001131 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001132 int bit;
Theodore Ts'o7827a7f2016-04-30 00:49:54 -04001133 struct buffer_head *bitmap_bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001134 struct inode *inode = NULL;
Theodore Ts'o7827a7f2016-04-30 00:49:54 -04001135 int err = -EFSCORRUPTED;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001136
Theodore Ts'o7827a7f2016-04-30 00:49:54 -04001137 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1138 goto bad_orphan;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001139
Mingming Cao617ba132006-10-11 01:20:53 -07001140 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1141 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001142 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
Darrick J. Wong9008a582015-10-17 21:33:24 -04001143 if (IS_ERR(bitmap_bh)) {
Theodore Ts'o7827a7f2016-04-30 00:49:54 -04001144 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1145 ino, PTR_ERR(bitmap_bh));
1146 return (struct inode *) bitmap_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001147 }
1148
1149 /* Having the inode bit set should be a 100% indicator that this
1150 * is a valid orphan (no e2fsck run on fs). Orphans also include
1151 * inodes that were being truncated, so we can't check i_nlink==0.
1152 */
David Howells1d1fe1e2008-02-07 00:15:37 -08001153 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1154 goto bad_orphan;
1155
1156 inode = ext4_iget(sb, ino);
Theodore Ts'o7827a7f2016-04-30 00:49:54 -04001157 if (IS_ERR(inode)) {
1158 err = PTR_ERR(inode);
1159 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1160 ino, err);
1161 return inode;
1162 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001163
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001164 /*
Theodore Ts'oc9eb13a2016-04-30 00:48:54 -04001165 * If the orphans has i_nlinks > 0 then it should be able to
1166 * be truncated, otherwise it won't be removed from the orphan
1167 * list during processing and an infinite loop will result.
1168 * Similarly, it must not be a bad inode.
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001169 */
Theodore Ts'oc9eb13a2016-04-30 00:48:54 -04001170 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1171 is_bad_inode(inode))
Duane Griffin91ef4ca2008-07-11 19:27:31 -04001172 goto bad_orphan;
1173
David Howells1d1fe1e2008-02-07 00:15:37 -08001174 if (NEXT_ORPHAN(inode) > max_ino)
1175 goto bad_orphan;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001176 brelse(bitmap_bh);
1177 return inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08001178
David Howells1d1fe1e2008-02-07 00:15:37 -08001179bad_orphan:
Theodore Ts'o7827a7f2016-04-30 00:49:54 -04001180 ext4_error(sb, "bad orphan inode %lu", ino);
1181 if (bitmap_bh)
1182 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1183 bit, (unsigned long long)bitmap_bh->b_blocknr,
1184 ext4_test_bit(bit, bitmap_bh->b_data));
David Howells1d1fe1e2008-02-07 00:15:37 -08001185 if (inode) {
Theodore Ts'o7827a7f2016-04-30 00:49:54 -04001186 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001187 is_bad_inode(inode));
Theodore Ts'o7827a7f2016-04-30 00:49:54 -04001188 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
David Howells1d1fe1e2008-02-07 00:15:37 -08001189 NEXT_ORPHAN(inode));
Theodore Ts'o7827a7f2016-04-30 00:49:54 -04001190 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1191 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
David Howells1d1fe1e2008-02-07 00:15:37 -08001192 /* Avoid freeing blocks if we got a bad deleted inode */
1193 if (inode->i_nlink == 0)
1194 inode->i_blocks = 0;
1195 iput(inode);
1196 }
1197 brelse(bitmap_bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08001198 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001199}
1200
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001201unsigned long ext4_count_free_inodes(struct super_block *sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001202{
1203 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001204 struct ext4_group_desc *gdp;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001205 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001206#ifdef EXT4FS_DEBUG
1207 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001208 unsigned long bitmap_count, x;
1209 struct buffer_head *bitmap_bh = NULL;
1210
Mingming Cao617ba132006-10-11 01:20:53 -07001211 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001212 desc_count = 0;
1213 bitmap_count = 0;
1214 gdp = NULL;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001215 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001216 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 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001220 brelse(bitmap_bh);
Eric Sandeene29d1cd2008-08-02 21:21:02 -04001221 bitmap_bh = ext4_read_inode_bitmap(sb, i);
Darrick J. Wong9008a582015-10-17 21:33:24 -04001222 if (IS_ERR(bitmap_bh)) {
1223 bitmap_bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001224 continue;
Darrick J. Wong9008a582015-10-17 21:33:24 -04001225 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001226
Theodore Ts'of6fb99c2012-06-30 19:14:57 -04001227 x = ext4_count_free(bitmap_bh->b_data,
1228 EXT4_INODES_PER_GROUP(sb) / 8);
Eric Sandeenc549a952008-01-28 23:58:27 -05001229 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
Peng Tao785b4b32009-07-27 21:44:40 -04001230 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001231 bitmap_count += x;
1232 }
1233 brelse(bitmap_bh);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001234 printk(KERN_DEBUG "ext4_count_free_inodes: "
1235 "stored = %u, computed = %lu, %lu\n",
1236 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001237 return desc_count;
1238#else
1239 desc_count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001240 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001241 gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001242 if (!gdp)
1243 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001244 desc_count += ext4_free_inodes_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001245 cond_resched();
1246 }
1247 return desc_count;
1248#endif
1249}
1250
1251/* Called at mount-time, super-block is locked */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001252unsigned long ext4_count_dirs(struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001253{
1254 unsigned long count = 0;
Theodore Ts'o8df96752009-05-01 08:50:38 -04001255 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001256
Theodore Ts'o8df96752009-05-01 08:50:38 -04001257 for (i = 0; i < ngroups; i++) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001258 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001259 if (!gdp)
1260 continue;
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001261 count += ext4_used_dirs_count(sb, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001262 }
1263 return count;
1264}
Lukas Czernerbfff6872010-10-27 21:30:05 -04001265
1266/*
1267 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1268 * inode table. Must be called without any spinlock held. The only place
1269 * where it is called from on active part of filesystem is ext4lazyinit
1270 * thread, so we do not need any special locks, however we have to prevent
1271 * inode allocation from the current group, so we take alloc_sem lock, to
Theodore Ts'o119c0d42012-02-06 20:12:03 -05001272 * block ext4_new_inode() until we are finished.
Lukas Czernerbfff6872010-10-27 21:30:05 -04001273 */
H Hartley Sweetene0cbee32011-10-18 10:57:51 -04001274int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
Lukas Czernerbfff6872010-10-27 21:30:05 -04001275 int barrier)
1276{
1277 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1278 struct ext4_sb_info *sbi = EXT4_SB(sb);
1279 struct ext4_group_desc *gdp = NULL;
1280 struct buffer_head *group_desc_bh;
1281 handle_t *handle;
1282 ext4_fsblk_t blk;
1283 int num, ret = 0, used_blks = 0;
Lukas Czernerbfff6872010-10-27 21:30:05 -04001284
1285 /* This should not happen, but just to be sure check this */
1286 if (sb->s_flags & MS_RDONLY) {
1287 ret = 1;
1288 goto out;
1289 }
1290
1291 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1292 if (!gdp)
1293 goto out;
1294
1295 /*
1296 * We do not need to lock this, because we are the only one
1297 * handling this flag.
1298 */
1299 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1300 goto out;
1301
Theodore Ts'o9924a922013-02-08 21:59:22 -05001302 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001303 if (IS_ERR(handle)) {
1304 ret = PTR_ERR(handle);
1305 goto out;
1306 }
1307
1308 down_write(&grp->alloc_sem);
1309 /*
1310 * If inode bitmap was already initialized there may be some
1311 * used inodes so we need to skip blocks with used inodes in
1312 * inode table.
1313 */
1314 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1315 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1316 ext4_itable_unused_count(sb, gdp)),
1317 sbi->s_inodes_per_block);
1318
Theodore Ts'o954e5722018-07-28 08:12:04 -04001319 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1320 ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1321 ext4_itable_unused_count(sb, gdp)) <
1322 EXT4_FIRST_INO(sb)))) {
Theodore Ts'o1084f252012-03-19 23:13:43 -04001323 ext4_error(sb, "Something is wrong with group %u: "
1324 "used itable blocks: %d; "
1325 "itable unused count: %u",
Lukas Czerner857ac882010-10-27 21:30:05 -04001326 group, used_blks,
1327 ext4_itable_unused_count(sb, gdp));
1328 ret = 1;
Yongqiang Yang33853a02011-08-01 06:32:19 -04001329 goto err_out;
Lukas Czerner857ac882010-10-27 21:30:05 -04001330 }
1331
Lukas Czernerbfff6872010-10-27 21:30:05 -04001332 blk = ext4_inode_table(sb, gdp) + used_blks;
1333 num = sbi->s_itb_per_group - used_blks;
1334
1335 BUFFER_TRACE(group_desc_bh, "get_write_access");
1336 ret = ext4_journal_get_write_access(handle,
1337 group_desc_bh);
1338 if (ret)
1339 goto err_out;
1340
Lukas Czernerbfff6872010-10-27 21:30:05 -04001341 /*
1342 * Skip zeroout if the inode table is full. But we set the ZEROED
1343 * flag anyway, because obviously, when it is full it does not need
1344 * further zeroing.
1345 */
1346 if (unlikely(num == 0))
1347 goto skip_zeroout;
1348
1349 ext4_debug("going to zero out inode table in group %d\n",
1350 group);
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001351 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001352 if (ret < 0)
1353 goto err_out;
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001354 if (barrier)
1355 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001356
1357skip_zeroout:
1358 ext4_lock_group(sb, group);
1359 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001360 ext4_group_desc_csum_set(sb, group, gdp);
Lukas Czernerbfff6872010-10-27 21:30:05 -04001361 ext4_unlock_group(sb, group);
1362
1363 BUFFER_TRACE(group_desc_bh,
1364 "call ext4_handle_dirty_metadata");
1365 ret = ext4_handle_dirty_metadata(handle, NULL,
1366 group_desc_bh);
1367
1368err_out:
1369 up_write(&grp->alloc_sem);
1370 ext4_journal_stop(handle);
1371out:
1372 return ret;
1373}