blob: c88b439ba5cd5838d264450746c450cc45245c16 [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>
Mingming Cao617ba132006-10-11 01:20:53 -070018#include <linux/ext4_fs.h>
Mingming Caodab291a2006-10-11 01:21:01 -070019#include <linux/ext4_jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070020#include <linux/stat.h>
21#include <linux/string.h>
22#include <linux/quotaops.h>
23#include <linux/buffer_head.h>
24#include <linux/random.h>
25#include <linux/bitops.h>
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070026#include <linux/blkdev.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070027#include <asm/byteorder.h>
28
29#include "xattr.h"
30#include "acl.h"
31
32/*
33 * ialloc.c contains the inodes allocation and deallocation routines
34 */
35
36/*
37 * The free inodes are managed by bitmaps. A file system contains several
38 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
39 * block for inodes, N blocks for the inode table and data blocks.
40 *
41 * The file system contains group descriptors which are located after the
42 * super block. Each descriptor contains the number of the bitmap block and
43 * the free blocks count in the block.
44 */
45
46
47/*
48 * Read the inode allocation bitmap for a given block_group, reading
49 * into the specified slot in the superblock's bitmap cache.
50 *
51 * Return buffer_head of bitmap on success or NULL.
52 */
53static struct buffer_head *
54read_inode_bitmap(struct super_block * sb, unsigned long block_group)
55{
Mingming Cao617ba132006-10-11 01:20:53 -070056 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070057 struct buffer_head *bh = NULL;
58
Mingming Cao617ba132006-10-11 01:20:53 -070059 desc = ext4_get_group_desc(sb, block_group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070060 if (!desc)
61 goto error_out;
62
Alexandre Ratchov8fadc142006-10-11 01:21:15 -070063 bh = sb_bread(sb, ext4_inode_bitmap(sb, desc));
Dave Kleikampac27a0e2006-10-11 01:20:50 -070064 if (!bh)
Mingming Cao617ba132006-10-11 01:20:53 -070065 ext4_error(sb, "read_inode_bitmap",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070066 "Cannot read inode bitmap - "
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070067 "block_group = %lu, inode_bitmap = %llu",
Alexandre Ratchov8fadc142006-10-11 01:21:15 -070068 block_group, ext4_inode_bitmap(sb, desc));
Dave Kleikampac27a0e2006-10-11 01:20:50 -070069error_out:
70 return bh;
71}
72
73/*
74 * NOTE! When we get the inode, we're the only people
75 * that have access to it, and as such there are no
76 * race conditions we have to worry about. The inode
77 * is not on the hash-lists, and it cannot be reached
78 * through the filesystem because the directory entry
79 * has been deleted earlier.
80 *
81 * HOWEVER: we must make sure that we get no aliases,
82 * which means that we have to call "clear_inode()"
83 * _before_ we mark the inode not in use in the inode
84 * bitmaps. Otherwise a newly created file might use
85 * the same inode number (not actually the same pointer
86 * though), and then we'd have two inodes sharing the
87 * same inode number and space on the harddisk.
88 */
Mingming Cao617ba132006-10-11 01:20:53 -070089void ext4_free_inode (handle_t *handle, struct inode * inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070090{
91 struct super_block * sb = inode->i_sb;
92 int is_directory;
93 unsigned long ino;
94 struct buffer_head *bitmap_bh = NULL;
95 struct buffer_head *bh2;
96 unsigned long block_group;
97 unsigned long bit;
Mingming Cao617ba132006-10-11 01:20:53 -070098 struct ext4_group_desc * gdp;
99 struct ext4_super_block * es;
100 struct ext4_sb_info *sbi;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700101 int fatal = 0, err;
102
103 if (atomic_read(&inode->i_count) > 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700104 printk ("ext4_free_inode: inode has count=%d\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700105 atomic_read(&inode->i_count));
106 return;
107 }
108 if (inode->i_nlink) {
Mingming Cao617ba132006-10-11 01:20:53 -0700109 printk ("ext4_free_inode: inode has nlink=%d\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700110 inode->i_nlink);
111 return;
112 }
113 if (!sb) {
Mingming Cao617ba132006-10-11 01:20:53 -0700114 printk("ext4_free_inode: inode on nonexistent device\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700115 return;
116 }
Mingming Cao617ba132006-10-11 01:20:53 -0700117 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700118
119 ino = inode->i_ino;
Mingming Cao617ba132006-10-11 01:20:53 -0700120 ext4_debug ("freeing inode %lu\n", ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700121
122 /*
123 * Note: we must free any quota before locking the superblock,
124 * as writing the quota to disk may need the lock as well.
125 */
126 DQUOT_INIT(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700127 ext4_xattr_delete_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700128 DQUOT_FREE_INODE(inode);
129 DQUOT_DROP(inode);
130
131 is_directory = S_ISDIR(inode->i_mode);
132
133 /* Do this BEFORE marking the inode not in use or returning an error */
134 clear_inode (inode);
135
Mingming Cao617ba132006-10-11 01:20:53 -0700136 es = EXT4_SB(sb)->s_es;
137 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
138 ext4_error (sb, "ext4_free_inode",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700139 "reserved or nonexistent inode %lu", ino);
140 goto error_return;
141 }
Mingming Cao617ba132006-10-11 01:20:53 -0700142 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
143 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700144 bitmap_bh = read_inode_bitmap(sb, block_group);
145 if (!bitmap_bh)
146 goto error_return;
147
148 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700149 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700150 if (fatal)
151 goto error_return;
152
153 /* Ok, now we can actually update the inode bitmaps.. */
Mingming Cao617ba132006-10-11 01:20:53 -0700154 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700155 bit, bitmap_bh->b_data))
Mingming Cao617ba132006-10-11 01:20:53 -0700156 ext4_error (sb, "ext4_free_inode",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700157 "bit already cleared for inode %lu", ino);
158 else {
Mingming Cao617ba132006-10-11 01:20:53 -0700159 gdp = ext4_get_group_desc (sb, block_group, &bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700160
161 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700162 fatal = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700163 if (fatal) goto error_return;
164
165 if (gdp) {
166 spin_lock(sb_bgl_lock(sbi, block_group));
167 gdp->bg_free_inodes_count = cpu_to_le16(
168 le16_to_cpu(gdp->bg_free_inodes_count) + 1);
169 if (is_directory)
170 gdp->bg_used_dirs_count = cpu_to_le16(
171 le16_to_cpu(gdp->bg_used_dirs_count) - 1);
172 spin_unlock(sb_bgl_lock(sbi, block_group));
173 percpu_counter_inc(&sbi->s_freeinodes_counter);
174 if (is_directory)
175 percpu_counter_dec(&sbi->s_dirs_counter);
176
177 }
Mingming Cao617ba132006-10-11 01:20:53 -0700178 BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
179 err = ext4_journal_dirty_metadata(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700180 if (!fatal) fatal = err;
181 }
Mingming Cao617ba132006-10-11 01:20:53 -0700182 BUFFER_TRACE(bitmap_bh, "call ext4_journal_dirty_metadata");
183 err = ext4_journal_dirty_metadata(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700184 if (!fatal)
185 fatal = err;
186 sb->s_dirt = 1;
187error_return:
188 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700189 ext4_std_error(sb, fatal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700190}
191
192/*
193 * There are two policies for allocating an inode. If the new inode is
194 * a directory, then a forward search is made for a block group with both
195 * free space and a low directory-to-inode ratio; if that fails, then of
196 * the groups with above-average free space, that group with the fewest
197 * directories already is chosen.
198 *
199 * For other inodes, search forward from the parent directory\'s block
200 * group to find a free inode.
201 */
202static int find_group_dir(struct super_block *sb, struct inode *parent)
203{
Mingming Cao617ba132006-10-11 01:20:53 -0700204 int ngroups = EXT4_SB(sb)->s_groups_count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700205 unsigned int freei, avefreei;
Mingming Cao617ba132006-10-11 01:20:53 -0700206 struct ext4_group_desc *desc, *best_desc = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700207 struct buffer_head *bh;
208 int group, best_group = -1;
209
Mingming Cao617ba132006-10-11 01:20:53 -0700210 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700211 avefreei = freei / ngroups;
212
213 for (group = 0; group < ngroups; group++) {
Mingming Cao617ba132006-10-11 01:20:53 -0700214 desc = ext4_get_group_desc (sb, group, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700215 if (!desc || !desc->bg_free_inodes_count)
216 continue;
217 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
218 continue;
219 if (!best_desc ||
220 (le16_to_cpu(desc->bg_free_blocks_count) >
221 le16_to_cpu(best_desc->bg_free_blocks_count))) {
222 best_group = group;
223 best_desc = desc;
224 }
225 }
226 return best_group;
227}
228
229/*
230 * Orlov's allocator for directories.
231 *
232 * We always try to spread first-level directories.
233 *
234 * If there are blockgroups with both free inodes and free blocks counts
235 * not worse than average we return one with smallest directory count.
236 * Otherwise we simply return a random group.
237 *
238 * For the rest rules look so:
239 *
240 * It's OK to put directory into a group unless
241 * it has too many directories already (max_dirs) or
242 * it has too few free inodes left (min_inodes) or
243 * it has too few free blocks left (min_blocks) or
244 * it's already running too large debt (max_debt).
245 * Parent's group is prefered, if it doesn't satisfy these
246 * conditions we search cyclically through the rest. If none
247 * of the groups look good we just look for a group with more
248 * free inodes than average (starting at parent's group).
249 *
250 * Debt is incremented each time we allocate a directory and decremented
251 * when we allocate an inode, within 0--255.
252 */
253
254#define INODE_COST 64
255#define BLOCK_COST 256
256
257static int find_group_orlov(struct super_block *sb, struct inode *parent)
258{
Mingming Cao617ba132006-10-11 01:20:53 -0700259 int parent_group = EXT4_I(parent)->i_block_group;
260 struct ext4_sb_info *sbi = EXT4_SB(sb);
261 struct ext4_super_block *es = sbi->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700262 int ngroups = sbi->s_groups_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700263 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700264 unsigned int freei, avefreei;
Mingming Cao617ba132006-10-11 01:20:53 -0700265 ext4_fsblk_t freeb, avefreeb;
266 ext4_fsblk_t blocks_per_dir;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700267 unsigned int ndirs;
268 int max_debt, max_dirs, min_inodes;
Mingming Cao617ba132006-10-11 01:20:53 -0700269 ext4_grpblk_t min_blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700270 int group = -1, i;
Mingming Cao617ba132006-10-11 01:20:53 -0700271 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700272 struct buffer_head *bh;
273
274 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
275 avefreei = freei / ngroups;
276 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700277 avefreeb = freeb;
Andrew Mortonf4e5bc22006-10-11 01:21:19 -0700278 do_div(avefreeb, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700279 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
280
281 if ((parent == sb->s_root->d_inode) ||
Mingming Cao617ba132006-10-11 01:20:53 -0700282 (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700283 int best_ndir = inodes_per_group;
284 int best_group = -1;
285
286 get_random_bytes(&group, sizeof(group));
287 parent_group = (unsigned)group % ngroups;
288 for (i = 0; i < ngroups; i++) {
289 group = (parent_group + i) % ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700290 desc = ext4_get_group_desc (sb, group, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291 if (!desc || !desc->bg_free_inodes_count)
292 continue;
293 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
294 continue;
295 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
296 continue;
297 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
298 continue;
299 best_group = group;
300 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
301 }
302 if (best_group >= 0)
303 return best_group;
304 goto fallback;
305 }
306
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700307 blocks_per_dir = ext4_blocks_count(es) - freeb;
Andrew Mortonf4e5bc22006-10-11 01:21:19 -0700308 do_div(blocks_per_dir, ndirs);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700309
310 max_dirs = ndirs / ngroups + inodes_per_group / 16;
311 min_inodes = avefreei - inodes_per_group / 4;
Mingming Cao617ba132006-10-11 01:20:53 -0700312 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700313
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700314 max_debt = EXT4_BLOCKS_PER_GROUP(sb);
Andrew Mortonf4e5bc22006-10-11 01:21:19 -0700315 max_debt /= max_t(int, blocks_per_dir, BLOCK_COST);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700316 if (max_debt * INODE_COST > inodes_per_group)
317 max_debt = inodes_per_group / INODE_COST;
318 if (max_debt > 255)
319 max_debt = 255;
320 if (max_debt == 0)
321 max_debt = 1;
322
323 for (i = 0; i < ngroups; i++) {
324 group = (parent_group + i) % ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700325 desc = ext4_get_group_desc (sb, group, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700326 if (!desc || !desc->bg_free_inodes_count)
327 continue;
328 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
329 continue;
330 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
331 continue;
332 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
333 continue;
334 return group;
335 }
336
337fallback:
338 for (i = 0; i < ngroups; i++) {
339 group = (parent_group + i) % ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700340 desc = ext4_get_group_desc (sb, group, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700341 if (!desc || !desc->bg_free_inodes_count)
342 continue;
343 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
344 return group;
345 }
346
347 if (avefreei) {
348 /*
349 * The free-inodes counter is approximate, and for really small
350 * filesystems the above test can fail to find any blockgroups
351 */
352 avefreei = 0;
353 goto fallback;
354 }
355
356 return -1;
357}
358
359static int find_group_other(struct super_block *sb, struct inode *parent)
360{
Mingming Cao617ba132006-10-11 01:20:53 -0700361 int parent_group = EXT4_I(parent)->i_block_group;
362 int ngroups = EXT4_SB(sb)->s_groups_count;
363 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700364 struct buffer_head *bh;
365 int group, i;
366
367 /*
368 * Try to place the inode in its parent directory
369 */
370 group = parent_group;
Mingming Cao617ba132006-10-11 01:20:53 -0700371 desc = ext4_get_group_desc (sb, group, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700372 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
373 le16_to_cpu(desc->bg_free_blocks_count))
374 return group;
375
376 /*
377 * We're going to place this inode in a different blockgroup from its
378 * parent. We want to cause files in a common directory to all land in
379 * the same blockgroup. But we want files which are in a different
380 * directory which shares a blockgroup with our parent to land in a
381 * different blockgroup.
382 *
383 * So add our directory's i_ino into the starting point for the hash.
384 */
385 group = (group + parent->i_ino) % ngroups;
386
387 /*
388 * Use a quadratic hash to find a group with a free inode and some free
389 * blocks.
390 */
391 for (i = 1; i < ngroups; i <<= 1) {
392 group += i;
393 if (group >= ngroups)
394 group -= ngroups;
Mingming Cao617ba132006-10-11 01:20:53 -0700395 desc = ext4_get_group_desc (sb, group, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700396 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
397 le16_to_cpu(desc->bg_free_blocks_count))
398 return group;
399 }
400
401 /*
402 * That failed: try linear search for a free inode, even if that group
403 * has no free blocks.
404 */
405 group = parent_group;
406 for (i = 0; i < ngroups; i++) {
407 if (++group >= ngroups)
408 group = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700409 desc = ext4_get_group_desc (sb, group, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700410 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
411 return group;
412 }
413
414 return -1;
415}
416
417/*
418 * There are two policies for allocating an inode. If the new inode is
419 * a directory, then a forward search is made for a block group with both
420 * free space and a low directory-to-inode ratio; if that fails, then of
421 * the groups with above-average free space, that group with the fewest
422 * directories already is chosen.
423 *
424 * For other inodes, search forward from the parent directory's block
425 * group to find a free inode.
426 */
Mingming Cao617ba132006-10-11 01:20:53 -0700427struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700428{
429 struct super_block *sb;
430 struct buffer_head *bitmap_bh = NULL;
431 struct buffer_head *bh2;
432 int group;
433 unsigned long ino = 0;
434 struct inode * inode;
Mingming Cao617ba132006-10-11 01:20:53 -0700435 struct ext4_group_desc * gdp = NULL;
436 struct ext4_super_block * es;
437 struct ext4_inode_info *ei;
438 struct ext4_sb_info *sbi;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700439 int err = 0;
440 struct inode *ret;
441 int i;
442
443 /* Cannot create files in a deleted directory */
444 if (!dir || !dir->i_nlink)
445 return ERR_PTR(-EPERM);
446
447 sb = dir->i_sb;
448 inode = new_inode(sb);
449 if (!inode)
450 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700451 ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700452
Mingming Cao617ba132006-10-11 01:20:53 -0700453 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700454 es = sbi->s_es;
455 if (S_ISDIR(mode)) {
456 if (test_opt (sb, OLDALLOC))
457 group = find_group_dir(sb, dir);
458 else
459 group = find_group_orlov(sb, dir);
460 } else
461 group = find_group_other(sb, dir);
462
463 err = -ENOSPC;
464 if (group == -1)
465 goto out;
466
467 for (i = 0; i < sbi->s_groups_count; i++) {
468 err = -EIO;
469
Mingming Cao617ba132006-10-11 01:20:53 -0700470 gdp = ext4_get_group_desc(sb, group, &bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700471 if (!gdp)
472 goto fail;
473
474 brelse(bitmap_bh);
475 bitmap_bh = read_inode_bitmap(sb, group);
476 if (!bitmap_bh)
477 goto fail;
478
479 ino = 0;
480
481repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700482 ino = ext4_find_next_zero_bit((unsigned long *)
483 bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino);
484 if (ino < EXT4_INODES_PER_GROUP(sb)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700485
486 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700487 err = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700488 if (err)
489 goto fail;
490
Mingming Cao617ba132006-10-11 01:20:53 -0700491 if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700492 ino, bitmap_bh->b_data)) {
493 /* we won it */
494 BUFFER_TRACE(bitmap_bh,
Mingming Cao617ba132006-10-11 01:20:53 -0700495 "call ext4_journal_dirty_metadata");
496 err = ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700497 bitmap_bh);
498 if (err)
499 goto fail;
500 goto got;
501 }
502 /* we lost it */
Mingming Caodab291a2006-10-11 01:21:01 -0700503 jbd2_journal_release_buffer(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700504
Mingming Cao617ba132006-10-11 01:20:53 -0700505 if (++ino < EXT4_INODES_PER_GROUP(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700506 goto repeat_in_this_group;
507 }
508
509 /*
510 * This case is possible in concurrent environment. It is very
511 * rare. We cannot repeat the find_group_xxx() call because
512 * that will simply return the same blockgroup, because the
513 * group descriptor metadata has not yet been updated.
514 * So we just go onto the next blockgroup.
515 */
516 if (++group == sbi->s_groups_count)
517 group = 0;
518 }
519 err = -ENOSPC;
520 goto out;
521
522got:
Mingming Cao617ba132006-10-11 01:20:53 -0700523 ino += group * EXT4_INODES_PER_GROUP(sb) + 1;
524 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
525 ext4_error (sb, "ext4_new_inode",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700526 "reserved inode or inode > inodes count - "
527 "block_group = %d, inode=%lu", group, ino);
528 err = -EIO;
529 goto fail;
530 }
531
532 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700533 err = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700534 if (err) goto fail;
535 spin_lock(sb_bgl_lock(sbi, group));
536 gdp->bg_free_inodes_count =
537 cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
538 if (S_ISDIR(mode)) {
539 gdp->bg_used_dirs_count =
540 cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
541 }
542 spin_unlock(sb_bgl_lock(sbi, group));
Mingming Cao617ba132006-10-11 01:20:53 -0700543 BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
544 err = ext4_journal_dirty_metadata(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700545 if (err) goto fail;
546
547 percpu_counter_dec(&sbi->s_freeinodes_counter);
548 if (S_ISDIR(mode))
549 percpu_counter_inc(&sbi->s_dirs_counter);
550 sb->s_dirt = 1;
551
552 inode->i_uid = current->fsuid;
553 if (test_opt (sb, GRPID))
554 inode->i_gid = dir->i_gid;
555 else if (dir->i_mode & S_ISGID) {
556 inode->i_gid = dir->i_gid;
557 if (S_ISDIR(mode))
558 mode |= S_ISGID;
559 } else
560 inode->i_gid = current->fsgid;
561 inode->i_mode = mode;
562
563 inode->i_ino = ino;
564 /* This is the optimal IO size (for stat), not the fs block size */
565 inode->i_blocks = 0;
566 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
567
568 memset(ei->i_data, 0, sizeof(ei->i_data));
569 ei->i_dir_start_lookup = 0;
570 ei->i_disksize = 0;
571
Mingming Cao617ba132006-10-11 01:20:53 -0700572 ei->i_flags = EXT4_I(dir)->i_flags & ~EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700573 if (S_ISLNK(mode))
Mingming Cao617ba132006-10-11 01:20:53 -0700574 ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700575 /* dirsync only applies to directories */
576 if (!S_ISDIR(mode))
Mingming Cao617ba132006-10-11 01:20:53 -0700577 ei->i_flags &= ~EXT4_DIRSYNC_FL;
578#ifdef EXT4_FRAGMENTS
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700579 ei->i_faddr = 0;
580 ei->i_frag_no = 0;
581 ei->i_frag_size = 0;
582#endif
583 ei->i_file_acl = 0;
584 ei->i_dir_acl = 0;
585 ei->i_dtime = 0;
586 ei->i_block_alloc_info = NULL;
587 ei->i_block_group = group;
588
Mingming Cao617ba132006-10-11 01:20:53 -0700589 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700590 if (IS_DIRSYNC(inode))
591 handle->h_sync = 1;
592 insert_inode_hash(inode);
593 spin_lock(&sbi->s_next_gen_lock);
594 inode->i_generation = sbi->s_next_generation++;
595 spin_unlock(&sbi->s_next_gen_lock);
596
Mingming Cao617ba132006-10-11 01:20:53 -0700597 ei->i_state = EXT4_STATE_NEW;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700598 ei->i_extra_isize =
Mingming Cao617ba132006-10-11 01:20:53 -0700599 (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) ?
600 sizeof(struct ext4_inode) - EXT4_GOOD_OLD_INODE_SIZE : 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700601
602 ret = inode;
603 if(DQUOT_ALLOC_INODE(inode)) {
604 err = -EDQUOT;
605 goto fail_drop;
606 }
607
Mingming Cao617ba132006-10-11 01:20:53 -0700608 err = ext4_init_acl(handle, inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700609 if (err)
610 goto fail_free_drop;
611
Mingming Cao617ba132006-10-11 01:20:53 -0700612 err = ext4_init_security(handle,inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700613 if (err)
614 goto fail_free_drop;
615
Mingming Cao617ba132006-10-11 01:20:53 -0700616 err = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700617 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -0700618 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700619 goto fail_free_drop;
620 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700621 if (test_opt(sb, EXTENTS)) {
622 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
623 ext4_ext_tree_init(handle, inode);
624 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
625 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
626 if (err) goto fail;
627 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS);
628 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "call ext4_journal_dirty_metadata");
629 err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
630 }
631 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700632
Mingming Cao617ba132006-10-11 01:20:53 -0700633 ext4_debug("allocating inode %lu\n", inode->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700634 goto really_out;
635fail:
Mingming Cao617ba132006-10-11 01:20:53 -0700636 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700637out:
638 iput(inode);
639 ret = ERR_PTR(err);
640really_out:
641 brelse(bitmap_bh);
642 return ret;
643
644fail_free_drop:
645 DQUOT_FREE_INODE(inode);
646
647fail_drop:
648 DQUOT_DROP(inode);
649 inode->i_flags |= S_NOQUOTA;
650 inode->i_nlink = 0;
651 iput(inode);
652 brelse(bitmap_bh);
653 return ERR_PTR(err);
654}
655
656/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -0700657struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700658{
Mingming Cao617ba132006-10-11 01:20:53 -0700659 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700660 unsigned long block_group;
661 int bit;
662 struct buffer_head *bitmap_bh = NULL;
663 struct inode *inode = NULL;
664
665 /* Error cases - e2fsck has already cleaned up for us */
666 if (ino > max_ino) {
Mingming Cao617ba132006-10-11 01:20:53 -0700667 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700668 "bad orphan ino %lu! e2fsck was run?", ino);
669 goto out;
670 }
671
Mingming Cao617ba132006-10-11 01:20:53 -0700672 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
673 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700674 bitmap_bh = read_inode_bitmap(sb, block_group);
675 if (!bitmap_bh) {
Mingming Cao617ba132006-10-11 01:20:53 -0700676 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700677 "inode bitmap error for orphan %lu", ino);
678 goto out;
679 }
680
681 /* Having the inode bit set should be a 100% indicator that this
682 * is a valid orphan (no e2fsck run on fs). Orphans also include
683 * inodes that were being truncated, so we can't check i_nlink==0.
684 */
Mingming Cao617ba132006-10-11 01:20:53 -0700685 if (!ext4_test_bit(bit, bitmap_bh->b_data) ||
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700686 !(inode = iget(sb, ino)) || is_bad_inode(inode) ||
687 NEXT_ORPHAN(inode) > max_ino) {
Mingming Cao617ba132006-10-11 01:20:53 -0700688 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700689 "bad orphan inode %lu! e2fsck was run?", ino);
Mingming Cao617ba132006-10-11 01:20:53 -0700690 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700691 bit, (unsigned long long)bitmap_bh->b_blocknr,
Mingming Cao617ba132006-10-11 01:20:53 -0700692 ext4_test_bit(bit, bitmap_bh->b_data));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700693 printk(KERN_NOTICE "inode=%p\n", inode);
694 if (inode) {
695 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
696 is_bad_inode(inode));
697 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
698 NEXT_ORPHAN(inode));
699 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
700 }
701 /* Avoid freeing blocks if we got a bad deleted inode */
702 if (inode && inode->i_nlink == 0)
703 inode->i_blocks = 0;
704 iput(inode);
705 inode = NULL;
706 }
707out:
708 brelse(bitmap_bh);
709 return inode;
710}
711
Mingming Cao617ba132006-10-11 01:20:53 -0700712unsigned long ext4_count_free_inodes (struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700713{
714 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700715 struct ext4_group_desc *gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700716 int i;
Mingming Cao617ba132006-10-11 01:20:53 -0700717#ifdef EXT4FS_DEBUG
718 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700719 unsigned long bitmap_count, x;
720 struct buffer_head *bitmap_bh = NULL;
721
Mingming Cao617ba132006-10-11 01:20:53 -0700722 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723 desc_count = 0;
724 bitmap_count = 0;
725 gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700726 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
727 gdp = ext4_get_group_desc (sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700728 if (!gdp)
729 continue;
730 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
731 brelse(bitmap_bh);
732 bitmap_bh = read_inode_bitmap(sb, i);
733 if (!bitmap_bh)
734 continue;
735
Mingming Cao617ba132006-10-11 01:20:53 -0700736 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700737 printk("group %d: stored = %d, counted = %lu\n",
738 i, le16_to_cpu(gdp->bg_free_inodes_count), x);
739 bitmap_count += x;
740 }
741 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700742 printk("ext4_count_free_inodes: stored = %u, computed = %lu, %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700743 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
744 return desc_count;
745#else
746 desc_count = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700747 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
748 gdp = ext4_get_group_desc (sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700749 if (!gdp)
750 continue;
751 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
752 cond_resched();
753 }
754 return desc_count;
755#endif
756}
757
758/* Called at mount-time, super-block is locked */
Mingming Cao617ba132006-10-11 01:20:53 -0700759unsigned long ext4_count_dirs (struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700760{
761 unsigned long count = 0;
762 int i;
763
Mingming Cao617ba132006-10-11 01:20:53 -0700764 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
765 struct ext4_group_desc *gdp = ext4_get_group_desc (sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700766 if (!gdp)
767 continue;
768 count += le16_to_cpu(gdp->bg_used_dirs_count);
769 }
770 return count;
771}
772