blob: d0c7793d939395c2500def60e7aa01a680cf2704 [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 int group, best_group = -1;
208
Mingming Cao617ba132006-10-11 01:20:53 -0700209 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700210 avefreei = freei / ngroups;
211
212 for (group = 0; group < ngroups; group++) {
Eric Sandeenef2fb672007-10-16 23:26:30 -0700213 desc = ext4_get_group_desc (sb, group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700214 if (!desc || !desc->bg_free_inodes_count)
215 continue;
216 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
217 continue;
218 if (!best_desc ||
219 (le16_to_cpu(desc->bg_free_blocks_count) >
220 le16_to_cpu(best_desc->bg_free_blocks_count))) {
221 best_group = group;
222 best_desc = desc;
223 }
224 }
225 return best_group;
226}
227
228/*
229 * Orlov's allocator for directories.
230 *
231 * We always try to spread first-level directories.
232 *
233 * If there are blockgroups with both free inodes and free blocks counts
234 * not worse than average we return one with smallest directory count.
235 * Otherwise we simply return a random group.
236 *
237 * For the rest rules look so:
238 *
239 * It's OK to put directory into a group unless
240 * it has too many directories already (max_dirs) or
241 * it has too few free inodes left (min_inodes) or
242 * it has too few free blocks left (min_blocks) or
243 * it's already running too large debt (max_debt).
244 * Parent's group is prefered, if it doesn't satisfy these
245 * conditions we search cyclically through the rest. If none
246 * of the groups look good we just look for a group with more
247 * free inodes than average (starting at parent's group).
248 *
249 * Debt is incremented each time we allocate a directory and decremented
250 * when we allocate an inode, within 0--255.
251 */
252
253#define INODE_COST 64
254#define BLOCK_COST 256
255
256static int find_group_orlov(struct super_block *sb, struct inode *parent)
257{
Mingming Cao617ba132006-10-11 01:20:53 -0700258 int parent_group = EXT4_I(parent)->i_block_group;
259 struct ext4_sb_info *sbi = EXT4_SB(sb);
260 struct ext4_super_block *es = sbi->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700261 int ngroups = sbi->s_groups_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700262 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700263 unsigned int freei, avefreei;
Mingming Cao617ba132006-10-11 01:20:53 -0700264 ext4_fsblk_t freeb, avefreeb;
265 ext4_fsblk_t blocks_per_dir;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700266 unsigned int ndirs;
267 int max_debt, max_dirs, min_inodes;
Mingming Cao617ba132006-10-11 01:20:53 -0700268 ext4_grpblk_t min_blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700269 int group = -1, i;
Mingming Cao617ba132006-10-11 01:20:53 -0700270 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700271
272 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
273 avefreei = freei / ngroups;
274 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700275 avefreeb = freeb;
Andrew Mortonf4e5bc22006-10-11 01:21:19 -0700276 do_div(avefreeb, ngroups);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700277 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
278
279 if ((parent == sb->s_root->d_inode) ||
Mingming Cao617ba132006-10-11 01:20:53 -0700280 (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700281 int best_ndir = inodes_per_group;
282 int best_group = -1;
283
284 get_random_bytes(&group, sizeof(group));
285 parent_group = (unsigned)group % ngroups;
286 for (i = 0; i < ngroups; i++) {
287 group = (parent_group + i) % ngroups;
Eric Sandeenef2fb672007-10-16 23:26:30 -0700288 desc = ext4_get_group_desc (sb, group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700289 if (!desc || !desc->bg_free_inodes_count)
290 continue;
291 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
292 continue;
293 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
294 continue;
295 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
296 continue;
297 best_group = group;
298 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
299 }
300 if (best_group >= 0)
301 return best_group;
302 goto fallback;
303 }
304
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700305 blocks_per_dir = ext4_blocks_count(es) - freeb;
Andrew Mortonf4e5bc22006-10-11 01:21:19 -0700306 do_div(blocks_per_dir, ndirs);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700307
308 max_dirs = ndirs / ngroups + inodes_per_group / 16;
309 min_inodes = avefreei - inodes_per_group / 4;
Mingming Cao617ba132006-10-11 01:20:53 -0700310 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700311
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700312 max_debt = EXT4_BLOCKS_PER_GROUP(sb);
Andrew Mortonf4e5bc22006-10-11 01:21:19 -0700313 max_debt /= max_t(int, blocks_per_dir, BLOCK_COST);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700314 if (max_debt * INODE_COST > inodes_per_group)
315 max_debt = inodes_per_group / INODE_COST;
316 if (max_debt > 255)
317 max_debt = 255;
318 if (max_debt == 0)
319 max_debt = 1;
320
321 for (i = 0; i < ngroups; i++) {
322 group = (parent_group + i) % ngroups;
Eric Sandeenef2fb672007-10-16 23:26:30 -0700323 desc = ext4_get_group_desc (sb, group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700324 if (!desc || !desc->bg_free_inodes_count)
325 continue;
326 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
327 continue;
328 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
329 continue;
330 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
331 continue;
332 return group;
333 }
334
335fallback:
336 for (i = 0; i < ngroups; i++) {
337 group = (parent_group + i) % ngroups;
Eric Sandeenef2fb672007-10-16 23:26:30 -0700338 desc = ext4_get_group_desc (sb, group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700339 if (!desc || !desc->bg_free_inodes_count)
340 continue;
341 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
342 return group;
343 }
344
345 if (avefreei) {
346 /*
347 * The free-inodes counter is approximate, and for really small
348 * filesystems the above test can fail to find any blockgroups
349 */
350 avefreei = 0;
351 goto fallback;
352 }
353
354 return -1;
355}
356
357static int find_group_other(struct super_block *sb, struct inode *parent)
358{
Mingming Cao617ba132006-10-11 01:20:53 -0700359 int parent_group = EXT4_I(parent)->i_block_group;
360 int ngroups = EXT4_SB(sb)->s_groups_count;
361 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700362 int group, i;
363
364 /*
365 * Try to place the inode in its parent directory
366 */
367 group = parent_group;
Eric Sandeenef2fb672007-10-16 23:26:30 -0700368 desc = ext4_get_group_desc (sb, group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700369 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
370 le16_to_cpu(desc->bg_free_blocks_count))
371 return group;
372
373 /*
374 * We're going to place this inode in a different blockgroup from its
375 * parent. We want to cause files in a common directory to all land in
376 * the same blockgroup. But we want files which are in a different
377 * directory which shares a blockgroup with our parent to land in a
378 * different blockgroup.
379 *
380 * So add our directory's i_ino into the starting point for the hash.
381 */
382 group = (group + parent->i_ino) % ngroups;
383
384 /*
385 * Use a quadratic hash to find a group with a free inode and some free
386 * blocks.
387 */
388 for (i = 1; i < ngroups; i <<= 1) {
389 group += i;
390 if (group >= ngroups)
391 group -= ngroups;
Eric Sandeenef2fb672007-10-16 23:26:30 -0700392 desc = ext4_get_group_desc (sb, group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700393 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
394 le16_to_cpu(desc->bg_free_blocks_count))
395 return group;
396 }
397
398 /*
399 * That failed: try linear search for a free inode, even if that group
400 * has no free blocks.
401 */
402 group = parent_group;
403 for (i = 0; i < ngroups; i++) {
404 if (++group >= ngroups)
405 group = 0;
Eric Sandeenef2fb672007-10-16 23:26:30 -0700406 desc = ext4_get_group_desc (sb, group, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700407 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
408 return group;
409 }
410
411 return -1;
412}
413
414/*
415 * There are two policies for allocating an inode. If the new inode is
416 * a directory, then a forward search is made for a block group with both
417 * free space and a low directory-to-inode ratio; if that fails, then of
418 * the groups with above-average free space, that group with the fewest
419 * directories already is chosen.
420 *
421 * For other inodes, search forward from the parent directory's block
422 * group to find a free inode.
423 */
Mingming Cao617ba132006-10-11 01:20:53 -0700424struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700425{
426 struct super_block *sb;
427 struct buffer_head *bitmap_bh = NULL;
428 struct buffer_head *bh2;
429 int group;
430 unsigned long ino = 0;
431 struct inode * inode;
Mingming Cao617ba132006-10-11 01:20:53 -0700432 struct ext4_group_desc * gdp = NULL;
433 struct ext4_super_block * es;
434 struct ext4_inode_info *ei;
435 struct ext4_sb_info *sbi;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700436 int err = 0;
437 struct inode *ret;
438 int i;
439
440 /* Cannot create files in a deleted directory */
441 if (!dir || !dir->i_nlink)
442 return ERR_PTR(-EPERM);
443
444 sb = dir->i_sb;
445 inode = new_inode(sb);
446 if (!inode)
447 return ERR_PTR(-ENOMEM);
Mingming Cao617ba132006-10-11 01:20:53 -0700448 ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700449
Mingming Cao617ba132006-10-11 01:20:53 -0700450 sbi = EXT4_SB(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700451 es = sbi->s_es;
452 if (S_ISDIR(mode)) {
453 if (test_opt (sb, OLDALLOC))
454 group = find_group_dir(sb, dir);
455 else
456 group = find_group_orlov(sb, dir);
457 } else
458 group = find_group_other(sb, dir);
459
460 err = -ENOSPC;
461 if (group == -1)
462 goto out;
463
464 for (i = 0; i < sbi->s_groups_count; i++) {
465 err = -EIO;
466
Mingming Cao617ba132006-10-11 01:20:53 -0700467 gdp = ext4_get_group_desc(sb, group, &bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700468 if (!gdp)
469 goto fail;
470
471 brelse(bitmap_bh);
472 bitmap_bh = read_inode_bitmap(sb, group);
473 if (!bitmap_bh)
474 goto fail;
475
476 ino = 0;
477
478repeat_in_this_group:
Mingming Cao617ba132006-10-11 01:20:53 -0700479 ino = ext4_find_next_zero_bit((unsigned long *)
480 bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino);
481 if (ino < EXT4_INODES_PER_GROUP(sb)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700482
483 BUFFER_TRACE(bitmap_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700484 err = ext4_journal_get_write_access(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700485 if (err)
486 goto fail;
487
Mingming Cao617ba132006-10-11 01:20:53 -0700488 if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700489 ino, bitmap_bh->b_data)) {
490 /* we won it */
491 BUFFER_TRACE(bitmap_bh,
Mingming Cao617ba132006-10-11 01:20:53 -0700492 "call ext4_journal_dirty_metadata");
493 err = ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700494 bitmap_bh);
495 if (err)
496 goto fail;
497 goto got;
498 }
499 /* we lost it */
Mingming Caodab291a2006-10-11 01:21:01 -0700500 jbd2_journal_release_buffer(handle, bitmap_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700501
Mingming Cao617ba132006-10-11 01:20:53 -0700502 if (++ino < EXT4_INODES_PER_GROUP(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700503 goto repeat_in_this_group;
504 }
505
506 /*
507 * This case is possible in concurrent environment. It is very
508 * rare. We cannot repeat the find_group_xxx() call because
509 * that will simply return the same blockgroup, because the
510 * group descriptor metadata has not yet been updated.
511 * So we just go onto the next blockgroup.
512 */
513 if (++group == sbi->s_groups_count)
514 group = 0;
515 }
516 err = -ENOSPC;
517 goto out;
518
519got:
Mingming Cao617ba132006-10-11 01:20:53 -0700520 ino += group * EXT4_INODES_PER_GROUP(sb) + 1;
521 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
522 ext4_error (sb, "ext4_new_inode",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700523 "reserved inode or inode > inodes count - "
524 "block_group = %d, inode=%lu", group, ino);
525 err = -EIO;
526 goto fail;
527 }
528
529 BUFFER_TRACE(bh2, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700530 err = ext4_journal_get_write_access(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700531 if (err) goto fail;
532 spin_lock(sb_bgl_lock(sbi, group));
533 gdp->bg_free_inodes_count =
534 cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
535 if (S_ISDIR(mode)) {
536 gdp->bg_used_dirs_count =
537 cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
538 }
539 spin_unlock(sb_bgl_lock(sbi, group));
Mingming Cao617ba132006-10-11 01:20:53 -0700540 BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
541 err = ext4_journal_dirty_metadata(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700542 if (err) goto fail;
543
544 percpu_counter_dec(&sbi->s_freeinodes_counter);
545 if (S_ISDIR(mode))
546 percpu_counter_inc(&sbi->s_dirs_counter);
547 sb->s_dirt = 1;
548
549 inode->i_uid = current->fsuid;
550 if (test_opt (sb, GRPID))
551 inode->i_gid = dir->i_gid;
552 else if (dir->i_mode & S_ISGID) {
553 inode->i_gid = dir->i_gid;
554 if (S_ISDIR(mode))
555 mode |= S_ISGID;
556 } else
557 inode->i_gid = current->fsgid;
558 inode->i_mode = mode;
559
560 inode->i_ino = ino;
561 /* This is the optimal IO size (for stat), not the fs block size */
562 inode->i_blocks = 0;
Kalpak Shahef7f3832007-07-18 09:15:20 -0400563 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
564 ext4_current_time(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700565
566 memset(ei->i_data, 0, sizeof(ei->i_data));
567 ei->i_dir_start_lookup = 0;
568 ei->i_disksize = 0;
569
Mingming Cao617ba132006-10-11 01:20:53 -0700570 ei->i_flags = EXT4_I(dir)->i_flags & ~EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700571 if (S_ISLNK(mode))
Mingming Cao617ba132006-10-11 01:20:53 -0700572 ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700573 /* dirsync only applies to directories */
574 if (!S_ISDIR(mode))
Mingming Cao617ba132006-10-11 01:20:53 -0700575 ei->i_flags &= ~EXT4_DIRSYNC_FL;
576#ifdef EXT4_FRAGMENTS
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 ei->i_faddr = 0;
578 ei->i_frag_no = 0;
579 ei->i_frag_size = 0;
580#endif
581 ei->i_file_acl = 0;
582 ei->i_dir_acl = 0;
583 ei->i_dtime = 0;
584 ei->i_block_alloc_info = NULL;
585 ei->i_block_group = group;
586
Mingming Cao617ba132006-10-11 01:20:53 -0700587 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700588 if (IS_DIRSYNC(inode))
589 handle->h_sync = 1;
590 insert_inode_hash(inode);
591 spin_lock(&sbi->s_next_gen_lock);
592 inode->i_generation = sbi->s_next_generation++;
593 spin_unlock(&sbi->s_next_gen_lock);
594
Mingming Cao617ba132006-10-11 01:20:53 -0700595 ei->i_state = EXT4_STATE_NEW;
Kalpak Shahef7f3832007-07-18 09:15:20 -0400596
597 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700598
599 ret = inode;
600 if(DQUOT_ALLOC_INODE(inode)) {
601 err = -EDQUOT;
602 goto fail_drop;
603 }
604
Mingming Cao617ba132006-10-11 01:20:53 -0700605 err = ext4_init_acl(handle, inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700606 if (err)
607 goto fail_free_drop;
608
Mingming Cao617ba132006-10-11 01:20:53 -0700609 err = ext4_init_security(handle,inode, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700610 if (err)
611 goto fail_free_drop;
612
Mingming Cao617ba132006-10-11 01:20:53 -0700613 err = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700614 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -0700615 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700616 goto fail_free_drop;
617 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700618 if (test_opt(sb, EXTENTS)) {
619 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
620 ext4_ext_tree_init(handle, inode);
621 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
622 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
623 if (err) goto fail;
624 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS);
625 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "call ext4_journal_dirty_metadata");
626 err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
627 }
628 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700629
Mingming Cao617ba132006-10-11 01:20:53 -0700630 ext4_debug("allocating inode %lu\n", inode->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700631 goto really_out;
632fail:
Mingming Cao617ba132006-10-11 01:20:53 -0700633 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700634out:
635 iput(inode);
636 ret = ERR_PTR(err);
637really_out:
638 brelse(bitmap_bh);
639 return ret;
640
641fail_free_drop:
642 DQUOT_FREE_INODE(inode);
643
644fail_drop:
645 DQUOT_DROP(inode);
646 inode->i_flags |= S_NOQUOTA;
647 inode->i_nlink = 0;
648 iput(inode);
649 brelse(bitmap_bh);
650 return ERR_PTR(err);
651}
652
653/* Verify that we are loading a valid orphan from disk */
Mingming Cao617ba132006-10-11 01:20:53 -0700654struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700655{
Mingming Cao617ba132006-10-11 01:20:53 -0700656 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700657 unsigned long block_group;
658 int bit;
659 struct buffer_head *bitmap_bh = NULL;
660 struct inode *inode = NULL;
661
662 /* Error cases - e2fsck has already cleaned up for us */
663 if (ino > max_ino) {
Mingming Cao617ba132006-10-11 01:20:53 -0700664 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700665 "bad orphan ino %lu! e2fsck was run?", ino);
666 goto out;
667 }
668
Mingming Cao617ba132006-10-11 01:20:53 -0700669 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
670 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700671 bitmap_bh = read_inode_bitmap(sb, block_group);
672 if (!bitmap_bh) {
Mingming Cao617ba132006-10-11 01:20:53 -0700673 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700674 "inode bitmap error for orphan %lu", ino);
675 goto out;
676 }
677
678 /* Having the inode bit set should be a 100% indicator that this
679 * is a valid orphan (no e2fsck run on fs). Orphans also include
680 * inodes that were being truncated, so we can't check i_nlink==0.
681 */
Mingming Cao617ba132006-10-11 01:20:53 -0700682 if (!ext4_test_bit(bit, bitmap_bh->b_data) ||
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700683 !(inode = iget(sb, ino)) || is_bad_inode(inode) ||
684 NEXT_ORPHAN(inode) > max_ino) {
Mingming Cao617ba132006-10-11 01:20:53 -0700685 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700686 "bad orphan inode %lu! e2fsck was run?", ino);
Mingming Cao617ba132006-10-11 01:20:53 -0700687 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700688 bit, (unsigned long long)bitmap_bh->b_blocknr,
Mingming Cao617ba132006-10-11 01:20:53 -0700689 ext4_test_bit(bit, bitmap_bh->b_data));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700690 printk(KERN_NOTICE "inode=%p\n", inode);
691 if (inode) {
692 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
693 is_bad_inode(inode));
694 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
695 NEXT_ORPHAN(inode));
696 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
697 }
698 /* Avoid freeing blocks if we got a bad deleted inode */
699 if (inode && inode->i_nlink == 0)
700 inode->i_blocks = 0;
701 iput(inode);
702 inode = NULL;
703 }
704out:
705 brelse(bitmap_bh);
706 return inode;
707}
708
Mingming Cao617ba132006-10-11 01:20:53 -0700709unsigned long ext4_count_free_inodes (struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700710{
711 unsigned long desc_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700712 struct ext4_group_desc *gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700713 int i;
Mingming Cao617ba132006-10-11 01:20:53 -0700714#ifdef EXT4FS_DEBUG
715 struct ext4_super_block *es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700716 unsigned long bitmap_count, x;
717 struct buffer_head *bitmap_bh = NULL;
718
Mingming Cao617ba132006-10-11 01:20:53 -0700719 es = EXT4_SB(sb)->s_es;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700720 desc_count = 0;
721 bitmap_count = 0;
722 gdp = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700723 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
724 gdp = ext4_get_group_desc (sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700725 if (!gdp)
726 continue;
727 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
728 brelse(bitmap_bh);
729 bitmap_bh = read_inode_bitmap(sb, i);
730 if (!bitmap_bh)
731 continue;
732
Mingming Cao617ba132006-10-11 01:20:53 -0700733 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700734 printk("group %d: stored = %d, counted = %lu\n",
735 i, le16_to_cpu(gdp->bg_free_inodes_count), x);
736 bitmap_count += x;
737 }
738 brelse(bitmap_bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700739 printk("ext4_count_free_inodes: stored = %u, computed = %lu, %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700740 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
741 return desc_count;
742#else
743 desc_count = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700744 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
745 gdp = ext4_get_group_desc (sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700746 if (!gdp)
747 continue;
748 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
749 cond_resched();
750 }
751 return desc_count;
752#endif
753}
754
755/* Called at mount-time, super-block is locked */
Mingming Cao617ba132006-10-11 01:20:53 -0700756unsigned long ext4_count_dirs (struct super_block * sb)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700757{
758 unsigned long count = 0;
759 int i;
760
Mingming Cao617ba132006-10-11 01:20:53 -0700761 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
762 struct ext4_group_desc *gdp = ext4_get_group_desc (sb, i, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700763 if (!gdp)
764 continue;
765 count += le16_to_cpu(gdp->bg_used_dirs_count);
766 }
767 return count;
768}
769