xref: /linux/fs/ext4/ialloc.c (revision 3a5b2ecdd1fa63a8f25bd769223bc1c2564ce45d)
1ac27a0ecSDave Kleikamp /*
2617ba13bSMingming Cao  *  linux/fs/ext4/ialloc.c
3ac27a0ecSDave Kleikamp  *
4ac27a0ecSDave Kleikamp  * Copyright (C) 1992, 1993, 1994, 1995
5ac27a0ecSDave Kleikamp  * Remy Card (card@masi.ibp.fr)
6ac27a0ecSDave Kleikamp  * Laboratoire MASI - Institut Blaise Pascal
7ac27a0ecSDave Kleikamp  * Universite Pierre et Marie Curie (Paris VI)
8ac27a0ecSDave Kleikamp  *
9ac27a0ecSDave Kleikamp  *  BSD ufs-inspired inode and directory allocation by
10ac27a0ecSDave Kleikamp  *  Stephen Tweedie (sct@redhat.com), 1993
11ac27a0ecSDave Kleikamp  *  Big-endian to little-endian byte-swapping/bitmaps by
12ac27a0ecSDave Kleikamp  *        David S. Miller (davem@caip.rutgers.edu), 1995
13ac27a0ecSDave Kleikamp  */
14ac27a0ecSDave Kleikamp 
15ac27a0ecSDave Kleikamp #include <linux/time.h>
16ac27a0ecSDave Kleikamp #include <linux/fs.h>
17dab291afSMingming Cao #include <linux/jbd2.h>
18617ba13bSMingming Cao #include <linux/ext4_fs.h>
19dab291afSMingming Cao #include <linux/ext4_jbd2.h>
20ac27a0ecSDave Kleikamp #include <linux/stat.h>
21ac27a0ecSDave Kleikamp #include <linux/string.h>
22ac27a0ecSDave Kleikamp #include <linux/quotaops.h>
23ac27a0ecSDave Kleikamp #include <linux/buffer_head.h>
24ac27a0ecSDave Kleikamp #include <linux/random.h>
25ac27a0ecSDave Kleikamp #include <linux/bitops.h>
26*3a5b2ecdSMingming Cao #include <linux/blkdev.h>
27ac27a0ecSDave Kleikamp #include <asm/byteorder.h>
28ac27a0ecSDave Kleikamp 
29ac27a0ecSDave Kleikamp #include "xattr.h"
30ac27a0ecSDave Kleikamp #include "acl.h"
31ac27a0ecSDave Kleikamp 
32ac27a0ecSDave Kleikamp /*
33ac27a0ecSDave Kleikamp  * ialloc.c contains the inodes allocation and deallocation routines
34ac27a0ecSDave Kleikamp  */
35ac27a0ecSDave Kleikamp 
36ac27a0ecSDave Kleikamp /*
37ac27a0ecSDave Kleikamp  * The free inodes are managed by bitmaps.  A file system contains several
38ac27a0ecSDave Kleikamp  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
39ac27a0ecSDave Kleikamp  * block for inodes, N blocks for the inode table and data blocks.
40ac27a0ecSDave Kleikamp  *
41ac27a0ecSDave Kleikamp  * The file system contains group descriptors which are located after the
42ac27a0ecSDave Kleikamp  * super block.  Each descriptor contains the number of the bitmap block and
43ac27a0ecSDave Kleikamp  * the free blocks count in the block.
44ac27a0ecSDave Kleikamp  */
45ac27a0ecSDave Kleikamp 
46ac27a0ecSDave Kleikamp 
47ac27a0ecSDave Kleikamp /*
48ac27a0ecSDave Kleikamp  * Read the inode allocation bitmap for a given block_group, reading
49ac27a0ecSDave Kleikamp  * into the specified slot in the superblock's bitmap cache.
50ac27a0ecSDave Kleikamp  *
51ac27a0ecSDave Kleikamp  * Return buffer_head of bitmap on success or NULL.
52ac27a0ecSDave Kleikamp  */
53ac27a0ecSDave Kleikamp static struct buffer_head *
54ac27a0ecSDave Kleikamp read_inode_bitmap(struct super_block * sb, unsigned long block_group)
55ac27a0ecSDave Kleikamp {
56617ba13bSMingming Cao 	struct ext4_group_desc *desc;
57ac27a0ecSDave Kleikamp 	struct buffer_head *bh = NULL;
58ac27a0ecSDave Kleikamp 
59617ba13bSMingming Cao 	desc = ext4_get_group_desc(sb, block_group, NULL);
60ac27a0ecSDave Kleikamp 	if (!desc)
61ac27a0ecSDave Kleikamp 		goto error_out;
62ac27a0ecSDave Kleikamp 
63ac27a0ecSDave Kleikamp 	bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
64ac27a0ecSDave Kleikamp 	if (!bh)
65617ba13bSMingming Cao 		ext4_error(sb, "read_inode_bitmap",
66ac27a0ecSDave Kleikamp 			    "Cannot read inode bitmap - "
67ac27a0ecSDave Kleikamp 			    "block_group = %lu, inode_bitmap = %u",
68ac27a0ecSDave Kleikamp 			    block_group, le32_to_cpu(desc->bg_inode_bitmap));
69ac27a0ecSDave Kleikamp error_out:
70ac27a0ecSDave Kleikamp 	return bh;
71ac27a0ecSDave Kleikamp }
72ac27a0ecSDave Kleikamp 
73ac27a0ecSDave Kleikamp /*
74ac27a0ecSDave Kleikamp  * NOTE! When we get the inode, we're the only people
75ac27a0ecSDave Kleikamp  * that have access to it, and as such there are no
76ac27a0ecSDave Kleikamp  * race conditions we have to worry about. The inode
77ac27a0ecSDave Kleikamp  * is not on the hash-lists, and it cannot be reached
78ac27a0ecSDave Kleikamp  * through the filesystem because the directory entry
79ac27a0ecSDave Kleikamp  * has been deleted earlier.
80ac27a0ecSDave Kleikamp  *
81ac27a0ecSDave Kleikamp  * HOWEVER: we must make sure that we get no aliases,
82ac27a0ecSDave Kleikamp  * which means that we have to call "clear_inode()"
83ac27a0ecSDave Kleikamp  * _before_ we mark the inode not in use in the inode
84ac27a0ecSDave Kleikamp  * bitmaps. Otherwise a newly created file might use
85ac27a0ecSDave Kleikamp  * the same inode number (not actually the same pointer
86ac27a0ecSDave Kleikamp  * though), and then we'd have two inodes sharing the
87ac27a0ecSDave Kleikamp  * same inode number and space on the harddisk.
88ac27a0ecSDave Kleikamp  */
89617ba13bSMingming Cao void ext4_free_inode (handle_t *handle, struct inode * inode)
90ac27a0ecSDave Kleikamp {
91ac27a0ecSDave Kleikamp 	struct super_block * sb = inode->i_sb;
92ac27a0ecSDave Kleikamp 	int is_directory;
93ac27a0ecSDave Kleikamp 	unsigned long ino;
94ac27a0ecSDave Kleikamp 	struct buffer_head *bitmap_bh = NULL;
95ac27a0ecSDave Kleikamp 	struct buffer_head *bh2;
96ac27a0ecSDave Kleikamp 	unsigned long block_group;
97ac27a0ecSDave Kleikamp 	unsigned long bit;
98617ba13bSMingming Cao 	struct ext4_group_desc * gdp;
99617ba13bSMingming Cao 	struct ext4_super_block * es;
100617ba13bSMingming Cao 	struct ext4_sb_info *sbi;
101ac27a0ecSDave Kleikamp 	int fatal = 0, err;
102ac27a0ecSDave Kleikamp 
103ac27a0ecSDave Kleikamp 	if (atomic_read(&inode->i_count) > 1) {
104617ba13bSMingming Cao 		printk ("ext4_free_inode: inode has count=%d\n",
105ac27a0ecSDave Kleikamp 					atomic_read(&inode->i_count));
106ac27a0ecSDave Kleikamp 		return;
107ac27a0ecSDave Kleikamp 	}
108ac27a0ecSDave Kleikamp 	if (inode->i_nlink) {
109617ba13bSMingming Cao 		printk ("ext4_free_inode: inode has nlink=%d\n",
110ac27a0ecSDave Kleikamp 			inode->i_nlink);
111ac27a0ecSDave Kleikamp 		return;
112ac27a0ecSDave Kleikamp 	}
113ac27a0ecSDave Kleikamp 	if (!sb) {
114617ba13bSMingming Cao 		printk("ext4_free_inode: inode on nonexistent device\n");
115ac27a0ecSDave Kleikamp 		return;
116ac27a0ecSDave Kleikamp 	}
117617ba13bSMingming Cao 	sbi = EXT4_SB(sb);
118ac27a0ecSDave Kleikamp 
119ac27a0ecSDave Kleikamp 	ino = inode->i_ino;
120617ba13bSMingming Cao 	ext4_debug ("freeing inode %lu\n", ino);
121ac27a0ecSDave Kleikamp 
122ac27a0ecSDave Kleikamp 	/*
123ac27a0ecSDave Kleikamp 	 * Note: we must free any quota before locking the superblock,
124ac27a0ecSDave Kleikamp 	 * as writing the quota to disk may need the lock as well.
125ac27a0ecSDave Kleikamp 	 */
126ac27a0ecSDave Kleikamp 	DQUOT_INIT(inode);
127617ba13bSMingming Cao 	ext4_xattr_delete_inode(handle, inode);
128ac27a0ecSDave Kleikamp 	DQUOT_FREE_INODE(inode);
129ac27a0ecSDave Kleikamp 	DQUOT_DROP(inode);
130ac27a0ecSDave Kleikamp 
131ac27a0ecSDave Kleikamp 	is_directory = S_ISDIR(inode->i_mode);
132ac27a0ecSDave Kleikamp 
133ac27a0ecSDave Kleikamp 	/* Do this BEFORE marking the inode not in use or returning an error */
134ac27a0ecSDave Kleikamp 	clear_inode (inode);
135ac27a0ecSDave Kleikamp 
136617ba13bSMingming Cao 	es = EXT4_SB(sb)->s_es;
137617ba13bSMingming Cao 	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
138617ba13bSMingming Cao 		ext4_error (sb, "ext4_free_inode",
139ac27a0ecSDave Kleikamp 			    "reserved or nonexistent inode %lu", ino);
140ac27a0ecSDave Kleikamp 		goto error_return;
141ac27a0ecSDave Kleikamp 	}
142617ba13bSMingming Cao 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
143617ba13bSMingming Cao 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
144ac27a0ecSDave Kleikamp 	bitmap_bh = read_inode_bitmap(sb, block_group);
145ac27a0ecSDave Kleikamp 	if (!bitmap_bh)
146ac27a0ecSDave Kleikamp 		goto error_return;
147ac27a0ecSDave Kleikamp 
148ac27a0ecSDave Kleikamp 	BUFFER_TRACE(bitmap_bh, "get_write_access");
149617ba13bSMingming Cao 	fatal = ext4_journal_get_write_access(handle, bitmap_bh);
150ac27a0ecSDave Kleikamp 	if (fatal)
151ac27a0ecSDave Kleikamp 		goto error_return;
152ac27a0ecSDave Kleikamp 
153ac27a0ecSDave Kleikamp 	/* Ok, now we can actually update the inode bitmaps.. */
154617ba13bSMingming Cao 	if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
155ac27a0ecSDave Kleikamp 					bit, bitmap_bh->b_data))
156617ba13bSMingming Cao 		ext4_error (sb, "ext4_free_inode",
157ac27a0ecSDave Kleikamp 			      "bit already cleared for inode %lu", ino);
158ac27a0ecSDave Kleikamp 	else {
159617ba13bSMingming Cao 		gdp = ext4_get_group_desc (sb, block_group, &bh2);
160ac27a0ecSDave Kleikamp 
161ac27a0ecSDave Kleikamp 		BUFFER_TRACE(bh2, "get_write_access");
162617ba13bSMingming Cao 		fatal = ext4_journal_get_write_access(handle, bh2);
163ac27a0ecSDave Kleikamp 		if (fatal) goto error_return;
164ac27a0ecSDave Kleikamp 
165ac27a0ecSDave Kleikamp 		if (gdp) {
166ac27a0ecSDave Kleikamp 			spin_lock(sb_bgl_lock(sbi, block_group));
167ac27a0ecSDave Kleikamp 			gdp->bg_free_inodes_count = cpu_to_le16(
168ac27a0ecSDave Kleikamp 				le16_to_cpu(gdp->bg_free_inodes_count) + 1);
169ac27a0ecSDave Kleikamp 			if (is_directory)
170ac27a0ecSDave Kleikamp 				gdp->bg_used_dirs_count = cpu_to_le16(
171ac27a0ecSDave Kleikamp 				  le16_to_cpu(gdp->bg_used_dirs_count) - 1);
172ac27a0ecSDave Kleikamp 			spin_unlock(sb_bgl_lock(sbi, block_group));
173ac27a0ecSDave Kleikamp 			percpu_counter_inc(&sbi->s_freeinodes_counter);
174ac27a0ecSDave Kleikamp 			if (is_directory)
175ac27a0ecSDave Kleikamp 				percpu_counter_dec(&sbi->s_dirs_counter);
176ac27a0ecSDave Kleikamp 
177ac27a0ecSDave Kleikamp 		}
178617ba13bSMingming Cao 		BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
179617ba13bSMingming Cao 		err = ext4_journal_dirty_metadata(handle, bh2);
180ac27a0ecSDave Kleikamp 		if (!fatal) fatal = err;
181ac27a0ecSDave Kleikamp 	}
182617ba13bSMingming Cao 	BUFFER_TRACE(bitmap_bh, "call ext4_journal_dirty_metadata");
183617ba13bSMingming Cao 	err = ext4_journal_dirty_metadata(handle, bitmap_bh);
184ac27a0ecSDave Kleikamp 	if (!fatal)
185ac27a0ecSDave Kleikamp 		fatal = err;
186ac27a0ecSDave Kleikamp 	sb->s_dirt = 1;
187ac27a0ecSDave Kleikamp error_return:
188ac27a0ecSDave Kleikamp 	brelse(bitmap_bh);
189617ba13bSMingming Cao 	ext4_std_error(sb, fatal);
190ac27a0ecSDave Kleikamp }
191ac27a0ecSDave Kleikamp 
192ac27a0ecSDave Kleikamp /*
193ac27a0ecSDave Kleikamp  * There are two policies for allocating an inode.  If the new inode is
194ac27a0ecSDave Kleikamp  * a directory, then a forward search is made for a block group with both
195ac27a0ecSDave Kleikamp  * free space and a low directory-to-inode ratio; if that fails, then of
196ac27a0ecSDave Kleikamp  * the groups with above-average free space, that group with the fewest
197ac27a0ecSDave Kleikamp  * directories already is chosen.
198ac27a0ecSDave Kleikamp  *
199ac27a0ecSDave Kleikamp  * For other inodes, search forward from the parent directory\'s block
200ac27a0ecSDave Kleikamp  * group to find a free inode.
201ac27a0ecSDave Kleikamp  */
202ac27a0ecSDave Kleikamp static int find_group_dir(struct super_block *sb, struct inode *parent)
203ac27a0ecSDave Kleikamp {
204617ba13bSMingming Cao 	int ngroups = EXT4_SB(sb)->s_groups_count;
205ac27a0ecSDave Kleikamp 	unsigned int freei, avefreei;
206617ba13bSMingming Cao 	struct ext4_group_desc *desc, *best_desc = NULL;
207ac27a0ecSDave Kleikamp 	struct buffer_head *bh;
208ac27a0ecSDave Kleikamp 	int group, best_group = -1;
209ac27a0ecSDave Kleikamp 
210617ba13bSMingming Cao 	freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
211ac27a0ecSDave Kleikamp 	avefreei = freei / ngroups;
212ac27a0ecSDave Kleikamp 
213ac27a0ecSDave Kleikamp 	for (group = 0; group < ngroups; group++) {
214617ba13bSMingming Cao 		desc = ext4_get_group_desc (sb, group, &bh);
215ac27a0ecSDave Kleikamp 		if (!desc || !desc->bg_free_inodes_count)
216ac27a0ecSDave Kleikamp 			continue;
217ac27a0ecSDave Kleikamp 		if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
218ac27a0ecSDave Kleikamp 			continue;
219ac27a0ecSDave Kleikamp 		if (!best_desc ||
220ac27a0ecSDave Kleikamp 		    (le16_to_cpu(desc->bg_free_blocks_count) >
221ac27a0ecSDave Kleikamp 		     le16_to_cpu(best_desc->bg_free_blocks_count))) {
222ac27a0ecSDave Kleikamp 			best_group = group;
223ac27a0ecSDave Kleikamp 			best_desc = desc;
224ac27a0ecSDave Kleikamp 		}
225ac27a0ecSDave Kleikamp 	}
226ac27a0ecSDave Kleikamp 	return best_group;
227ac27a0ecSDave Kleikamp }
228ac27a0ecSDave Kleikamp 
229ac27a0ecSDave Kleikamp /*
230ac27a0ecSDave Kleikamp  * Orlov's allocator for directories.
231ac27a0ecSDave Kleikamp  *
232ac27a0ecSDave Kleikamp  * We always try to spread first-level directories.
233ac27a0ecSDave Kleikamp  *
234ac27a0ecSDave Kleikamp  * If there are blockgroups with both free inodes and free blocks counts
235ac27a0ecSDave Kleikamp  * not worse than average we return one with smallest directory count.
236ac27a0ecSDave Kleikamp  * Otherwise we simply return a random group.
237ac27a0ecSDave Kleikamp  *
238ac27a0ecSDave Kleikamp  * For the rest rules look so:
239ac27a0ecSDave Kleikamp  *
240ac27a0ecSDave Kleikamp  * It's OK to put directory into a group unless
241ac27a0ecSDave Kleikamp  * it has too many directories already (max_dirs) or
242ac27a0ecSDave Kleikamp  * it has too few free inodes left (min_inodes) or
243ac27a0ecSDave Kleikamp  * it has too few free blocks left (min_blocks) or
244ac27a0ecSDave Kleikamp  * it's already running too large debt (max_debt).
245ac27a0ecSDave Kleikamp  * Parent's group is prefered, if it doesn't satisfy these
246ac27a0ecSDave Kleikamp  * conditions we search cyclically through the rest. If none
247ac27a0ecSDave Kleikamp  * of the groups look good we just look for a group with more
248ac27a0ecSDave Kleikamp  * free inodes than average (starting at parent's group).
249ac27a0ecSDave Kleikamp  *
250ac27a0ecSDave Kleikamp  * Debt is incremented each time we allocate a directory and decremented
251ac27a0ecSDave Kleikamp  * when we allocate an inode, within 0--255.
252ac27a0ecSDave Kleikamp  */
253ac27a0ecSDave Kleikamp 
254ac27a0ecSDave Kleikamp #define INODE_COST 64
255ac27a0ecSDave Kleikamp #define BLOCK_COST 256
256ac27a0ecSDave Kleikamp 
257ac27a0ecSDave Kleikamp static int find_group_orlov(struct super_block *sb, struct inode *parent)
258ac27a0ecSDave Kleikamp {
259617ba13bSMingming Cao 	int parent_group = EXT4_I(parent)->i_block_group;
260617ba13bSMingming Cao 	struct ext4_sb_info *sbi = EXT4_SB(sb);
261617ba13bSMingming Cao 	struct ext4_super_block *es = sbi->s_es;
262ac27a0ecSDave Kleikamp 	int ngroups = sbi->s_groups_count;
263617ba13bSMingming Cao 	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
264ac27a0ecSDave Kleikamp 	unsigned int freei, avefreei;
265617ba13bSMingming Cao 	ext4_fsblk_t freeb, avefreeb;
266617ba13bSMingming Cao 	ext4_fsblk_t blocks_per_dir;
267ac27a0ecSDave Kleikamp 	unsigned int ndirs;
268ac27a0ecSDave Kleikamp 	int max_debt, max_dirs, min_inodes;
269617ba13bSMingming Cao 	ext4_grpblk_t min_blocks;
270ac27a0ecSDave Kleikamp 	int group = -1, i;
271617ba13bSMingming Cao 	struct ext4_group_desc *desc;
272ac27a0ecSDave Kleikamp 	struct buffer_head *bh;
273ac27a0ecSDave Kleikamp 
274ac27a0ecSDave Kleikamp 	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
275ac27a0ecSDave Kleikamp 	avefreei = freei / ngroups;
276ac27a0ecSDave Kleikamp 	freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
277*3a5b2ecdSMingming Cao 	avefreeb = freeb;
278*3a5b2ecdSMingming Cao 	sector_div(avefreeb, ngroups);
279ac27a0ecSDave Kleikamp 	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
280ac27a0ecSDave Kleikamp 
281ac27a0ecSDave Kleikamp 	if ((parent == sb->s_root->d_inode) ||
282617ba13bSMingming Cao 	    (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) {
283ac27a0ecSDave Kleikamp 		int best_ndir = inodes_per_group;
284ac27a0ecSDave Kleikamp 		int best_group = -1;
285ac27a0ecSDave Kleikamp 
286ac27a0ecSDave Kleikamp 		get_random_bytes(&group, sizeof(group));
287ac27a0ecSDave Kleikamp 		parent_group = (unsigned)group % ngroups;
288ac27a0ecSDave Kleikamp 		for (i = 0; i < ngroups; i++) {
289ac27a0ecSDave Kleikamp 			group = (parent_group + i) % ngroups;
290617ba13bSMingming Cao 			desc = ext4_get_group_desc (sb, group, &bh);
291ac27a0ecSDave Kleikamp 			if (!desc || !desc->bg_free_inodes_count)
292ac27a0ecSDave Kleikamp 				continue;
293ac27a0ecSDave Kleikamp 			if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
294ac27a0ecSDave Kleikamp 				continue;
295ac27a0ecSDave Kleikamp 			if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
296ac27a0ecSDave Kleikamp 				continue;
297ac27a0ecSDave Kleikamp 			if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
298ac27a0ecSDave Kleikamp 				continue;
299ac27a0ecSDave Kleikamp 			best_group = group;
300ac27a0ecSDave Kleikamp 			best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
301ac27a0ecSDave Kleikamp 		}
302ac27a0ecSDave Kleikamp 		if (best_group >= 0)
303ac27a0ecSDave Kleikamp 			return best_group;
304ac27a0ecSDave Kleikamp 		goto fallback;
305ac27a0ecSDave Kleikamp 	}
306ac27a0ecSDave Kleikamp 
307*3a5b2ecdSMingming Cao 	blocks_per_dir = le32_to_cpu(es->s_blocks_count) - freeb;
308*3a5b2ecdSMingming Cao 	sector_div(blocks_per_dir, ndirs);
309ac27a0ecSDave Kleikamp 
310ac27a0ecSDave Kleikamp 	max_dirs = ndirs / ngroups + inodes_per_group / 16;
311ac27a0ecSDave Kleikamp 	min_inodes = avefreei - inodes_per_group / 4;
312617ba13bSMingming Cao 	min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4;
313ac27a0ecSDave Kleikamp 
314*3a5b2ecdSMingming Cao 	max_debt = EXT4_BLOCKS_PER_GROUP(sb);
315*3a5b2ecdSMingming Cao 	sector_div(max_debt, max(blocks_per_dir, (ext4_fsblk_t)BLOCK_COST));
316ac27a0ecSDave Kleikamp 	if (max_debt * INODE_COST > inodes_per_group)
317ac27a0ecSDave Kleikamp 		max_debt = inodes_per_group / INODE_COST;
318ac27a0ecSDave Kleikamp 	if (max_debt > 255)
319ac27a0ecSDave Kleikamp 		max_debt = 255;
320ac27a0ecSDave Kleikamp 	if (max_debt == 0)
321ac27a0ecSDave Kleikamp 		max_debt = 1;
322ac27a0ecSDave Kleikamp 
323ac27a0ecSDave Kleikamp 	for (i = 0; i < ngroups; i++) {
324ac27a0ecSDave Kleikamp 		group = (parent_group + i) % ngroups;
325617ba13bSMingming Cao 		desc = ext4_get_group_desc (sb, group, &bh);
326ac27a0ecSDave Kleikamp 		if (!desc || !desc->bg_free_inodes_count)
327ac27a0ecSDave Kleikamp 			continue;
328ac27a0ecSDave Kleikamp 		if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
329ac27a0ecSDave Kleikamp 			continue;
330ac27a0ecSDave Kleikamp 		if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
331ac27a0ecSDave Kleikamp 			continue;
332ac27a0ecSDave Kleikamp 		if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
333ac27a0ecSDave Kleikamp 			continue;
334ac27a0ecSDave Kleikamp 		return group;
335ac27a0ecSDave Kleikamp 	}
336ac27a0ecSDave Kleikamp 
337ac27a0ecSDave Kleikamp fallback:
338ac27a0ecSDave Kleikamp 	for (i = 0; i < ngroups; i++) {
339ac27a0ecSDave Kleikamp 		group = (parent_group + i) % ngroups;
340617ba13bSMingming Cao 		desc = ext4_get_group_desc (sb, group, &bh);
341ac27a0ecSDave Kleikamp 		if (!desc || !desc->bg_free_inodes_count)
342ac27a0ecSDave Kleikamp 			continue;
343ac27a0ecSDave Kleikamp 		if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
344ac27a0ecSDave Kleikamp 			return group;
345ac27a0ecSDave Kleikamp 	}
346ac27a0ecSDave Kleikamp 
347ac27a0ecSDave Kleikamp 	if (avefreei) {
348ac27a0ecSDave Kleikamp 		/*
349ac27a0ecSDave Kleikamp 		 * The free-inodes counter is approximate, and for really small
350ac27a0ecSDave Kleikamp 		 * filesystems the above test can fail to find any blockgroups
351ac27a0ecSDave Kleikamp 		 */
352ac27a0ecSDave Kleikamp 		avefreei = 0;
353ac27a0ecSDave Kleikamp 		goto fallback;
354ac27a0ecSDave Kleikamp 	}
355ac27a0ecSDave Kleikamp 
356ac27a0ecSDave Kleikamp 	return -1;
357ac27a0ecSDave Kleikamp }
358ac27a0ecSDave Kleikamp 
359ac27a0ecSDave Kleikamp static int find_group_other(struct super_block *sb, struct inode *parent)
360ac27a0ecSDave Kleikamp {
361617ba13bSMingming Cao 	int parent_group = EXT4_I(parent)->i_block_group;
362617ba13bSMingming Cao 	int ngroups = EXT4_SB(sb)->s_groups_count;
363617ba13bSMingming Cao 	struct ext4_group_desc *desc;
364ac27a0ecSDave Kleikamp 	struct buffer_head *bh;
365ac27a0ecSDave Kleikamp 	int group, i;
366ac27a0ecSDave Kleikamp 
367ac27a0ecSDave Kleikamp 	/*
368ac27a0ecSDave Kleikamp 	 * Try to place the inode in its parent directory
369ac27a0ecSDave Kleikamp 	 */
370ac27a0ecSDave Kleikamp 	group = parent_group;
371617ba13bSMingming Cao 	desc = ext4_get_group_desc (sb, group, &bh);
372ac27a0ecSDave Kleikamp 	if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
373ac27a0ecSDave Kleikamp 			le16_to_cpu(desc->bg_free_blocks_count))
374ac27a0ecSDave Kleikamp 		return group;
375ac27a0ecSDave Kleikamp 
376ac27a0ecSDave Kleikamp 	/*
377ac27a0ecSDave Kleikamp 	 * We're going to place this inode in a different blockgroup from its
378ac27a0ecSDave Kleikamp 	 * parent.  We want to cause files in a common directory to all land in
379ac27a0ecSDave Kleikamp 	 * the same blockgroup.  But we want files which are in a different
380ac27a0ecSDave Kleikamp 	 * directory which shares a blockgroup with our parent to land in a
381ac27a0ecSDave Kleikamp 	 * different blockgroup.
382ac27a0ecSDave Kleikamp 	 *
383ac27a0ecSDave Kleikamp 	 * So add our directory's i_ino into the starting point for the hash.
384ac27a0ecSDave Kleikamp 	 */
385ac27a0ecSDave Kleikamp 	group = (group + parent->i_ino) % ngroups;
386ac27a0ecSDave Kleikamp 
387ac27a0ecSDave Kleikamp 	/*
388ac27a0ecSDave Kleikamp 	 * Use a quadratic hash to find a group with a free inode and some free
389ac27a0ecSDave Kleikamp 	 * blocks.
390ac27a0ecSDave Kleikamp 	 */
391ac27a0ecSDave Kleikamp 	for (i = 1; i < ngroups; i <<= 1) {
392ac27a0ecSDave Kleikamp 		group += i;
393ac27a0ecSDave Kleikamp 		if (group >= ngroups)
394ac27a0ecSDave Kleikamp 			group -= ngroups;
395617ba13bSMingming Cao 		desc = ext4_get_group_desc (sb, group, &bh);
396ac27a0ecSDave Kleikamp 		if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
397ac27a0ecSDave Kleikamp 				le16_to_cpu(desc->bg_free_blocks_count))
398ac27a0ecSDave Kleikamp 			return group;
399ac27a0ecSDave Kleikamp 	}
400ac27a0ecSDave Kleikamp 
401ac27a0ecSDave Kleikamp 	/*
402ac27a0ecSDave Kleikamp 	 * That failed: try linear search for a free inode, even if that group
403ac27a0ecSDave Kleikamp 	 * has no free blocks.
404ac27a0ecSDave Kleikamp 	 */
405ac27a0ecSDave Kleikamp 	group = parent_group;
406ac27a0ecSDave Kleikamp 	for (i = 0; i < ngroups; i++) {
407ac27a0ecSDave Kleikamp 		if (++group >= ngroups)
408ac27a0ecSDave Kleikamp 			group = 0;
409617ba13bSMingming Cao 		desc = ext4_get_group_desc (sb, group, &bh);
410ac27a0ecSDave Kleikamp 		if (desc && le16_to_cpu(desc->bg_free_inodes_count))
411ac27a0ecSDave Kleikamp 			return group;
412ac27a0ecSDave Kleikamp 	}
413ac27a0ecSDave Kleikamp 
414ac27a0ecSDave Kleikamp 	return -1;
415ac27a0ecSDave Kleikamp }
416ac27a0ecSDave Kleikamp 
417ac27a0ecSDave Kleikamp /*
418ac27a0ecSDave Kleikamp  * There are two policies for allocating an inode.  If the new inode is
419ac27a0ecSDave Kleikamp  * a directory, then a forward search is made for a block group with both
420ac27a0ecSDave Kleikamp  * free space and a low directory-to-inode ratio; if that fails, then of
421ac27a0ecSDave Kleikamp  * the groups with above-average free space, that group with the fewest
422ac27a0ecSDave Kleikamp  * directories already is chosen.
423ac27a0ecSDave Kleikamp  *
424ac27a0ecSDave Kleikamp  * For other inodes, search forward from the parent directory's block
425ac27a0ecSDave Kleikamp  * group to find a free inode.
426ac27a0ecSDave Kleikamp  */
427617ba13bSMingming Cao struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
428ac27a0ecSDave Kleikamp {
429ac27a0ecSDave Kleikamp 	struct super_block *sb;
430ac27a0ecSDave Kleikamp 	struct buffer_head *bitmap_bh = NULL;
431ac27a0ecSDave Kleikamp 	struct buffer_head *bh2;
432ac27a0ecSDave Kleikamp 	int group;
433ac27a0ecSDave Kleikamp 	unsigned long ino = 0;
434ac27a0ecSDave Kleikamp 	struct inode * inode;
435617ba13bSMingming Cao 	struct ext4_group_desc * gdp = NULL;
436617ba13bSMingming Cao 	struct ext4_super_block * es;
437617ba13bSMingming Cao 	struct ext4_inode_info *ei;
438617ba13bSMingming Cao 	struct ext4_sb_info *sbi;
439ac27a0ecSDave Kleikamp 	int err = 0;
440ac27a0ecSDave Kleikamp 	struct inode *ret;
441ac27a0ecSDave Kleikamp 	int i;
442ac27a0ecSDave Kleikamp 
443ac27a0ecSDave Kleikamp 	/* Cannot create files in a deleted directory */
444ac27a0ecSDave Kleikamp 	if (!dir || !dir->i_nlink)
445ac27a0ecSDave Kleikamp 		return ERR_PTR(-EPERM);
446ac27a0ecSDave Kleikamp 
447ac27a0ecSDave Kleikamp 	sb = dir->i_sb;
448ac27a0ecSDave Kleikamp 	inode = new_inode(sb);
449ac27a0ecSDave Kleikamp 	if (!inode)
450ac27a0ecSDave Kleikamp 		return ERR_PTR(-ENOMEM);
451617ba13bSMingming Cao 	ei = EXT4_I(inode);
452ac27a0ecSDave Kleikamp 
453617ba13bSMingming Cao 	sbi = EXT4_SB(sb);
454ac27a0ecSDave Kleikamp 	es = sbi->s_es;
455ac27a0ecSDave Kleikamp 	if (S_ISDIR(mode)) {
456ac27a0ecSDave Kleikamp 		if (test_opt (sb, OLDALLOC))
457ac27a0ecSDave Kleikamp 			group = find_group_dir(sb, dir);
458ac27a0ecSDave Kleikamp 		else
459ac27a0ecSDave Kleikamp 			group = find_group_orlov(sb, dir);
460ac27a0ecSDave Kleikamp 	} else
461ac27a0ecSDave Kleikamp 		group = find_group_other(sb, dir);
462ac27a0ecSDave Kleikamp 
463ac27a0ecSDave Kleikamp 	err = -ENOSPC;
464ac27a0ecSDave Kleikamp 	if (group == -1)
465ac27a0ecSDave Kleikamp 		goto out;
466ac27a0ecSDave Kleikamp 
467ac27a0ecSDave Kleikamp 	for (i = 0; i < sbi->s_groups_count; i++) {
468ac27a0ecSDave Kleikamp 		err = -EIO;
469ac27a0ecSDave Kleikamp 
470617ba13bSMingming Cao 		gdp = ext4_get_group_desc(sb, group, &bh2);
471ac27a0ecSDave Kleikamp 		if (!gdp)
472ac27a0ecSDave Kleikamp 			goto fail;
473ac27a0ecSDave Kleikamp 
474ac27a0ecSDave Kleikamp 		brelse(bitmap_bh);
475ac27a0ecSDave Kleikamp 		bitmap_bh = read_inode_bitmap(sb, group);
476ac27a0ecSDave Kleikamp 		if (!bitmap_bh)
477ac27a0ecSDave Kleikamp 			goto fail;
478ac27a0ecSDave Kleikamp 
479ac27a0ecSDave Kleikamp 		ino = 0;
480ac27a0ecSDave Kleikamp 
481ac27a0ecSDave Kleikamp repeat_in_this_group:
482617ba13bSMingming Cao 		ino = ext4_find_next_zero_bit((unsigned long *)
483617ba13bSMingming Cao 				bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino);
484617ba13bSMingming Cao 		if (ino < EXT4_INODES_PER_GROUP(sb)) {
485ac27a0ecSDave Kleikamp 
486ac27a0ecSDave Kleikamp 			BUFFER_TRACE(bitmap_bh, "get_write_access");
487617ba13bSMingming Cao 			err = ext4_journal_get_write_access(handle, bitmap_bh);
488ac27a0ecSDave Kleikamp 			if (err)
489ac27a0ecSDave Kleikamp 				goto fail;
490ac27a0ecSDave Kleikamp 
491617ba13bSMingming Cao 			if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group),
492ac27a0ecSDave Kleikamp 						ino, bitmap_bh->b_data)) {
493ac27a0ecSDave Kleikamp 				/* we won it */
494ac27a0ecSDave Kleikamp 				BUFFER_TRACE(bitmap_bh,
495617ba13bSMingming Cao 					"call ext4_journal_dirty_metadata");
496617ba13bSMingming Cao 				err = ext4_journal_dirty_metadata(handle,
497ac27a0ecSDave Kleikamp 								bitmap_bh);
498ac27a0ecSDave Kleikamp 				if (err)
499ac27a0ecSDave Kleikamp 					goto fail;
500ac27a0ecSDave Kleikamp 				goto got;
501ac27a0ecSDave Kleikamp 			}
502ac27a0ecSDave Kleikamp 			/* we lost it */
503dab291afSMingming Cao 			jbd2_journal_release_buffer(handle, bitmap_bh);
504ac27a0ecSDave Kleikamp 
505617ba13bSMingming Cao 			if (++ino < EXT4_INODES_PER_GROUP(sb))
506ac27a0ecSDave Kleikamp 				goto repeat_in_this_group;
507ac27a0ecSDave Kleikamp 		}
508ac27a0ecSDave Kleikamp 
509ac27a0ecSDave Kleikamp 		/*
510ac27a0ecSDave Kleikamp 		 * This case is possible in concurrent environment.  It is very
511ac27a0ecSDave Kleikamp 		 * rare.  We cannot repeat the find_group_xxx() call because
512ac27a0ecSDave Kleikamp 		 * that will simply return the same blockgroup, because the
513ac27a0ecSDave Kleikamp 		 * group descriptor metadata has not yet been updated.
514ac27a0ecSDave Kleikamp 		 * So we just go onto the next blockgroup.
515ac27a0ecSDave Kleikamp 		 */
516ac27a0ecSDave Kleikamp 		if (++group == sbi->s_groups_count)
517ac27a0ecSDave Kleikamp 			group = 0;
518ac27a0ecSDave Kleikamp 	}
519ac27a0ecSDave Kleikamp 	err = -ENOSPC;
520ac27a0ecSDave Kleikamp 	goto out;
521ac27a0ecSDave Kleikamp 
522ac27a0ecSDave Kleikamp got:
523617ba13bSMingming Cao 	ino += group * EXT4_INODES_PER_GROUP(sb) + 1;
524617ba13bSMingming Cao 	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
525617ba13bSMingming Cao 		ext4_error (sb, "ext4_new_inode",
526ac27a0ecSDave Kleikamp 			    "reserved inode or inode > inodes count - "
527ac27a0ecSDave Kleikamp 			    "block_group = %d, inode=%lu", group, ino);
528ac27a0ecSDave Kleikamp 		err = -EIO;
529ac27a0ecSDave Kleikamp 		goto fail;
530ac27a0ecSDave Kleikamp 	}
531ac27a0ecSDave Kleikamp 
532ac27a0ecSDave Kleikamp 	BUFFER_TRACE(bh2, "get_write_access");
533617ba13bSMingming Cao 	err = ext4_journal_get_write_access(handle, bh2);
534ac27a0ecSDave Kleikamp 	if (err) goto fail;
535ac27a0ecSDave Kleikamp 	spin_lock(sb_bgl_lock(sbi, group));
536ac27a0ecSDave Kleikamp 	gdp->bg_free_inodes_count =
537ac27a0ecSDave Kleikamp 		cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
538ac27a0ecSDave Kleikamp 	if (S_ISDIR(mode)) {
539ac27a0ecSDave Kleikamp 		gdp->bg_used_dirs_count =
540ac27a0ecSDave Kleikamp 			cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
541ac27a0ecSDave Kleikamp 	}
542ac27a0ecSDave Kleikamp 	spin_unlock(sb_bgl_lock(sbi, group));
543617ba13bSMingming Cao 	BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
544617ba13bSMingming Cao 	err = ext4_journal_dirty_metadata(handle, bh2);
545ac27a0ecSDave Kleikamp 	if (err) goto fail;
546ac27a0ecSDave Kleikamp 
547ac27a0ecSDave Kleikamp 	percpu_counter_dec(&sbi->s_freeinodes_counter);
548ac27a0ecSDave Kleikamp 	if (S_ISDIR(mode))
549ac27a0ecSDave Kleikamp 		percpu_counter_inc(&sbi->s_dirs_counter);
550ac27a0ecSDave Kleikamp 	sb->s_dirt = 1;
551ac27a0ecSDave Kleikamp 
552ac27a0ecSDave Kleikamp 	inode->i_uid = current->fsuid;
553ac27a0ecSDave Kleikamp 	if (test_opt (sb, GRPID))
554ac27a0ecSDave Kleikamp 		inode->i_gid = dir->i_gid;
555ac27a0ecSDave Kleikamp 	else if (dir->i_mode & S_ISGID) {
556ac27a0ecSDave Kleikamp 		inode->i_gid = dir->i_gid;
557ac27a0ecSDave Kleikamp 		if (S_ISDIR(mode))
558ac27a0ecSDave Kleikamp 			mode |= S_ISGID;
559ac27a0ecSDave Kleikamp 	} else
560ac27a0ecSDave Kleikamp 		inode->i_gid = current->fsgid;
561ac27a0ecSDave Kleikamp 	inode->i_mode = mode;
562ac27a0ecSDave Kleikamp 
563ac27a0ecSDave Kleikamp 	inode->i_ino = ino;
564ac27a0ecSDave Kleikamp 	/* This is the optimal IO size (for stat), not the fs block size */
565ac27a0ecSDave Kleikamp 	inode->i_blocks = 0;
566ac27a0ecSDave Kleikamp 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
567ac27a0ecSDave Kleikamp 
568ac27a0ecSDave Kleikamp 	memset(ei->i_data, 0, sizeof(ei->i_data));
569ac27a0ecSDave Kleikamp 	ei->i_dir_start_lookup = 0;
570ac27a0ecSDave Kleikamp 	ei->i_disksize = 0;
571ac27a0ecSDave Kleikamp 
572617ba13bSMingming Cao 	ei->i_flags = EXT4_I(dir)->i_flags & ~EXT4_INDEX_FL;
573ac27a0ecSDave Kleikamp 	if (S_ISLNK(mode))
574617ba13bSMingming Cao 		ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL);
575ac27a0ecSDave Kleikamp 	/* dirsync only applies to directories */
576ac27a0ecSDave Kleikamp 	if (!S_ISDIR(mode))
577617ba13bSMingming Cao 		ei->i_flags &= ~EXT4_DIRSYNC_FL;
578617ba13bSMingming Cao #ifdef EXT4_FRAGMENTS
579ac27a0ecSDave Kleikamp 	ei->i_faddr = 0;
580ac27a0ecSDave Kleikamp 	ei->i_frag_no = 0;
581ac27a0ecSDave Kleikamp 	ei->i_frag_size = 0;
582ac27a0ecSDave Kleikamp #endif
583ac27a0ecSDave Kleikamp 	ei->i_file_acl = 0;
584ac27a0ecSDave Kleikamp 	ei->i_dir_acl = 0;
585ac27a0ecSDave Kleikamp 	ei->i_dtime = 0;
586ac27a0ecSDave Kleikamp 	ei->i_block_alloc_info = NULL;
587ac27a0ecSDave Kleikamp 	ei->i_block_group = group;
588ac27a0ecSDave Kleikamp 
589617ba13bSMingming Cao 	ext4_set_inode_flags(inode);
590ac27a0ecSDave Kleikamp 	if (IS_DIRSYNC(inode))
591ac27a0ecSDave Kleikamp 		handle->h_sync = 1;
592ac27a0ecSDave Kleikamp 	insert_inode_hash(inode);
593ac27a0ecSDave Kleikamp 	spin_lock(&sbi->s_next_gen_lock);
594ac27a0ecSDave Kleikamp 	inode->i_generation = sbi->s_next_generation++;
595ac27a0ecSDave Kleikamp 	spin_unlock(&sbi->s_next_gen_lock);
596ac27a0ecSDave Kleikamp 
597617ba13bSMingming Cao 	ei->i_state = EXT4_STATE_NEW;
598ac27a0ecSDave Kleikamp 	ei->i_extra_isize =
599617ba13bSMingming Cao 		(EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) ?
600617ba13bSMingming Cao 		sizeof(struct ext4_inode) - EXT4_GOOD_OLD_INODE_SIZE : 0;
601ac27a0ecSDave Kleikamp 
602ac27a0ecSDave Kleikamp 	ret = inode;
603ac27a0ecSDave Kleikamp 	if(DQUOT_ALLOC_INODE(inode)) {
604ac27a0ecSDave Kleikamp 		err = -EDQUOT;
605ac27a0ecSDave Kleikamp 		goto fail_drop;
606ac27a0ecSDave Kleikamp 	}
607ac27a0ecSDave Kleikamp 
608617ba13bSMingming Cao 	err = ext4_init_acl(handle, inode, dir);
609ac27a0ecSDave Kleikamp 	if (err)
610ac27a0ecSDave Kleikamp 		goto fail_free_drop;
611ac27a0ecSDave Kleikamp 
612617ba13bSMingming Cao 	err = ext4_init_security(handle,inode, dir);
613ac27a0ecSDave Kleikamp 	if (err)
614ac27a0ecSDave Kleikamp 		goto fail_free_drop;
615ac27a0ecSDave Kleikamp 
616617ba13bSMingming Cao 	err = ext4_mark_inode_dirty(handle, inode);
617ac27a0ecSDave Kleikamp 	if (err) {
618617ba13bSMingming Cao 		ext4_std_error(sb, err);
619ac27a0ecSDave Kleikamp 		goto fail_free_drop;
620ac27a0ecSDave Kleikamp 	}
621a86c6181SAlex Tomas 	if (test_opt(sb, EXTENTS)) {
622a86c6181SAlex Tomas 		EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
623a86c6181SAlex Tomas 		ext4_ext_tree_init(handle, inode);
624a86c6181SAlex Tomas 		if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
625a86c6181SAlex Tomas 			err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
626a86c6181SAlex Tomas 			if (err) goto fail;
627a86c6181SAlex Tomas 			EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS);
628a86c6181SAlex Tomas 			BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "call ext4_journal_dirty_metadata");
629a86c6181SAlex Tomas 			err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
630a86c6181SAlex Tomas 		}
631a86c6181SAlex Tomas 	}
632ac27a0ecSDave Kleikamp 
633617ba13bSMingming Cao 	ext4_debug("allocating inode %lu\n", inode->i_ino);
634ac27a0ecSDave Kleikamp 	goto really_out;
635ac27a0ecSDave Kleikamp fail:
636617ba13bSMingming Cao 	ext4_std_error(sb, err);
637ac27a0ecSDave Kleikamp out:
638ac27a0ecSDave Kleikamp 	iput(inode);
639ac27a0ecSDave Kleikamp 	ret = ERR_PTR(err);
640ac27a0ecSDave Kleikamp really_out:
641ac27a0ecSDave Kleikamp 	brelse(bitmap_bh);
642ac27a0ecSDave Kleikamp 	return ret;
643ac27a0ecSDave Kleikamp 
644ac27a0ecSDave Kleikamp fail_free_drop:
645ac27a0ecSDave Kleikamp 	DQUOT_FREE_INODE(inode);
646ac27a0ecSDave Kleikamp 
647ac27a0ecSDave Kleikamp fail_drop:
648ac27a0ecSDave Kleikamp 	DQUOT_DROP(inode);
649ac27a0ecSDave Kleikamp 	inode->i_flags |= S_NOQUOTA;
650ac27a0ecSDave Kleikamp 	inode->i_nlink = 0;
651ac27a0ecSDave Kleikamp 	iput(inode);
652ac27a0ecSDave Kleikamp 	brelse(bitmap_bh);
653ac27a0ecSDave Kleikamp 	return ERR_PTR(err);
654ac27a0ecSDave Kleikamp }
655ac27a0ecSDave Kleikamp 
656ac27a0ecSDave Kleikamp /* Verify that we are loading a valid orphan from disk */
657617ba13bSMingming Cao struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
658ac27a0ecSDave Kleikamp {
659617ba13bSMingming Cao 	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
660ac27a0ecSDave Kleikamp 	unsigned long block_group;
661ac27a0ecSDave Kleikamp 	int bit;
662ac27a0ecSDave Kleikamp 	struct buffer_head *bitmap_bh = NULL;
663ac27a0ecSDave Kleikamp 	struct inode *inode = NULL;
664ac27a0ecSDave Kleikamp 
665ac27a0ecSDave Kleikamp 	/* Error cases - e2fsck has already cleaned up for us */
666ac27a0ecSDave Kleikamp 	if (ino > max_ino) {
667617ba13bSMingming Cao 		ext4_warning(sb, __FUNCTION__,
668ac27a0ecSDave Kleikamp 			     "bad orphan ino %lu!  e2fsck was run?", ino);
669ac27a0ecSDave Kleikamp 		goto out;
670ac27a0ecSDave Kleikamp 	}
671ac27a0ecSDave Kleikamp 
672617ba13bSMingming Cao 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
673617ba13bSMingming Cao 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
674ac27a0ecSDave Kleikamp 	bitmap_bh = read_inode_bitmap(sb, block_group);
675ac27a0ecSDave Kleikamp 	if (!bitmap_bh) {
676617ba13bSMingming Cao 		ext4_warning(sb, __FUNCTION__,
677ac27a0ecSDave Kleikamp 			     "inode bitmap error for orphan %lu", ino);
678ac27a0ecSDave Kleikamp 		goto out;
679ac27a0ecSDave Kleikamp 	}
680ac27a0ecSDave Kleikamp 
681ac27a0ecSDave Kleikamp 	/* Having the inode bit set should be a 100% indicator that this
682ac27a0ecSDave Kleikamp 	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
683ac27a0ecSDave Kleikamp 	 * inodes that were being truncated, so we can't check i_nlink==0.
684ac27a0ecSDave Kleikamp 	 */
685617ba13bSMingming Cao 	if (!ext4_test_bit(bit, bitmap_bh->b_data) ||
686ac27a0ecSDave Kleikamp 			!(inode = iget(sb, ino)) || is_bad_inode(inode) ||
687ac27a0ecSDave Kleikamp 			NEXT_ORPHAN(inode) > max_ino) {
688617ba13bSMingming Cao 		ext4_warning(sb, __FUNCTION__,
689ac27a0ecSDave Kleikamp 			     "bad orphan inode %lu!  e2fsck was run?", ino);
690617ba13bSMingming Cao 		printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
691ac27a0ecSDave Kleikamp 		       bit, (unsigned long long)bitmap_bh->b_blocknr,
692617ba13bSMingming Cao 		       ext4_test_bit(bit, bitmap_bh->b_data));
693ac27a0ecSDave Kleikamp 		printk(KERN_NOTICE "inode=%p\n", inode);
694ac27a0ecSDave Kleikamp 		if (inode) {
695ac27a0ecSDave Kleikamp 			printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
696ac27a0ecSDave Kleikamp 			       is_bad_inode(inode));
697ac27a0ecSDave Kleikamp 			printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
698ac27a0ecSDave Kleikamp 			       NEXT_ORPHAN(inode));
699ac27a0ecSDave Kleikamp 			printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
700ac27a0ecSDave Kleikamp 		}
701ac27a0ecSDave Kleikamp 		/* Avoid freeing blocks if we got a bad deleted inode */
702ac27a0ecSDave Kleikamp 		if (inode && inode->i_nlink == 0)
703ac27a0ecSDave Kleikamp 			inode->i_blocks = 0;
704ac27a0ecSDave Kleikamp 		iput(inode);
705ac27a0ecSDave Kleikamp 		inode = NULL;
706ac27a0ecSDave Kleikamp 	}
707ac27a0ecSDave Kleikamp out:
708ac27a0ecSDave Kleikamp 	brelse(bitmap_bh);
709ac27a0ecSDave Kleikamp 	return inode;
710ac27a0ecSDave Kleikamp }
711ac27a0ecSDave Kleikamp 
712617ba13bSMingming Cao unsigned long ext4_count_free_inodes (struct super_block * sb)
713ac27a0ecSDave Kleikamp {
714ac27a0ecSDave Kleikamp 	unsigned long desc_count;
715617ba13bSMingming Cao 	struct ext4_group_desc *gdp;
716ac27a0ecSDave Kleikamp 	int i;
717617ba13bSMingming Cao #ifdef EXT4FS_DEBUG
718617ba13bSMingming Cao 	struct ext4_super_block *es;
719ac27a0ecSDave Kleikamp 	unsigned long bitmap_count, x;
720ac27a0ecSDave Kleikamp 	struct buffer_head *bitmap_bh = NULL;
721ac27a0ecSDave Kleikamp 
722617ba13bSMingming Cao 	es = EXT4_SB(sb)->s_es;
723ac27a0ecSDave Kleikamp 	desc_count = 0;
724ac27a0ecSDave Kleikamp 	bitmap_count = 0;
725ac27a0ecSDave Kleikamp 	gdp = NULL;
726617ba13bSMingming Cao 	for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
727617ba13bSMingming Cao 		gdp = ext4_get_group_desc (sb, i, NULL);
728ac27a0ecSDave Kleikamp 		if (!gdp)
729ac27a0ecSDave Kleikamp 			continue;
730ac27a0ecSDave Kleikamp 		desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
731ac27a0ecSDave Kleikamp 		brelse(bitmap_bh);
732ac27a0ecSDave Kleikamp 		bitmap_bh = read_inode_bitmap(sb, i);
733ac27a0ecSDave Kleikamp 		if (!bitmap_bh)
734ac27a0ecSDave Kleikamp 			continue;
735ac27a0ecSDave Kleikamp 
736617ba13bSMingming Cao 		x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
737ac27a0ecSDave Kleikamp 		printk("group %d: stored = %d, counted = %lu\n",
738ac27a0ecSDave Kleikamp 			i, le16_to_cpu(gdp->bg_free_inodes_count), x);
739ac27a0ecSDave Kleikamp 		bitmap_count += x;
740ac27a0ecSDave Kleikamp 	}
741ac27a0ecSDave Kleikamp 	brelse(bitmap_bh);
742617ba13bSMingming Cao 	printk("ext4_count_free_inodes: stored = %u, computed = %lu, %lu\n",
743ac27a0ecSDave Kleikamp 		le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
744ac27a0ecSDave Kleikamp 	return desc_count;
745ac27a0ecSDave Kleikamp #else
746ac27a0ecSDave Kleikamp 	desc_count = 0;
747617ba13bSMingming Cao 	for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
748617ba13bSMingming Cao 		gdp = ext4_get_group_desc (sb, i, NULL);
749ac27a0ecSDave Kleikamp 		if (!gdp)
750ac27a0ecSDave Kleikamp 			continue;
751ac27a0ecSDave Kleikamp 		desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
752ac27a0ecSDave Kleikamp 		cond_resched();
753ac27a0ecSDave Kleikamp 	}
754ac27a0ecSDave Kleikamp 	return desc_count;
755ac27a0ecSDave Kleikamp #endif
756ac27a0ecSDave Kleikamp }
757ac27a0ecSDave Kleikamp 
758ac27a0ecSDave Kleikamp /* Called at mount-time, super-block is locked */
759617ba13bSMingming Cao unsigned long ext4_count_dirs (struct super_block * sb)
760ac27a0ecSDave Kleikamp {
761ac27a0ecSDave Kleikamp 	unsigned long count = 0;
762ac27a0ecSDave Kleikamp 	int i;
763ac27a0ecSDave Kleikamp 
764617ba13bSMingming Cao 	for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
765617ba13bSMingming Cao 		struct ext4_group_desc *gdp = ext4_get_group_desc (sb, i, NULL);
766ac27a0ecSDave Kleikamp 		if (!gdp)
767ac27a0ecSDave Kleikamp 			continue;
768ac27a0ecSDave Kleikamp 		count += le16_to_cpu(gdp->bg_used_dirs_count);
769ac27a0ecSDave Kleikamp 	}
770ac27a0ecSDave Kleikamp 	return count;
771ac27a0ecSDave Kleikamp }
772ac27a0ecSDave Kleikamp 
773