x86, mrst: use a temporary variable for SFI irq
[linux-btrfs-devel.git] / fs / hfs / btree.c
blob3ebc437736febb4e0ef37a3612ea6501c7eac59a
1 /*
2 * linux/fs/hfs/btree.c
4 * Copyright (C) 2001
5 * Brad Boyer (flar@allandria.com)
6 * (C) 2003 Ardis Technologies <roman@ardistech.com>
8 * Handle opening/closing btree
9 */
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/log2.h>
15 #include "btree.h"
17 /* Get a reference to a B*Tree and do some initial checks */
18 struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id, btree_keycmp keycmp)
20 struct hfs_btree *tree;
21 struct hfs_btree_header_rec *head;
22 struct address_space *mapping;
23 struct page *page;
24 unsigned int size;
26 tree = kzalloc(sizeof(*tree), GFP_KERNEL);
27 if (!tree)
28 return NULL;
30 mutex_init(&tree->tree_lock);
31 spin_lock_init(&tree->hash_lock);
32 /* Set the correct compare function */
33 tree->sb = sb;
34 tree->cnid = id;
35 tree->keycmp = keycmp;
37 tree->inode = iget_locked(sb, id);
38 if (!tree->inode)
39 goto free_tree;
40 BUG_ON(!(tree->inode->i_state & I_NEW));
42 struct hfs_mdb *mdb = HFS_SB(sb)->mdb;
43 HFS_I(tree->inode)->flags = 0;
44 mutex_init(&HFS_I(tree->inode)->extents_lock);
45 switch (id) {
46 case HFS_EXT_CNID:
47 hfs_inode_read_fork(tree->inode, mdb->drXTExtRec, mdb->drXTFlSize,
48 mdb->drXTFlSize, be32_to_cpu(mdb->drXTClpSiz));
49 tree->inode->i_mapping->a_ops = &hfs_btree_aops;
50 break;
51 case HFS_CAT_CNID:
52 hfs_inode_read_fork(tree->inode, mdb->drCTExtRec, mdb->drCTFlSize,
53 mdb->drCTFlSize, be32_to_cpu(mdb->drCTClpSiz));
54 tree->inode->i_mapping->a_ops = &hfs_btree_aops;
55 break;
56 default:
57 BUG();
60 unlock_new_inode(tree->inode);
62 if (!HFS_I(tree->inode)->first_blocks) {
63 printk(KERN_ERR "hfs: invalid btree extent records (0 size).\n");
64 goto free_inode;
67 mapping = tree->inode->i_mapping;
68 page = read_mapping_page(mapping, 0, NULL);
69 if (IS_ERR(page))
70 goto free_inode;
72 /* Load the header */
73 head = (struct hfs_btree_header_rec *)(kmap(page) + sizeof(struct hfs_bnode_desc));
74 tree->root = be32_to_cpu(head->root);
75 tree->leaf_count = be32_to_cpu(head->leaf_count);
76 tree->leaf_head = be32_to_cpu(head->leaf_head);
77 tree->leaf_tail = be32_to_cpu(head->leaf_tail);
78 tree->node_count = be32_to_cpu(head->node_count);
79 tree->free_nodes = be32_to_cpu(head->free_nodes);
80 tree->attributes = be32_to_cpu(head->attributes);
81 tree->node_size = be16_to_cpu(head->node_size);
82 tree->max_key_len = be16_to_cpu(head->max_key_len);
83 tree->depth = be16_to_cpu(head->depth);
85 size = tree->node_size;
86 if (!is_power_of_2(size))
87 goto fail_page;
88 if (!tree->node_count)
89 goto fail_page;
90 switch (id) {
91 case HFS_EXT_CNID:
92 if (tree->max_key_len != HFS_MAX_EXT_KEYLEN) {
93 printk(KERN_ERR "hfs: invalid extent max_key_len %d\n",
94 tree->max_key_len);
95 goto fail_page;
97 break;
98 case HFS_CAT_CNID:
99 if (tree->max_key_len != HFS_MAX_CAT_KEYLEN) {
100 printk(KERN_ERR "hfs: invalid catalog max_key_len %d\n",
101 tree->max_key_len);
102 goto fail_page;
104 break;
105 default:
106 BUG();
109 tree->node_size_shift = ffs(size) - 1;
110 tree->pages_per_bnode = (tree->node_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
112 kunmap(page);
113 page_cache_release(page);
114 return tree;
116 fail_page:
117 page_cache_release(page);
118 free_inode:
119 tree->inode->i_mapping->a_ops = &hfs_aops;
120 iput(tree->inode);
121 free_tree:
122 kfree(tree);
123 return NULL;
126 /* Release resources used by a btree */
127 void hfs_btree_close(struct hfs_btree *tree)
129 struct hfs_bnode *node;
130 int i;
132 if (!tree)
133 return;
135 for (i = 0; i < NODE_HASH_SIZE; i++) {
136 while ((node = tree->node_hash[i])) {
137 tree->node_hash[i] = node->next_hash;
138 if (atomic_read(&node->refcnt))
139 printk(KERN_ERR "hfs: node %d:%d still has %d user(s)!\n",
140 node->tree->cnid, node->this, atomic_read(&node->refcnt));
141 hfs_bnode_free(node);
142 tree->node_hash_cnt--;
145 iput(tree->inode);
146 kfree(tree);
149 void hfs_btree_write(struct hfs_btree *tree)
151 struct hfs_btree_header_rec *head;
152 struct hfs_bnode *node;
153 struct page *page;
155 node = hfs_bnode_find(tree, 0);
156 if (IS_ERR(node))
157 /* panic? */
158 return;
159 /* Load the header */
160 page = node->page[0];
161 head = (struct hfs_btree_header_rec *)(kmap(page) + sizeof(struct hfs_bnode_desc));
163 head->root = cpu_to_be32(tree->root);
164 head->leaf_count = cpu_to_be32(tree->leaf_count);
165 head->leaf_head = cpu_to_be32(tree->leaf_head);
166 head->leaf_tail = cpu_to_be32(tree->leaf_tail);
167 head->node_count = cpu_to_be32(tree->node_count);
168 head->free_nodes = cpu_to_be32(tree->free_nodes);
169 head->attributes = cpu_to_be32(tree->attributes);
170 head->depth = cpu_to_be16(tree->depth);
172 kunmap(page);
173 set_page_dirty(page);
174 hfs_bnode_put(node);
177 static struct hfs_bnode *hfs_bmap_new_bmap(struct hfs_bnode *prev, u32 idx)
179 struct hfs_btree *tree = prev->tree;
180 struct hfs_bnode *node;
181 struct hfs_bnode_desc desc;
182 __be32 cnid;
184 node = hfs_bnode_create(tree, idx);
185 if (IS_ERR(node))
186 return node;
188 if (!tree->free_nodes)
189 panic("FIXME!!!");
190 tree->free_nodes--;
191 prev->next = idx;
192 cnid = cpu_to_be32(idx);
193 hfs_bnode_write(prev, &cnid, offsetof(struct hfs_bnode_desc, next), 4);
195 node->type = HFS_NODE_MAP;
196 node->num_recs = 1;
197 hfs_bnode_clear(node, 0, tree->node_size);
198 desc.next = 0;
199 desc.prev = 0;
200 desc.type = HFS_NODE_MAP;
201 desc.height = 0;
202 desc.num_recs = cpu_to_be16(1);
203 desc.reserved = 0;
204 hfs_bnode_write(node, &desc, 0, sizeof(desc));
205 hfs_bnode_write_u16(node, 14, 0x8000);
206 hfs_bnode_write_u16(node, tree->node_size - 2, 14);
207 hfs_bnode_write_u16(node, tree->node_size - 4, tree->node_size - 6);
209 return node;
212 struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
214 struct hfs_bnode *node, *next_node;
215 struct page **pagep;
216 u32 nidx, idx;
217 unsigned off;
218 u16 off16;
219 u16 len;
220 u8 *data, byte, m;
221 int i;
223 while (!tree->free_nodes) {
224 struct inode *inode = tree->inode;
225 u32 count;
226 int res;
228 res = hfs_extend_file(inode);
229 if (res)
230 return ERR_PTR(res);
231 HFS_I(inode)->phys_size = inode->i_size =
232 (loff_t)HFS_I(inode)->alloc_blocks *
233 HFS_SB(tree->sb)->alloc_blksz;
234 HFS_I(inode)->fs_blocks = inode->i_size >>
235 tree->sb->s_blocksize_bits;
236 inode_set_bytes(inode, inode->i_size);
237 count = inode->i_size >> tree->node_size_shift;
238 tree->free_nodes = count - tree->node_count;
239 tree->node_count = count;
242 nidx = 0;
243 node = hfs_bnode_find(tree, nidx);
244 if (IS_ERR(node))
245 return node;
246 len = hfs_brec_lenoff(node, 2, &off16);
247 off = off16;
249 off += node->page_offset;
250 pagep = node->page + (off >> PAGE_CACHE_SHIFT);
251 data = kmap(*pagep);
252 off &= ~PAGE_CACHE_MASK;
253 idx = 0;
255 for (;;) {
256 while (len) {
257 byte = data[off];
258 if (byte != 0xff) {
259 for (m = 0x80, i = 0; i < 8; m >>= 1, i++) {
260 if (!(byte & m)) {
261 idx += i;
262 data[off] |= m;
263 set_page_dirty(*pagep);
264 kunmap(*pagep);
265 tree->free_nodes--;
266 mark_inode_dirty(tree->inode);
267 hfs_bnode_put(node);
268 return hfs_bnode_create(tree, idx);
272 if (++off >= PAGE_CACHE_SIZE) {
273 kunmap(*pagep);
274 data = kmap(*++pagep);
275 off = 0;
277 idx += 8;
278 len--;
280 kunmap(*pagep);
281 nidx = node->next;
282 if (!nidx) {
283 printk(KERN_DEBUG "hfs: create new bmap node...\n");
284 next_node = hfs_bmap_new_bmap(node, idx);
285 } else
286 next_node = hfs_bnode_find(tree, nidx);
287 hfs_bnode_put(node);
288 if (IS_ERR(next_node))
289 return next_node;
290 node = next_node;
292 len = hfs_brec_lenoff(node, 0, &off16);
293 off = off16;
294 off += node->page_offset;
295 pagep = node->page + (off >> PAGE_CACHE_SHIFT);
296 data = kmap(*pagep);
297 off &= ~PAGE_CACHE_MASK;
301 void hfs_bmap_free(struct hfs_bnode *node)
303 struct hfs_btree *tree;
304 struct page *page;
305 u16 off, len;
306 u32 nidx;
307 u8 *data, byte, m;
309 dprint(DBG_BNODE_MOD, "btree_free_node: %u\n", node->this);
310 tree = node->tree;
311 nidx = node->this;
312 node = hfs_bnode_find(tree, 0);
313 if (IS_ERR(node))
314 return;
315 len = hfs_brec_lenoff(node, 2, &off);
316 while (nidx >= len * 8) {
317 u32 i;
319 nidx -= len * 8;
320 i = node->next;
321 hfs_bnode_put(node);
322 if (!i) {
323 /* panic */;
324 printk(KERN_CRIT "hfs: unable to free bnode %u. bmap not found!\n", node->this);
325 return;
327 node = hfs_bnode_find(tree, i);
328 if (IS_ERR(node))
329 return;
330 if (node->type != HFS_NODE_MAP) {
331 /* panic */;
332 printk(KERN_CRIT "hfs: invalid bmap found! (%u,%d)\n", node->this, node->type);
333 hfs_bnode_put(node);
334 return;
336 len = hfs_brec_lenoff(node, 0, &off);
338 off += node->page_offset + nidx / 8;
339 page = node->page[off >> PAGE_CACHE_SHIFT];
340 data = kmap(page);
341 off &= ~PAGE_CACHE_MASK;
342 m = 1 << (~nidx & 7);
343 byte = data[off];
344 if (!(byte & m)) {
345 printk(KERN_CRIT "hfs: trying to free free bnode %u(%d)\n", node->this, node->type);
346 kunmap(page);
347 hfs_bnode_put(node);
348 return;
350 data[off] = byte & ~m;
351 set_page_dirty(page);
352 kunmap(page);
353 hfs_bnode_put(node);
354 tree->free_nodes++;
355 mark_inode_dirty(tree->inode);