fs/hfs/bnode.c

   1 /*
   2  *  linux/fs/hfs/bnode.c
   3  *
   4  * Copyright (C) 2001
   5  * Brad Boyer (flar@allandria.com)
   6  * (C) 2003 Ardis Technologies <roman@ardistech.com>
   7  *
   8  * Handle basic btree node operations
   9  */
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/slab.h>
  13 #include <linux/swap.h>
  14
  15 #include "btree.h"
  16
  17 void hfs_bnode_read(struct hfs_bnode *node, void *buf, int off, int len)
  18 {
  19         struct page *page;
  20         int pagenum;
  21         int bytes_read;
  22         int bytes_to_read;
  23         void *vaddr;
  24
  25         off += node->page_offset;
  26         pagenum = off >> PAGE_SHIFT;
  27         off &= ~PAGE_MASK; /* compute page offset for the first page */
  28
  29         for (bytes_read = 0; bytes_read < len; bytes_read += bytes_to_read) {
  30                 if (pagenum >= node->tree->pages_per_bnode)
  31                         break;
  32                 page = node->page[pagenum];
  33                 bytes_to_read = min_t(int, len - bytes_read, PAGE_SIZE - off);
  34
  35                 vaddr = kmap_atomic(page);
  36                 memcpy(buf + bytes_read, vaddr + off, bytes_to_read);
  37                 kunmap_atomic(vaddr);
  38
  39                 pagenum++;
  40                 off = 0; /* page offset only applies to the first page */
  41         }
  42 }
  43
  44 u16 hfs_bnode_read_u16(struct hfs_bnode *node, int off)
  45 {
  46         __be16 data;
  47         // optimize later...
  48         hfs_bnode_read(node, &data, off, 2);
  49         return be16_to_cpu(data);
  50 }
  51
  52 u8 hfs_bnode_read_u8(struct hfs_bnode *node, int off)
  53 {
  54         u8 data;
  55         // optimize later...
  56         hfs_bnode_read(node, &data, off, 1);
  57         return data;
  58 }
  59
  60 void hfs_bnode_read_key(struct hfs_bnode *node, void *key, int off)
  61 {
  62         struct hfs_btree *tree;
  63         int key_len;
  64
  65         tree = node->tree;
  66         if (node->type == HFS_NODE_LEAF ||
  67             tree->attributes & HFS_TREE_VARIDXKEYS)
  68                 key_len = hfs_bnode_read_u8(node, off) + 1;
  69         else
  70                 key_len = tree->max_key_len + 1;
  71
  72         hfs_bnode_read(node, key, off, key_len);
  73 }
  74
  75 void hfs_bnode_write(struct hfs_bnode *node, void *buf, int off, int len)
  76 {
  77         struct page *page;
  78
  79         off += node->page_offset;
  80         page = node->page[0];
  81
  82         memcpy(kmap(page) + off, buf, len);
  83         kunmap(page);
  84         set_page_dirty(page);
  85 }
  86
  87 void hfs_bnode_write_u16(struct hfs_bnode *node, int off, u16 data)
  88 {
  89         __be16 v = cpu_to_be16(data);
  90         // optimize later...
  91         hfs_bnode_write(node, &v, off, 2);
  92 }
  93
  94 void hfs_bnode_write_u8(struct hfs_bnode *node, int off, u8 data)
  95 {
  96         // optimize later...
  97         hfs_bnode_write(node, &data, off, 1);
  98 }
  99
 100 void hfs_bnode_clear(struct hfs_bnode *node, int off, int len)
 101 {
 102         struct page *page;
 103
 104         off += node->page_offset;
 105         page = node->page[0];
 106
 107         memset(kmap(page) + off, 0, len);
 108         kunmap(page);
 109         set_page_dirty(page);
 110 }
 111
 112 void hfs_bnode_copy(struct hfs_bnode *dst_node, int dst,
 113                 struct hfs_bnode *src_node, int src, int len)
 114 {
 115         struct hfs_btree *tree;
 116         struct page *src_page, *dst_page;
 117
 118         hfs_dbg(BNODE_MOD, "copybytes: %u,%u,%u\n", dst, src, len);
 119         if (!len)
 120                 return;
 121         tree = src_node->tree;
 122         src += src_node->page_offset;
 123         dst += dst_node->page_offset;
 124         src_page = src_node->page[0];
 125         dst_page = dst_node->page[0];
 126
 127         memcpy(kmap(dst_page) + dst, kmap(src_page) + src, len);
 128         kunmap(src_page);
 129         kunmap(dst_page);
 130         set_page_dirty(dst_page);
 131 }
 132
 133 void hfs_bnode_move(struct hfs_bnode *node, int dst, int src, int len)
 134 {
 135         struct page *page;
 136         void *ptr;
 137
 138         hfs_dbg(BNODE_MOD, "movebytes: %u,%u,%u\n", dst, src, len);
 139         if (!len)
 140                 return;
 141         src += node->page_offset;
 142         dst += node->page_offset;
 143         page = node->page[0];
 144         ptr = kmap(page);
 145         memmove(ptr + dst, ptr + src, len);
 146         kunmap(page);
 147         set_page_dirty(page);
 148 }
 149
 150 void hfs_bnode_dump(struct hfs_bnode *node)
 151 {
 152         struct hfs_bnode_desc desc;
 153         __be32 cnid;
 154         int i, off, key_off;
 155
 156         hfs_dbg(BNODE_MOD, "bnode: %d\n", node->this);
 157         hfs_bnode_read(node, &desc, 0, sizeof(desc));
 158         hfs_dbg(BNODE_MOD, "%d, %d, %d, %d, %d\n",
 159                 be32_to_cpu(desc.next), be32_to_cpu(desc.prev),
 160                 desc.type, desc.height, be16_to_cpu(desc.num_recs));
 161
 162         off = node->tree->node_size - 2;
 163         for (i = be16_to_cpu(desc.num_recs); i >= 0; off -= 2, i--) {
 164                 key_off = hfs_bnode_read_u16(node, off);
 165                 hfs_dbg_cont(BNODE_MOD, " %d", key_off);
 166                 if (i && node->type == HFS_NODE_INDEX) {
 167                         int tmp;
 168
 169                         if (node->tree->attributes & HFS_TREE_VARIDXKEYS)
 170                                 tmp = (hfs_bnode_read_u8(node, key_off) | 1) + 1;
 171                         else
 172                                 tmp = node->tree->max_key_len + 1;
 173                         hfs_dbg_cont(BNODE_MOD, " (%d,%d",
 174                                      tmp, hfs_bnode_read_u8(node, key_off));
 175                         hfs_bnode_read(node, &cnid, key_off + tmp, 4);
 176                         hfs_dbg_cont(BNODE_MOD, ",%d)", be32_to_cpu(cnid));
 177                 } else if (i && node->type == HFS_NODE_LEAF) {
 178                         int tmp;
 179
 180                         tmp = hfs_bnode_read_u8(node, key_off);
 181                         hfs_dbg_cont(BNODE_MOD, " (%d)", tmp);
 182                 }
 183         }
 184         hfs_dbg_cont(BNODE_MOD, "\n");
 185 }
 186
 187 void hfs_bnode_unlink(struct hfs_bnode *node)
 188 {
 189         struct hfs_btree *tree;
 190         struct hfs_bnode *tmp;
 191         __be32 cnid;
 192
 193         tree = node->tree;
 194         if (node->prev) {
 195                 tmp = hfs_bnode_find(tree, node->prev);
 196                 if (IS_ERR(tmp))
 197                         return;
 198                 tmp->next = node->next;
 199                 cnid = cpu_to_be32(tmp->next);
 200                 hfs_bnode_write(tmp, &cnid, offsetof(struct hfs_bnode_desc, next), 4);
 201                 hfs_bnode_put(tmp);
 202         } else if (node->type == HFS_NODE_LEAF)
 203                 tree->leaf_head = node->next;
 204
 205         if (node->next) {
 206                 tmp = hfs_bnode_find(tree, node->next);
 207                 if (IS_ERR(tmp))
 208                         return;
 209                 tmp->prev = node->prev;
 210                 cnid = cpu_to_be32(tmp->prev);
 211                 hfs_bnode_write(tmp, &cnid, offsetof(struct hfs_bnode_desc, prev), 4);
 212                 hfs_bnode_put(tmp);
 213         } else if (node->type == HFS_NODE_LEAF)
 214                 tree->leaf_tail = node->prev;
 215
 216         // move down?
 217         if (!node->prev && !node->next) {
 218                 printk(KERN_DEBUG "hfs_btree_del_level\n");
 219         }
 220         if (!node->parent) {
 221                 tree->root = 0;
 222                 tree->depth = 0;
 223         }
 224         set_bit(HFS_BNODE_DELETED, &node->flags);
 225 }
 226
 227 static inline int hfs_bnode_hash(u32 num)
 228 {
 229         num = (num >> 16) + num;
 230         num += num >> 8;
 231         return num & (NODE_HASH_SIZE - 1);
 232 }
 233
 234 struct hfs_bnode *hfs_bnode_findhash(struct hfs_btree *tree, u32 cnid)
 235 {
 236         struct hfs_bnode *node;
 237
 238         if (cnid >= tree->node_count) {
 239                 pr_err("request for non-existent node %d in B*Tree\n", cnid);
 240                 return NULL;
 241         }
 242
 243         for (node = tree->node_hash[hfs_bnode_hash(cnid)];
 244              node; node = node->next_hash) {
 245                 if (node->this == cnid) {
 246                         return node;
 247                 }
 248         }
 249         return NULL;
 250 }
 251
 252 static struct hfs_bnode *__hfs_bnode_create(struct hfs_btree *tree, u32 cnid)
 253 {
 254         struct super_block *sb;
 255         struct hfs_bnode *node, *node2;
 256         struct address_space *mapping;
 257         struct page *page;
 258         int size, block, i, hash;
 259         loff_t off;
 260
 261         if (cnid >= tree->node_count) {
 262                 pr_err("request for non-existent node %d in B*Tree\n", cnid);
 263                 return NULL;
 264         }
 265
 266         sb = tree->inode->i_sb;
 267         size = sizeof(struct hfs_bnode) + tree->pages_per_bnode *
 268                 sizeof(struct page *);
 269         node = kzalloc(size, GFP_KERNEL);
 270         if (!node)
 271                 return NULL;
 272         node->tree = tree;
 273         node->this = cnid;
 274         set_bit(HFS_BNODE_NEW, &node->flags);
 275         atomic_set(&node->refcnt, 1);
 276         hfs_dbg(BNODE_REFS, "new_node(%d:%d): 1\n",
 277                 node->tree->cnid, node->this);
 278         init_waitqueue_head(&node->lock_wq);
 279         spin_lock(&tree->hash_lock);
 280         node2 = hfs_bnode_findhash(tree, cnid);
 281         if (!node2) {
 282                 hash = hfs_bnode_hash(cnid);
 283                 node->next_hash = tree->node_hash[hash];
 284                 tree->node_hash[hash] = node;
 285                 tree->node_hash_cnt++;
 286         } else {
 287                 spin_unlock(&tree->hash_lock);
 288                 kfree(node);
 289                 wait_event(node2->lock_wq, !test_bit(HFS_BNODE_NEW, &node2->flags));
 290                 return node2;
 291         }
 292         spin_unlock(&tree->hash_lock);
 293
 294         mapping = tree->inode->i_mapping;
 295         off = (loff_t)cnid * tree->node_size;
 296         block = off >> PAGE_CACHE_SHIFT;
 297         node->page_offset = off & ~PAGE_CACHE_MASK;
 298         for (i = 0; i < tree->pages_per_bnode; i++) {
 299                 page = read_mapping_page(mapping, block++, NULL);
 300                 if (IS_ERR(page))
 301                         goto fail;
 302                 if (PageError(page)) {
 303                         page_cache_release(page);
 304                         goto fail;
 305                 }
 306                 node->page[i] = page;
 307         }
 308
 309         return node;
 310 fail:
 311         set_bit(HFS_BNODE_ERROR, &node->flags);
 312         return node;
 313 }
 314
 315 void hfs_bnode_unhash(struct hfs_bnode *node)
 316 {
 317         struct hfs_bnode **p;
 318
 319         hfs_dbg(BNODE_REFS, "remove_node(%d:%d): %d\n",
 320                 node->tree->cnid, node->this, atomic_read(&node->refcnt));
 321         for (p = &node->tree->node_hash[hfs_bnode_hash(node->this)];
 322              *p && *p != node; p = &(*p)->next_hash)
 323                 ;
 324         BUG_ON(!*p);
 325         *p = node->next_hash;
 326         node->tree->node_hash_cnt--;
 327 }
 328
 329 /* Load a particular node out of a tree */
 330 struct hfs_bnode *hfs_bnode_find(struct hfs_btree *tree, u32 num)
 331 {
 332         struct hfs_bnode *node;
 333         struct hfs_bnode_desc *desc;
 334         int i, rec_off, off, next_off;
 335         int entry_size, key_size;
 336
 337         spin_lock(&tree->hash_lock);
 338         node = hfs_bnode_findhash(tree, num);
 339         if (node) {
 340                 hfs_bnode_get(node);
 341                 spin_unlock(&tree->hash_lock);
 342                 wait_event(node->lock_wq, !test_bit(HFS_BNODE_NEW, &node->flags));
 343                 if (test_bit(HFS_BNODE_ERROR, &node->flags))
 344                         goto node_error;
 345                 return node;
 346         }
 347         spin_unlock(&tree->hash_lock);
 348         node = __hfs_bnode_create(tree, num);
 349         if (!node)
 350                 return ERR_PTR(-ENOMEM);
 351         if (test_bit(HFS_BNODE_ERROR, &node->flags))
 352                 goto node_error;
 353         if (!test_bit(HFS_BNODE_NEW, &node->flags))
 354                 return node;
 355
 356         desc = (struct hfs_bnode_desc *)(kmap(node->page[0]) + node->page_offset);
 357         node->prev = be32_to_cpu(desc->prev);
 358         node->next = be32_to_cpu(desc->next);
 359         node->num_recs = be16_to_cpu(desc->num_recs);
 360         node->type = desc->type;
 361         node->height = desc->height;
 362         kunmap(node->page[0]);
 363
 364         switch (node->type) {
 365         case HFS_NODE_HEADER:
 366         case HFS_NODE_MAP:
 367                 if (node->height != 0)
 368                         goto node_error;
 369                 break;
 370         case HFS_NODE_LEAF:
 371                 if (node->height != 1)
 372                         goto node_error;
 373                 break;
 374         case HFS_NODE_INDEX:
 375                 if (node->height <= 1 || node->height > tree->depth)
 376                         goto node_error;
 377                 break;
 378         default:
 379                 goto node_error;
 380         }
 381
 382         rec_off = tree->node_size - 2;
 383         off = hfs_bnode_read_u16(node, rec_off);
 384         if (off != sizeof(struct hfs_bnode_desc))
 385                 goto node_error;
 386         for (i = 1; i <= node->num_recs; off = next_off, i++) {
 387                 rec_off -= 2;
 388                 next_off = hfs_bnode_read_u16(node, rec_off);
 389                 if (next_off <= off ||
 390                     next_off > tree->node_size ||
 391                     next_off & 1)
 392                         goto node_error;
 393                 entry_size = next_off - off;
 394                 if (node->type != HFS_NODE_INDEX &&
 395                     node->type != HFS_NODE_LEAF)
 396                         continue;
 397                 key_size = hfs_bnode_read_u8(node, off) + 1;
 398                 if (key_size >= entry_size /*|| key_size & 1*/)
 399                         goto node_error;
 400         }
 401         clear_bit(HFS_BNODE_NEW, &node->flags);
 402         wake_up(&node->lock_wq);
 403         return node;
 404
 405 node_error:
 406         set_bit(HFS_BNODE_ERROR, &node->flags);
 407         clear_bit(HFS_BNODE_NEW, &node->flags);
 408         wake_up(&node->lock_wq);
 409         hfs_bnode_put(node);
 410         return ERR_PTR(-EIO);
 411 }
 412
 413 void hfs_bnode_free(struct hfs_bnode *node)
 414 {
 415         int i;
 416
 417         for (i = 0; i < node->tree->pages_per_bnode; i++)
 418                 if (node->page[i])
 419                         page_cache_release(node->page[i]);
 420         kfree(node);
 421 }
 422
 423 struct hfs_bnode *hfs_bnode_create(struct hfs_btree *tree, u32 num)
 424 {
 425         struct hfs_bnode *node;
 426         struct page **pagep;
 427         int i;
 428
 429         spin_lock(&tree->hash_lock);
 430         node = hfs_bnode_findhash(tree, num);
 431         spin_unlock(&tree->hash_lock);
 432         if (node) {
 433                 pr_crit("new node %u already hashed?\n", num);
 434                 WARN_ON(1);
 435                 return node;
 436         }
 437         node = __hfs_bnode_create(tree, num);
 438         if (!node)
 439                 return ERR_PTR(-ENOMEM);
 440         if (test_bit(HFS_BNODE_ERROR, &node->flags)) {
 441                 hfs_bnode_put(node);
 442                 return ERR_PTR(-EIO);
 443         }
 444
 445         pagep = node->page;
 446         memset(kmap(*pagep) + node->page_offset, 0,
 447                min((int)PAGE_CACHE_SIZE, (int)tree->node_size));
 448         set_page_dirty(*pagep);
 449         kunmap(*pagep);
 450         for (i = 1; i < tree->pages_per_bnode; i++) {
 451                 memset(kmap(*++pagep), 0, PAGE_CACHE_SIZE);
 452                 set_page_dirty(*pagep);
 453                 kunmap(*pagep);
 454         }
 455         clear_bit(HFS_BNODE_NEW, &node->flags);
 456         wake_up(&node->lock_wq);
 457
 458         return node;
 459 }
 460
 461 void hfs_bnode_get(struct hfs_bnode *node)
 462 {
 463         if (node) {
 464                 atomic_inc(&node->refcnt);
 465                 hfs_dbg(BNODE_REFS, "get_node(%d:%d): %d\n",
 466                         node->tree->cnid, node->this,
 467                         atomic_read(&node->refcnt));
 468         }
 469 }
 470
 471 /* Dispose of resources used by a node */
 472 void hfs_bnode_put(struct hfs_bnode *node)
 473 {
 474         if (node) {
 475                 struct hfs_btree *tree = node->tree;
 476                 int i;
 477
 478                 hfs_dbg(BNODE_REFS, "put_node(%d:%d): %d\n",
 479                         node->tree->cnid, node->this,
 480                         atomic_read(&node->refcnt));
 481                 BUG_ON(!atomic_read(&node->refcnt));
 482                 if (!atomic_dec_and_lock(&node->refcnt, &tree->hash_lock))
 483                         return;
 484                 for (i = 0; i < tree->pages_per_bnode; i++) {
 485                         if (!node->page[i])
 486                                 continue;
 487                         mark_page_accessed(node->page[i]);
 488                 }
 489
 490                 if (test_bit(HFS_BNODE_DELETED, &node->flags)) {
 491                         hfs_bnode_unhash(node);
 492                         spin_unlock(&tree->hash_lock);
 493                         hfs_bmap_free(node);
 494                         hfs_bnode_free(node);
 495                         return;
 496                 }
 497                 spin_unlock(&tree->hash_lock);
 498         }
 499 }