fs/hfsplus/btree.c

   1 /*
   2  *  linux/fs/hfsplus/btree.c
   3  *
   4  * Copyright (C) 2001
   5  * Brad Boyer (flar@allandria.com)
   6  * (C) 2003 Ardis Technologies <roman@ardistech.com>
   7  *
   8  * Handle opening/closing btree
   9  */
  10
  11 #include <linux/slab.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/log2.h>
  14
  15 #include "hfsplus_fs.h"
  16 #include "hfsplus_raw.h"
  17
  18 /*
  19  * Initial source code of clump size calculation is gotten
  20  * from http://opensource.apple.com/tarballs/diskdev_cmds/
  21  */
  22 #define CLUMP_ENTRIES   15
  23
  24 static short clumptbl[CLUMP_ENTRIES * 3] = {
  25 /*
  26  *          Volume      Attributes       Catalog         Extents
  27  *           Size       Clump (MB)      Clump (MB)      Clump (MB)
  28  */
  29         /*   1GB */       4,              4,             4,
  30         /*   2GB */       6,              6,             4,
  31         /*   4GB */       8,              8,             4,
  32         /*   8GB */      11,             11,             5,
  33         /*
  34          * For volumes 16GB and larger, we want to make sure that a full OS
  35          * install won't require fragmentation of the Catalog or Attributes
  36          * B-trees.  We do this by making the clump sizes sufficiently large,
  37          * and by leaving a gap after the B-trees for them to grow into.
  38          *
  39          * For SnowLeopard 10A298, a FullNetInstall with all packages selected
  40          * results in:
  41          * Catalog B-tree Header
  42          *      nodeSize:          8192
  43          *      totalNodes:       31616
  44          *      freeNodes:         1978
  45          * (used = 231.55 MB)
  46          * Attributes B-tree Header
  47          *      nodeSize:          8192
  48          *      totalNodes:       63232
  49          *      freeNodes:          958
  50          * (used = 486.52 MB)
  51          *
  52          * We also want Time Machine backup volumes to have a sufficiently
  53          * large clump size to reduce fragmentation.
  54          *
  55          * The series of numbers for Catalog and Attribute form a geometric
  56          * series. For Catalog (16GB to 512GB), each term is 8**(1/5) times
  57          * the previous term.  For Attributes (16GB to 512GB), each term is
  58          * 4**(1/5) times the previous term.  For 1TB to 16TB, each term is
  59          * 2**(1/5) times the previous term.
  60          */
  61         /*  16GB */      64,             32,             5,
  62         /*  32GB */      84,             49,             6,
  63         /*  64GB */     111,             74,             7,
  64         /* 128GB */     147,            111,             8,
  65         /* 256GB */     194,            169,             9,
  66         /* 512GB */     256,            256,            11,
  67         /*   1TB */     294,            294,            14,
  68         /*   2TB */     338,            338,            16,
  69         /*   4TB */     388,            388,            20,
  70         /*   8TB */     446,            446,            25,
  71         /*  16TB */     512,            512,            32
  72 };
  73
  74 u32 hfsplus_calc_btree_clump_size(u32 block_size, u32 node_size,
  75                                         u64 sectors, int file_id)
  76 {
  77         u32 mod = max(node_size, block_size);
  78         u32 clump_size;
  79         int column;
  80         int i;
  81
  82         /* Figure out which column of the above table to use for this file. */
  83         switch (file_id) {
  84         case HFSPLUS_ATTR_CNID:
  85                 column = 0;
  86                 break;
  87         case HFSPLUS_CAT_CNID:
  88                 column = 1;
  89                 break;
  90         default:
  91                 column = 2;
  92                 break;
  93         }
  94
  95         /*
  96          * The default clump size is 0.8% of the volume size. And
  97          * it must also be a multiple of the node and block size.
  98          */
  99         if (sectors < 0x200000) {
 100                 clump_size = sectors << 2;      /*  0.8 %  */
 101                 if (clump_size < (8 * node_size))
 102                         clump_size = 8 * node_size;
 103         } else {
 104                 /* turn exponent into table index... */
 105                 for (i = 0, sectors = sectors >> 22;
 106                      sectors && (i < CLUMP_ENTRIES - 1);
 107                      ++i, sectors = sectors >> 1) {
 108                         /* empty body */
 109                 }
 110
 111                 clump_size = clumptbl[column + (i) * 3] * 1024 * 1024;
 112         }
 113
 114         /*
 115          * Round the clump size to a multiple of node and block size.
 116          * NOTE: This rounds down.
 117          */
 118         clump_size /= mod;
 119         clump_size *= mod;
 120
 121         /*
 122          * Rounding down could have rounded down to 0 if the block size was
 123          * greater than the clump size.  If so, just use one block or node.
 124          */
 125         if (clump_size == 0)
 126                 clump_size = mod;
 127
 128         return clump_size;
 129 }
 130
 131 /* Get a reference to a B*Tree and do some initial checks */
 132 struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
 133 {
 134         struct hfs_btree *tree;
 135         struct hfs_btree_header_rec *head;
 136         struct address_space *mapping;
 137         struct inode *inode;
 138         struct page *page;
 139         unsigned int size;
 140
 141         tree = kzalloc(sizeof(*tree), GFP_KERNEL);
 142         if (!tree)
 143                 return NULL;
 144
 145         mutex_init(&tree->tree_lock);
 146         spin_lock_init(&tree->hash_lock);
 147         tree->sb = sb;
 148         tree->cnid = id;
 149         inode = hfsplus_iget(sb, id);
 150         if (IS_ERR(inode))
 151                 goto free_tree;
 152         tree->inode = inode;
 153
 154         if (!HFSPLUS_I(tree->inode)->first_blocks) {
 155                 pr_err("invalid btree extent records (0 size)\n");
 156                 goto free_inode;
 157         }
 158
 159         mapping = tree->inode->i_mapping;
 160         page = read_mapping_page(mapping, 0, NULL);
 161         if (IS_ERR(page))
 162                 goto free_inode;
 163
 164         /* Load the header */
 165         head = (struct hfs_btree_header_rec *)(kmap(page) +
 166                 sizeof(struct hfs_bnode_desc));
 167         tree->root = be32_to_cpu(head->root);
 168         tree->leaf_count = be32_to_cpu(head->leaf_count);
 169         tree->leaf_head = be32_to_cpu(head->leaf_head);
 170         tree->leaf_tail = be32_to_cpu(head->leaf_tail);
 171         tree->node_count = be32_to_cpu(head->node_count);
 172         tree->free_nodes = be32_to_cpu(head->free_nodes);
 173         tree->attributes = be32_to_cpu(head->attributes);
 174         tree->node_size = be16_to_cpu(head->node_size);
 175         tree->max_key_len = be16_to_cpu(head->max_key_len);
 176         tree->depth = be16_to_cpu(head->depth);
 177
 178         /* Verify the tree and set the correct compare function */
 179         switch (id) {
 180         case HFSPLUS_EXT_CNID:
 181                 if (tree->max_key_len != HFSPLUS_EXT_KEYLEN - sizeof(u16)) {
 182                         pr_err("invalid extent max_key_len %d\n",
 183                                 tree->max_key_len);
 184                         goto fail_page;
 185                 }
 186                 if (tree->attributes & HFS_TREE_VARIDXKEYS) {
 187                         pr_err("invalid extent btree flag\n");
 188                         goto fail_page;
 189                 }
 190
 191                 tree->keycmp = hfsplus_ext_cmp_key;
 192                 break;
 193         case HFSPLUS_CAT_CNID:
 194                 if (tree->max_key_len != HFSPLUS_CAT_KEYLEN - sizeof(u16)) {
 195                         pr_err("invalid catalog max_key_len %d\n",
 196                                 tree->max_key_len);
 197                         goto fail_page;
 198                 }
 199                 if (!(tree->attributes & HFS_TREE_VARIDXKEYS)) {
 200                         pr_err("invalid catalog btree flag\n");
 201                         goto fail_page;
 202                 }
 203
 204                 if (test_bit(HFSPLUS_SB_HFSX, &HFSPLUS_SB(sb)->flags) &&
 205                     (head->key_type == HFSPLUS_KEY_BINARY))
 206                         tree->keycmp = hfsplus_cat_bin_cmp_key;
 207                 else {
 208                         tree->keycmp = hfsplus_cat_case_cmp_key;
 209                         set_bit(HFSPLUS_SB_CASEFOLD, &HFSPLUS_SB(sb)->flags);
 210                 }
 211                 break;
 212         case HFSPLUS_ATTR_CNID:
 213                 if (tree->max_key_len != HFSPLUS_ATTR_KEYLEN - sizeof(u16)) {
 214                         pr_err("invalid attributes max_key_len %d\n",
 215                                 tree->max_key_len);
 216                         goto fail_page;
 217                 }
 218                 tree->keycmp = hfsplus_attr_bin_cmp_key;
 219                 break;
 220         default:
 221                 pr_err("unknown B*Tree requested\n");
 222                 goto fail_page;
 223         }
 224
 225         if (!(tree->attributes & HFS_TREE_BIGKEYS)) {
 226                 pr_err("invalid btree flag\n");
 227                 goto fail_page;
 228         }
 229
 230         size = tree->node_size;
 231         if (!is_power_of_2(size))
 232                 goto fail_page;
 233         if (!tree->node_count)
 234                 goto fail_page;
 235
 236         tree->node_size_shift = ffs(size) - 1;
 237
 238         tree->pages_per_bnode =
 239                 (tree->node_size + PAGE_SIZE - 1) >>
 240                 PAGE_SHIFT;
 241
 242         kunmap(page);
 243         put_page(page);
 244         return tree;
 245
 246  fail_page:
 247         put_page(page);
 248  free_inode:
 249         tree->inode->i_mapping->a_ops = &hfsplus_aops;
 250         iput(tree->inode);
 251  free_tree:
 252         kfree(tree);
 253         return NULL;
 254 }
 255
 256 /* Release resources used by a btree */
 257 void hfs_btree_close(struct hfs_btree *tree)
 258 {
 259         struct hfs_bnode *node;
 260         int i;
 261
 262         if (!tree)
 263                 return;
 264
 265         for (i = 0; i < NODE_HASH_SIZE; i++) {
 266                 while ((node = tree->node_hash[i])) {
 267                         tree->node_hash[i] = node->next_hash;
 268                         if (atomic_read(&node->refcnt))
 269                                 pr_crit("node %d:%d "
 270                                                 "still has %d user(s)!\n",
 271                                         node->tree->cnid, node->this,
 272                                         atomic_read(&node->refcnt));
 273                         hfs_bnode_free(node);
 274                         tree->node_hash_cnt--;
 275                 }
 276         }
 277         iput(tree->inode);
 278         kfree(tree);
 279 }
 280
 281 int hfs_btree_write(struct hfs_btree *tree)
 282 {
 283         struct hfs_btree_header_rec *head;
 284         struct hfs_bnode *node;
 285         struct page *page;
 286
 287         node = hfs_bnode_find(tree, 0);
 288         if (IS_ERR(node))
 289                 /* panic? */
 290                 return -EIO;
 291         /* Load the header */
 292         page = node->page[0];
 293         head = (struct hfs_btree_header_rec *)(kmap(page) +
 294                 sizeof(struct hfs_bnode_desc));
 295
 296         head->root = cpu_to_be32(tree->root);
 297         head->leaf_count = cpu_to_be32(tree->leaf_count);
 298         head->leaf_head = cpu_to_be32(tree->leaf_head);
 299         head->leaf_tail = cpu_to_be32(tree->leaf_tail);
 300         head->node_count = cpu_to_be32(tree->node_count);
 301         head->free_nodes = cpu_to_be32(tree->free_nodes);
 302         head->attributes = cpu_to_be32(tree->attributes);
 303         head->depth = cpu_to_be16(tree->depth);
 304
 305         kunmap(page);
 306         set_page_dirty(page);
 307         hfs_bnode_put(node);
 308         return 0;
 309 }
 310
 311 static struct hfs_bnode *hfs_bmap_new_bmap(struct hfs_bnode *prev, u32 idx)
 312 {
 313         struct hfs_btree *tree = prev->tree;
 314         struct hfs_bnode *node;
 315         struct hfs_bnode_desc desc;
 316         __be32 cnid;
 317
 318         node = hfs_bnode_create(tree, idx);
 319         if (IS_ERR(node))
 320                 return node;
 321
 322         tree->free_nodes--;
 323         prev->next = idx;
 324         cnid = cpu_to_be32(idx);
 325         hfs_bnode_write(prev, &cnid, offsetof(struct hfs_bnode_desc, next), 4);
 326
 327         node->type = HFS_NODE_MAP;
 328         node->num_recs = 1;
 329         hfs_bnode_clear(node, 0, tree->node_size);
 330         desc.next = 0;
 331         desc.prev = 0;
 332         desc.type = HFS_NODE_MAP;
 333         desc.height = 0;
 334         desc.num_recs = cpu_to_be16(1);
 335         desc.reserved = 0;
 336         hfs_bnode_write(node, &desc, 0, sizeof(desc));
 337         hfs_bnode_write_u16(node, 14, 0x8000);
 338         hfs_bnode_write_u16(node, tree->node_size - 2, 14);
 339         hfs_bnode_write_u16(node, tree->node_size - 4, tree->node_size - 6);
 340
 341         return node;
 342 }
 343
 344 /* Make sure @tree has enough space for the @rsvd_nodes */
 345 int hfs_bmap_reserve(struct hfs_btree *tree, int rsvd_nodes)
 346 {
 347         struct inode *inode = tree->inode;
 348         struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
 349         u32 count;
 350         int res;
 351
 352         if (rsvd_nodes <= 0)
 353                 return 0;
 354
 355         while (tree->free_nodes < rsvd_nodes) {
 356                 res = hfsplus_file_extend(inode, hfs_bnode_need_zeroout(tree));
 357                 if (res)
 358                         return res;
 359                 hip->phys_size = inode->i_size =
 360                         (loff_t)hip->alloc_blocks <<
 361                                 HFSPLUS_SB(tree->sb)->alloc_blksz_shift;
 362                 hip->fs_blocks =
 363                         hip->alloc_blocks << HFSPLUS_SB(tree->sb)->fs_shift;
 364                 inode_set_bytes(inode, inode->i_size);
 365                 count = inode->i_size >> tree->node_size_shift;
 366                 tree->free_nodes += count - tree->node_count;
 367                 tree->node_count = count;
 368         }
 369         return 0;
 370 }
 371
 372 struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
 373 {
 374         struct hfs_bnode *node, *next_node;
 375         struct page **pagep;
 376         u32 nidx, idx;
 377         unsigned off;
 378         u16 off16;
 379         u16 len;
 380         u8 *data, byte, m;
 381         int i, res;
 382
 383         res = hfs_bmap_reserve(tree, 1);
 384         if (res)
 385                 return ERR_PTR(res);
 386
 387         nidx = 0;
 388         node = hfs_bnode_find(tree, nidx);
 389         if (IS_ERR(node))
 390                 return node;
 391         len = hfs_brec_lenoff(node, 2, &off16);
 392         off = off16;
 393
 394         off += node->page_offset;
 395         pagep = node->page + (off >> PAGE_SHIFT);
 396         data = kmap(*pagep);
 397         off &= ~PAGE_MASK;
 398         idx = 0;
 399
 400         for (;;) {
 401                 while (len) {
 402                         byte = data[off];
 403                         if (byte != 0xff) {
 404                                 for (m = 0x80, i = 0; i < 8; m >>= 1, i++) {
 405                                         if (!(byte & m)) {
 406                                                 idx += i;
 407                                                 data[off] |= m;
 408                                                 set_page_dirty(*pagep);
 409                                                 kunmap(*pagep);
 410                                                 tree->free_nodes--;
 411                                                 mark_inode_dirty(tree->inode);
 412                                                 hfs_bnode_put(node);
 413                                                 return hfs_bnode_create(tree,
 414                                                         idx);
 415                                         }
 416                                 }
 417                         }
 418                         if (++off >= PAGE_SIZE) {
 419                                 kunmap(*pagep);
 420                                 data = kmap(*++pagep);
 421                                 off = 0;
 422                         }
 423                         idx += 8;
 424                         len--;
 425                 }
 426                 kunmap(*pagep);
 427                 nidx = node->next;
 428                 if (!nidx) {
 429                         hfs_dbg(BNODE_MOD, "create new bmap node\n");
 430                         next_node = hfs_bmap_new_bmap(node, idx);
 431                 } else
 432                         next_node = hfs_bnode_find(tree, nidx);
 433                 hfs_bnode_put(node);
 434                 if (IS_ERR(next_node))
 435                         return next_node;
 436                 node = next_node;
 437
 438                 len = hfs_brec_lenoff(node, 0, &off16);
 439                 off = off16;
 440                 off += node->page_offset;
 441                 pagep = node->page + (off >> PAGE_SHIFT);
 442                 data = kmap(*pagep);
 443                 off &= ~PAGE_MASK;
 444         }
 445 }
 446
 447 void hfs_bmap_free(struct hfs_bnode *node)
 448 {
 449         struct hfs_btree *tree;
 450         struct page *page;
 451         u16 off, len;
 452         u32 nidx;
 453         u8 *data, byte, m;
 454
 455         hfs_dbg(BNODE_MOD, "btree_free_node: %u\n", node->this);
 456         BUG_ON(!node->this);
 457         tree = node->tree;
 458         nidx = node->this;
 459         node = hfs_bnode_find(tree, 0);
 460         if (IS_ERR(node))
 461                 return;
 462         len = hfs_brec_lenoff(node, 2, &off);
 463         while (nidx >= len * 8) {
 464                 u32 i;
 465
 466                 nidx -= len * 8;
 467                 i = node->next;
 468                 if (!i) {
 469                         /* panic */;
 470                         pr_crit("unable to free bnode %u. "
 471                                         "bmap not found!\n",
 472                                 node->this);
 473                         hfs_bnode_put(node);
 474                         return;
 475                 }
 476                 hfs_bnode_put(node);
 477                 node = hfs_bnode_find(tree, i);
 478                 if (IS_ERR(node))
 479                         return;
 480                 if (node->type != HFS_NODE_MAP) {
 481                         /* panic */;
 482                         pr_crit("invalid bmap found! "
 483                                         "(%u,%d)\n",
 484                                 node->this, node->type);
 485                         hfs_bnode_put(node);
 486                         return;
 487                 }
 488                 len = hfs_brec_lenoff(node, 0, &off);
 489         }
 490         off += node->page_offset + nidx / 8;
 491         page = node->page[off >> PAGE_SHIFT];
 492         data = kmap(page);
 493         off &= ~PAGE_MASK;
 494         m = 1 << (~nidx & 7);
 495         byte = data[off];
 496         if (!(byte & m)) {
 497                 pr_crit("trying to free free bnode "
 498                                 "%u(%d)\n",
 499                         node->this, node->type);
 500                 kunmap(page);
 501                 hfs_bnode_put(node);
 502                 return;
 503         }
 504         data[off] = byte & ~m;
 505         set_page_dirty(page);
 506         kunmap(page);
 507         hfs_bnode_put(node);
 508         tree->free_nodes++;
 509         mark_inode_dirty(tree->inode);
 510 }