fs/hfsplus/btree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/hfsplus/btree.c
   4  *
   5  * Copyright (C) 2001
   6  * Brad Boyer (flar@allandria.com)
   7  * (C) 2003 Ardis Technologies <roman@ardistech.com>
   8  *
   9  * Handle opening/closing btree
  10  */
  11
  12 #include <linux/slab.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/log2.h>
  15
  16 #include "hfsplus_fs.h"
  17 #include "hfsplus_raw.h"
  18
  19 /*
  20  * Initial source code of clump size calculation is gotten
  21  * from http://opensource.apple.com/tarballs/diskdev_cmds/
  22  */
  23 #define CLUMP_ENTRIES   15
  24
  25 static short clumptbl[CLUMP_ENTRIES * 3] = {
  26 /*
  27  *          Volume      Attributes       Catalog         Extents
  28  *           Size       Clump (MB)      Clump (MB)      Clump (MB)
  29  */
  30         /*   1GB */       4,              4,             4,
  31         /*   2GB */       6,              6,             4,
  32         /*   4GB */       8,              8,             4,
  33         /*   8GB */      11,             11,             5,
  34         /*
  35          * For volumes 16GB and larger, we want to make sure that a full OS
  36          * install won't require fragmentation of the Catalog or Attributes
  37          * B-trees.  We do this by making the clump sizes sufficiently large,
  38          * and by leaving a gap after the B-trees for them to grow into.
  39          *
  40          * For SnowLeopard 10A298, a FullNetInstall with all packages selected
  41          * results in:
  42          * Catalog B-tree Header
  43          *      nodeSize:          8192
  44          *      totalNodes:       31616
  45          *      freeNodes:         1978
  46          * (used = 231.55 MB)
  47          * Attributes B-tree Header
  48          *      nodeSize:          8192
  49          *      totalNodes:       63232
  50          *      freeNodes:          958
  51          * (used = 486.52 MB)
  52          *
  53          * We also want Time Machine backup volumes to have a sufficiently
  54          * large clump size to reduce fragmentation.
  55          *
  56          * The series of numbers for Catalog and Attribute form a geometric
  57          * series. For Catalog (16GB to 512GB), each term is 8**(1/5) times
  58          * the previous term.  For Attributes (16GB to 512GB), each term is
  59          * 4**(1/5) times the previous term.  For 1TB to 16TB, each term is
  60          * 2**(1/5) times the previous term.
  61          */
  62         /*  16GB */      64,             32,             5,
  63         /*  32GB */      84,             49,             6,
  64         /*  64GB */     111,             74,             7,
  65         /* 128GB */     147,            111,             8,
  66         /* 256GB */     194,            169,             9,
  67         /* 512GB */     256,            256,            11,
  68         /*   1TB */     294,            294,            14,
  69         /*   2TB */     338,            338,            16,
  70         /*   4TB */     388,            388,            20,
  71         /*   8TB */     446,            446,            25,
  72         /*  16TB */     512,            512,            32
  73 };
  74
  75 u32 hfsplus_calc_btree_clump_size(u32 block_size, u32 node_size,
  76                                         u64 sectors, int file_id)
  77 {
  78         u32 mod = max(node_size, block_size);
  79         u32 clump_size;
  80         int column;
  81         int i;
  82
  83         /* Figure out which column of the above table to use for this file. */
  84         switch (file_id) {
  85         case HFSPLUS_ATTR_CNID:
  86                 column = 0;
  87                 break;
  88         case HFSPLUS_CAT_CNID:
  89                 column = 1;
  90                 break;
  91         default:
  92                 column = 2;
  93                 break;
  94         }
  95
  96         /*
  97          * The default clump size is 0.8% of the volume size. And
  98          * it must also be a multiple of the node and block size.
  99          */
 100         if (sectors < 0x200000) {
 101                 clump_size = sectors << 2;      /*  0.8 %  */
 102                 if (clump_size < (8 * node_size))
 103                         clump_size = 8 * node_size;
 104         } else {
 105                 /* turn exponent into table index... */
 106                 for (i = 0, sectors = sectors >> 22;
 107                      sectors && (i < CLUMP_ENTRIES - 1);
 108                      ++i, sectors = sectors >> 1) {
 109                         /* empty body */
 110                 }
 111
 112                 clump_size = clumptbl[column + (i) * 3] * 1024 * 1024;
 113         }
 114
 115         /*
 116          * Round the clump size to a multiple of node and block size.
 117          * NOTE: This rounds down.
 118          */
 119         clump_size /= mod;
 120         clump_size *= mod;
 121
 122         /*
 123          * Rounding down could have rounded down to 0 if the block size was
 124          * greater than the clump size.  If so, just use one block or node.
 125          */
 126         if (clump_size == 0)
 127                 clump_size = mod;
 128
 129         return clump_size;
 130 }
 131
 132 /* Get a reference to a B*Tree and do some initial checks */
 133 struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
 134 {
 135         struct hfs_btree *tree;
 136         struct hfs_btree_header_rec *head;
 137         struct address_space *mapping;
 138         struct inode *inode;
 139         struct page *page;
 140         unsigned int size;
 141
 142         tree = kzalloc(sizeof(*tree), GFP_KERNEL);
 143         if (!tree)
 144                 return NULL;
 145
 146         mutex_init(&tree->tree_lock);
 147         spin_lock_init(&tree->hash_lock);
 148         tree->sb = sb;
 149         tree->cnid = id;
 150         inode = hfsplus_iget(sb, id);
 151         if (IS_ERR(inode))
 152                 goto free_tree;
 153         tree->inode = inode;
 154
 155         if (!HFSPLUS_I(tree->inode)->first_blocks) {
 156                 pr_err("invalid btree extent records (0 size)\n");
 157                 goto free_inode;
 158         }
 159
 160         mapping = tree->inode->i_mapping;
 161         page = read_mapping_page(mapping, 0, NULL);
 162         if (IS_ERR(page))
 163                 goto free_inode;
 164
 165         /* Load the header */
 166         head = (struct hfs_btree_header_rec *)(kmap(page) +
 167                 sizeof(struct hfs_bnode_desc));
 168         tree->root = be32_to_cpu(head->root);
 169         tree->leaf_count = be32_to_cpu(head->leaf_count);
 170         tree->leaf_head = be32_to_cpu(head->leaf_head);
 171         tree->leaf_tail = be32_to_cpu(head->leaf_tail);
 172         tree->node_count = be32_to_cpu(head->node_count);
 173         tree->free_nodes = be32_to_cpu(head->free_nodes);
 174         tree->attributes = be32_to_cpu(head->attributes);
 175         tree->node_size = be16_to_cpu(head->node_size);
 176         tree->max_key_len = be16_to_cpu(head->max_key_len);
 177         tree->depth = be16_to_cpu(head->depth);
 178
 179         /* Verify the tree and set the correct compare function */
 180         switch (id) {
 181         case HFSPLUS_EXT_CNID:
 182                 if (tree->max_key_len != HFSPLUS_EXT_KEYLEN - sizeof(u16)) {
 183                         pr_err("invalid extent max_key_len %d\n",
 184                                 tree->max_key_len);
 185                         goto fail_page;
 186                 }
 187                 if (tree->attributes & HFS_TREE_VARIDXKEYS) {
 188                         pr_err("invalid extent btree flag\n");
 189                         goto fail_page;
 190                 }
 191
 192                 tree->keycmp = hfsplus_ext_cmp_key;
 193                 break;
 194         case HFSPLUS_CAT_CNID:
 195                 if (tree->max_key_len != HFSPLUS_CAT_KEYLEN - sizeof(u16)) {
 196                         pr_err("invalid catalog max_key_len %d\n",
 197                                 tree->max_key_len);
 198                         goto fail_page;
 199                 }
 200                 if (!(tree->attributes & HFS_TREE_VARIDXKEYS)) {
 201                         pr_err("invalid catalog btree flag\n");
 202                         goto fail_page;
 203                 }
 204
 205                 if (test_bit(HFSPLUS_SB_HFSX, &HFSPLUS_SB(sb)->flags) &&
 206                     (head->key_type == HFSPLUS_KEY_BINARY))
 207                         tree->keycmp = hfsplus_cat_bin_cmp_key;
 208                 else {
 209                         tree->keycmp = hfsplus_cat_case_cmp_key;
 210                         set_bit(HFSPLUS_SB_CASEFOLD, &HFSPLUS_SB(sb)->flags);
 211                 }
 212                 break;
 213         case HFSPLUS_ATTR_CNID:
 214                 if (tree->max_key_len != HFSPLUS_ATTR_KEYLEN - sizeof(u16)) {
 215                         pr_err("invalid attributes max_key_len %d\n",
 216                                 tree->max_key_len);
 217                         goto fail_page;
 218                 }
 219                 tree->keycmp = hfsplus_attr_bin_cmp_key;
 220                 break;
 221         default:
 222                 pr_err("unknown B*Tree requested\n");
 223                 goto fail_page;
 224         }
 225
 226         if (!(tree->attributes & HFS_TREE_BIGKEYS)) {
 227                 pr_err("invalid btree flag\n");
 228                 goto fail_page;
 229         }
 230
 231         size = tree->node_size;
 232         if (!is_power_of_2(size))
 233                 goto fail_page;
 234         if (!tree->node_count)
 235                 goto fail_page;
 236
 237         tree->node_size_shift = ffs(size) - 1;
 238
 239         tree->pages_per_bnode =
 240                 (tree->node_size + PAGE_SIZE - 1) >>
 241                 PAGE_SHIFT;
 242
 243         kunmap(page);
 244         put_page(page);
 245         return tree;
 246
 247  fail_page:
 248         put_page(page);
 249  free_inode:
 250         tree->inode->i_mapping->a_ops = &hfsplus_aops;
 251         iput(tree->inode);
 252  free_tree:
 253         kfree(tree);
 254         return NULL;
 255 }
 256
 257 /* Release resources used by a btree */
 258 void hfs_btree_close(struct hfs_btree *tree)
 259 {
 260         struct hfs_bnode *node;
 261         int i;
 262
 263         if (!tree)
 264                 return;
 265
 266         for (i = 0; i < NODE_HASH_SIZE; i++) {
 267                 while ((node = tree->node_hash[i])) {
 268                         tree->node_hash[i] = node->next_hash;
 269                         if (atomic_read(&node->refcnt))
 270                                 pr_crit("node %d:%d "
 271                                                 "still has %d user(s)!\n",
 272                                         node->tree->cnid, node->this,
 273                                         atomic_read(&node->refcnt));
 274                         hfs_bnode_free(node);
 275                         tree->node_hash_cnt--;
 276                 }
 277         }
 278         iput(tree->inode);
 279         kfree(tree);
 280 }
 281
 282 int hfs_btree_write(struct hfs_btree *tree)
 283 {
 284         struct hfs_btree_header_rec *head;
 285         struct hfs_bnode *node;
 286         struct page *page;
 287
 288         node = hfs_bnode_find(tree, 0);
 289         if (IS_ERR(node))
 290                 /* panic? */
 291                 return -EIO;
 292         /* Load the header */
 293         page = node->page[0];
 294         head = (struct hfs_btree_header_rec *)(kmap(page) +
 295                 sizeof(struct hfs_bnode_desc));
 296
 297         head->root = cpu_to_be32(tree->root);
 298         head->leaf_count = cpu_to_be32(tree->leaf_count);
 299         head->leaf_head = cpu_to_be32(tree->leaf_head);
 300         head->leaf_tail = cpu_to_be32(tree->leaf_tail);
 301         head->node_count = cpu_to_be32(tree->node_count);
 302         head->free_nodes = cpu_to_be32(tree->free_nodes);
 303         head->attributes = cpu_to_be32(tree->attributes);
 304         head->depth = cpu_to_be16(tree->depth);
 305
 306         kunmap(page);
 307         set_page_dirty(page);
 308         hfs_bnode_put(node);
 309         return 0;
 310 }
 311
 312 static struct hfs_bnode *hfs_bmap_new_bmap(struct hfs_bnode *prev, u32 idx)
 313 {
 314         struct hfs_btree *tree = prev->tree;
 315         struct hfs_bnode *node;
 316         struct hfs_bnode_desc desc;
 317         __be32 cnid;
 318
 319         node = hfs_bnode_create(tree, idx);
 320         if (IS_ERR(node))
 321                 return node;
 322
 323         tree->free_nodes--;
 324         prev->next = idx;
 325         cnid = cpu_to_be32(idx);
 326         hfs_bnode_write(prev, &cnid, offsetof(struct hfs_bnode_desc, next), 4);
 327
 328         node->type = HFS_NODE_MAP;
 329         node->num_recs = 1;
 330         hfs_bnode_clear(node, 0, tree->node_size);
 331         desc.next = 0;
 332         desc.prev = 0;
 333         desc.type = HFS_NODE_MAP;
 334         desc.height = 0;
 335         desc.num_recs = cpu_to_be16(1);
 336         desc.reserved = 0;
 337         hfs_bnode_write(node, &desc, 0, sizeof(desc));
 338         hfs_bnode_write_u16(node, 14, 0x8000);
 339         hfs_bnode_write_u16(node, tree->node_size - 2, 14);
 340         hfs_bnode_write_u16(node, tree->node_size - 4, tree->node_size - 6);
 341
 342         return node;
 343 }
 344
 345 /* Make sure @tree has enough space for the @rsvd_nodes */
 346 int hfs_bmap_reserve(struct hfs_btree *tree, int rsvd_nodes)
 347 {
 348         struct inode *inode = tree->inode;
 349         struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
 350         u32 count;
 351         int res;
 352
 353         if (rsvd_nodes <= 0)
 354                 return 0;
 355
 356         while (tree->free_nodes < rsvd_nodes) {
 357                 res = hfsplus_file_extend(inode, hfs_bnode_need_zeroout(tree));
 358                 if (res)
 359                         return res;
 360                 hip->phys_size = inode->i_size =
 361                         (loff_t)hip->alloc_blocks <<
 362                                 HFSPLUS_SB(tree->sb)->alloc_blksz_shift;
 363                 hip->fs_blocks =
 364                         hip->alloc_blocks << HFSPLUS_SB(tree->sb)->fs_shift;
 365                 inode_set_bytes(inode, inode->i_size);
 366                 count = inode->i_size >> tree->node_size_shift;
 367                 tree->free_nodes += count - tree->node_count;
 368                 tree->node_count = count;
 369         }
 370         return 0;
 371 }
 372
 373 struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
 374 {
 375         struct hfs_bnode *node, *next_node;
 376         struct page **pagep;
 377         u32 nidx, idx;
 378         unsigned off;
 379         u16 off16;
 380         u16 len;
 381         u8 *data, byte, m;
 382         int i, res;
 383
 384         res = hfs_bmap_reserve(tree, 1);
 385         if (res)
 386                 return ERR_PTR(res);
 387
 388         nidx = 0;
 389         node = hfs_bnode_find(tree, nidx);
 390         if (IS_ERR(node))
 391                 return node;
 392         len = hfs_brec_lenoff(node, 2, &off16);
 393         off = off16;
 394
 395         off += node->page_offset;
 396         pagep = node->page + (off >> PAGE_SHIFT);
 397         data = kmap(*pagep);
 398         off &= ~PAGE_MASK;
 399         idx = 0;
 400
 401         for (;;) {
 402                 while (len) {
 403                         byte = data[off];
 404                         if (byte != 0xff) {
 405                                 for (m = 0x80, i = 0; i < 8; m >>= 1, i++) {
 406                                         if (!(byte & m)) {
 407                                                 idx += i;
 408                                                 data[off] |= m;
 409                                                 set_page_dirty(*pagep);
 410                                                 kunmap(*pagep);
 411                                                 tree->free_nodes--;
 412                                                 mark_inode_dirty(tree->inode);
 413                                                 hfs_bnode_put(node);
 414                                                 return hfs_bnode_create(tree,
 415                                                         idx);
 416                                         }
 417                                 }
 418                         }
 419                         if (++off >= PAGE_SIZE) {
 420                                 kunmap(*pagep);
 421                                 data = kmap(*++pagep);
 422                                 off = 0;
 423                         }
 424                         idx += 8;
 425                         len--;
 426                 }
 427                 kunmap(*pagep);
 428                 nidx = node->next;
 429                 if (!nidx) {
 430                         hfs_dbg(BNODE_MOD, "create new bmap node\n");
 431                         next_node = hfs_bmap_new_bmap(node, idx);
 432                 } else
 433                         next_node = hfs_bnode_find(tree, nidx);
 434                 hfs_bnode_put(node);
 435                 if (IS_ERR(next_node))
 436                         return next_node;
 437                 node = next_node;
 438
 439                 len = hfs_brec_lenoff(node, 0, &off16);
 440                 off = off16;
 441                 off += node->page_offset;
 442                 pagep = node->page + (off >> PAGE_SHIFT);
 443                 data = kmap(*pagep);
 444                 off &= ~PAGE_MASK;
 445         }
 446 }
 447
 448 void hfs_bmap_free(struct hfs_bnode *node)
 449 {
 450         struct hfs_btree *tree;
 451         struct page *page;
 452         u16 off, len;
 453         u32 nidx;
 454         u8 *data, byte, m;
 455
 456         hfs_dbg(BNODE_MOD, "btree_free_node: %u\n", node->this);
 457         BUG_ON(!node->this);
 458         tree = node->tree;
 459         nidx = node->this;
 460         node = hfs_bnode_find(tree, 0);
 461         if (IS_ERR(node))
 462                 return;
 463         len = hfs_brec_lenoff(node, 2, &off);
 464         while (nidx >= len * 8) {
 465                 u32 i;
 466
 467                 nidx -= len * 8;
 468                 i = node->next;
 469                 if (!i) {
 470                         /* panic */;
 471                         pr_crit("unable to free bnode %u. "
 472                                         "bmap not found!\n",
 473                                 node->this);
 474                         hfs_bnode_put(node);
 475                         return;
 476                 }
 477                 hfs_bnode_put(node);
 478                 node = hfs_bnode_find(tree, i);
 479                 if (IS_ERR(node))
 480                         return;
 481                 if (node->type != HFS_NODE_MAP) {
 482                         /* panic */;
 483                         pr_crit("invalid bmap found! "
 484                                         "(%u,%d)\n",
 485                                 node->this, node->type);
 486                         hfs_bnode_put(node);
 487                         return;
 488                 }
 489                 len = hfs_brec_lenoff(node, 0, &off);
 490         }
 491         off += node->page_offset + nidx / 8;
 492         page = node->page[off >> PAGE_SHIFT];
 493         data = kmap(page);
 494         off &= ~PAGE_MASK;
 495         m = 1 << (~nidx & 7);
 496         byte = data[off];
 497         if (!(byte & m)) {
 498                 pr_crit("trying to free free bnode "
 499                                 "%u(%d)\n",
 500                         node->this, node->type);
 501                 kunmap(page);
 502                 hfs_bnode_put(node);
 503                 return;
 504         }
 505         data[off] = byte & ~m;
 506         set_page_dirty(page);
 507         kunmap(page);
 508         hfs_bnode_put(node);
 509         tree->free_nodes++;
 510         mark_inode_dirty(tree->inode);
 511 }