fs/hfsplus/btree.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *  linux/fs/hfsplus/btree.c
0004  *
0005  * Copyright (C) 2001
0006  * Brad Boyer (flar@allandria.com)
0007  * (C) 2003 Ardis Technologies <roman@ardistech.com>
0008  *
0009  * Handle opening/closing btree
0010  */
0011
0012 #include <linux/slab.h>
0013 #include <linux/pagemap.h>
0014 #include <linux/log2.h>
0015
0016 #include "hfsplus_fs.h"
0017 #include "hfsplus_raw.h"
0018
0019 /*
0020  * Initial source code of clump size calculation is gotten
0021  * from http://opensource.apple.com/tarballs/diskdev_cmds/
0022  */
0023 #define CLUMP_ENTRIES   15
0024
0025 static short clumptbl[CLUMP_ENTRIES * 3] = {
0026 /*
0027  *      Volume  Attributes   Catalog     Extents
0028  *       Size   Clump (MB)  Clump (MB)  Clump (MB)
0029  */
0030     /*   1GB */   4,          4,         4,
0031     /*   2GB */   6,          6,         4,
0032     /*   4GB */   8,          8,         4,
0033     /*   8GB */  11,         11,         5,
0034     /*
0035      * For volumes 16GB and larger, we want to make sure that a full OS
0036      * install won't require fragmentation of the Catalog or Attributes
0037      * B-trees.  We do this by making the clump sizes sufficiently large,
0038      * and by leaving a gap after the B-trees for them to grow into.
0039      *
0040      * For SnowLeopard 10A298, a FullNetInstall with all packages selected
0041      * results in:
0042      * Catalog B-tree Header
0043      *  nodeSize:          8192
0044      *  totalNodes:       31616
0045      *  freeNodes:         1978
0046      * (used = 231.55 MB)
0047      * Attributes B-tree Header
0048      *  nodeSize:          8192
0049      *  totalNodes:       63232
0050      *  freeNodes:          958
0051      * (used = 486.52 MB)
0052      *
0053      * We also want Time Machine backup volumes to have a sufficiently
0054      * large clump size to reduce fragmentation.
0055      *
0056      * The series of numbers for Catalog and Attribute form a geometric
0057      * series. For Catalog (16GB to 512GB), each term is 8**(1/5) times
0058      * the previous term.  For Attributes (16GB to 512GB), each term is
0059      * 4**(1/5) times the previous term.  For 1TB to 16TB, each term is
0060      * 2**(1/5) times the previous term.
0061      */
0062     /*  16GB */  64,         32,         5,
0063     /*  32GB */  84,         49,         6,
0064     /*  64GB */ 111,         74,         7,
0065     /* 128GB */ 147,        111,         8,
0066     /* 256GB */ 194,        169,         9,
0067     /* 512GB */ 256,        256,        11,
0068     /*   1TB */ 294,        294,        14,
0069     /*   2TB */ 338,        338,        16,
0070     /*   4TB */ 388,        388,        20,
0071     /*   8TB */ 446,        446,        25,
0072     /*  16TB */ 512,        512,        32
0073 };
0074
0075 u32 hfsplus_calc_btree_clump_size(u32 block_size, u32 node_size,
0076                     u64 sectors, int file_id)
0077 {
0078     u32 mod = max(node_size, block_size);
0079     u32 clump_size;
0080     int column;
0081     int i;
0082
0083     /* Figure out which column of the above table to use for this file. */
0084     switch (file_id) {
0085     case HFSPLUS_ATTR_CNID:
0086         column = 0;
0087         break;
0088     case HFSPLUS_CAT_CNID:
0089         column = 1;
0090         break;
0091     default:
0092         column = 2;
0093         break;
0094     }
0095
0096     /*
0097      * The default clump size is 0.8% of the volume size. And
0098      * it must also be a multiple of the node and block size.
0099      */
0100     if (sectors < 0x200000) {
0101         clump_size = sectors << 2;  /*  0.8 %  */
0102         if (clump_size < (8 * node_size))
0103             clump_size = 8 * node_size;
0104     } else {
0105         /* turn exponent into table index... */
0106         for (i = 0, sectors = sectors >> 22;
0107              sectors && (i < CLUMP_ENTRIES - 1);
0108              ++i, sectors = sectors >> 1) {
0109             /* empty body */
0110         }
0111
0112         clump_size = clumptbl[column + (i) * 3] * 1024 * 1024;
0113     }
0114
0115     /*
0116      * Round the clump size to a multiple of node and block size.
0117      * NOTE: This rounds down.
0118      */
0119     clump_size /= mod;
0120     clump_size *= mod;
0121
0122     /*
0123      * Rounding down could have rounded down to 0 if the block size was
0124      * greater than the clump size.  If so, just use one block or node.
0125      */
0126     if (clump_size == 0)
0127         clump_size = mod;
0128
0129     return clump_size;
0130 }
0131
0132 /* Get a reference to a B*Tree and do some initial checks */
0133 struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
0134 {
0135     struct hfs_btree *tree;
0136     struct hfs_btree_header_rec *head;
0137     struct address_space *mapping;
0138     struct inode *inode;
0139     struct page *page;
0140     unsigned int size;
0141
0142     tree = kzalloc(sizeof(*tree), GFP_KERNEL);
0143     if (!tree)
0144         return NULL;
0145
0146     mutex_init(&tree->tree_lock);
0147     spin_lock_init(&tree->hash_lock);
0148     tree->sb = sb;
0149     tree->cnid = id;
0150     inode = hfsplus_iget(sb, id);
0151     if (IS_ERR(inode))
0152         goto free_tree;
0153     tree->inode = inode;
0154
0155     if (!HFSPLUS_I(tree->inode)->first_blocks) {
0156         pr_err("invalid btree extent records (0 size)\n");
0157         goto free_inode;
0158     }
0159
0160     mapping = tree->inode->i_mapping;
0161     page = read_mapping_page(mapping, 0, NULL);
0162     if (IS_ERR(page))
0163         goto free_inode;
0164
0165     /* Load the header */
0166     head = (struct hfs_btree_header_rec *)(kmap(page) +
0167         sizeof(struct hfs_bnode_desc));
0168     tree->root = be32_to_cpu(head->root);
0169     tree->leaf_count = be32_to_cpu(head->leaf_count);
0170     tree->leaf_head = be32_to_cpu(head->leaf_head);
0171     tree->leaf_tail = be32_to_cpu(head->leaf_tail);
0172     tree->node_count = be32_to_cpu(head->node_count);
0173     tree->free_nodes = be32_to_cpu(head->free_nodes);
0174     tree->attributes = be32_to_cpu(head->attributes);
0175     tree->node_size = be16_to_cpu(head->node_size);
0176     tree->max_key_len = be16_to_cpu(head->max_key_len);
0177     tree->depth = be16_to_cpu(head->depth);
0178
0179     /* Verify the tree and set the correct compare function */
0180     switch (id) {
0181     case HFSPLUS_EXT_CNID:
0182         if (tree->max_key_len != HFSPLUS_EXT_KEYLEN - sizeof(u16)) {
0183             pr_err("invalid extent max_key_len %d\n",
0184                 tree->max_key_len);
0185             goto fail_page;
0186         }
0187         if (tree->attributes & HFS_TREE_VARIDXKEYS) {
0188             pr_err("invalid extent btree flag\n");
0189             goto fail_page;
0190         }
0191
0192         tree->keycmp = hfsplus_ext_cmp_key;
0193         break;
0194     case HFSPLUS_CAT_CNID:
0195         if (tree->max_key_len != HFSPLUS_CAT_KEYLEN - sizeof(u16)) {
0196             pr_err("invalid catalog max_key_len %d\n",
0197                 tree->max_key_len);
0198             goto fail_page;
0199         }
0200         if (!(tree->attributes & HFS_TREE_VARIDXKEYS)) {
0201             pr_err("invalid catalog btree flag\n");
0202             goto fail_page;
0203         }
0204
0205         if (test_bit(HFSPLUS_SB_HFSX, &HFSPLUS_SB(sb)->flags) &&
0206             (head->key_type == HFSPLUS_KEY_BINARY))
0207             tree->keycmp = hfsplus_cat_bin_cmp_key;
0208         else {
0209             tree->keycmp = hfsplus_cat_case_cmp_key;
0210             set_bit(HFSPLUS_SB_CASEFOLD, &HFSPLUS_SB(sb)->flags);
0211         }
0212         break;
0213     case HFSPLUS_ATTR_CNID:
0214         if (tree->max_key_len != HFSPLUS_ATTR_KEYLEN - sizeof(u16)) {
0215             pr_err("invalid attributes max_key_len %d\n",
0216                 tree->max_key_len);
0217             goto fail_page;
0218         }
0219         tree->keycmp = hfsplus_attr_bin_cmp_key;
0220         break;
0221     default:
0222         pr_err("unknown B*Tree requested\n");
0223         goto fail_page;
0224     }
0225
0226     if (!(tree->attributes & HFS_TREE_BIGKEYS)) {
0227         pr_err("invalid btree flag\n");
0228         goto fail_page;
0229     }
0230
0231     size = tree->node_size;
0232     if (!is_power_of_2(size))
0233         goto fail_page;
0234     if (!tree->node_count)
0235         goto fail_page;
0236
0237     tree->node_size_shift = ffs(size) - 1;
0238
0239     tree->pages_per_bnode =
0240         (tree->node_size + PAGE_SIZE - 1) >>
0241         PAGE_SHIFT;
0242
0243     kunmap(page);
0244     put_page(page);
0245     return tree;
0246
0247  fail_page:
0248     put_page(page);
0249  free_inode:
0250     tree->inode->i_mapping->a_ops = &hfsplus_aops;
0251     iput(tree->inode);
0252  free_tree:
0253     kfree(tree);
0254     return NULL;
0255 }
0256
0257 /* Release resources used by a btree */
0258 void hfs_btree_close(struct hfs_btree *tree)
0259 {
0260     struct hfs_bnode *node;
0261     int i;
0262
0263     if (!tree)
0264         return;
0265
0266     for (i = 0; i < NODE_HASH_SIZE; i++) {
0267         while ((node = tree->node_hash[i])) {
0268             tree->node_hash[i] = node->next_hash;
0269             if (atomic_read(&node->refcnt))
0270                 pr_crit("node %d:%d "
0271                         "still has %d user(s)!\n",
0272                     node->tree->cnid, node->this,
0273                     atomic_read(&node->refcnt));
0274             hfs_bnode_free(node);
0275             tree->node_hash_cnt--;
0276         }
0277     }
0278     iput(tree->inode);
0279     kfree(tree);
0280 }
0281
0282 int hfs_btree_write(struct hfs_btree *tree)
0283 {
0284     struct hfs_btree_header_rec *head;
0285     struct hfs_bnode *node;
0286     struct page *page;
0287
0288     node = hfs_bnode_find(tree, 0);
0289     if (IS_ERR(node))
0290         /* panic? */
0291         return -EIO;
0292     /* Load the header */
0293     page = node->page[0];
0294     head = (struct hfs_btree_header_rec *)(kmap(page) +
0295         sizeof(struct hfs_bnode_desc));
0296
0297     head->root = cpu_to_be32(tree->root);
0298     head->leaf_count = cpu_to_be32(tree->leaf_count);
0299     head->leaf_head = cpu_to_be32(tree->leaf_head);
0300     head->leaf_tail = cpu_to_be32(tree->leaf_tail);
0301     head->node_count = cpu_to_be32(tree->node_count);
0302     head->free_nodes = cpu_to_be32(tree->free_nodes);
0303     head->attributes = cpu_to_be32(tree->attributes);
0304     head->depth = cpu_to_be16(tree->depth);
0305
0306     kunmap(page);
0307     set_page_dirty(page);
0308     hfs_bnode_put(node);
0309     return 0;
0310 }
0311
0312 static struct hfs_bnode *hfs_bmap_new_bmap(struct hfs_bnode *prev, u32 idx)
0313 {
0314     struct hfs_btree *tree = prev->tree;
0315     struct hfs_bnode *node;
0316     struct hfs_bnode_desc desc;
0317     __be32 cnid;
0318
0319     node = hfs_bnode_create(tree, idx);
0320     if (IS_ERR(node))
0321         return node;
0322
0323     tree->free_nodes--;
0324     prev->next = idx;
0325     cnid = cpu_to_be32(idx);
0326     hfs_bnode_write(prev, &cnid, offsetof(struct hfs_bnode_desc, next), 4);
0327
0328     node->type = HFS_NODE_MAP;
0329     node->num_recs = 1;
0330     hfs_bnode_clear(node, 0, tree->node_size);
0331     desc.next = 0;
0332     desc.prev = 0;
0333     desc.type = HFS_NODE_MAP;
0334     desc.height = 0;
0335     desc.num_recs = cpu_to_be16(1);
0336     desc.reserved = 0;
0337     hfs_bnode_write(node, &desc, 0, sizeof(desc));
0338     hfs_bnode_write_u16(node, 14, 0x8000);
0339     hfs_bnode_write_u16(node, tree->node_size - 2, 14);
0340     hfs_bnode_write_u16(node, tree->node_size - 4, tree->node_size - 6);
0341
0342     return node;
0343 }
0344
0345 /* Make sure @tree has enough space for the @rsvd_nodes */
0346 int hfs_bmap_reserve(struct hfs_btree *tree, int rsvd_nodes)
0347 {
0348     struct inode *inode = tree->inode;
0349     struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
0350     u32 count;
0351     int res;
0352
0353     if (rsvd_nodes <= 0)
0354         return 0;
0355
0356     while (tree->free_nodes < rsvd_nodes) {
0357         res = hfsplus_file_extend(inode, hfs_bnode_need_zeroout(tree));
0358         if (res)
0359             return res;
0360         hip->phys_size = inode->i_size =
0361             (loff_t)hip->alloc_blocks <<
0362                 HFSPLUS_SB(tree->sb)->alloc_blksz_shift;
0363         hip->fs_blocks =
0364             hip->alloc_blocks << HFSPLUS_SB(tree->sb)->fs_shift;
0365         inode_set_bytes(inode, inode->i_size);
0366         count = inode->i_size >> tree->node_size_shift;
0367         tree->free_nodes += count - tree->node_count;
0368         tree->node_count = count;
0369     }
0370     return 0;
0371 }
0372
0373 struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
0374 {
0375     struct hfs_bnode *node, *next_node;
0376     struct page **pagep;
0377     u32 nidx, idx;
0378     unsigned off;
0379     u16 off16;
0380     u16 len;
0381     u8 *data, byte, m;
0382     int i, res;
0383
0384     res = hfs_bmap_reserve(tree, 1);
0385     if (res)
0386         return ERR_PTR(res);
0387
0388     nidx = 0;
0389     node = hfs_bnode_find(tree, nidx);
0390     if (IS_ERR(node))
0391         return node;
0392     len = hfs_brec_lenoff(node, 2, &off16);
0393     off = off16;
0394
0395     off += node->page_offset;
0396     pagep = node->page + (off >> PAGE_SHIFT);
0397     data = kmap(*pagep);
0398     off &= ~PAGE_MASK;
0399     idx = 0;
0400
0401     for (;;) {
0402         while (len) {
0403             byte = data[off];
0404             if (byte != 0xff) {
0405                 for (m = 0x80, i = 0; i < 8; m >>= 1, i++) {
0406                     if (!(byte & m)) {
0407                         idx += i;
0408                         data[off] |= m;
0409                         set_page_dirty(*pagep);
0410                         kunmap(*pagep);
0411                         tree->free_nodes--;
0412                         mark_inode_dirty(tree->inode);
0413                         hfs_bnode_put(node);
0414                         return hfs_bnode_create(tree,
0415                             idx);
0416                     }
0417                 }
0418             }
0419             if (++off >= PAGE_SIZE) {
0420                 kunmap(*pagep);
0421                 data = kmap(*++pagep);
0422                 off = 0;
0423             }
0424             idx += 8;
0425             len--;
0426         }
0427         kunmap(*pagep);
0428         nidx = node->next;
0429         if (!nidx) {
0430             hfs_dbg(BNODE_MOD, "create new bmap node\n");
0431             next_node = hfs_bmap_new_bmap(node, idx);
0432         } else
0433             next_node = hfs_bnode_find(tree, nidx);
0434         hfs_bnode_put(node);
0435         if (IS_ERR(next_node))
0436             return next_node;
0437         node = next_node;
0438
0439         len = hfs_brec_lenoff(node, 0, &off16);
0440         off = off16;
0441         off += node->page_offset;
0442         pagep = node->page + (off >> PAGE_SHIFT);
0443         data = kmap(*pagep);
0444         off &= ~PAGE_MASK;
0445     }
0446 }
0447
0448 void hfs_bmap_free(struct hfs_bnode *node)
0449 {
0450     struct hfs_btree *tree;
0451     struct page *page;
0452     u16 off, len;
0453     u32 nidx;
0454     u8 *data, byte, m;
0455
0456     hfs_dbg(BNODE_MOD, "btree_free_node: %u\n", node->this);
0457     BUG_ON(!node->this);
0458     tree = node->tree;
0459     nidx = node->this;
0460     node = hfs_bnode_find(tree, 0);
0461     if (IS_ERR(node))
0462         return;
0463     len = hfs_brec_lenoff(node, 2, &off);
0464     while (nidx >= len * 8) {
0465         u32 i;
0466
0467         nidx -= len * 8;
0468         i = node->next;
0469         if (!i) {
0470             /* panic */;
0471             pr_crit("unable to free bnode %u. "
0472                     "bmap not found!\n",
0473                 node->this);
0474             hfs_bnode_put(node);
0475             return;
0476         }
0477         hfs_bnode_put(node);
0478         node = hfs_bnode_find(tree, i);
0479         if (IS_ERR(node))
0480             return;
0481         if (node->type != HFS_NODE_MAP) {
0482             /* panic */;
0483             pr_crit("invalid bmap found! "
0484                     "(%u,%d)\n",
0485                 node->this, node->type);
0486             hfs_bnode_put(node);
0487             return;
0488         }
0489         len = hfs_brec_lenoff(node, 0, &off);
0490     }
0491     off += node->page_offset + nidx / 8;
0492     page = node->page[off >> PAGE_SHIFT];
0493     data = kmap(page);
0494     off &= ~PAGE_MASK;
0495     m = 1 << (~nidx & 7);
0496     byte = data[off];
0497     if (!(byte & m)) {
0498         pr_crit("trying to free free bnode "
0499                 "%u(%d)\n",
0500             node->this, node->type);
0501         kunmap(page);
0502         hfs_bnode_put(node);
0503         return;
0504     }
0505     data[off] = byte & ~m;
0506     set_page_dirty(page);
0507     kunmap(page);
0508     hfs_bnode_put(node);
0509     tree->free_nodes++;
0510     mark_inode_dirty(tree->inode);
0511 }