Squashfs: directory lookup operations
authorPhillip Lougher <phillip@lougher.demon.co.uk>
Mon, 5 Jan 2009 08:46:23 +0000 (08:46 +0000)
committerPhillip Lougher <phillip@lougher.demon.co.uk>
Mon, 5 Jan 2009 08:46:23 +0000 (08:46 +0000)
Signed-off-by: Phillip Lougher <phillip@lougher.demon.co.uk>
fs/squashfs/namei.c [new file with mode: 0644]

diff --git a/fs/squashfs/namei.c b/fs/squashfs/namei.c
new file mode 100644 (file)
index 0000000..9e39865
--- /dev/null
@@ -0,0 +1,242 @@
+/*
+ * Squashfs - a compressed read only filesystem for Linux
+ *
+ * Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008
+ * Phillip Lougher <phillip@lougher.demon.co.uk>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2,
+ * or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * namei.c
+ */
+
+/*
+ * This file implements code to do filename lookup in directories.
+ *
+ * Like inodes, directories are packed into compressed metadata blocks, stored
+ * in a directory table.  Directories are accessed using the start address of
+ * the metablock containing the directory and the offset into the
+ * decompressed block (<block, offset>).
+ *
+ * Directories are organised in a slightly complex way, and are not simply
+ * a list of file names.  The organisation takes advantage of the
+ * fact that (in most cases) the inodes of the files will be in the same
+ * compressed metadata block, and therefore, can share the start block.
+ * Directories are therefore organised in a two level list, a directory
+ * header containing the shared start block value, and a sequence of directory
+ * entries, each of which share the shared start block.  A new directory header
+ * is written once/if the inode start block changes.  The directory
+ * header/directory entry list is repeated as many times as necessary.
+ *
+ * Directories are sorted, and can contain a directory index to speed up
+ * file lookup.  Directory indexes store one entry per metablock, each entry
+ * storing the index/filename mapping to the first directory header
+ * in each metadata block.  Directories are sorted in alphabetical order,
+ * and at lookup the index is scanned linearly looking for the first filename
+ * alphabetically larger than the filename being looked up.  At this point the
+ * location of the metadata block the filename is in has been found.
+ * The general idea of the index is ensure only one metadata block needs to be
+ * decompressed to do a lookup irrespective of the length of the directory.
+ * This scheme has the advantage that it doesn't require extra memory overhead
+ * and doesn't require much extra storage on disk.
+ */
+
+#include <linux/fs.h>
+#include <linux/vfs.h>
+#include <linux/slab.h>
+#include <linux/string.h>
+#include <linux/dcache.h>
+#include <linux/zlib.h>
+
+#include "squashfs_fs.h"
+#include "squashfs_fs_sb.h"
+#include "squashfs_fs_i.h"
+#include "squashfs.h"
+
+/*
+ * Lookup name in the directory index, returning the location of the metadata
+ * block containing it, and the directory index this represents.
+ *
+ * If we get an error reading the index then return the part of the index
+ * (if any) we have managed to read - the index isn't essential, just
+ * quicker.
+ */
+static int get_dir_index_using_name(struct super_block *sb,
+                       u64 *next_block, int *next_offset, u64 index_start,
+                       int index_offset, int i_count, const char *name,
+                       int len)
+{
+       struct squashfs_sb_info *msblk = sb->s_fs_info;
+       int i, size, length = 0, err;
+       struct squashfs_dir_index *index;
+       char *str;
+
+       TRACE("Entered get_dir_index_using_name, i_count %d\n", i_count);
+
+       index = kmalloc(sizeof(*index) + SQUASHFS_NAME_LEN * 2 + 2, GFP_KERNEL);
+       if (index == NULL) {
+               ERROR("Failed to allocate squashfs_dir_index\n");
+               goto out;
+       }
+
+       str = &index->name[SQUASHFS_NAME_LEN + 1];
+       strncpy(str, name, len);
+       str[len] = '\0';
+
+       for (i = 0; i < i_count; i++) {
+               err = squashfs_read_metadata(sb, index, &index_start,
+                                       &index_offset, sizeof(*index));
+               if (err < 0)
+                       break;
+
+
+               size = le32_to_cpu(index->size) + 1;
+
+               err = squashfs_read_metadata(sb, index->name, &index_start,
+                                       &index_offset, size);
+               if (err < 0)
+                       break;
+
+               index->name[size] = '\0';
+
+               if (strcmp(index->name, str) > 0)
+                       break;
+
+               length = le32_to_cpu(index->index);
+               *next_block = le32_to_cpu(index->start_block) +
+                                       msblk->directory_table;
+       }
+
+       *next_offset = (length + *next_offset) % SQUASHFS_METADATA_SIZE;
+       kfree(index);
+
+out:
+       /*
+        * Return index (f_pos) of the looked up metadata block.  Translate
+        * from internal f_pos to external f_pos which is offset by 3 because
+        * we invent "." and ".." entries which are not actually stored in the
+        * directory.
+        */
+       return length + 3;
+}
+
+
+static struct dentry *squashfs_lookup(struct inode *dir, struct dentry *dentry,
+                                struct nameidata *nd)
+{
+       const unsigned char *name = dentry->d_name.name;
+       int len = dentry->d_name.len;
+       struct inode *inode = NULL;
+       struct squashfs_sb_info *msblk = dir->i_sb->s_fs_info;
+       struct squashfs_dir_header dirh;
+       struct squashfs_dir_entry *dire;
+       u64 block = squashfs_i(dir)->start + msblk->directory_table;
+       int offset = squashfs_i(dir)->offset;
+       int err, length = 0, dir_count, size;
+
+       TRACE("Entered squashfs_lookup [%llx:%x]\n", block, offset);
+
+       dire = kmalloc(sizeof(*dire) + SQUASHFS_NAME_LEN + 1, GFP_KERNEL);
+       if (dire == NULL) {
+               ERROR("Failed to allocate squashfs_dir_entry\n");
+               return ERR_PTR(-ENOMEM);
+       }
+
+       if (len > SQUASHFS_NAME_LEN) {
+               err = -ENAMETOOLONG;
+               goto failed;
+       }
+
+       length = get_dir_index_using_name(dir->i_sb, &block, &offset,
+                               squashfs_i(dir)->dir_idx_start,
+                               squashfs_i(dir)->dir_idx_offset,
+                               squashfs_i(dir)->dir_idx_cnt, name, len);
+
+       while (length < i_size_read(dir)) {
+               /*
+                * Read directory header.
+                */
+               err = squashfs_read_metadata(dir->i_sb, &dirh, &block,
+                               &offset, sizeof(dirh));
+               if (err < 0)
+                       goto read_failure;
+
+               length += sizeof(dirh);
+
+               dir_count = le32_to_cpu(dirh.count) + 1;
+               while (dir_count--) {
+                       /*
+                        * Read directory entry.
+                        */
+                       err = squashfs_read_metadata(dir->i_sb, dire, &block,
+                                       &offset, sizeof(*dire));
+                       if (err < 0)
+                               goto read_failure;
+
+                       size = le16_to_cpu(dire->size) + 1;
+
+                       err = squashfs_read_metadata(dir->i_sb, dire->name,
+                                       &block, &offset, size);
+                       if (err < 0)
+                               goto read_failure;
+
+                       length += sizeof(*dire) + size;
+
+                       if (name[0] < dire->name[0])
+                               goto exit_lookup;
+
+                       if (len == size && !strncmp(name, dire->name, len)) {
+                               unsigned int blk, off, ino_num;
+                               long long ino;
+                               blk = le32_to_cpu(dirh.start_block);
+                               off = le16_to_cpu(dire->offset);
+                               ino_num = le32_to_cpu(dirh.inode_number) +
+                                       (short) le16_to_cpu(dire->inode_number);
+                               ino = SQUASHFS_MKINODE(blk, off);
+
+                               TRACE("calling squashfs_iget for directory "
+                                       "entry %s, inode  %x:%x, %d\n", name,
+                                       blk, off, ino_num);
+
+                               inode = squashfs_iget(dir->i_sb, ino, ino_num);
+                               if (IS_ERR(inode)) {
+                                       err = PTR_ERR(inode);
+                                       goto failed;
+                               }
+
+                               goto exit_lookup;
+                       }
+               }
+       }
+
+exit_lookup:
+       kfree(dire);
+       if (inode)
+               return d_splice_alias(inode, dentry);
+       d_add(dentry, inode);
+       return ERR_PTR(0);
+
+read_failure:
+       ERROR("Unable to read directory block [%llx:%x]\n",
+               squashfs_i(dir)->start + msblk->directory_table,
+               squashfs_i(dir)->offset);
+failed:
+       kfree(dire);
+       return ERR_PTR(err);
+}
+
+
+const struct inode_operations squashfs_dir_inode_ops = {
+       .lookup = squashfs_lookup
+};