[RFC][PATCH 2/7] omfs: inode and superblock routines

[Bob Copeland <me@bobcopeland.com>]

This patch contains routines for reading and writing OMFS inodes
and the filesystem superblock.

---

 fs/omfs/inode.c |  453 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 453 insertions(+), 0 deletions(-)
 create mode 100644 fs/omfs/inode.c

5a57cd0c9df8444bb16354bddf0142744f672b08
diff --git a/fs/omfs/inode.c b/fs/omfs/inode.c
new file mode 100644
index 0000000..ac0febb
--- /dev/null
+++ b/fs/omfs/inode.c
@@ -0,0 +1,453 @@
+/*
+ * Optimized MPEG FS - inode and super operations.
+ * Copyright (C) 2005 Bob Copeland <me@bobcopeland.com>
+ */
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/vfs.h>
+#include <linux/buffer_head.h>
+#include <linux/vmalloc.h>
+#include "omfs.h"
+
+MODULE_AUTHOR("Bob Copeland <me@bobcopeland.com>");
+MODULE_DESCRIPTION("OMFS (ReplayTV/Karma) Filesystem for Linux");
+MODULE_LICENSE("GPL");
+
+static kmem_cache_t *omfs_inode_cachep;
+
+static struct inode *omfs_alloc_inode(struct super_block *sb)
+{
+	struct omfs_inode_info *oinf;
+	oinf = (struct omfs_inode_info *) kmem_cache_alloc(omfs_inode_cachep, 
+			SLAB_KERNEL);
+	if (!oinf)
+		return NULL;
+	return &oinf->vfs_inode;
+}
+
+static void omfs_destroy_inode(struct inode *inode)
+{
+	kmem_cache_free(omfs_inode_cachep, OMFS_I(inode));
+}
+
+static void init_once(void *p, kmem_cache_t *cachep, unsigned long flags)
+{
+	struct omfs_inode_info *oinf = (struct omfs_inode_info *) p;
+
+	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
+	    SLAB_CTOR_CONSTRUCTOR)
+		inode_init_once(&oinf->vfs_inode);
+}
+ 
+static int init_inodecache(void)
+{
+	omfs_inode_cachep = kmem_cache_create("omfs_inode_cache",
+					     sizeof(struct omfs_inode_info),
+					     0, SLAB_RECLAIM_ACCOUNT,
+					     init_once, NULL);
+	if (!omfs_inode_cachep)
+		return -ENOMEM;
+	return 0;
+}
+
+static void destroy_inodecache(void)
+{
+	if (kmem_cache_destroy(omfs_inode_cachep))
+		printk(KERN_INFO "omfs_inode_cache: not all structures were freed\n");
+}
+
+#ifdef OMFS_WRITE
+struct inode *omfs_new_inode(struct inode *dir, int mode)
+{
+	struct inode *inode;
+	u64 new_block;
+	int res;
+	int len;
+	struct omfs_sb_info *sbi = OMFS_SB(dir->i_sb);
+	
+	inode = new_inode(dir->i_sb);
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+
+	res = omfs_allocate_range(dir->i_sb, sbi->s_mirrors, sbi->s_mirrors, 
+			&new_block, &len);
+	if (res)
+		return ERR_PTR(res);
+
+	inode->i_ino = new_block;
+	inode->i_mode = mode;
+	inode->i_uid = current->fsuid;
+	inode->i_gid = current->fsgid;
+	inode->i_blksize = PAGE_SIZE;
+	inode->i_blocks = 0;
+	inode->i_mapping->a_ops = &omfs_aops;
+
+	OMFS_I(inode)->i_state = OMFS_STATE_NEW;
+
+	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+	switch (mode & S_IFMT) {
+	case S_IFDIR:
+		inode->i_op = &omfs_dir_inops;
+		inode->i_fop = &omfs_dir_operations;
+		inode->i_size = sbi->s_sys_blocksize;
+		inode->i_nlink++;  /* "." entry */
+		break;
+	case S_IFREG:
+		inode->i_fop = &omfs_file_operations;
+		inode->i_size = 0;
+		break;
+	}
+
+	insert_inode_hash(inode);
+	printk(KERN_DEBUG "omfs: new marked dirty: %lx\n", inode->i_ino);
+	mark_inode_dirty(inode);
+	return inode;
+}
+
+static int omfs_write_inode(struct inode *inode, int wait)
+{
+	struct omfs_inode *oi;
+	struct omfs_sb_info *sbi = OMFS_SB(inode->i_sb);
+	struct buffer_head *bh, *bh2;
+	unsigned int block;
+	u64 ctime;
+	int i;
+
+	printk(KERN_DEBUG "omfs: writing inode %lx\n", inode->i_ino);
+
+	/* get current inode since we may have written sibling ptrs etc. */
+	block = clus_to_blk(sbi, inode->i_ino);
+	bh = sb_bread(inode->i_sb, block);
+	if (!bh) 
+		return -EIO;
+	
+	oi = (struct omfs_inode *) bh->b_data;
+	if (OMFS_I(inode)->i_state & OMFS_STATE_NEW)
+	{
+		memset(oi, 0, sizeof(struct omfs_header));
+		OMFS_I(inode)->i_state &= ~OMFS_STATE_NEW;
+	}
+
+	oi->i_head.h_self = cpu_to_be64(inode->i_ino);
+	if (S_ISDIR(inode->i_mode)) {
+		oi->i_type = OMFS_DIR;
+	} else if (S_ISREG(inode->i_mode)) {
+		oi->i_type = OMFS_FILE;
+	} else 
+		BUG();
+
+	oi->i_head.h_body_size = cpu_to_be32(sbi->s_sys_blocksize - 
+		sizeof(struct omfs_header));
+	oi->i_head.h_version = 1;
+	oi->i_head.h_type = OMFS_INODE_NORMAL;
+	oi->i_head.h_magic = OMFS_IMAGIC;
+	oi->i_size = cpu_to_be64(inode->i_size);
+
+	ctime = inode->i_ctime.tv_sec * 1000LL + 
+		((inode->i_ctime.tv_nsec + 999)/1000);
+	oi->i_ctime = cpu_to_be64(ctime);
+
+	if (omfs_update_checksums(oi, inode->i_sb, inode->i_ino) != 0)
+		return -EIO;
+
+	mark_buffer_dirty(bh);
+	brelse(bh);
+
+	// if mirroring writes, copy to next fsblock
+	for (i = 0; i < sbi->s_mirrors; i++)
+	{
+		bh = sb_bread(inode->i_sb, block);
+		if (!bh)
+			return -EIO;
+		bh2 = sb_bread(inode->i_sb, block + i * 
+			(sbi->s_blocksize / sbi->s_sys_blocksize));
+		if (!bh2)
+		{
+			brelse(bh);
+			return -EIO;
+		}
+		memcpy(bh2->b_data, bh->b_data, bh->b_size);
+		mark_buffer_dirty(bh2);
+		brelse(bh);
+		brelse(bh2);
+	}
+
+	// do_sync?
+	return 0;
+}
+#endif
+
+void omfs_read_inode(struct inode *inode)
+{
+	struct omfs_inode *oi;
+	struct omfs_inode_info *oinf = OMFS_I(inode);
+	struct buffer_head *bh;
+	unsigned int block;
+	u64 ctime;
+	unsigned long nsecs;
+	ino_t ino = inode->i_ino;
+
+	// check against s_num_blocks
+	block = clus_to_blk(OMFS_SB(inode->i_sb), ino);
+	bh = sb_bread(inode->i_sb, block);
+	if (!bh) {
+		make_bad_inode(inode);
+		return;
+	}
+
+	oi = (struct omfs_inode *)bh->b_data;
+
+	// check self
+	if (ino != be64_to_cpu(oi->i_head.h_self)) {
+		make_bad_inode(inode);
+		return;
+	}
+
+	inode->i_uid = 0;
+	inode->i_gid = 0;
+
+	ctime = be64_to_cpu(oi->i_ctime);
+	nsecs = do_div(ctime, 1000) * 1000L;
+
+	inode->i_atime.tv_sec = ctime;
+	inode->i_mtime.tv_sec = ctime;
+	inode->i_ctime.tv_sec = ctime;
+	inode->i_atime.tv_nsec = nsecs;
+	inode->i_mtime.tv_nsec = nsecs;
+	inode->i_ctime.tv_nsec = nsecs;
+
+	oinf->i_state = 0;
+
+	if (oi->i_type == OMFS_DIR) {
+		inode->i_mode = S_IFDIR | S_IRUGO | S_IWUGO | S_IXUGO;
+		inode->i_op = &omfs_dir_inops;
+		inode->i_fop = &omfs_dir_operations;
+		inode->i_size = be32_to_cpu(oi->i_head.h_body_size) +
+			sizeof(struct omfs_header);
+	} else if (oi->i_type == OMFS_FILE) {
+		inode->i_mode = S_IFREG | S_IRUGO;
+		inode->i_mapping->a_ops = &omfs_aops;
+		inode->i_fop = &omfs_file_operations;
+		inode->i_size = be64_to_cpu(oi->i_size);
+	}
+	brelse(bh);
+}
+
+
+static void omfs_put_super(struct super_block *sb)
+{
+	struct omfs_sb_info *sbi = OMFS_SB(sb);
+	if (sbi)
+	{
+		kfree(sbi->s_imap);
+		kfree(sbi);
+	}
+	sb->s_fs_info = NULL;
+}
+
+static int omfs_statfs(struct super_block *s, struct kstatfs *buf)
+{
+	struct omfs_sb_info *sbi = OMFS_SB(s);
+	buf->f_type = OMFS_MAGIC;
+	buf->f_bsize = sbi->s_blocksize;
+	buf->f_blocks = sbi->s_num_blocks;
+	buf->f_files = sbi->s_num_blocks;
+	buf->f_namelen = OMFS_NAMELEN;
+
+	buf->f_bfree = buf->f_bavail = buf->f_ffree = 
+		omfs_count_free(s);
+	return 0;
+}
+
+struct super_operations omfs_sops = {
+	.alloc_inode	= omfs_alloc_inode,
+	.destroy_inode	= omfs_destroy_inode,
+#ifdef OMFS_WRITE
+	.write_inode	= omfs_write_inode, 
+#endif
+	.read_inode	= omfs_read_inode,
+	.put_super	= omfs_put_super,
+	.statfs		= omfs_statfs,
+};
+
+/*
+ * For Rio Karma, there is an on-disk free bitmap whose location is
+ * stored in the root block.  For ReplayTV, there is no such free bitmap
+ * so we have to walk the tree.  Both inodes and file data are allocated
+ * from the same map.  This array can be big (300k) so we allocate 
+ * in units of the blocksize.
+ */
+static int omfs_get_imap(struct super_block *sb)
+{
+	int bitmap_size;
+	int array_size;
+	int count;
+	struct omfs_sb_info *sbi = OMFS_SB(sb);
+	struct buffer_head *bh;
+	unsigned long **ptr;
+	sector_t block;
+
+	bitmap_size = (sbi->s_num_blocks + 7) / 8;
+	array_size = (bitmap_size + sb->s_blocksize - 1) / sb->s_blocksize;
+
+	printk(KERN_DEBUG "omfs: trying to alloc %d bytes\n", array_size);
+	printk(KERN_DEBUG "omfs: bitmap size: %d / blocksize %ld \n", 
+		bitmap_size, sb->s_blocksize);
+	sbi->s_imap_size = array_size;
+	sbi->s_imap = kzalloc(array_size * sizeof(unsigned long), GFP_KERNEL);
+	if (!sbi->s_imap)
+		return -ENOMEM;
+
+	block = clus_to_blk(sbi, sbi->s_bitmap_ino);
+	ptr = sbi->s_imap;
+	for (count = bitmap_size; count > 0; count -= sb->s_blocksize) {
+		bh = sb_bread(sb, block++);
+		if (!bh)
+			goto nomem;
+		*ptr = kmalloc(sb->s_blocksize, GFP_KERNEL);
+		memcpy(*ptr, bh->b_data, sb->s_blocksize);
+		brelse(bh);
+		ptr ++;
+	}
+	return 0;
+nomem:
+	for (count=0; count<array_size; count++)
+	{
+		printk(KERN_DEBUG "Freeing %d\n", count);
+		kfree(sbi->s_imap[count]);
+	}
+	printk(KERN_DEBUG "Freeing imap\n");
+	kfree(sbi->s_imap);
+	sbi->s_imap = NULL;
+	return -ENOMEM;
+}
+
+static void set_block_shift(struct omfs_sb_info *sbi)
+{
+	unsigned int scale = sbi->s_blocksize / sbi->s_sys_blocksize;
+	sbi->s_block_shift = 0;
+	for (scale>>=1; scale; scale>>=1)
+		sbi->s_block_shift++;
+}
+
+static int omfs_fill_super(struct super_block *sb, void *data, int silent)
+{
+	struct buffer_head *bh=NULL, *bh2=NULL;
+	struct omfs_super_block *omfs_sb;
+	struct omfs_root_block *omfs_rb;
+	struct omfs_sb_info *sbi;
+	struct inode *root;
+	sector_t start;
+	int ret = 0;
+
+	sbi = kzalloc(sizeof(struct omfs_sb_info), GFP_KERNEL);
+	if (!sbi)
+		return -ENOMEM;
+	sb->s_fs_info = sbi;
+
+#ifndef OMFS_WRITE
+	sb->s_flags |= MS_RDONLY; 
+#endif
+	sb->s_maxbytes = 0xffffffff;
+
+	sb_set_blocksize(sb, 0x200);
+
+	if (!(bh = sb_bread(sb, 0)))
+		goto out;
+
+	omfs_sb = (struct omfs_super_block *)bh->b_data;
+
+	if (be32_to_cpu(omfs_sb->s_magic) != OMFS_MAGIC) {
+		if (!silent)
+			printk(KERN_ERR "omfs: Invalid superblock (%x)\n",
+				   omfs_sb->s_magic);
+		goto out;
+	}
+	sb->s_magic = OMFS_MAGIC;
+
+	sbi->s_num_blocks = be64_to_cpu(omfs_sb->s_num_blocks);
+	sbi->s_blocksize = be32_to_cpu(omfs_sb->s_blocksize);
+	sbi->s_mirrors = be32_to_cpu(omfs_sb->s_mirrors);
+	sbi->s_root_ino = be64_to_cpu(omfs_sb->s_root_block);
+	sbi->s_sys_blocksize = be32_to_cpu(omfs_sb->s_sys_blocksize);
+
+	// Use sys_blocksize as the fs block since it is smaller than a 
+	// page while the fs blocksize can be larger.
+	sb_set_blocksize(sb, sbi->s_sys_blocksize);
+
+	// and the difference goes into a shift.  sys_blocksize is always
+	// a power of two factor of blocksize.
+	set_block_shift(sbi);
+
+	start = clus_to_blk(sbi, be64_to_cpu(omfs_sb->s_root_block));
+	if (!(bh2 = sb_bread(sb, start)))
+		goto out;
+
+	omfs_rb = (struct omfs_root_block *)bh2->b_data;
+	printk(KERN_DEBUG "omfs: volume: %s\n", omfs_rb->r_name);
+
+	sbi->s_bitmap_ino = be64_to_cpu(omfs_rb->r_bitmap);
+	sbi->s_clustersize = be32_to_cpu(omfs_rb->r_clustersize);
+
+	ret = omfs_get_imap(sb);
+	if (ret)
+		goto end;
+
+	sb->s_op = &omfs_sops;
+	root = iget(sb, be64_to_cpu(omfs_rb->r_root_dir));
+
+	sb->s_root = d_alloc_root(root);
+	if (!sb->s_root) {
+		iput(root);
+		goto out;
+	}
+	goto end;
+
+out:
+	ret = -EINVAL;
+
+end:
+	if (bh2)
+		brelse(bh2);
+	if (bh)
+		brelse(bh);
+	return ret;
+}
+
+static struct super_block *omfs_get_sb(struct file_system_type *fs_type,
+					   int flags, const char *dev_name,
+					   void *data)
+{
+	return get_sb_bdev(fs_type, flags, dev_name, data, omfs_fill_super);
+}
+
+static struct file_system_type omfs_fs_type = {
+	.owner = THIS_MODULE,
+	.name = "omfs",
+	.get_sb = omfs_get_sb,
+	.kill_sb = kill_block_super,
+	.fs_flags = FS_REQUIRES_DEV,
+};
+
+static int __init init_omfs_fs(void)
+{
+	int err = init_inodecache();
+	if (err)
+		goto out;
+
+	err = register_filesystem(&omfs_fs_type);
+	if (err)
+		destroy_inodecache();
+out:
+	return err;
+}
+
+static void __exit exit_omfs_fs(void)
+{
+	unregister_filesystem(&omfs_fs_type);
+	destroy_inodecache();
+}
+
+module_init(init_omfs_fs);
+module_exit(exit_omfs_fs);
-- 
1.2.1