[PATCH] e2fsprogs: Add undo I/O manager [ Was Re: E2fsprogs 1.40 release imminent! ]

["Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>]

Theodore Ts'o wrote:
> The hg repository located at:
> 
> 	http://thunk.org/hg/e2fsprogs
> 
> ... contains what I hope to be the e2fsprogs 1.40 release.  If folks
> could test it and let me know if they find any embarassing bugs, I would
> greatly appreciate it.  It can also be found at:
> 
> ftp://ftp.kernel.org/pub/linux/kernel/people/tytso/e2fsprogs/e2fsprogs-1.40-rc1.tar.gz
> 
> Over the next few days, I'll be working on converting e2fsprogs SCM from
> hg to git.  Assuming no one finds any problems, I'll be releasing
> e2fsprogs 1.40 by the end of the week.
> 
>

What about undo I/O manager ? 


This I/O manager saves the contents of the location being overwritten
to a tdb database. This helps in undoing the changes done to the
file system.

The call sequence involve

set_undo_io_backing_manager(unix_io_manager);
set_undo_io_backup_file("/tmp/test.tdb");
retval = ext2fs_open2(dev_name, 0, flags,
		superblock, block_size, undo_io_manager,
		&current_fs);

The patch is on top of e2fsprogs-upstream mercurial repo

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
----

diff -r fe55db3e508c lib/ext2fs/Makefile.in
--- a/lib/ext2fs/Makefile.in	Sun Jun 24 17:32:47 2007 -0400
+++ b/lib/ext2fs/Makefile.in	Mon Jun 25 20:59:36 2007 +0530
@@ -66,7 +66,8 @@ OBJS= $(DEBUGFS_LIB_OBJS) $(RESIZE_LIB_O
 	unix_io.o \
 	unlink.o \
 	valid_blk.o \
-	version.o
+	version.o \
+	undo_io.o
 
 SRCS= ext2_err.c \
 	$(srcdir)/alloc.c \
@@ -134,7 +135,8 @@ SRCS= ext2_err.c \
 	$(srcdir)/tst_bitops.c \
 	$(srcdir)/tst_byteswap.c \
 	$(srcdir)/tst_getsize.c \
-	$(srcdir)/tst_iscan.c
+	$(srcdir)/tst_iscan.c \
+	$(srcdir)/undo_io.c
 
 HFILES= bitops.h ext2fs.h ext2_io.h ext2_fs.h ext2_ext_attr.h ext3_extents.h \
 	tdb.h
@@ -583,3 +585,4 @@ tst_iscan.o: $(srcdir)/tst_iscan.c $(src
  $(top_builddir)/lib/ext2fs/ext2_types.h $(srcdir)/ext2fs.h \
  $(srcdir)/ext2_fs.h $(srcdir)/ext3_extents.h $(top_srcdir)/lib/et/com_err.h \
  $(srcdir)/ext2_io.h $(top_builddir)/lib/ext2fs/ext2_err.h $(srcdir)/bitops.h
+undo_io.o: $(srcdir)/undo_io.c $(srcdir)/ext2_fs.h $(srcdir)/ext2fs.h
diff -r fe55db3e508c lib/ext2fs/ext2_io.h
--- a/lib/ext2fs/ext2_io.h	Sun Jun 24 17:32:47 2007 -0400
+++ b/lib/ext2fs/ext2_io.h	Mon Jun 25 20:59:36 2007 +0530
@@ -96,6 +96,11 @@ extern errcode_t io_channel_write_byte(i
 /* unix_io.c */
 extern io_manager unix_io_manager;
 
+/* undo_io.c */
+extern io_manager undo_io_manager;
+extern errcode_t set_undo_io_backing_manager(io_manager manager);
+extern errcode_t set_undo_io_backup_file(char *file_name);
+
 /* test_io.c */
 extern io_manager test_io_manager, test_io_backing_manager;
 extern void (*test_io_cb_read_blk)
diff -r fe55db3e508c lib/ext2fs/undo_io.c
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/lib/ext2fs/undo_io.c	Mon Jun 25 21:02:24 2007 +0530
@@ -0,0 +1,478 @@
+/*
+ * undo_io.c --- This is the undo io manager that copies the old data that
+ * copies the old data being overwritten into a tdb database
+ *
+ * Copyright IBM Corporation, 2007
+ * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
+ *
+ * %Begin-Header%
+ * This file may be redistributed under the terms of the GNU Public
+ * License.
+ * %End-Header%
+ */
+
+#define _LARGEFILE_SOURCE
+#define _LARGEFILE64_SOURCE
+
+#include <stdio.h>
+#include <string.h>
+#if HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#if HAVE_ERRNO_H
+#include <errno.h>
+#endif
+#include <fcntl.h>
+#include <time.h>
+#ifdef __linux__
+#include <sys/utsname.h>
+#endif
+#if HAVE_SYS_STAT_H
+#include <sys/stat.h>
+#endif
+#if HAVE_SYS_TYPES_H
+#include <sys/types.h>
+#endif
+#if HAVE_SYS_RESOURCE_H
+#include <sys/resource.h>
+#endif
+
+#include "tdb.h"
+
+#include "ext2_fs.h"
+#include "ext2fs.h"
+
+/*
+ * For checking structure magic numbers...
+ */
+
+#define EXT2_CHECK_MAGIC(struct, code) \
+	  if ((struct)->magic != (code)) return (code)
+
+
+
+struct undo_private_data {
+	int	magic;
+	TDB_CONTEXT *tdb;
+	char *tdb_file;
+
+	/* The backing io channel */
+	io_channel real;
+
+	/* to support offset in unix I/O manager */
+	ext2_loff_t offset;
+};
+
+static errcode_t undo_open(const char *name, int flags, io_channel *channel);
+static errcode_t undo_close(io_channel channel);
+static errcode_t undo_set_blksize(io_channel channel, int blksize);
+static errcode_t undo_read_blk(io_channel channel, unsigned long block,
+			       int count, void *data);
+static errcode_t undo_write_blk(io_channel channel, unsigned long block,
+				int count, const void *data);
+static errcode_t undo_flush(io_channel channel);
+static errcode_t undo_write_byte(io_channel channel, unsigned long offset,
+				int size, const void *data);
+static errcode_t undo_set_option(io_channel channel, const char *option,
+				 const char *arg);
+
+static struct struct_io_manager struct_undo_manager = {
+	EXT2_ET_MAGIC_IO_MANAGER,
+	"Undo I/O Manager",
+	undo_open,
+	undo_close,
+	undo_set_blksize,
+	undo_read_blk,
+	undo_write_blk,
+	undo_flush,
+	undo_write_byte,
+	undo_set_option
+};
+
+io_manager undo_io_manager = &struct_undo_manager;
+static io_manager undo_io_backing_manager ;
+static  char *tdb_file ;
+static int tdb_data_size = 0;
+
+errcode_t set_undo_io_backing_manager(io_manager manager)
+{
+	/*
+	 * We may want to do some validation later
+	 */
+	undo_io_backing_manager = manager;
+	return 0;
+}
+
+errcode_t set_undo_io_backup_file(char *file_name)
+{
+	tdb_file = strdup(file_name);
+
+	if (tdb_file == NULL) {
+		return EXT2_ET_NO_MEMORY;
+	}
+
+	return 0;
+}
+
+
+static errcode_t undo_write_tdb(io_channel channel,
+					unsigned long block, int count)
+
+{
+	int size, loop_count = 0, i;
+	unsigned long block_num, backing_blk_num;
+	errcode_t retval = 0;
+	ext2_loff_t offset;
+	struct undo_private_data *data;
+	TDB_DATA tdb_key, tdb_data;
+	char *read_ptr;
+
+	data = (struct undo_private_data *) channel->private_data;
+
+
+	if (data->tdb == NULL) {
+		/*
+		 * Transaction database not initialized
+		 */
+		return 0;
+	}
+
+	/*
+	 * Set the block size used to read for tdb
+	 */
+	if (!tdb_data_size)
+		tdb_data_size = channel->block_size;
+
+	if (count == 1)
+		size = channel->block_size;
+	else {
+		if (count < 0)
+			size = -count;
+		else
+			size = count * channel->block_size;
+	}
+
+	/*
+	 * Data is stored in tdb database as blocks of tdb_data_size size
+	 * This helps in efficient lookup further.
+	 *
+	 * We divide the disk to blocks of tdb_data_size.
+	 */
+
+	block_num = ((block*channel->block_size)+data->offset)/tdb_data_size;
+
+
+	loop_count = (size + tdb_data_size -1)/tdb_data_size;
+
+	tdb_transaction_start(data->tdb);
+	for (i = 0; i < loop_count; i++) {
+
+		block_num += i;
+		tdb_key.dptr = (unsigned char *)&block_num;
+		tdb_key.dsize = sizeof(block_num);
+
+		/*
+		 * Check if we have the record already
+		 */
+		if (tdb_exists(data->tdb, tdb_key)) {
+
+			/* Try the next block */
+			continue;
+		}
+
+		/*
+		 * Read one block using the backing I/O manager
+		 * The backing I/O manager block size may be
+		 * different from the tdb_data_size.
+		 * Also we need to recalcuate the block number with respect
+		 * to the backing I/O manager.
+		 */
+
+		offset = block_num * tdb_data_size;
+		backing_blk_num = (offset - data->offset) / channel->block_size;
+
+		count = tdb_data_size +
+				((offset - data->offset) % channel->block_size);
+
+		retval = ext2fs_get_mem(count, &read_ptr);
+		if (retval)
+			return retval;
+
+		memset(read_ptr, 0, count);
+
+		retval = io_channel_read_blk(data->real,
+						backing_blk_num,
+						-count, read_ptr);
+		if (retval) {
+			free(read_ptr);
+			return retval;
+		}
+
+
+		tdb_data.dptr = read_ptr +
+				((offset - data->offset) % channel->block_size);
+
+		tdb_data.dsize = tdb_data_size;
+
+#ifdef DEBUG
+		printf("Printing with key %ld data %x and size %d\n",
+						block_num,
+						tdb_data.dptr, tdb_data_size);
+#endif
+
+		retval = tdb_store(data->tdb, tdb_key, tdb_data, TDB_INSERT);
+		if (retval == -1) {
+			/*
+			 * TDB_ERR_EXISTS cannot happen because we
+			 * have already verified it doesn't exist
+			 */
+			tdb_transaction_cancel(data->tdb);
+			retval = EXT2_ET_TDB_ERR_IO;
+			free(read_ptr);
+			return retval;
+
+		}
+		free(read_ptr);
+	}
+
+	tdb_transaction_commit(data->tdb);
+
+	return retval;
+
+}
+
+static TDB_CONTEXT *undo_setup_tdb(char *tdb_file,
+				struct undo_private_data *data)
+{
+	errcode_t	retval;
+
+	data->tdb = tdb_open(tdb_file, 0, TDB_CLEAR_IF_FIRST,
+			   O_RDWR | O_CREAT | O_TRUNC, 0600);
+	return data->tdb;
+
+}
+static errcode_t undo_open(const char *name, int flags, io_channel *channel)
+{
+	io_channel	io = NULL;
+	struct undo_private_data *data = NULL;
+	errcode_t	retval;
+	int		open_flags;
+	struct stat	st;
+
+	if (name == 0)
+		return EXT2_ET_BAD_DEVICE_NAME;
+	retval = ext2fs_get_mem(sizeof(struct struct_io_channel), &io);
+	if (retval)
+		return retval;
+	memset(io, 0, sizeof(struct struct_io_channel));
+	io->magic = EXT2_ET_MAGIC_IO_CHANNEL;
+	retval = ext2fs_get_mem(sizeof(struct undo_private_data), &data);
+	if (retval)
+		goto cleanup;
+
+	io->manager = undo_io_manager;
+	retval = ext2fs_get_mem(strlen(name)+1, &io->name);
+	if (retval)
+		goto cleanup;
+
+	strcpy(io->name, name);
+	io->private_data = data;
+	io->block_size = 1024;
+	io->read_error = 0;
+	io->write_error = 0;
+	io->refcount = 1;
+
+	memset(data, 0, sizeof(struct undo_private_data));
+	data->magic = EXT2_ET_MAGIC_UNIX_IO_CHANNEL;
+
+	if (undo_io_backing_manager) {
+		retval = undo_io_backing_manager->open(name, flags,
+						       &data->real);
+		if (retval)
+			goto cleanup;
+	} else {
+		data->real = 0;
+	}
+
+	/* setup the tdb file */
+	if (undo_setup_tdb(tdb_file, data) == NULL )
+		goto cleanup;
+
+	*channel = io;
+	return 0;
+
+cleanup:
+	if (data->real)
+		io_channel_close(data->real);
+
+	if (data)
+		ext2fs_free_mem(&data);
+
+	if (io)
+		ext2fs_free_mem(&io);
+
+	return retval;
+}
+
+static errcode_t undo_close(io_channel channel)
+{
+	struct undo_private_data *data;
+	errcode_t	retval = 0;
+
+	EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+	data = (struct undo_private_data *) channel->private_data;
+	EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+	if (--channel->refcount > 0)
+		return 0;
+
+	if (data->real)
+		retval = io_channel_close(data->real);
+
+	if (data->tdb)
+		tdb_close(data->tdb);
+
+	ext2fs_free_mem(&channel->private_data);
+	if (channel->name)
+		ext2fs_free_mem(&channel->name);
+	ext2fs_free_mem(&channel);
+
+	return retval;
+}
+
+static errcode_t undo_set_blksize(io_channel channel, int blksize)
+{
+	struct undo_private_data *data;
+	errcode_t		retval;
+
+	EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+	data = (struct undo_private_data *) channel->private_data;
+	EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+	if (data->real)
+		retval = io_channel_set_blksize(data->real, blksize);
+
+	channel->block_size = blksize;
+	return 0;
+}
+
+
+static errcode_t undo_read_blk(io_channel channel, unsigned long block,
+			       int count, void *buf)
+{
+	errcode_t	retval;
+	struct undo_private_data *data;
+
+	EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+	data = (struct undo_private_data *) channel->private_data;
+	EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+	if (data->real)
+		retval = io_channel_read_blk(data->real, block, count, buf);
+
+	return retval;
+}
+
+static errcode_t undo_write_blk(io_channel channel, unsigned long block,
+				int count, const void *buf)
+{
+	struct undo_private_data *data;
+	errcode_t	retval = 0;
+
+	EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+	data = (struct undo_private_data *) channel->private_data;
+	EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+	/*
+	 * First write the existing content into database
+	 */
+	retval = undo_write_tdb(channel, block, count);
+	if (retval)
+		 return retval;
+
+
+	if (data->real)
+		retval = io_channel_write_blk(data->real, block, count, buf);
+
+	return retval;
+
+}
+
+static errcode_t undo_write_byte(io_channel channel, unsigned long offset,
+				 int size, const void *buf)
+{
+	struct undo_private_data *data;
+	errcode_t	retval = 0;
+	ssize_t		actual;
+	ext2_loff_t	location;
+	unsigned long blk_num, count;;
+
+	EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+	data = (struct undo_private_data *) channel->private_data;
+	EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+	location = offset + data->offset;
+	blk_num = location/channel->block_size;
+	/*
+	 * the size specified may spread across multiple blocks
+	 * also make sure we account for the fact that block start
+	 * offset for tdb is different from the backing I/O manager
+	 * due to possible different block size
+	 */
+	count = (size + (location % channel->block_size) +
+			channel->block_size  -1)/channel->block_size;
+
+	retval = undo_write_tdb(channel, blk_num, count);
+	if (retval)
+		return retval;
+
+	if (data->real && data->real->manager->write_byte)
+		retval = io_channel_write_byte(data->real, offset, size, buf);
+
+	return retval;
+}
+
+/*
+ * Flush data buffers to disk.
+ */
+static errcode_t undo_flush(io_channel channel)
+{
+	errcode_t	retval = 0;
+	struct undo_private_data *data;
+
+	EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+	data = (struct undo_private_data *) channel->private_data;
+	EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+	if (data->real)
+		retval = io_channel_flush(data->real);
+
+	return retval;
+}
+
+static errcode_t undo_set_option(io_channel channel, const char *option,
+				 const char *arg)
+{
+	struct undo_private_data *data;
+	unsigned long tmp;
+	char *end;
+
+	EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+	data = (struct undo_private_data *) channel->private_data;
+	EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+	/*
+	 * Need to support offset option to work with
+	 * Unix I/O manager
+	 */
+	if (!strcmp(option, "offset")) {
+		if (!arg)
+			return EXT2_ET_INVALID_ARGUMENT;
+
+		tmp = strtoul(arg, &end, 0);
+		if (*end)
+			return EXT2_ET_INVALID_ARGUMENT;
+		data->offset = tmp;
+		return 0;
+	}
+}