[LogFS] add new flash file system

This is a new flash file system. See
Documentation/filesystems/logfs.txt
Signed-off-by: NJoern Engel <joern@logfs.org>

Documentation/filesystems/00-INDEX

@@ -62,6 +62,8 @@ jfs.txt
 	- info and mount options for the JFS filesystem.
 locks.txt
 	- info on file locking implementations, flock() vs. fcntl(), etc.
+logfs.txt
+	- info on the LogFS flash filesystem.
 mandatory-locking.txt
 	- info on the Linux implementation of Sys V mandatory file locking.
 ncpfs.txt

Documentation/filesystems/logfs.txt

+
+The LogFS Flash Filesystem
+==========================
+
+Specification
+=============
+
+Superblocks
+-----------
+
+Two superblocks exist at the beginning and end of the filesystem.
+Each superblock is 256 Bytes large, with another 3840 Bytes reserved
+for future purposes, making a total of 4096 Bytes.
+
+Superblock locations may differ for MTD and block devices.  On MTD the
+first non-bad block contains a superblock in the first 4096 Bytes and
+the last non-bad block contains a superblock in the last 4096 Bytes.
+On block devices, the first 4096 Bytes of the device contain the first
+superblock and the last aligned 4096 Byte-block contains the second
+superblock.
+
+For the most part, the superblocks can be considered read-only.  They
+are written only to correct errors detected within the superblocks,
+move the journal and change the filesystem parameters through tunefs.
+As a result, the superblock does not contain any fields that require
+constant updates, like the amount of free space, etc.
+
+Segments
+--------
+
+The space in the device is split up into equal-sized segments.
+Segments are the primary write unit of LogFS.  Within each segments,
+writes happen from front (low addresses) to back (high addresses.  If
+only a partial segment has been written, the segment number, the
+current position within and optionally a write buffer are stored in
+the journal.
+
+Segments are erased as a whole.  Therefore Garbage Collection may be
+required to completely free a segment before doing so.
+
+Journal
+--------
+
+The journal contains all global information about the filesystem that
+is subject to frequent change.  At mount time, it has to be scanned
+for the most recent commit entry, which contains a list of pointers to
+all currently valid entries.
+
+Object Store
+------------
+
+All space except for the superblocks and journal is part of the object
+store.  Each segment contains a segment header and a number of
+objects, each consisting of the object header and the payload.
+Objects are either inodes, directory entries (dentries), file data
+blocks or indirect blocks.
+
+Levels
+------
+
+Garbage collection (GC) may fail if all data is written
+indiscriminately.  One requirement of GC is that data is seperated
+roughly according to the distance between the tree root and the data.
+Effectively that means all file data is on level 0, indirect blocks
+are on levels 1, 2, 3 4 or 5 for 1x, 2x, 3x, 4x or 5x indirect blocks,
+respectively.  Inode file data is on level 6 for the inodes and 7-11
+for indirect blocks.
+
+Each segment contains objects of a single level only.  As a result,
+each level requires its own seperate segment to be open for writing.
+
+Inode File
+----------
+
+All inodes are stored in a special file, the inode file.  Single
+exception is the inode file's inode (master inode) which for obvious
+reasons is stored in the journal instead.  Instead of data blocks, the
+leaf nodes of the inode files are inodes.
+
+Aliases
+-------
+
+Writes in LogFS are done by means of a wandering tree.  A naïve
+implementation would require that for each write or a block, all
+parent blocks are written as well, since the block pointers have
+changed.  Such an implementation would not be very efficient.
+
+In LogFS, the block pointer changes are cached in the journal by means
+of alias entries.  Each alias consists of its logical address - inode
+number, block index, level and child number (index into block) - and
+the changed data.  Any 8-byte word can be changes in this manner.
+
+Currently aliases are used for block pointers, file size, file used
+bytes and the height of an inodes indirect tree.
+
+Segment Aliases
+---------------
+
+Related to regular aliases, these are used to handle bad blocks.
+Initially, bad blocks are handled by moving the affected segment
+content to a spare segment and noting this move in the journal with a
+segment alias, a simple (to, from) tupel.  GC will later empty this
+segment and the alias can be removed again.  This is used on MTD only.
+
+Vim
+---
+
+By cleverly predicting the life time of data, it is possible to
+seperate long-living data from short-living data and thereby reduce
+the GC overhead later.  Each type of distinc life expectency (vim) can
+have a seperate segment open for writing.  Each (level, vim) tupel can
+be open just once.  If an open segment with unknown vim is encountered
+at mount time, it is closed and ignored henceforth.
+
+Indirect Tree
+-------------
+
+Inodes in LogFS are similar to FFS-style filesystems with direct and
+indirect block pointers.  One difference is that LogFS uses a single
+indirect pointer that can be either a 1x, 2x, etc. indirect pointer.
+A height field in the inode defines the height of the indirect tree
+and thereby the indirection of the pointer.
+
+Another difference is the addressing of indirect blocks.  In LogFS,
+the first 16 pointers in the first indirect block are left empty,
+corresponding to the 16 direct pointers in the inode.  In ext2 (maybe
+others as well) the first pointer in the first indirect block
+corresponds to logical block 12, skipping the 12 direct pointers.
+So where ext2 is using arithmetic to better utilize space, LogFS keeps
+arithmetic simple and uses compression to save space.
+
+Compression
+-----------
+
+Both file data and metadata can be compressed.  Compression for file
+data can be enabled with chattr +c and disabled with chattr -c.  Doing
+so has no effect on existing data, but new data will be stored
+accordingly.  New inodes will inherit the compression flag of the
+parent directory.
+
+Metadata is always compressed.  However, the space accounting ignores
+this and charges for the uncompressed size.  Failing to do so could
+result in GC failures when, after moving some data, indirect blocks
+compress worse than previously.  Even on a 100% full medium, GC may
+not consume any extra space, so the compression gains are lost space
+to the user.
+
+However, they are not lost space to the filesystem internals.  By
+cheating the user for those bytes, the filesystem gained some slack
+space and GC will run less often and faster.
+
+Garbage Collection and Wear Leveling
+------------------------------------
+
+Garbage collection is invoked whenever the number of free segments
+falls below a threshold.  The best (known) candidate is picked based
+on the least amount of valid data contained in the segment.  All
+remaining valid data is copied elsewhere, thereby invalidating it.
+
+The GC code also checks for aliases and writes then back if their
+number gets too large.
+
+Wear leveling is done by occasionally picking a suboptimal segment for
+garbage collection.  If a stale segments erase count is significantly
+lower than the active segments' erase counts, it will be picked.  Wear
+leveling is rate limited, so it will never monopolize the device for
+more than one segment worth at a time.
+
+Values for "occasionally", "significantly lower" are compile time
+constants.
+
+Hashed directories
+------------------
+
+To satisfy efficient lookup(), directory entries are hashed and
+located based on the hash.  In order to both support large directories
+and not be overly inefficient for small directories, several hash
+tables of increasing size are used.  For each table, the hash value
+modulo the table size gives the table index.
+
+Tables sizes are chosen to limit the number of indirect blocks with a
+fully populated table to 0, 1, 2 or 3 respectively.  So the first
+table contains 16 entries, the second 512-16, etc.
+
+The last table is special in several ways.  First its size depends on
+the effective 32bit limit on telldir/seekdir cookies.  Since logfs
+uses the upper half of the address space for indirect blocks, the size
+is limited to 2^31.  Secondly the table contains hash buckets with 16
+entries each.
+
+Using single-entry buckets would result in birthday "attacks".  At
+just 2^16 used entries, hash collisions would be likely (P >= 0.5).
+My math skills are insufficient to do the combinatorics for the 17x
+collisions necessary to overflow a bucket, but testing showed that in
+10,000 runs the lowest directory fill before a bucket overflow was
+188,057,130 entries with an average of 315,149,915 entries.  So for
+directory sizes of up to a million, bucket overflows should be
+virtually impossible under normal circumstances.
+
+With carefully chosen filenames, it is obviously possible to cause an
+overflow with just 21 entries (4 higher tables + 16 entries + 1).  So
+there may be a security concern if a malicious user has write access
+to a directory.
+
+Open For Discussion
+===================
+
+Device Address Space
+--------------------
+
+A device address space is used for caching.  Both block devices and
+MTD provide functions to either read a single page or write a segment.
+Partial segments may be written for data integrity, but where possible
+complete segments are written for performance on simple block device
+flash media.
+
+Meta Inodes
+-----------
+
+Inodes are stored in the inode file, which is just a regular file for
+most purposes.  At umount time, however, the inode file needs to
+remain open until all dirty inodes are written.  So
+generic_shutdown_super() may not close this inode, but shouldn't
+complain about remaining inodes due to the inode file either.  Same
+goes for mapping inode of the device address space.
+
+Currently logfs uses a hack that essentially copies part of fs/inode.c
+code over.  A general solution would be preferred.
+
+Indirect block mapping
+----------------------
+
+With compression, the block device (or mapping inode) cannot be used
+to cache indirect blocks.  Some other place is required.  Currently
+logfs uses the top half of each inode's address space.  The low 8TB
+(on 32bit) are filled with file data, the high 8TB are used for
+indirect blocks.
+
+One problem is that 16TB files created on 64bit systems actually have
+data in the top 8TB.  But files >16TB would cause problems anyway, so
+only the limit has changed.

fs/Kconfig

@@ -177,6 +177,7 @@ source "fs/efs/Kconfig"
 source "fs/jffs2/Kconfig"
 # UBIFS File system configuration
 source "fs/ubifs/Kconfig"
+source "fs/logfs/Kconfig"
 source "fs/cramfs/Kconfig"
 source "fs/squashfs/Kconfig"
 source "fs/freevxfs/Kconfig"

fs/Makefile

@@ -99,6 +99,7 @@ obj-$(CONFIG_NTFS_FS)		+= ntfs/
 obj-$(CONFIG_UFS_FS)		+= ufs/
 obj-$(CONFIG_EFS_FS)		+= efs/
 obj-$(CONFIG_JFFS2_FS)		+= jffs2/
+obj-$(CONFIG_LOGFS)		+= logfs/
 obj-$(CONFIG_UBIFS_FS)		+= ubifs/
 obj-$(CONFIG_AFFS_FS)		+= affs/
 obj-$(CONFIG_ROMFS_FS)		+= romfs/

fs/logfs/Kconfig

+config LOGFS
+	tristate "LogFS file system (EXPERIMENTAL)"
+	depends on (MTD || BLOCK) && EXPERIMENTAL
+	select ZLIB_INFLATE
+	select ZLIB_DEFLATE
+	select CRC32
+	select BTREE
+	help
+	  Flash filesystem aimed to scale efficiently to large devices.
+	  In comparison to JFFS2 it offers significantly faster mount
+	  times and potentially less RAM usage, although the latter has
+	  not been measured yet.
+
+	  In its current state it is still very experimental and should
+	  not be used for other than testing purposes.
+
+	  If unsure, say N.

fs/logfs/Makefile

+obj-$(CONFIG_LOGFS)	+= logfs.o
+
+logfs-y	+= compr.o
+logfs-y	+= dir.o
+logfs-y	+= file.o
+logfs-y	+= gc.o
+logfs-y	+= inode.o
+logfs-y	+= journal.o
+logfs-y	+= readwrite.o
+logfs-y	+= segment.o
+logfs-y	+= super.o
+logfs-$(CONFIG_BLOCK)	+= dev_bdev.o
+logfs-$(CONFIG_MTD)	+= dev_mtd.o

fs/logfs/compr.c

+/*
+ * fs/logfs/compr.c	- compression routines
+ *
+ * As should be obvious for Linux kernel code, license is GPLv2
+ *
+ * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
+ */
+#include "logfs.h"
+#include <linux/vmalloc.h>
+#include <linux/zlib.h>
+
+#define COMPR_LEVEL 3
+
+static DEFINE_MUTEX(compr_mutex);
+static struct z_stream_s stream;
+
+int logfs_compress(void *in, void *out, size_t inlen, size_t outlen)
+{
+	int err, ret;
+
+	ret = -EIO;
+	mutex_lock(&compr_mutex);
+	err = zlib_deflateInit(&stream, COMPR_LEVEL);
+	if (err != Z_OK)
+		goto error;
+
+	stream.next_in = in;
+	stream.avail_in = inlen;
+	stream.total_in = 0;
+	stream.next_out = out;
+	stream.avail_out = outlen;
+	stream.total_out = 0;
+
+	err = zlib_deflate(&stream, Z_FINISH);
+	if (err != Z_STREAM_END)
+		goto error;
+
+	err = zlib_deflateEnd(&stream);
+	if (err != Z_OK)
+		goto error;
+
+	if (stream.total_out >= stream.total_in)
+		goto error;
+
+	ret = stream.total_out;
+error:
+	mutex_unlock(&compr_mutex);
+	return ret;
+}
+
+int logfs_uncompress(void *in, void *out, size_t inlen, size_t outlen)
+{
+	int err, ret;
+
+	ret = -EIO;
+	mutex_lock(&compr_mutex);
+	err = zlib_inflateInit(&stream);
+	if (err != Z_OK)
+		goto error;
+
+	stream.next_in = in;
+	stream.avail_in = inlen;
+	stream.total_in = 0;
+	stream.next_out = out;
+	stream.avail_out = outlen;
+	stream.total_out = 0;
+
+	err = zlib_inflate(&stream, Z_FINISH);
+	if (err != Z_STREAM_END)
+		goto error;
+
+	err = zlib_inflateEnd(&stream);
+	if (err != Z_OK)
+		goto error;
+
+	ret = 0;
+error:
+	mutex_unlock(&compr_mutex);
+	return ret;
+}
+
+int __init logfs_compr_init(void)
+{
+	size_t size = max(zlib_deflate_workspacesize(),
+			zlib_inflate_workspacesize());
+	stream.workspace = vmalloc(size);
+	if (!stream.workspace)
+		return -ENOMEM;
+	return 0;
+}
+
+void logfs_compr_exit(void)
+{
+	vfree(stream.workspace);
+}

fs/logfs/dev_bdev.c

+/*
+ * fs/logfs/dev_bdev.c	- Device access methods for block devices
+ *
+ * As should be obvious for Linux kernel code, license is GPLv2
+ *
+ * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
+ */
+#include "logfs.h"
+#include <linux/bio.h>
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+
+#define PAGE_OFS(ofs) ((ofs) & (PAGE_SIZE-1))
+
+static void request_complete(struct bio *bio, int err)
+{
+	complete((struct completion *)bio->bi_private);
+}
+
+static int sync_request(struct page *page, struct block_device *bdev, int rw)
+{
+	struct bio bio;
+	struct bio_vec bio_vec;
+	struct completion complete;
+
+	bio_init(&bio);
+	bio.bi_io_vec = &bio_vec;
+	bio_vec.bv_page = page;
+	bio_vec.bv_len = PAGE_SIZE;
+	bio_vec.bv_offset = 0;
+	bio.bi_vcnt = 1;
+	bio.bi_idx = 0;
+	bio.bi_size = PAGE_SIZE;
+	bio.bi_bdev = bdev;
+	bio.bi_sector = page->index * (PAGE_SIZE >> 9);
+	init_completion(&complete);
+	bio.bi_private = &complete;
+	bio.bi_end_io = request_complete;
+
+	submit_bio(rw, &bio);
+	generic_unplug_device(bdev_get_queue(bdev));
+	wait_for_completion(&complete);
+	return test_bit(BIO_UPTODATE, &bio.bi_flags) ? 0 : -EIO;
+}
+
+static int bdev_readpage(void *_sb, struct page *page)
+{
+	struct super_block *sb = _sb;
+	struct block_device *bdev = logfs_super(sb)->s_bdev;
+	int err;
+
+	err = sync_request(page, bdev, READ);
+	if (err) {
+		ClearPageUptodate(page);
+		SetPageError(page);
+	} else {
+		SetPageUptodate(page);
+		ClearPageError(page);
+	}
+	unlock_page(page);
+	return err;
+}
+
+static DECLARE_WAIT_QUEUE_HEAD(wq);
+
+static void writeseg_end_io(struct bio *bio, int err)
+{
+	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
+	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
+	struct super_block *sb = bio->bi_private;
+	struct logfs_super *super = logfs_super(sb);
+	struct page *page;
+
+	BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
+	BUG_ON(err);
+	BUG_ON(bio->bi_vcnt == 0);
+	do {
+		page = bvec->bv_page;
+		if (--bvec >= bio->bi_io_vec)
+			prefetchw(&bvec->bv_page->flags);
+
+		end_page_writeback(page);
+	} while (bvec >= bio->bi_io_vec);
+	bio_put(bio);
+	if (atomic_dec_and_test(&super->s_pending_writes))
+		wake_up(&wq);
+}
+
+static int __bdev_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
+		size_t nr_pages)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct address_space *mapping = super->s_mapping_inode->i_mapping;
+	struct bio *bio;
+	struct page *page;
+	struct request_queue *q = bdev_get_queue(sb->s_bdev);
+	unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
+	int i;
+
+	bio = bio_alloc(GFP_NOFS, max_pages);
+	BUG_ON(!bio); /* FIXME: handle this */
+
+	for (i = 0; i < nr_pages; i++) {
+		if (i >= max_pages) {
+			/* Block layer cannot split bios :( */
+			bio->bi_vcnt = i;
+			bio->bi_idx = 0;
+			bio->bi_size = i * PAGE_SIZE;
+			bio->bi_bdev = super->s_bdev;
+			bio->bi_sector = ofs >> 9;
+			bio->bi_private = sb;
+			bio->bi_end_io = writeseg_end_io;
+			atomic_inc(&super->s_pending_writes);
+			submit_bio(WRITE, bio);
+
+			ofs += i * PAGE_SIZE;
+			index += i;
+			nr_pages -= i;
+			i = 0;
+
+			bio = bio_alloc(GFP_NOFS, max_pages);
+			BUG_ON(!bio);
+		}
+		page = find_lock_page(mapping, index + i);
+		BUG_ON(!page);
+		bio->bi_io_vec[i].bv_page = page;
+		bio->bi_io_vec[i].bv_len = PAGE_SIZE;
+		bio->bi_io_vec[i].bv_offset = 0;
+
+		BUG_ON(PageWriteback(page));
+		set_page_writeback(page);
+		unlock_page(page);
+	}
+	bio->bi_vcnt = nr_pages;
+	bio->bi_idx = 0;
+	bio->bi_size = nr_pages * PAGE_SIZE;
+	bio->bi_bdev = super->s_bdev;
+	bio->bi_sector = ofs >> 9;
+	bio->bi_private = sb;
+	bio->bi_end_io = writeseg_end_io;
+	atomic_inc(&super->s_pending_writes);
+	submit_bio(WRITE, bio);
+	return 0;
+}
+
+static void bdev_writeseg(struct super_block *sb, u64 ofs, size_t len)
+{
+	struct logfs_super *super = logfs_super(sb);
+	int head;
+
+	BUG_ON(super->s_flags & LOGFS_SB_FLAG_RO);
+
+	if (len == 0) {
+		/* This can happen when the object fit perfectly into a
+		 * segment, the segment gets written per sync and subsequently
+		 * closed.
+		 */
+		return;
+	}
+	head = ofs & (PAGE_SIZE - 1);
+	if (head) {
+		ofs -= head;
+		len += head;
+	}
+	len = PAGE_ALIGN(len);
+	__bdev_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
+	generic_unplug_device(bdev_get_queue(logfs_super(sb)->s_bdev));
+}
+
+static int bdev_erase(struct super_block *sb, loff_t to, size_t len)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct address_space *mapping = super->s_mapping_inode->i_mapping;
+	struct page *page;
+	pgoff_t index = to >> PAGE_SHIFT;
+	int i, nr_pages = len >> PAGE_SHIFT;
+
+	BUG_ON(to & (PAGE_SIZE - 1));
+	BUG_ON(len & (PAGE_SIZE - 1));
+
+	if (logfs_super(sb)->s_flags & LOGFS_SB_FLAG_RO)
+		return -EROFS;
+
+	for (i = 0; i < nr_pages; i++) {
+		page = find_get_page(mapping, index + i);
+		if (page) {
+			memset(page_address(page), 0xFF, PAGE_SIZE);
+			page_cache_release(page);
+		}
+	}
+	return 0;
+}
+
+static void bdev_sync(struct super_block *sb)
+{
+	struct logfs_super *super = logfs_super(sb);
+
+	wait_event(wq, atomic_read(&super->s_pending_writes) == 0);
+}
+
+static struct page *bdev_find_first_sb(struct super_block *sb, u64 *ofs)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct address_space *mapping = super->s_mapping_inode->i_mapping;
+	filler_t *filler = bdev_readpage;
+
+	*ofs = 0;
+	return read_cache_page(mapping, 0, filler, sb);
+}
+
+static struct page *bdev_find_last_sb(struct super_block *sb, u64 *ofs)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct address_space *mapping = super->s_mapping_inode->i_mapping;
+	filler_t *filler = bdev_readpage;
+	u64 pos = (super->s_bdev->bd_inode->i_size & ~0xfffULL) - 0x1000;
+	pgoff_t index = pos >> PAGE_SHIFT;
+
+	*ofs = pos;
+	return read_cache_page(mapping, index, filler, sb);
+}
+
+static int bdev_write_sb(struct super_block *sb, struct page *page)
+{
+	struct block_device *bdev = logfs_super(sb)->s_bdev;
+
+	/* Nothing special to do for block devices. */
+	return sync_request(page, bdev, WRITE);
+}
+
+static void bdev_put_device(struct super_block *sb)
+{
+	close_bdev_exclusive(logfs_super(sb)->s_bdev, FMODE_READ|FMODE_WRITE);
+}
+
+static const struct logfs_device_ops bd_devops = {
+	.find_first_sb	= bdev_find_first_sb,
+	.find_last_sb	= bdev_find_last_sb,
+	.write_sb	= bdev_write_sb,
+	.readpage	= bdev_readpage,
+	.writeseg	= bdev_writeseg,
+	.erase		= bdev_erase,
+	.sync		= bdev_sync,
+	.put_device	= bdev_put_device,
+};
+
+int logfs_get_sb_bdev(struct file_system_type *type, int flags,
+		const char *devname, struct vfsmount *mnt)
+{
+	struct block_device *bdev;
+
+	bdev = open_bdev_exclusive(devname, FMODE_READ|FMODE_WRITE, type);
+	if (IS_ERR(bdev))
+		return PTR_ERR(bdev);
+
+	if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
+		int mtdnr = MINOR(bdev->bd_dev);
+		close_bdev_exclusive(bdev, FMODE_READ|FMODE_WRITE);
+		return logfs_get_sb_mtd(type, flags, mtdnr, mnt);
+	}
+
+	return logfs_get_sb_device(type, flags, NULL, bdev, &bd_devops, mnt);
+}

fs/logfs/dev_mtd.c

+/*
+ * fs/logfs/dev_mtd.c	- Device access methods for MTD
+ *
+ * As should be obvious for Linux kernel code, license is GPLv2
+ *
+ * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
+ */
+#include "logfs.h"
+#include <linux/completion.h>
+#include <linux/mount.h>
+#include <linux/sched.h>
+
+#define PAGE_OFS(ofs) ((ofs) & (PAGE_SIZE-1))
+
+static int mtd_read(struct super_block *sb, loff_t ofs, size_t len, void *buf)
+{
+	struct mtd_info *mtd = logfs_super(sb)->s_mtd;
+	size_t retlen;
+	int ret;
+
+	ret = mtd->read(mtd, ofs, len, &retlen, buf);
+	BUG_ON(ret == -EINVAL);
+	if (ret)
+		return ret;
+
+	/* Not sure if we should loop instead. */
+	if (retlen != len)
+		return -EIO;
+
+	return 0;
+}
+
+static int mtd_write(struct super_block *sb, loff_t ofs, size_t len, void *buf)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct mtd_info *mtd = super->s_mtd;
+	size_t retlen;
+	loff_t page_start, page_end;
+	int ret;
+
+	if (super->s_flags & LOGFS_SB_FLAG_RO)
+		return -EROFS;
+
+	BUG_ON((ofs >= mtd->size) || (len > mtd->size - ofs));
+	BUG_ON(ofs != (ofs >> super->s_writeshift) << super->s_writeshift);
+	BUG_ON(len > PAGE_CACHE_SIZE);
+	page_start = ofs & PAGE_CACHE_MASK;
+	page_end = PAGE_CACHE_ALIGN(ofs + len) - 1;
+	ret = mtd->write(mtd, ofs, len, &retlen, buf);
+	if (ret || (retlen != len))
+		return -EIO;
+
+	return 0;
+}
+
+/*
+ * For as long as I can remember (since about 2001) mtd->erase has been an
+ * asynchronous interface lacking the first driver to actually use the
+ * asynchronous properties.  So just to prevent the first implementor of such
+ * a thing from breaking logfs in 2350, we do the usual pointless dance to
+ * declare a completion variable and wait for completion before returning
+ * from mtd_erase().  What an excercise in futility!
+ */
+static void logfs_erase_callback(struct erase_info *ei)
+{
+	complete((struct completion *)ei->priv);
+}
+
+static int mtd_erase_mapping(struct super_block *sb, loff_t ofs, size_t len)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct address_space *mapping = super->s_mapping_inode->i_mapping;
+	struct page *page;
+	pgoff_t index = ofs >> PAGE_SHIFT;
+
+	for (index = ofs >> PAGE_SHIFT; index < (ofs + len) >> PAGE_SHIFT; index++) {
+		page = find_get_page(mapping, index);
+		if (!page)
+			continue;
+		memset(page_address(page), 0xFF, PAGE_SIZE);
+		page_cache_release(page);
+	}
+	return 0;
+}
+
+static int mtd_erase(struct super_block *sb, loff_t ofs, size_t len)
+{
+	struct mtd_info *mtd = logfs_super(sb)->s_mtd;
+	struct erase_info ei;
+	DECLARE_COMPLETION_ONSTACK(complete);
+	int ret;
+
+	BUG_ON(len % mtd->erasesize);
+	if (logfs_super(sb)->s_flags & LOGFS_SB_FLAG_RO)
+		return -EROFS;
+
+	memset(&ei, 0, sizeof(ei));
+	ei.mtd = mtd;
+	ei.addr = ofs;
+	ei.len = len;
+	ei.callback = logfs_erase_callback;
+	ei.priv = (long)&complete;
+	ret = mtd->erase(mtd, &ei);
+	if (ret)
+		return -EIO;
+
+	wait_for_completion(&complete);
+	if (ei.state != MTD_ERASE_DONE)
+		return -EIO;
+	return mtd_erase_mapping(sb, ofs, len);
+}
+
+static void mtd_sync(struct super_block *sb)
+{
+	struct mtd_info *mtd = logfs_super(sb)->s_mtd;
+
+	if (mtd->sync)
+		mtd->sync(mtd);
+}
+
+static int mtd_readpage(void *_sb, struct page *page)
+{
+	struct super_block *sb = _sb;
+	int err;
+
+	err = mtd_read(sb, page->index << PAGE_SHIFT, PAGE_SIZE,
+			page_address(page));
+	if (err == -EUCLEAN) {
+		err = 0;
+		/* FIXME: force GC this segment */
+	}
+	if (err) {
+		ClearPageUptodate(page);
+		SetPageError(page);
+	} else {
+		SetPageUptodate(page);
+		ClearPageError(page);
+	}
+	unlock_page(page);
+	return err;
+}
+
+static struct page *mtd_find_first_sb(struct super_block *sb, u64 *ofs)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct address_space *mapping = super->s_mapping_inode->i_mapping;
+	filler_t *filler = mtd_readpage;
+	struct mtd_info *mtd = super->s_mtd;
+
+	if (!mtd->block_isbad)
+		return NULL;
+
+	*ofs = 0;
+	while (mtd->block_isbad(mtd, *ofs)) {
+		*ofs += mtd->erasesize;
+		if (*ofs >= mtd->size)
+			return NULL;
+	}
+	BUG_ON(*ofs & ~PAGE_MASK);
+	return read_cache_page(mapping, *ofs >> PAGE_SHIFT, filler, sb);
+}
+
+static struct page *mtd_find_last_sb(struct super_block *sb, u64 *ofs)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct address_space *mapping = super->s_mapping_inode->i_mapping;
+	filler_t *filler = mtd_readpage;
+	struct mtd_info *mtd = super->s_mtd;
+
+	if (!mtd->block_isbad)
+		return NULL;
+
+	*ofs = mtd->size - mtd->erasesize;
+	while (mtd->block_isbad(mtd, *ofs)) {
+		*ofs -= mtd->erasesize;
+		if (*ofs <= 0)
+			return NULL;
+	}
+	*ofs = *ofs + mtd->erasesize - 0x1000;
+	BUG_ON(*ofs & ~PAGE_MASK);
+	return read_cache_page(mapping, *ofs >> PAGE_SHIFT, filler, sb);
+}
+
+static int __mtd_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
+		size_t nr_pages)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct address_space *mapping = super->s_mapping_inode->i_mapping;
+	struct page *page;
+	int i, err;
+
+	for (i = 0; i < nr_pages; i++) {
+		page = find_lock_page(mapping, index + i);
+		BUG_ON(!page);
+
+		err = mtd_write(sb, page->index << PAGE_SHIFT, PAGE_SIZE,
+				page_address(page));
+		unlock_page(page);
+		page_cache_release(page);
+		if (err)
+			return err;
+	}
+	return 0;
+}
+
+static void mtd_writeseg(struct super_block *sb, u64 ofs, size_t len)
+{
+	struct logfs_super *super = logfs_super(sb);
+	int head;
+
+	if (super->s_flags & LOGFS_SB_FLAG_RO)
+		return;
+
+	if (len == 0) {
+		/* This can happen when the object fit perfectly into a
+		 * segment, the segment gets written per sync and subsequently
+		 * closed.
+		 */
+		return;
+	}
+	head = ofs & (PAGE_SIZE - 1);
+	if (head) {
+		ofs -= head;
+		len += head;
+	}
+	len = PAGE_ALIGN(len);
+	__mtd_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
+}
+
+static void mtd_put_device(struct super_block *sb)
+{
+	put_mtd_device(logfs_super(sb)->s_mtd);
+}
+
+static const struct logfs_device_ops mtd_devops = {
+	.find_first_sb	= mtd_find_first_sb,
+	.find_last_sb	= mtd_find_last_sb,
+	.readpage	= mtd_readpage,
+	.writeseg	= mtd_writeseg,
+	.erase		= mtd_erase,
+	.sync		= mtd_sync,
+	.put_device	= mtd_put_device,
+};
+
+int logfs_get_sb_mtd(struct file_system_type *type, int flags,
+		int mtdnr, struct vfsmount *mnt)
+{
+	struct mtd_info *mtd;
+	const struct logfs_device_ops *devops = &mtd_devops;
+
+	mtd = get_mtd_device(NULL, mtdnr);
+	return logfs_get_sb_device(type, flags, mtd, NULL, devops, mnt);
+}

fs/logfs/file.c

+/*
+ * fs/logfs/file.c	- prepare_write, commit_write and friends
+ *
+ * As should be obvious for Linux kernel code, license is GPLv2
+ *
+ * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
+ */
+#include "logfs.h"
+#include <linux/sched.h>
+#include <linux/writeback.h>
+
+static int logfs_write_begin(struct file *file, struct address_space *mapping,
+		loff_t pos, unsigned len, unsigned flags,
+		struct page **pagep, void **fsdata)
+{
+	struct inode *inode = mapping->host;
+	struct page *page;
+	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
+
+	page = grab_cache_page_write_begin(mapping, index, flags);
+	if (!page)
+		return -ENOMEM;
+	*pagep = page;
+
+	if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
+		return 0;
+	if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
+		unsigned start = pos & (PAGE_CACHE_SIZE - 1);
+		unsigned end = start + len;
+
+		/* Reading beyond i_size is simple: memset to zero */
+		zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
+		return 0;
+	}
+	return logfs_readpage_nolock(page);
+}
+
+static int logfs_write_end(struct file *file, struct address_space *mapping,
+		loff_t pos, unsigned len, unsigned copied, struct page *page,
+		void *fsdata)
+{
+	struct inode *inode = mapping->host;
+	pgoff_t index = page->index;
+	unsigned start = pos & (PAGE_CACHE_SIZE - 1);
+	unsigned end = start + copied;
+	int ret = 0;
+
+	BUG_ON(PAGE_CACHE_SIZE != inode->i_sb->s_blocksize);
+	BUG_ON(page->index > I3_BLOCKS);
+
+	if (copied < len) {
+		/*
+		 * Short write of a non-initialized paged.  Just tell userspace
+		 * to retry the entire page.
+		 */
+		if (!PageUptodate(page)) {
+			copied = 0;
+			goto out;
+		}
+	}
+	if (copied == 0)
+		goto out; /* FIXME: do we need to update inode? */
+
+	if (i_size_read(inode) < (index << PAGE_CACHE_SHIFT) + end) {
+		i_size_write(inode, (index << PAGE_CACHE_SHIFT) + end);
+		mark_inode_dirty_sync(inode);
+	}
+
+	SetPageUptodate(page);
+	if (!PageDirty(page)) {
+		if (!get_page_reserve(inode, page))
+			__set_page_dirty_nobuffers(page);
+		else
+			ret = logfs_write_buf(inode, page, WF_LOCK);
+	}
+out:
+	unlock_page(page);
+	page_cache_release(page);
+	return ret ? ret : copied;
+}
+
+int logfs_readpage(struct file *file, struct page *page)
+{
+	int ret;
+
+	ret = logfs_readpage_nolock(page);
+	unlock_page(page);
+	return ret;
+}
+
+/* Clear the page's dirty flag in the radix tree. */
+/* TODO: mucking with PageWriteback is silly.  Add a generic function to clear
+ * the dirty bit from the radix tree for filesystems that don't have to wait
+ * for page writeback to finish (i.e. any compressing filesystem).
+ */
+static void clear_radix_tree_dirty(struct page *page)
+{
+	BUG_ON(PagePrivate(page) || page->private);
+	set_page_writeback(page);
+	end_page_writeback(page);
+}
+
+static int __logfs_writepage(struct page *page)
+{
+	struct inode *inode = page->mapping->host;
+	int err;
+
+	err = logfs_write_buf(inode, page, WF_LOCK);
+	if (err)
+		set_page_dirty(page);
+	else
+		clear_radix_tree_dirty(page);
+	unlock_page(page);
+	return err;
+}
+
+static int logfs_writepage(struct page *page, struct writeback_control *wbc)
+{
+	struct inode *inode = page->mapping->host;
+	loff_t i_size = i_size_read(inode);
+	pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
+	unsigned offset;
+	u64 bix;
+	level_t level;
+
+	log_file("logfs_writepage(%lx, %lx, %p)\n", inode->i_ino, page->index,
+			page);
+
+	logfs_unpack_index(page->index, &bix, &level);
+
+	/* Indirect blocks are never truncated */
+	if (level != 0)
+		return __logfs_writepage(page);
+
+	/*
+	 * TODO: everything below is a near-verbatim copy of nobh_writepage().
+	 * The relevant bits should be factored out after logfs is merged.
+	 */
+
+	/* Is the page fully inside i_size? */
+	if (bix < end_index)
+		return __logfs_writepage(page);
+
+	 /* Is the page fully outside i_size? (truncate in progress) */
+	offset = i_size & (PAGE_CACHE_SIZE-1);
+	if (bix > end_index || offset == 0) {
+		unlock_page(page);
+		return 0; /* don't care */
+	}
+
+	/*
+	 * The page straddles i_size.  It must be zeroed out on each and every
+	 * writepage invokation because it may be mmapped.  "A file is mapped
+	 * in multiples of the page size.  For a file that is not a multiple of
+	 * the  page size, the remaining memory is zeroed when mapped, and
+	 * writes to that region are not written out to the file."
+	 */
+	zero_user_segment(page, offset, PAGE_CACHE_SIZE);
+	return __logfs_writepage(page);
+}
+
+static void logfs_invalidatepage(struct page *page, unsigned long offset)
+{
+	move_page_to_btree(page);
+	BUG_ON(PagePrivate(page) || page->private);
+}
+
+static int logfs_releasepage(struct page *page, gfp_t only_xfs_uses_this)
+{
+	return 0; /* None of these are easy to release */
+}
+
+
+int logfs_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
+		unsigned long arg)
+{
+	struct logfs_inode *li = logfs_inode(inode);
+	unsigned int oldflags, flags;
+	int err;
+
+	switch (cmd) {
+	case FS_IOC_GETFLAGS:
+		flags = li->li_flags & LOGFS_FL_USER_VISIBLE;
+		return put_user(flags, (int __user *)arg);
+	case FS_IOC_SETFLAGS:
+		if (IS_RDONLY(inode))
+			return -EROFS;
+
+		if (!is_owner_or_cap(inode))
+			return -EACCES;
+
+		err = get_user(flags, (int __user *)arg);
+		if (err)
+			return err;
+
+		mutex_lock(&inode->i_mutex);
+		oldflags = li->li_flags;
+		flags &= LOGFS_FL_USER_MODIFIABLE;
+		flags |= oldflags & ~LOGFS_FL_USER_MODIFIABLE;
+		li->li_flags = flags;
+		mutex_unlock(&inode->i_mutex);
+
+		inode->i_ctime = CURRENT_TIME;
+		mark_inode_dirty_sync(inode);
+		return 0;
+
+	default:
+		return -ENOTTY;
+	}
+}
+
+int logfs_fsync(struct file *file, struct dentry *dentry, int datasync)
+{
+	struct super_block *sb = dentry->d_inode->i_sb;
+	struct logfs_super *super = logfs_super(sb);
+
+	/* FIXME: write anchor */
+	super->s_devops->sync(sb);
+	return 0;
+}
+
+static int logfs_setattr(struct dentry *dentry, struct iattr *attr)
+{
+	struct inode *inode = dentry->d_inode;
+	int err = 0;
+
+	if (attr->ia_valid & ATTR_SIZE)
+		err = logfs_truncate(inode, attr->ia_size);
+	attr->ia_valid &= ~ATTR_SIZE;
+
+	if (!err)
+		err = inode_change_ok(inode, attr);
+	if (!err)
+		err = inode_setattr(inode, attr);
+	return err;
+}
+
+const struct inode_operations logfs_reg_iops = {
+	.setattr	= logfs_setattr,
+};
+
+const struct file_operations logfs_reg_fops = {
+	.aio_read	= generic_file_aio_read,
+	.aio_write	= generic_file_aio_write,
+	.fsync		= logfs_fsync,
+	.ioctl		= logfs_ioctl,
+	.llseek		= generic_file_llseek,
+	.mmap		= generic_file_readonly_mmap,
+	.open		= generic_file_open,
+	.read		= do_sync_read,
+	.write		= do_sync_write,
+};
+
+const struct address_space_operations logfs_reg_aops = {
+	.invalidatepage	= logfs_invalidatepage,
+	.readpage	= logfs_readpage,
+	.releasepage	= logfs_releasepage,
+	.set_page_dirty	= __set_page_dirty_nobuffers,
+	.writepage	= logfs_writepage,
+	.writepages	= generic_writepages,
+	.write_begin	= logfs_write_begin,
+	.write_end	= logfs_write_end,
+};

fs/logfs/inode.c

+/*
+ * fs/logfs/inode.c	- inode handling code
+ *
+ * As should be obvious for Linux kernel code, license is GPLv2
+ *
+ * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
+ */
+#include "logfs.h"
+#include <linux/writeback.h>
+#include <linux/backing-dev.h>
+
+/*
+ * How soon to reuse old inode numbers?  LogFS doesn't store deleted inodes
+ * on the medium.  It therefore also lacks a method to store the previous
+ * generation number for deleted inodes.  Instead a single generation number
+ * is stored which will be used for new inodes.  Being just a 32bit counter,
+ * this can obvious wrap relatively quickly.  So we only reuse inodes if we
+ * know that a fair number of inodes can be created before we have to increment
+ * the generation again - effectively adding some bits to the counter.
+ * But being too aggressive here means we keep a very large and very sparse
+ * inode file, wasting space on indirect blocks.
+ * So what is a good value?  Beats me.  64k seems moderately bad on both
+ * fronts, so let's use that for now...
+ *
+ * NFS sucks, as everyone already knows.
+ */
+#define INOS_PER_WRAP (0x10000)
+
+/*
+ * Logfs' requirement to read inodes for garbage collection makes life a bit
+ * harder.  GC may have to read inodes that are in I_FREEING state, when they
+ * are being written out - and waiting for GC to make progress, naturally.
+ *
+ * So we cannot just call iget() or some variant of it, but first have to check
+ * wether the inode in question might be in I_FREEING state.  Therefore we
+ * maintain our own per-sb list of "almost deleted" inodes and check against
+ * that list first.  Normally this should be at most 1-2 entries long.
+ *
+ * Also, inodes have logfs-specific reference counting on top of what the vfs
+ * does.  When .destroy_inode is called, normally the reference count will drop
+ * to zero and the inode gets deleted.  But if GC accessed the inode, its
+ * refcount will remain nonzero and final deletion will have to wait.
+ *
+ * As a result we have two sets of functions to get/put inodes:
+ * logfs_safe_iget/logfs_safe_iput	- safe to call from GC context
+ * logfs_iget/iput			- normal version
+ */
+static struct kmem_cache *logfs_inode_cache;
+
+static DEFINE_SPINLOCK(logfs_inode_lock);
+
+static void logfs_inode_setops(struct inode *inode)
+{
+	switch (inode->i_mode & S_IFMT) {
+	case S_IFDIR:
+		inode->i_op = &logfs_dir_iops;
+		inode->i_fop = &logfs_dir_fops;
+		inode->i_mapping->a_ops = &logfs_reg_aops;
+		break;
+	case S_IFREG:
+		inode->i_op = &logfs_reg_iops;
+		inode->i_fop = &logfs_reg_fops;
+		inode->i_mapping->a_ops = &logfs_reg_aops;
+		break;
+	case S_IFLNK:
+		inode->i_op = &logfs_symlink_iops;
+		inode->i_mapping->a_ops = &logfs_reg_aops;
+		break;
+	case S_IFSOCK:	/* fall through */
+	case S_IFBLK:	/* fall through */
+	case S_IFCHR:	/* fall through */
+	case S_IFIFO:
+		init_special_inode(inode, inode->i_mode, inode->i_rdev);
+		break;
+	default:
+		BUG();
+	}
+}
+
+static struct inode *__logfs_iget(struct super_block *sb, ino_t ino)
+{
+	struct inode *inode = iget_locked(sb, ino);
+	int err;
+
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+	if (!(inode->i_state & I_NEW))
+		return inode;
+
+	err = logfs_read_inode(inode);
+	if (err || inode->i_nlink == 0) {
+		/* inode->i_nlink == 0 can be true when called from
+		 * block validator */
+		/* set i_nlink to 0 to prevent caching */
+		inode->i_nlink = 0;
+		logfs_inode(inode)->li_flags |= LOGFS_IF_ZOMBIE;
+		iget_failed(inode);
+		if (!err)
+			err = -ENOENT;
+		return ERR_PTR(err);
+	}
+
+	logfs_inode_setops(inode);
+	unlock_new_inode(inode);
+	return inode;
+}
+
+struct inode *logfs_iget(struct super_block *sb, ino_t ino)
+{
+	BUG_ON(ino == LOGFS_INO_MASTER);
+	BUG_ON(ino == LOGFS_INO_SEGFILE);
+	return __logfs_iget(sb, ino);
+}
+
+/*
+ * is_cached is set to 1 if we hand out a cached inode, 0 otherwise.
+ * this allows logfs_iput to do the right thing later
+ */
+struct inode *logfs_safe_iget(struct super_block *sb, ino_t ino, int *is_cached)
+{
+	struct logfs_super *super = logfs_super(sb);
+	struct logfs_inode *li;
+
+	if (ino == LOGFS_INO_MASTER)
+		return super->s_master_inode;
+	if (ino == LOGFS_INO_SEGFILE)
+		return super->s_segfile_inode;
+
+	spin_lock(&logfs_inode_lock);
+	list_for_each_entry(li, &super->s_freeing_list, li_freeing_list)
+		if (li->vfs_inode.i_ino == ino) {
+			li->li_refcount++;
+			spin_unlock(&logfs_inode_lock);
+			*is_cached = 1;
+			return &li->vfs_inode;
+		}
+	spin_unlock(&logfs_inode_lock);
+
+	*is_cached = 0;
+	return __logfs_iget(sb, ino);
+}
+
+static void __logfs_destroy_inode(struct inode *inode)
+{
+	struct logfs_inode *li = logfs_inode(inode);
+
+	BUG_ON(li->li_block);
+	list_del(&li->li_freeing_list);
+	kmem_cache_free(logfs_inode_cache, li);
+}
+
+static void logfs_destroy_inode(struct inode *inode)
+{
+	struct logfs_inode *li = logfs_inode(inode);
+
+	BUG_ON(list_empty(&li->li_freeing_list));
+	spin_lock(&logfs_inode_lock);
+	li->li_refcount--;
+	if (li->li_refcount == 0)
+		__logfs_destroy_inode(inode);
+	spin_unlock(&logfs_inode_lock);
+}
+
+void logfs_safe_iput(struct inode *inode, int is_cached)
+{
+	if (inode->i_ino == LOGFS_INO_MASTER)
+		return;
+	if (inode->i_ino == LOGFS_INO_SEGFILE)
+		return;
+
+	if (is_cached) {
+		logfs_destroy_inode(inode);
+		return;
+	}
+
+	iput(inode);
+}
+
+static void logfs_init_inode(struct super_block *sb, struct inode *inode)
+{
+	struct logfs_inode *li = logfs_inode(inode);
+	int i;
+
+	li->li_flags	= 0;
+	li->li_height	= 0;
+	li->li_used_bytes = 0;
+	li->li_block	= NULL;
+	inode->i_uid	= 0;
+	inode->i_gid	= 0;
+	inode->i_size	= 0;
+	inode->i_blocks	= 0;
+	inode->i_ctime	= CURRENT_TIME;
+	inode->i_mtime	= CURRENT_TIME;
+	inode->i_nlink	= 1;
+	INIT_LIST_HEAD(&li->li_freeing_list);
+
+	for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
+		li->li_data[i] = 0;
+
+	return;
+}
+
+static struct inode *logfs_alloc_inode(struct super_block *sb)
+{
+	struct logfs_inode *li;
+
+	li = kmem_cache_alloc(logfs_inode_cache, GFP_NOFS);
+	if (!li)
+		return NULL;
+	logfs_init_inode(sb, &li->vfs_inode);
+	return &li->vfs_inode;
+}
+
+/*
+ * In logfs inodes are written to an inode file.  The inode file, like any
+ * other file, is managed with a inode.  The inode file's inode, aka master
+ * inode, requires special handling in several respects.  First, it cannot be
+ * written to the inode file, so it is stored in the journal instead.
+ *
+ * Secondly, this inode cannot be written back and destroyed before all other
+ * inodes have been written.  The ordering is important.  Linux' VFS is happily
+ * unaware of the ordering constraint and would ordinarily destroy the master
+ * inode at umount time while other inodes are still in use and dirty.  Not
+ * good.
+ *
+ * So logfs makes sure the master inode is not written until all other inodes
+ * have been destroyed.  Sadly, this method has another side-effect.  The VFS
+ * will notice one remaining inode and print a frightening warning message.
+ * Worse, it is impossible to judge whether such a warning was caused by the
+ * master inode or any other inodes have leaked as well.
+ *
+ * Our attempt of solving this is with logfs_new_meta_inode() below.  Its
+ * purpose is to create a new inode that will not trigger the warning if such
+ * an inode is still in use.  An ugly hack, no doubt.  Suggections for
+ * improvement are welcome.
+ */
+struct inode *logfs_new_meta_inode(struct super_block *sb, u64 ino)
+{
+	struct inode *inode;
+
+	inode = logfs_alloc_inode(sb);
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+
+	inode->i_mode = S_IFREG;
+	inode->i_ino = ino;
+	inode->i_sb = sb;
+
+	/* This is a blatant copy of alloc_inode code.  We'd need alloc_inode
+	 * to be nonstatic, alas. */
+	{
+		struct address_space * const mapping = &inode->i_data;
+
+		mapping->a_ops = &logfs_reg_aops;
+		mapping->host = inode;
+		mapping->flags = 0;
+		mapping_set_gfp_mask(mapping, GFP_NOFS);
+		mapping->assoc_mapping = NULL;
+		mapping->backing_dev_info = &default_backing_dev_info;
+		inode->i_mapping = mapping;
+		inode->i_nlink = 1;
+	}
+
+	return inode;
+}
+
+struct inode *logfs_read_meta_inode(struct super_block *sb, u64 ino)
+{
+	struct inode *inode;
+	int err;
+
+	inode = logfs_new_meta_inode(sb, ino);
+	if (IS_ERR(inode))
+		return inode;
+
+	err = logfs_read_inode(inode);
+	if (err) {
+		destroy_meta_inode(inode);
+		return ERR_PTR(err);
+	}
+	logfs_inode_setops(inode);
+	return inode;
+}
+
+static int logfs_write_inode(struct inode *inode, int do_sync)
+{
+	int ret;
+	long flags = WF_LOCK;
+
+	/* Can only happen if creat() failed.  Safe to skip. */
+	if (logfs_inode(inode)->li_flags & LOGFS_IF_STILLBORN)
+		return 0;
+
+	ret = __logfs_write_inode(inode, flags);
+	LOGFS_BUG_ON(ret, inode->i_sb);
+	return ret;
+}
+
+void destroy_meta_inode(struct inode *inode)
+{
+	if (inode) {
+		if (inode->i_data.nrpages)
+			truncate_inode_pages(&inode->i_data, 0);
+		logfs_clear_inode(inode);
+		kmem_cache_free(logfs_inode_cache, logfs_inode(inode));
+	}
+}
+
+/* called with inode_lock held */
+static void logfs_drop_inode(struct inode *inode)
+{
+	struct logfs_super *super = logfs_super(inode->i_sb);
+	struct logfs_inode *li = logfs_inode(inode);
+
+	spin_lock(&logfs_inode_lock);
+	list_move(&li->li_freeing_list, &super->s_freeing_list);
+	spin_unlock(&logfs_inode_lock);
+	generic_drop_inode(inode);
+}
+
+static void logfs_set_ino_generation(struct super_block *sb,
+		struct inode *inode)
+{
+	struct logfs_super *super = logfs_super(sb);
+	u64 ino;
+
+	mutex_lock(&super->s_journal_mutex);
+	ino = logfs_seek_hole(super->s_master_inode, super->s_last_ino);
+	super->s_last_ino = ino;
+	super->s_inos_till_wrap--;
+	if (super->s_inos_till_wrap < 0) {
+		super->s_last_ino = LOGFS_RESERVED_INOS;
+		super->s_generation++;
+		super->s_inos_till_wrap = INOS_PER_WRAP;
+	}
+	inode->i_ino = ino;
+	inode->i_generation = super->s_generation;
+	mutex_unlock(&super->s_journal_mutex);
+}
+
+struct inode *logfs_new_inode(struct inode *dir, int mode)
+{
+	struct super_block *sb = dir->i_sb;
+	struct inode *inode;
+
+	inode = new_inode(sb);
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+
+	logfs_init_inode(sb, inode);
+
+	/* inherit parent flags */
+	logfs_inode(inode)->li_flags |=
+		logfs_inode(dir)->li_flags & LOGFS_FL_INHERITED;
+
+	inode->i_mode = mode;
+	logfs_set_ino_generation(sb, inode);
+
+	inode->i_uid = current_fsuid();
+	inode->i_gid = current_fsgid();
+	if (dir->i_mode & S_ISGID) {
+		inode->i_gid = dir->i_gid;
+		if (S_ISDIR(mode))
+			inode->i_mode |= S_ISGID;
+	}
+
+	logfs_inode_setops(inode);
+	insert_inode_hash(inode);
+
+	return inode;
+}
+
+static void logfs_init_once(void *_li)
+{
+	struct logfs_inode *li = _li;
+	int i;
+
+	li->li_flags = 0;
+	li->li_used_bytes = 0;
+	li->li_refcount = 1;
+	for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
+		li->li_data[i] = 0;
+	inode_init_once(&li->vfs_inode);
+}
+
+static int logfs_sync_fs(struct super_block *sb, int wait)
+{
+	/* FIXME: write anchor */
+	logfs_super(sb)->s_devops->sync(sb);
+	return 0;
+}
+
+const struct super_operations logfs_super_operations = {
+	.alloc_inode	= logfs_alloc_inode,
+	.clear_inode	= logfs_clear_inode,
+	.delete_inode	= logfs_delete_inode,
+	.destroy_inode	= logfs_destroy_inode,
+	.drop_inode	= logfs_drop_inode,
+	.write_inode	= logfs_write_inode,
+	.statfs		= logfs_statfs,
+	.sync_fs	= logfs_sync_fs,
+};
+
+int logfs_init_inode_cache(void)
+{
+	logfs_inode_cache = kmem_cache_create("logfs_inode_cache",
+			sizeof(struct logfs_inode), 0, SLAB_RECLAIM_ACCOUNT,
+			logfs_init_once);
+	if (!logfs_inode_cache)
+		return -ENOMEM;
+	return 0;
+}
+
+void logfs_destroy_inode_cache(void)
+{
+	kmem_cache_destroy(logfs_inode_cache);
+}

include/linux/btree-128.h

+extern struct btree_geo btree_geo128;
+
+struct btree_head128 { struct btree_head h; };
+
+static inline void btree_init_mempool128(struct btree_head128 *head,
+					 mempool_t *mempool)
+{
+	btree_init_mempool(&head->h, mempool);
+}
+
+static inline int btree_init128(struct btree_head128 *head)
+{
+	return btree_init(&head->h);
+}
+
+static inline void btree_destroy128(struct btree_head128 *head)
+{
+	btree_destroy(&head->h);
+}
+
+static inline void *btree_lookup128(struct btree_head128 *head, u64 k1, u64 k2)
+{
+	u64 key[2] = {k1, k2};
+	return btree_lookup(&head->h, &btree_geo128, (unsigned long *)&key);
+}
+
+static inline void *btree_get_prev128(struct btree_head128 *head,
+				      u64 *k1, u64 *k2)
+{
+	u64 key[2] = {*k1, *k2};
+	void *val;
+
+	val = btree_get_prev(&head->h, &btree_geo128,
+			     (unsigned long *)&key);
+	*k1 = key[0];
+	*k2 = key[1];
+	return val;
+}
+
+static inline int btree_insert128(struct btree_head128 *head, u64 k1, u64 k2,
+				  void *val, gfp_t gfp)
+{
+	u64 key[2] = {k1, k2};
+	return btree_insert(&head->h, &btree_geo128,
+			    (unsigned long *)&key, val, gfp);
+}
+
+static inline int btree_update128(struct btree_head128 *head, u64 k1, u64 k2,
+				  void *val)
+{
+	u64 key[2] = {k1, k2};
+	return btree_update(&head->h, &btree_geo128,
+			    (unsigned long *)&key, val);
+}
+
+static inline void *btree_remove128(struct btree_head128 *head, u64 k1, u64 k2)
+{
+	u64 key[2] = {k1, k2};
+	return btree_remove(&head->h, &btree_geo128, (unsigned long *)&key);
+}
+
+static inline void *btree_last128(struct btree_head128 *head, u64 *k1, u64 *k2)
+{
+	u64 key[2];
+	void *val;
+
+	val = btree_last(&head->h, &btree_geo128, (unsigned long *)&key[0]);
+	if (val) {
+		*k1 = key[0];
+		*k2 = key[1];
+	}
+
+	return val;
+}
+
+static inline int btree_merge128(struct btree_head128 *target,
+				 struct btree_head128 *victim,
+				 gfp_t gfp)
+{
+	return btree_merge(&target->h, &victim->h, &btree_geo128, gfp);
+}
+
+void visitor128(void *elem, unsigned long opaque, unsigned long *__key,
+		size_t index, void *__func);
+
+typedef void (*visitor128_t)(void *elem, unsigned long opaque,
+			     u64 key1, u64 key2, size_t index);
+
+static inline size_t btree_visitor128(struct btree_head128 *head,
+				      unsigned long opaque,
+				      visitor128_t func2)
+{
+	return btree_visitor(&head->h, &btree_geo128, opaque,
+			     visitor128, func2);
+}
+
+static inline size_t btree_grim_visitor128(struct btree_head128 *head,
+					   unsigned long opaque,
+					   visitor128_t func2)
+{
+	return btree_grim_visitor(&head->h, &btree_geo128, opaque,
+				  visitor128, func2);
+}
+
+#define btree_for_each_safe128(head, k1, k2, val)	\
+	for (val = btree_last128(head, &k1, &k2);	\
+	     val;					\
+	     val = btree_get_prev128(head, &k1, &k2))
+

include/linux/btree-type.h

+#define __BTREE_TP(pfx, type, sfx)	pfx ## type ## sfx
+#define _BTREE_TP(pfx, type, sfx)	__BTREE_TP(pfx, type, sfx)
+#define BTREE_TP(pfx)			_BTREE_TP(pfx, BTREE_TYPE_SUFFIX,)
+#define BTREE_FN(name)			BTREE_TP(btree_ ## name)
+#define BTREE_TYPE_HEAD			BTREE_TP(struct btree_head)
+#define VISITOR_FN			BTREE_TP(visitor)
+#define VISITOR_FN_T			_BTREE_TP(visitor, BTREE_TYPE_SUFFIX, _t)
+
+BTREE_TYPE_HEAD {
+	struct btree_head h;
+};
+
+static inline void BTREE_FN(init_mempool)(BTREE_TYPE_HEAD *head,
+					  mempool_t *mempool)
+{
+	btree_init_mempool(&head->h, mempool);
+}
+
+static inline int BTREE_FN(init)(BTREE_TYPE_HEAD *head)
+{
+	return btree_init(&head->h);
+}
+
+static inline void BTREE_FN(destroy)(BTREE_TYPE_HEAD *head)
+{
+	btree_destroy(&head->h);
+}
+
+static inline int BTREE_FN(merge)(BTREE_TYPE_HEAD *target,
+				  BTREE_TYPE_HEAD *victim,
+				  gfp_t gfp)
+{
+	return btree_merge(&target->h, &victim->h, BTREE_TYPE_GEO, gfp);
+}
+
+#if (BITS_PER_LONG > BTREE_TYPE_BITS)
+static inline void *BTREE_FN(lookup)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE key)
+{
+	unsigned long _key = key;
+	return btree_lookup(&head->h, BTREE_TYPE_GEO, &_key);
+}
+
+static inline int BTREE_FN(insert)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE key,
+				   void *val, gfp_t gfp)
+{
+	unsigned long _key = key;
+	return btree_insert(&head->h, BTREE_TYPE_GEO, &_key, val, gfp);
+}
+
+static inline int BTREE_FN(update)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE key,
+		void *val)
+{
+	unsigned long _key = key;
+	return btree_update(&head->h, BTREE_TYPE_GEO, &_key, val);
+}
+
+static inline void *BTREE_FN(remove)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE key)
+{
+	unsigned long _key = key;
+	return btree_remove(&head->h, BTREE_TYPE_GEO, &_key);
+}
+
+static inline void *BTREE_FN(last)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE *key)
+{
+	unsigned long _key;
+	void *val = btree_last(&head->h, BTREE_TYPE_GEO, &_key);
+	if (val)
+		*key = _key;
+	return val;
+}
+
+static inline void *BTREE_FN(get_prev)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE *key)
+{
+	unsigned long _key = *key;
+	void *val = btree_get_prev(&head->h, BTREE_TYPE_GEO, &_key);
+	if (val)
+		*key = _key;
+	return val;
+}
+#else
+static inline void *BTREE_FN(lookup)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE key)
+{
+	return btree_lookup(&head->h, BTREE_TYPE_GEO, (unsigned long *)&key);
+}
+
+static inline int BTREE_FN(insert)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE key,
+			   void *val, gfp_t gfp)
+{
+	return btree_insert(&head->h, BTREE_TYPE_GEO, (unsigned long *)&key,
+			    val, gfp);
+}
+
+static inline int BTREE_FN(update)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE key,
+		void *val)
+{
+	return btree_update(&head->h, BTREE_TYPE_GEO, (unsigned long *)&key, val);
+}
+
+static inline void *BTREE_FN(remove)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE key)
+{
+	return btree_remove(&head->h, BTREE_TYPE_GEO, (unsigned long *)&key);
+}
+
+static inline void *BTREE_FN(last)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE *key)
+{
+	return btree_last(&head->h, BTREE_TYPE_GEO, (unsigned long *)key);
+}
+
+static inline void *BTREE_FN(get_prev)(BTREE_TYPE_HEAD *head, BTREE_KEYTYPE *key)
+{
+	return btree_get_prev(&head->h, BTREE_TYPE_GEO, (unsigned long *)key);
+}
+#endif
+
+void VISITOR_FN(void *elem, unsigned long opaque, unsigned long *key,
+		size_t index, void *__func);
+
+typedef void (*VISITOR_FN_T)(void *elem, unsigned long opaque,
+			     BTREE_KEYTYPE key, size_t index);
+
+static inline size_t BTREE_FN(visitor)(BTREE_TYPE_HEAD *head,
+				       unsigned long opaque,
+				       VISITOR_FN_T func2)
+{
+	return btree_visitor(&head->h, BTREE_TYPE_GEO, opaque,
+			     visitorl, func2);
+}
+
+static inline size_t BTREE_FN(grim_visitor)(BTREE_TYPE_HEAD *head,
+					    unsigned long opaque,
+					    VISITOR_FN_T func2)
+{
+	return btree_grim_visitor(&head->h, BTREE_TYPE_GEO, opaque,
+				  visitorl, func2);
+}
+
+#undef VISITOR_FN
+#undef VISITOR_FN_T
+#undef __BTREE_TP
+#undef _BTREE_TP
+#undef BTREE_TP
+#undef BTREE_FN
+#undef BTREE_TYPE_HEAD
+#undef BTREE_TYPE_SUFFIX
+#undef BTREE_TYPE_GEO
+#undef BTREE_KEYTYPE
+#undef BTREE_TYPE_BITS

lib/Kconfig

@@ -156,6 +156,9 @@ config TEXTSEARCH_BM
 config TEXTSEARCH_FSM
 	tristate
 
+config BTREE
+	boolean
+
 config HAS_IOMEM
 	boolean
 	depends on !NO_IOMEM