[deliverable/linux.git] / fs / logfs / super.c

/*
 * fs/logfs/super.c
 *
 * As should be obvious for Linux kernel code, license is GPLv2
 *
 * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
 *
 * Generally contains mount/umount code and also serves as a dump area for
 * any functions that don't fit elsewhere and neither justify a file of their
 * own.
 */
#include "logfs.h"
#include <linux/bio.h>
#include <linux/slab.h>
#include <linux/blkdev.h>
#include <linux/mtd/mtd.h>
#include <linux/statfs.h>
#include <linux/buffer_head.h>

static DEFINE_MUTEX(emergency_mutex);
static struct page *emergency_page;

struct page *emergency_read_begin(struct address_space *mapping, pgoff_t index)
{
	filler_t *filler = (filler_t *)mapping->a_ops->readpage;
	struct page *page;
	int err;

	page = read_cache_page(mapping, index, filler, NULL);
	if (page)
		return page;

	/* No more pages available, switch to emergency page */
	printk(KERN_INFO"Logfs: Using emergency page\n");
	mutex_lock(&emergency_mutex);
	err = filler(NULL, emergency_page);
	if (err) {
		mutex_unlock(&emergency_mutex);
		printk(KERN_EMERG"Logfs: Error reading emergency page\n");
		return ERR_PTR(err);
	}
	return emergency_page;
}

void emergency_read_end(struct page *page)
{
	if (page == emergency_page)
		mutex_unlock(&emergency_mutex);
	else
		page_cache_release(page);
}

static void dump_segfile(struct super_block *sb)
{
	struct logfs_super *super = logfs_super(sb);
	struct logfs_segment_entry se;
	u32 segno;

	for (segno = 0; segno < super->s_no_segs; segno++) {
		logfs_get_segment_entry(sb, segno, &se);
		printk("%3x: %6x %8x", segno, be32_to_cpu(se.ec_level),
				be32_to_cpu(se.valid));
		if (++segno < super->s_no_segs) {
			logfs_get_segment_entry(sb, segno, &se);
			printk(" %6x %8x", be32_to_cpu(se.ec_level),
					be32_to_cpu(se.valid));
		}
		if (++segno < super->s_no_segs) {
			logfs_get_segment_entry(sb, segno, &se);
			printk(" %6x %8x", be32_to_cpu(se.ec_level),
					be32_to_cpu(se.valid));
		}
		if (++segno < super->s_no_segs) {
			logfs_get_segment_entry(sb, segno, &se);
			printk(" %6x %8x", be32_to_cpu(se.ec_level),
					be32_to_cpu(se.valid));
		}
		printk("\n");
	}
}

/*
 * logfs_crash_dump - dump debug information to device
 *
 * The LogFS superblock only occupies part of a segment.  This function will
 * write as much debug information as it can gather into the spare space.
 */
void logfs_crash_dump(struct super_block *sb)
{
	dump_segfile(sb);
}

/*
 * FIXME: There should be a reserve for root, similar to ext2.
 */
int logfs_statfs(struct dentry *dentry, struct kstatfs *stats)
{
	struct super_block *sb = dentry->d_sb;
	struct logfs_super *super = logfs_super(sb);

	stats->f_type		= LOGFS_MAGIC_U32;
	stats->f_bsize		= sb->s_blocksize;
	stats->f_blocks		= super->s_size >> LOGFS_BLOCK_BITS >> 3;
	stats->f_bfree		= super->s_free_bytes >> sb->s_blocksize_bits;
	stats->f_bavail		= super->s_free_bytes >> sb->s_blocksize_bits;
	stats->f_files		= 0;
	stats->f_ffree		= 0;
	stats->f_namelen	= LOGFS_MAX_NAMELEN;
	return 0;
}

static int logfs_sb_set(struct super_block *sb, void *_super)
{
	struct logfs_super *super = _super;

	sb->s_fs_info = super;
	sb->s_mtd = super->s_mtd;
	sb->s_bdev = super->s_bdev;
#ifdef CONFIG_BLOCK
	if (sb->s_bdev)
		sb->s_bdi = &bdev_get_queue(sb->s_bdev)->backing_dev_info;
#endif
#ifdef CONFIG_MTD
	if (sb->s_mtd)
		sb->s_bdi = sb->s_mtd->backing_dev_info;
#endif
	return 0;
}

static int logfs_sb_test(struct super_block *sb, void *_super)
{
	struct logfs_super *super = _super;
	struct mtd_info *mtd = super->s_mtd;

	if (mtd && sb->s_mtd == mtd)
		return 1;
	if (super->s_bdev && sb->s_bdev == super->s_bdev)
		return 1;
	return 0;
}

static void set_segment_header(struct logfs_segment_header *sh, u8 type,
		u8 level, u32 segno, u32 ec)
{
	sh->pad = 0;
	sh->type = type;
	sh->level = level;
	sh->segno = cpu_to_be32(segno);
	sh->ec = cpu_to_be32(ec);
	sh->gec = cpu_to_be64(segno);
	sh->crc = logfs_crc32(sh, LOGFS_SEGMENT_HEADERSIZE, 4);
}

static void logfs_write_ds(struct super_block *sb, struct logfs_disk_super *ds,
		u32 segno, u32 ec)
{
	struct logfs_super *super = logfs_super(sb);
	struct logfs_segment_header *sh = &ds->ds_sh;
	int i;

	memset(ds, 0, sizeof(*ds));
	set_segment_header(sh, SEG_SUPER, 0, segno, ec);

	ds->ds_ifile_levels	= super->s_ifile_levels;
	ds->ds_iblock_levels	= super->s_iblock_levels;
	ds->ds_data_levels	= super->s_data_levels; /* XXX: Remove */
	ds->ds_segment_shift	= super->s_segshift;
	ds->ds_block_shift	= sb->s_blocksize_bits;
	ds->ds_write_shift	= super->s_writeshift;
	ds->ds_filesystem_size	= cpu_to_be64(super->s_size);
	ds->ds_segment_size	= cpu_to_be32(super->s_segsize);
	ds->ds_bad_seg_reserve	= cpu_to_be32(super->s_bad_seg_reserve);
	ds->ds_feature_incompat	= cpu_to_be64(super->s_feature_incompat);
	ds->ds_feature_ro_compat= cpu_to_be64(super->s_feature_ro_compat);
	ds->ds_feature_compat	= cpu_to_be64(super->s_feature_compat);
	ds->ds_feature_flags	= cpu_to_be64(super->s_feature_flags);
	ds->ds_root_reserve	= cpu_to_be64(super->s_root_reserve);
	ds->ds_speed_reserve	= cpu_to_be64(super->s_speed_reserve);
	journal_for_each(i)
		ds->ds_journal_seg[i] = cpu_to_be32(super->s_journal_seg[i]);
	ds->ds_magic		= cpu_to_be64(LOGFS_MAGIC);
	ds->ds_crc = logfs_crc32(ds, sizeof(*ds),
			LOGFS_SEGMENT_HEADERSIZE + 12);
}

static int write_one_sb(struct super_block *sb,
		struct page *(*find_sb)(struct super_block *sb, u64 *ofs))
{
	struct logfs_super *super = logfs_super(sb);
	struct logfs_disk_super *ds;
	struct logfs_segment_entry se;
	struct page *page;
	u64 ofs;
	u32 ec, segno;
	int err;

	page = find_sb(sb, &ofs);
	if (!page)
		return -EIO;
	ds = page_address(page);
	segno = seg_no(sb, ofs);
	logfs_get_segment_entry(sb, segno, &se);
	ec = be32_to_cpu(se.ec_level) >> 4;
	ec++;
	logfs_set_segment_erased(sb, segno, ec, 0);
	logfs_write_ds(sb, ds, segno, ec);
	err = super->s_devops->write_sb(sb, page);
	page_cache_release(page);
	return err;
}

int logfs_write_sb(struct super_block *sb)
{
	struct logfs_super *super = logfs_super(sb);
	int err;

	/* First superblock */
	err = write_one_sb(sb, super->s_devops->find_first_sb);
	if (err)
		return err;

	/* Last superblock */
	err = write_one_sb(sb, super->s_devops->find_last_sb);
	if (err)
		return err;
	return 0;
}

static int ds_cmp(const void *ds0, const void *ds1)
{
	size_t len = sizeof(struct logfs_disk_super);

	/* We know the segment headers differ, so ignore them */
	len -= LOGFS_SEGMENT_HEADERSIZE;
	ds0 += LOGFS_SEGMENT_HEADERSIZE;
	ds1 += LOGFS_SEGMENT_HEADERSIZE;
	return memcmp(ds0, ds1, len);
}

static int logfs_recover_sb(struct super_block *sb)
{
	struct logfs_super *super = logfs_super(sb);
	struct logfs_disk_super _ds0, *ds0 = &_ds0;
	struct logfs_disk_super _ds1, *ds1 = &_ds1;
	int err, valid0, valid1;

	/* read first superblock */
	err = wbuf_read(sb, super->s_sb_ofs[0], sizeof(*ds0), ds0);
	if (err)
		return err;
	/* read last superblock */
	err = wbuf_read(sb, super->s_sb_ofs[1], sizeof(*ds1), ds1);
	if (err)
		return err;
	valid0 = logfs_check_ds(ds0) == 0;
	valid1 = logfs_check_ds(ds1) == 0;

	if (!valid0 && valid1) {
		printk(KERN_INFO"First superblock is invalid - fixing.\n");
		return write_one_sb(sb, super->s_devops->find_first_sb);
	}
	if (valid0 && !valid1) {
		printk(KERN_INFO"Last superblock is invalid - fixing.\n");
		return write_one_sb(sb, super->s_devops->find_last_sb);
	}
	if (valid0 && valid1 && ds_cmp(ds0, ds1)) {
		printk(KERN_INFO"Superblocks don't match - fixing.\n");
		return logfs_write_sb(sb);
	}
	/* If neither is valid now, something's wrong.  Didn't we properly
	 * check them before?!? */
	BUG_ON(!valid0 && !valid1);
	return 0;
}

static int logfs_make_writeable(struct super_block *sb)
{
	int err;

	err = logfs_open_segfile(sb);
	if (err)
		return err;

	/* Repair any broken superblock copies */
	err = logfs_recover_sb(sb);
	if (err)
		return err;

	/* Check areas for trailing unaccounted data */
	err = logfs_check_areas(sb);
	if (err)
		return err;

	/* Do one GC pass before any data gets dirtied */
	logfs_gc_pass(sb);

	/* after all initializations are done, replay the journal
	 * for rw-mounts, if necessary */
	err = logfs_replay_journal(sb);
	if (err)
		return err;

	return 0;
}

static int logfs_get_sb_final(struct super_block *sb)
{
	struct logfs_super *super = logfs_super(sb);
	struct inode *rootdir;
	int err;

	/* root dir */
	rootdir = logfs_iget(sb, LOGFS_INO_ROOT);
	if (IS_ERR(rootdir))
		goto fail;

	sb->s_root = d_alloc_root(rootdir);
	if (!sb->s_root) {
		iput(rootdir);
		goto fail;
	}

	/* at that point we know that ->put_super() will be called */
	super->s_erase_page = alloc_pages(GFP_KERNEL, 0);
	if (!super->s_erase_page)
		return -ENOMEM;
	memset(page_address(super->s_erase_page), 0xFF, PAGE_SIZE);

	/* FIXME: check for read-only mounts */
	err = logfs_make_writeable(sb);
	if (err) {
		__free_page(super->s_erase_page);
		return err;
	}

	log_super("LogFS: Finished mounting\n");
	return 0;

fail:
	iput(super->s_master_inode);
	iput(super->s_segfile_inode);
	iput(super->s_mapping_inode);
	return -EIO;
}

int logfs_check_ds(struct logfs_disk_super *ds)
{
	struct logfs_segment_header *sh = &ds->ds_sh;

	if (ds->ds_magic != cpu_to_be64(LOGFS_MAGIC))
		return -EINVAL;
	if (sh->crc != logfs_crc32(sh, LOGFS_SEGMENT_HEADERSIZE, 4))
		return -EINVAL;
	if (ds->ds_crc != logfs_crc32(ds, sizeof(*ds),
				LOGFS_SEGMENT_HEADERSIZE + 12))
		return -EINVAL;
	return 0;
}

static struct page *find_super_block(struct super_block *sb)
{
	struct logfs_super *super = logfs_super(sb);
	struct page *first, *last;

	first = super->s_devops->find_first_sb(sb, &super->s_sb_ofs[0]);
	if (!first || IS_ERR(first))
		return NULL;
	last = super->s_devops->find_last_sb(sb, &super->s_sb_ofs[1]);
	if (!last || IS_ERR(last)) {
		page_cache_release(first);
		return NULL;
	}

	if (!logfs_check_ds(page_address(first))) {
		page_cache_release(last);
		return first;
	}

	/* First one didn't work, try the second superblock */
	if (!logfs_check_ds(page_address(last))) {
		page_cache_release(first);
		return last;
	}

	/* Neither worked, sorry folks */
	page_cache_release(first);
	page_cache_release(last);
	return NULL;
}

static int __logfs_read_sb(struct super_block *sb)
{
	struct logfs_super *super = logfs_super(sb);
	struct page *page;
	struct logfs_disk_super *ds;
	int i;

	page = find_super_block(sb);
	if (!page)
		return -EINVAL;

	ds = page_address(page);
	super->s_size = be64_to_cpu(ds->ds_filesystem_size);
	super->s_root_reserve = be64_to_cpu(ds->ds_root_reserve);
	super->s_speed_reserve = be64_to_cpu(ds->ds_speed_reserve);
	super->s_bad_seg_reserve = be32_to_cpu(ds->ds_bad_seg_reserve);
	super->s_segsize = 1 << ds->ds_segment_shift;
	super->s_segmask = (1 << ds->ds_segment_shift) - 1;
	super->s_segshift = ds->ds_segment_shift;
	sb->s_blocksize = 1 << ds->ds_block_shift;
	sb->s_blocksize_bits = ds->ds_block_shift;
	super->s_writesize = 1 << ds->ds_write_shift;
	super->s_writeshift = ds->ds_write_shift;
	super->s_no_segs = super->s_size >> super->s_segshift;
	super->s_no_blocks = super->s_segsize >> sb->s_blocksize_bits;
	super->s_feature_incompat = be64_to_cpu(ds->ds_feature_incompat);
	super->s_feature_ro_compat = be64_to_cpu(ds->ds_feature_ro_compat);
	super->s_feature_compat = be64_to_cpu(ds->ds_feature_compat);
	super->s_feature_flags = be64_to_cpu(ds->ds_feature_flags);

	journal_for_each(i)
		super->s_journal_seg[i] = be32_to_cpu(ds->ds_journal_seg[i]);

	super->s_ifile_levels = ds->ds_ifile_levels;
	super->s_iblock_levels = ds->ds_iblock_levels;
	super->s_data_levels = ds->ds_data_levels;
	super->s_total_levels = super->s_ifile_levels + super->s_iblock_levels
		+ super->s_data_levels;
	page_cache_release(page);
	return 0;
}

static int logfs_read_sb(struct super_block *sb, int read_only)
{
	struct logfs_super *super = logfs_super(sb);
	int ret;

	super->s_btree_pool = mempool_create(32, btree_alloc, btree_free, NULL);
	if (!super->s_btree_pool)
		return -ENOMEM;

	btree_init_mempool64(&super->s_shadow_tree.new, super->s_btree_pool);
	btree_init_mempool64(&super->s_shadow_tree.old, super->s_btree_pool);
	btree_init_mempool32(&super->s_shadow_tree.segment_map,
			super->s_btree_pool);

	ret = logfs_init_mapping(sb);
	if (ret)
		return ret;

	ret = __logfs_read_sb(sb);
	if (ret)
		return ret;

	if (super->s_feature_incompat & ~LOGFS_FEATURES_INCOMPAT)
		return -EIO;
	if ((super->s_feature_ro_compat & ~LOGFS_FEATURES_RO_COMPAT) &&
			!read_only)
		return -EIO;

	ret = logfs_init_rw(sb);
	if (ret)
		return ret;

	ret = logfs_init_areas(sb);
	if (ret)
		return ret;

	ret = logfs_init_gc(sb);
	if (ret)
		return ret;

	ret = logfs_init_journal(sb);
	if (ret)
		return ret;

	return 0;
}

static void logfs_kill_sb(struct super_block *sb)
{
	struct logfs_super *super = logfs_super(sb);

	log_super("LogFS: Start unmounting\n");
	/* Alias entries slow down mount, so evict as many as possible */
	sync_filesystem(sb);
	logfs_write_anchor(sb);

	/*
	 * From this point on alias entries are simply dropped - and any
	 * writes to the object store are considered bugs.
	 */
	super->s_flags |= LOGFS_SB_FLAG_SHUTDOWN;
	log_super("LogFS: Now in shutdown\n");
	generic_shutdown_super(sb);

	BUG_ON(super->s_dirty_used_bytes || super->s_dirty_free_bytes);

	logfs_cleanup_gc(sb);
	logfs_cleanup_journal(sb);
	logfs_cleanup_areas(sb);
	logfs_cleanup_rw(sb);
	if (super->s_erase_page)
		__free_page(super->s_erase_page);
	super->s_devops->put_device(super);
	logfs_mempool_destroy(super->s_btree_pool);
	logfs_mempool_destroy(super->s_alias_pool);
	kfree(super);
	log_super("LogFS: Finished unmounting\n");
}

static struct dentry *logfs_get_sb_device(struct logfs_super *super,
		struct file_system_type *type, int flags)
{
	struct super_block *sb;
	int err = -ENOMEM;
	static int mount_count;

	log_super("LogFS: Start mount %x\n", mount_count++);

	err = -EINVAL;
	sb = sget(type, logfs_sb_test, logfs_sb_set, super);
	if (IS_ERR(sb)) {
		super->s_devops->put_device(super);
		kfree(super);
		return ERR_CAST(sb);
	}

	if (sb->s_root) {
		/* Device is already in use */
		super->s_devops->put_device(super);
		kfree(super);
		return dget(sb->s_root);
	}

	/*
	 * sb->s_maxbytes is limited to 8TB.  On 32bit systems, the page cache
	 * only covers 16TB and the upper 8TB are used for indirect blocks.
	 * On 64bit system we could bump up the limit, but that would make
	 * the filesystem incompatible with 32bit systems.
	 */
	sb->s_maxbytes	= (1ull << 43) - 1;
	sb->s_op	= &logfs_super_operations;
	sb->s_flags	= flags | MS_NOATIME;

	err = logfs_read_sb(sb, sb->s_flags & MS_RDONLY);
	if (err)
		goto err1;

	sb->s_flags |= MS_ACTIVE;
	err = logfs_get_sb_final(sb);
	if (err) {
		deactivate_locked_super(sb);
		return ERR_PTR(err);
	}
	return dget(sb->s_root);

err1:
	/* no ->s_root, no ->put_super() */
	iput(super->s_master_inode);
	iput(super->s_segfile_inode);
	iput(super->s_mapping_inode);
	deactivate_locked_super(sb);
	return ERR_PTR(err);
}

static struct dentry *logfs_mount(struct file_system_type *type, int flags,
		const char *devname, void *data)
{
	ulong mtdnr;
	struct logfs_super *super;
	int err;

	super = kzalloc(sizeof(*super), GFP_KERNEL);
	if (!super)
		return ERR_PTR(-ENOMEM);

	mutex_init(&super->s_dirop_mutex);
	mutex_init(&super->s_object_alias_mutex);
	INIT_LIST_HEAD(&super->s_freeing_list);

	if (!devname)
		err = logfs_get_sb_bdev(super, type, devname);
	else if (strncmp(devname, "mtd", 3))
		err = logfs_get_sb_bdev(super, type, devname);
	else {
		char *garbage;
		mtdnr = simple_strtoul(devname+3, &garbage, 0);
		if (*garbage)
			err = -EINVAL;
		else
			err = logfs_get_sb_mtd(super, mtdnr);
	}

	if (err) {
		kfree(super);
		return ERR_PTR(err);
	}

	return logfs_get_sb_device(super, type, flags);
}

static struct file_system_type logfs_fs_type = {
	.owner		= THIS_MODULE,
	.name		= "logfs",
	.mount		= logfs_mount,
	.kill_sb	= logfs_kill_sb,
	.fs_flags	= FS_REQUIRES_DEV,

};

static int __init logfs_init(void)
{
	int ret;

	emergency_page = alloc_pages(GFP_KERNEL, 0);
	if (!emergency_page)
		return -ENOMEM;

	ret = logfs_compr_init();
	if (ret)
		goto out1;

	ret = logfs_init_inode_cache();
	if (ret)
		goto out2;

	return register_filesystem(&logfs_fs_type);
out2:
	logfs_compr_exit();
out1:
	__free_pages(emergency_page, 0);
	return ret;
}

static void __exit logfs_exit(void)
{
	unregister_filesystem(&logfs_fs_type);
	logfs_destroy_inode_cache();
	logfs_compr_exit();
	__free_pages(emergency_page, 0);
}

module_init(logfs_init);
module_exit(logfs_exit);

MODULE_LICENSE("GPL v2");
MODULE_AUTHOR("Joern Engel <joern@logfs.org>");
MODULE_DESCRIPTION("scalable flash filesystem");
Commit	Line	Data
5db53f3e JE	1	/*
	2	* fs/logfs/super.c
	3	*
	4	* As should be obvious for Linux kernel code, license is GPLv2
	5	*
	6	* Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
	7	*
	8	* Generally contains mount/umount code and also serves as a dump area for
	9	* any functions that don't fit elsewhere and neither justify a file of their
	10	* own.
	11	*/
	12	#include "logfs.h"
	13	#include <linux/bio.h>
5a0e3ad6	14	#include <linux/slab.h>
b8639077	15	#include <linux/blkdev.h>
5db53f3e JE	16	#include <linux/mtd/mtd.h>
	17	#include <linux/statfs.h>
	18	#include <linux/buffer_head.h>
	19
	20	static DEFINE_MUTEX(emergency_mutex);
	21	static struct page *emergency_page;
	22
	23	struct page emergency_read_begin(struct address_space mapping, pgoff_t index)
	24	{
	25	filler_t filler = (filler_t )mapping->a_ops->readpage;
	26	struct page *page;
	27	int err;
	28
	29	page = read_cache_page(mapping, index, filler, NULL);
	30	if (page)
	31	return page;
	32
	33	/* No more pages available, switch to emergency page */
	34	printk(KERN_INFO"Logfs: Using emergency page\n");
	35	mutex_lock(&emergency_mutex);
	36	err = filler(NULL, emergency_page);
	37	if (err) {
	38	mutex_unlock(&emergency_mutex);
	39	printk(KERN_EMERG"Logfs: Error reading emergency page\n");
	40	return ERR_PTR(err);
	41	}
	42	return emergency_page;
	43	}
	44
	45	void emergency_read_end(struct page *page)
	46	{
	47	if (page == emergency_page)
	48	mutex_unlock(&emergency_mutex);
	49	else
	50	page_cache_release(page);
	51	}
	52
	53	static void dump_segfile(struct super_block *sb)
	54	{
	55	struct logfs_super *super = logfs_super(sb);
	56	struct logfs_segment_entry se;
	57	u32 segno;
	58
	59	for (segno = 0; segno < super->s_no_segs; segno++) {
	60	logfs_get_segment_entry(sb, segno, &se);
	61	printk("%3x: %6x %8x", segno, be32_to_cpu(se.ec_level),
	62	be32_to_cpu(se.valid));
	63	if (++segno < super->s_no_segs) {
	64	logfs_get_segment_entry(sb, segno, &se);
	65	printk(" %6x %8x", be32_to_cpu(se.ec_level),
	66	be32_to_cpu(se.valid));
	67	}
	68	if (++segno < super->s_no_segs) {
	69	logfs_get_segment_entry(sb, segno, &se);
	70	printk(" %6x %8x", be32_to_cpu(se.ec_level),
	71	be32_to_cpu(se.valid));
	72	}
	73	if (++segno < super->s_no_segs) {
	74	logfs_get_segment_entry(sb, segno, &se);
	75	printk(" %6x %8x", be32_to_cpu(se.ec_level),
	76	be32_to_cpu(se.valid));
	77	}
	78	printk("\n");
	79	}
80	}
81
82	/*
83	* logfs_crash_dump - dump debug information to device
84	*
85	* The LogFS superblock only occupies part of a segment. This function will
86	* write as much debug information as it can gather into the spare space.
87	*/
88	void logfs_crash_dump(struct super_block *sb)
89	{
90	dump_segfile(sb);
91	}
92
5db53f3e JE	93	/*
	94	* FIXME: There should be a reserve for root, similar to ext2.
	95	*/
	96	int logfs_statfs(struct dentry dentry, struct kstatfs stats)
	97	{
	98	struct super_block *sb = dentry->d_sb;
	99	struct logfs_super *super = logfs_super(sb);
	100
	101	stats->f_type = LOGFS_MAGIC_U32;
	102	stats->f_bsize = sb->s_blocksize;
	103	stats->f_blocks = super->s_size >> LOGFS_BLOCK_BITS >> 3;
	104	stats->f_bfree = super->s_free_bytes >> sb->s_blocksize_bits;
	105	stats->f_bavail = super->s_free_bytes >> sb->s_blocksize_bits;
	106	stats->f_files = 0;
	107	stats->f_ffree = 0;
	108	stats->f_namelen = LOGFS_MAX_NAMELEN;
	109	return 0;
	110	}
	111
	112	static int logfs_sb_set(struct super_block sb, void _super)
	113	{
	114	struct logfs_super *super = _super;
	115
	116	sb->s_fs_info = super;
	117	sb->s_mtd = super->s_mtd;
	118	sb->s_bdev = super->s_bdev;
bba0b5c2	119	#ifdef CONFIG_BLOCK
b8639077 JE	120	if (sb->s_bdev)
b8639077 JE	121	sb->s_bdi = &bdev_get_queue(sb->s_bdev)->backing_dev_info;
bba0b5c2 JE	122	#endif
bba0b5c2 JE	123	#ifdef CONFIG_MTD
b8639077 JE	124	if (sb->s_mtd)
b8639077 JE	125	sb->s_bdi = sb->s_mtd->backing_dev_info;
bba0b5c2	126	#endif
5db53f3e JE	127	return 0;
	128	}
	129
	130	static int logfs_sb_test(struct super_block sb, void _super)
	131	{
	132	struct logfs_super *super = _super;
	133	struct mtd_info *mtd = super->s_mtd;
	134
	135	if (mtd && sb->s_mtd == mtd)
	136	return 1;
	137	if (super->s_bdev && sb->s_bdev == super->s_bdev)
	138	return 1;
	139	return 0;
	140	}
	141
	142	static void set_segment_header(struct logfs_segment_header *sh, u8 type,
	143	u8 level, u32 segno, u32 ec)
	144	{
	145	sh->pad = 0;
	146	sh->type = type;
	147	sh->level = level;
	148	sh->segno = cpu_to_be32(segno);
	149	sh->ec = cpu_to_be32(ec);
	150	sh->gec = cpu_to_be64(segno);
	151	sh->crc = logfs_crc32(sh, LOGFS_SEGMENT_HEADERSIZE, 4);
	152	}
	153
	154	static void logfs_write_ds(struct super_block sb, struct logfs_disk_super ds,
	155	u32 segno, u32 ec)
	156	{
	157	struct logfs_super *super = logfs_super(sb);
	158	struct logfs_segment_header *sh = &ds->ds_sh;
	159	int i;
	160
	161	memset(ds, 0, sizeof(*ds));
	162	set_segment_header(sh, SEG_SUPER, 0, segno, ec);
	163
	164	ds->ds_ifile_levels = super->s_ifile_levels;
	165	ds->ds_iblock_levels = super->s_iblock_levels;
	166	ds->ds_data_levels = super->s_data_levels; /* XXX: Remove */
	167	ds->ds_segment_shift = super->s_segshift;
	168	ds->ds_block_shift = sb->s_blocksize_bits;
	169	ds->ds_write_shift = super->s_writeshift;
	170	ds->ds_filesystem_size = cpu_to_be64(super->s_size);
	171	ds->ds_segment_size = cpu_to_be32(super->s_segsize);
	172	ds->ds_bad_seg_reserve = cpu_to_be32(super->s_bad_seg_reserve);
	173	ds->ds_feature_incompat = cpu_to_be64(super->s_feature_incompat);
	174	ds->ds_feature_ro_compat= cpu_to_be64(super->s_feature_ro_compat);
	175	ds->ds_feature_compat = cpu_to_be64(super->s_feature_compat);
	176	ds->ds_feature_flags = cpu_to_be64(super->s_feature_flags);
	177	ds->ds_root_reserve = cpu_to_be64(super->s_root_reserve);
	178	ds->ds_speed_reserve = cpu_to_be64(super->s_speed_reserve);
	179	journal_for_each(i)
	180	ds->ds_journal_seg[i] = cpu_to_be32(super->s_journal_seg[i]);
	181	ds->ds_magic = cpu_to_be64(LOGFS_MAGIC);
	182	ds->ds_crc = logfs_crc32(ds, sizeof(*ds),
	183	LOGFS_SEGMENT_HEADERSIZE + 12);
	184	}
	185
	186	static int write_one_sb(struct super_block *sb,
	187	struct page (find_sb)(struct super_block sb, u64 ofs))
	188	{
	189	struct logfs_super *super = logfs_super(sb);
	190	struct logfs_disk_super *ds;
191	struct logfs_segment_entry se;
192	struct page *page;
193	u64 ofs;
194	u32 ec, segno;
195	int err;
196
197	page = find_sb(sb, &ofs);
198	if (!page)
199	return -EIO;
200	ds = page_address(page);
201	segno = seg_no(sb, ofs);
202	logfs_get_segment_entry(sb, segno, &se);
203	ec = be32_to_cpu(se.ec_level) >> 4;
204	ec++;
205	logfs_set_segment_erased(sb, segno, ec, 0);
206	logfs_write_ds(sb, ds, segno, ec);
207	err = super->s_devops->write_sb(sb, page);
208	page_cache_release(page);
209	return err;
210	}
211
212	int logfs_write_sb(struct super_block *sb)
213	{
214	struct logfs_super *super = logfs_super(sb);
215	int err;
216
217	/* First superblock */
218	err = write_one_sb(sb, super->s_devops->find_first_sb);
219	if (err)
220	return err;
221
222	/* Last superblock */
223	err = write_one_sb(sb, super->s_devops->find_last_sb);
224	if (err)
225	return err;
226	return 0;
227	}
228
229	static int ds_cmp(const void ds0, const void ds1)
230	{
231	size_t len = sizeof(struct logfs_disk_super);
232
233	/* We know the segment headers differ, so ignore them */
234	len -= LOGFS_SEGMENT_HEADERSIZE;
235	ds0 += LOGFS_SEGMENT_HEADERSIZE;
236	ds1 += LOGFS_SEGMENT_HEADERSIZE;
237	return memcmp(ds0, ds1, len);
238	}
239
240	static int logfs_recover_sb(struct super_block *sb)
241	{
242	struct logfs_super *super = logfs_super(sb);
243	struct logfs_disk_super _ds0, *ds0 = &_ds0;
244	struct logfs_disk_super _ds1, *ds1 = &_ds1;
245	int err, valid0, valid1;
246
247	/* read first superblock */
248	err = wbuf_read(sb, super->s_sb_ofs[0], sizeof(*ds0), ds0);
249	if (err)
250	return err;
251	/* read last superblock */
252	err = wbuf_read(sb, super->s_sb_ofs[1], sizeof(*ds1), ds1);
253	if (err)
254	return err;
255	valid0 = logfs_check_ds(ds0) == 0;
256	valid1 = logfs_check_ds(ds1) == 0;
257
258	if (!valid0 && valid1) {
259	printk(KERN_INFO"First superblock is invalid - fixing.\n");
260	return write_one_sb(sb, super->s_devops->find_first_sb);
261	}
262	if (valid0 && !valid1) {
263	printk(KERN_INFO"Last superblock is invalid - fixing.\n");
264	return write_one_sb(sb, super->s_devops->find_last_sb);
265	}
266	if (valid0 && valid1 && ds_cmp(ds0, ds1)) {
267	printk(KERN_INFO"Superblocks don't match - fixing.\n");
faaa27ab	268	return logfs_write_sb(sb);
5db53f3e JE	269	}
	270	/* If neither is valid now, something's wrong. Didn't we properly
	271	* check them before?!? */
	272	BUG_ON(!valid0 && !valid1);
	273	return 0;
	274	}
	275
	276	static int logfs_make_writeable(struct super_block *sb)
	277	{
	278	int err;
	279
49137f2e JE	280	err = logfs_open_segfile(sb);
	281	if (err)
	282	return err;
	283
5db53f3e JE	284	/* Repair any broken superblock copies */
	285	err = logfs_recover_sb(sb);
	286	if (err)
	287	return err;
	288
	289	/* Check areas for trailing unaccounted data */
	290	err = logfs_check_areas(sb);
	291	if (err)
	292	return err;
	293
5db53f3e JE	294	/* Do one GC pass before any data gets dirtied */
	295	logfs_gc_pass(sb);
	296
	297	/* after all initializations are done, replay the journal
	298	* for rw-mounts, if necessary */
	299	err = logfs_replay_journal(sb);
	300	if (err)
	301	return err;
	302
	303	return 0;
	304	}
	305
a1da9e8a	306	static int logfs_get_sb_final(struct super_block *sb)
5db53f3e	307	{
9421502b	308	struct logfs_super *super = logfs_super(sb);
5db53f3e JE	309	struct inode *rootdir;
	310	int err;
	311
	312	/* root dir */
	313	rootdir = logfs_iget(sb, LOGFS_INO_ROOT);
	314	if (IS_ERR(rootdir))
	315	goto fail;
	316
	317	sb->s_root = d_alloc_root(rootdir);
26562449 AV	318	if (!sb->s_root) {
	319	iput(rootdir);
	320	goto fail;
	321	}
5db53f3e	322
8e22c1a4	323	/* at that point we know that ->put_super() will be called */
9421502b JE	324	super->s_erase_page = alloc_pages(GFP_KERNEL, 0);
9421502b JE	325	if (!super->s_erase_page)
8e22c1a4	326	return -ENOMEM;
9421502b JE	327	memset(page_address(super->s_erase_page), 0xFF, PAGE_SIZE);
9421502b JE	328
5db53f3e JE	329	/* FIXME: check for read-only mounts */
5db53f3e JE	330	err = logfs_make_writeable(sb);
8e22c1a4 AV	331	if (err) {
	332	__free_page(super->s_erase_page);
	333	return err;
	334	}
5db53f3e JE	335
5db53f3e JE	336	log_super("LogFS: Finished mounting\n");
5db53f3e JE	337	return 0;
5db53f3e JE	338
5db53f3e	339	fail:
8e22c1a4 AV	340	iput(super->s_master_inode);
	341	iput(super->s_segfile_inode);
	342	iput(super->s_mapping_inode);
5db53f3e JE	343	return -EIO;
	344	}
	345
	346	int logfs_check_ds(struct logfs_disk_super *ds)
	347	{
	348	struct logfs_segment_header *sh = &ds->ds_sh;
	349
	350	if (ds->ds_magic != cpu_to_be64(LOGFS_MAGIC))
	351	return -EINVAL;
	352	if (sh->crc != logfs_crc32(sh, LOGFS_SEGMENT_HEADERSIZE, 4))
	353	return -EINVAL;
	354	if (ds->ds_crc != logfs_crc32(ds, sizeof(*ds),
	355	LOGFS_SEGMENT_HEADERSIZE + 12))
	356	return -EINVAL;
	357	return 0;
	358	}
	359
	360	static struct page find_super_block(struct super_block sb)
	361	{
	362	struct logfs_super *super = logfs_super(sb);
	363	struct page first, last;
	364
	365	first = super->s_devops->find_first_sb(sb, &super->s_sb_ofs[0]);
	366	if (!first \|\| IS_ERR(first))
	367	return NULL;
	368	last = super->s_devops->find_last_sb(sb, &super->s_sb_ofs[1]);
3272c8a5	369	if (!last \|\| IS_ERR(last)) {
5db53f3e JE	370	page_cache_release(first);
	371	return NULL;
	372	}
	373
	374	if (!logfs_check_ds(page_address(first))) {
	375	page_cache_release(last);
	376	return first;
	377	}
	378
	379	/* First one didn't work, try the second superblock */
	380	if (!logfs_check_ds(page_address(last))) {
	381	page_cache_release(first);
	382	return last;
	383	}
	384
	385	/* Neither worked, sorry folks */
	386	page_cache_release(first);
	387	page_cache_release(last);
	388	return NULL;
	389	}
	390
	391	static int __logfs_read_sb(struct super_block *sb)
	392	{
	393	struct logfs_super *super = logfs_super(sb);
	394	struct page *page;
	395	struct logfs_disk_super *ds;
	396	int i;
	397
	398	page = find_super_block(sb);
	399	if (!page)
ad342631	400	return -EINVAL;
5db53f3e JE	401
	402	ds = page_address(page);
	403	super->s_size = be64_to_cpu(ds->ds_filesystem_size);
	404	super->s_root_reserve = be64_to_cpu(ds->ds_root_reserve);
	405	super->s_speed_reserve = be64_to_cpu(ds->ds_speed_reserve);
	406	super->s_bad_seg_reserve = be32_to_cpu(ds->ds_bad_seg_reserve);
	407	super->s_segsize = 1 << ds->ds_segment_shift;
	408	super->s_segmask = (1 << ds->ds_segment_shift) - 1;
	409	super->s_segshift = ds->ds_segment_shift;
	410	sb->s_blocksize = 1 << ds->ds_block_shift;
	411	sb->s_blocksize_bits = ds->ds_block_shift;
	412	super->s_writesize = 1 << ds->ds_write_shift;
	413	super->s_writeshift = ds->ds_write_shift;
	414	super->s_no_segs = super->s_size >> super->s_segshift;
	415	super->s_no_blocks = super->s_segsize >> sb->s_blocksize_bits;
	416	super->s_feature_incompat = be64_to_cpu(ds->ds_feature_incompat);
	417	super->s_feature_ro_compat = be64_to_cpu(ds->ds_feature_ro_compat);
	418	super->s_feature_compat = be64_to_cpu(ds->ds_feature_compat);
	419	super->s_feature_flags = be64_to_cpu(ds->ds_feature_flags);
	420
	421	journal_for_each(i)
	422	super->s_journal_seg[i] = be32_to_cpu(ds->ds_journal_seg[i]);
	423
	424	super->s_ifile_levels = ds->ds_ifile_levels;
	425	super->s_iblock_levels = ds->ds_iblock_levels;
	426	super->s_data_levels = ds->ds_data_levels;
	427	super->s_total_levels = super->s_ifile_levels + super->s_iblock_levels
	428	+ super->s_data_levels;
	429	page_cache_release(page);
	430	return 0;
	431	}
	432
6a08ab84	433	static int logfs_read_sb(struct super_block *sb, int read_only)
5db53f3e JE	434	{
	435	struct logfs_super *super = logfs_super(sb);
	436	int ret;
	437
	438	super->s_btree_pool = mempool_create(32, btree_alloc, btree_free, NULL);
	439	if (!super->s_btree_pool)
	440	return -ENOMEM;
	441
	442	btree_init_mempool64(&super->s_shadow_tree.new, super->s_btree_pool);
	443	btree_init_mempool64(&super->s_shadow_tree.old, super->s_btree_pool);
032d8f72 JE	444	btree_init_mempool32(&super->s_shadow_tree.segment_map,
032d8f72 JE	445	super->s_btree_pool);
5db53f3e JE	446
	447	ret = logfs_init_mapping(sb);
	448	if (ret)
	449	return ret;
	450
	451	ret = __logfs_read_sb(sb);
	452	if (ret)
	453	return ret;
	454
6a08ab84 JE	455	if (super->s_feature_incompat & ~LOGFS_FEATURES_INCOMPAT)
	456	return -EIO;
	457	if ((super->s_feature_ro_compat & ~LOGFS_FEATURES_RO_COMPAT) &&
	458	!read_only)
	459	return -EIO;
	460
5db53f3e JE	461	ret = logfs_init_rw(sb);
	462	if (ret)
	463	return ret;
	464
	465	ret = logfs_init_areas(sb);
	466	if (ret)
	467	return ret;
	468
	469	ret = logfs_init_gc(sb);
	470	if (ret)
	471	return ret;
	472
	473	ret = logfs_init_journal(sb);
	474	if (ret)
	475	return ret;
	476
	477	return 0;
	478	}
	479
	480	static void logfs_kill_sb(struct super_block *sb)
	481	{
	482	struct logfs_super *super = logfs_super(sb);
	483
	484	log_super("LogFS: Start unmounting\n");
	485	/* Alias entries slow down mount, so evict as many as possible */
	486	sync_filesystem(sb);
c6d38301	487	logfs_write_anchor(sb);
5db53f3e JE	488
	489	/*
	490	* From this point on alias entries are simply dropped - and any
	491	* writes to the object store are considered bugs.
	492	*/
	493	super->s_flags \|= LOGFS_SB_FLAG_SHUTDOWN;
	494	log_super("LogFS: Now in shutdown\n");
	495	generic_shutdown_super(sb);
	496
	497	BUG_ON(super->s_dirty_used_bytes \|\| super->s_dirty_free_bytes);
	498
	499	logfs_cleanup_gc(sb);
	500	logfs_cleanup_journal(sb);
	501	logfs_cleanup_areas(sb);
	502	logfs_cleanup_rw(sb);
9421502b JE	503	if (super->s_erase_page)
9421502b JE	504	__free_page(super->s_erase_page);
e5a0726a	505	super->s_devops->put_device(super);
1f1b0008 JE	506	logfs_mempool_destroy(super->s_btree_pool);
1f1b0008 JE	507	logfs_mempool_destroy(super->s_alias_pool);
5db53f3e JE	508	kfree(super);
	509	log_super("LogFS: Finished unmounting\n");
	510	}
	511
a1da9e8a AV	512	static struct dentry logfs_get_sb_device(struct logfs_super super,
a1da9e8a AV	513	struct file_system_type *type, int flags)
5db53f3e	514	{
5db53f3e JE	515	struct super_block *sb;
	516	int err = -ENOMEM;
	517	static int mount_count;
	518
	519	log_super("LogFS: Start mount %x\n", mount_count++);
5db53f3e	520
5db53f3e JE	521	err = -EINVAL;
5db53f3e JE	522	sb = sget(type, logfs_sb_test, logfs_sb_set, super);
a1da9e8a AV	523	if (IS_ERR(sb)) {
	524	super->s_devops->put_device(super);
	525	kfree(super);
	526	return ERR_CAST(sb);
	527	}
5db53f3e JE	528
	529	if (sb->s_root) {
	530	/* Device is already in use */
a1da9e8a AV	531	super->s_devops->put_device(super);
	532	kfree(super);
	533	return dget(sb->s_root);
5db53f3e JE	534	}
5db53f3e JE	535
5db53f3e JE	536	/*
	537	* sb->s_maxbytes is limited to 8TB. On 32bit systems, the page cache
	538	* only covers 16TB and the upper 8TB are used for indirect blocks.
	539	* On 64bit system we could bump up the limit, but that would make
	540	* the filesystem incompatible with 32bit systems.
	541	*/
	542	sb->s_maxbytes = (1ull << 43) - 1;
	543	sb->s_op = &logfs_super_operations;
	544	sb->s_flags = flags \| MS_NOATIME;
	545
6a08ab84	546	err = logfs_read_sb(sb, sb->s_flags & MS_RDONLY);
5db53f3e JE	547	if (err)
	548	goto err1;
	549
	550	sb->s_flags \|= MS_ACTIVE;
a1da9e8a AV	551	err = logfs_get_sb_final(sb);
a1da9e8a AV	552	if (err) {
8e22c1a4	553	deactivate_locked_super(sb);
a1da9e8a AV	554	return ERR_PTR(err);
	555	}
	556	return dget(sb->s_root);
5db53f3e JE	557
5db53f3e JE	558	err1:
8e22c1a4 AV	559	/* no ->s_root, no ->put_super() */
	560	iput(super->s_master_inode);
	561	iput(super->s_segfile_inode);
	562	iput(super->s_mapping_inode);
6f2e9e6a	563	deactivate_locked_super(sb);
a1da9e8a	564	return ERR_PTR(err);
5db53f3e JE	565	}
5db53f3e JE	566
a1da9e8a AV	567	static struct dentry logfs_mount(struct file_system_type type, int flags,
a1da9e8a AV	568	const char devname, void data)
5db53f3e JE	569	{
5db53f3e JE	570	ulong mtdnr;
71a1c012	571	struct logfs_super *super;
7d945a3a	572	int err;
71a1c012 AV	573
	574	super = kzalloc(sizeof(*super), GFP_KERNEL);
	575	if (!super)
a1da9e8a	576	return ERR_PTR(-ENOMEM);
5db53f3e	577
cce2c56e LT	578	mutex_init(&super->s_dirop_mutex);
	579	mutex_init(&super->s_object_alias_mutex);
	580	INIT_LIST_HEAD(&super->s_freeing_list);
	581
5db53f3e	582	if (!devname)
7d945a3a AV	583	err = logfs_get_sb_bdev(super, type, devname);
	584	else if (strncmp(devname, "mtd", 3))
	585	err = logfs_get_sb_bdev(super, type, devname);
	586	else {
5db53f3e JE	587	char *garbage;
5db53f3e JE	588	mtdnr = simple_strtoul(devname+3, &garbage, 0);
7d945a3a AV	589	if (*garbage)
	590	err = -EINVAL;
	591	else
	592	err = logfs_get_sb_mtd(super, mtdnr);
	593	}
	594
	595	if (err) {
	596	kfree(super);
a1da9e8a	597	return ERR_PTR(err);
5db53f3e JE	598	}
5db53f3e JE	599
a1da9e8a	600	return logfs_get_sb_device(super, type, flags);
5db53f3e JE	601	}
	602
	603	static struct file_system_type logfs_fs_type = {
	604	.owner = THIS_MODULE,
	605	.name = "logfs",
a1da9e8a	606	.mount = logfs_mount,
5db53f3e JE	607	.kill_sb = logfs_kill_sb,
	608	.fs_flags = FS_REQUIRES_DEV,
	609
	610	};
	611
	612	static int __init logfs_init(void)
	613	{
	614	int ret;
	615
	616	emergency_page = alloc_pages(GFP_KERNEL, 0);
	617	if (!emergency_page)
	618	return -ENOMEM;
	619
	620	ret = logfs_compr_init();
	621	if (ret)
	622	goto out1;
	623
	624	ret = logfs_init_inode_cache();
	625	if (ret)
	626	goto out2;
	627
	628	return register_filesystem(&logfs_fs_type);
	629	out2:
	630	logfs_compr_exit();
	631	out1:
	632	__free_pages(emergency_page, 0);
	633	return ret;
	634	}
	635
	636	static void __exit logfs_exit(void)
	637	{
	638	unregister_filesystem(&logfs_fs_type);
	639	logfs_destroy_inode_cache();
	640	logfs_compr_exit();
	641	__free_pages(emergency_page, 0);
	642	}
	643
	644	module_init(logfs_init);
	645	module_exit(logfs_exit);
	646
	647	MODULE_LICENSE("GPL v2");
	648	MODULE_AUTHOR("Joern Engel <joern@logfs.org>");
	649	MODULE_DESCRIPTION("scalable flash filesystem");