nilfs2: segment constructor
authorRyusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Tue, 7 Apr 2009 02:01:37 +0000 (19:01 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 7 Apr 2009 15:31:15 +0000 (08:31 -0700)
This adds the segment constructor (also called log writer).

The segment constructor collects dirty buffers for every dirty inode,
makes summaries of the buffers, assigns disk block addresses to the
buffers, and then submits BIOs for the buffers.

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/nilfs2/file.c
fs/nilfs2/seglist.h [new file with mode: 0644]
fs/nilfs2/segment.c [new file with mode: 0644]
fs/nilfs2/segment.h [new file with mode: 0644]

index 7ddd42e24f77a8caf57d1bf938693b72f15cf184..8031086db8d54b01e9f4c7772cf6bbd5a8953412 100644 (file)
@@ -73,10 +73,66 @@ nilfs_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
        return ret;
 }
 
-static int nilfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
+static int nilfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
-       if (!(vma->vm_flags & (VM_WRITE | VM_MAYWRITE)))
-               return -EPERM;
+       struct page *page = vmf->page;
+       struct inode *inode = vma->vm_file->f_dentry->d_inode;
+       struct nilfs_transaction_info ti;
+       int ret;
+
+       if (unlikely(nilfs_near_disk_full(NILFS_SB(inode->i_sb)->s_nilfs)))
+               return VM_FAULT_SIGBUS; /* -ENOSPC */
+
+       lock_page(page);
+       if (page->mapping != inode->i_mapping ||
+           page_offset(page) >= i_size_read(inode) || !PageUptodate(page)) {
+               unlock_page(page);
+               return VM_FAULT_NOPAGE; /* make the VM retry the fault */
+       }
+
+       /*
+        * check to see if the page is mapped already (no holes)
+        */
+       if (PageMappedToDisk(page)) {
+               unlock_page(page);
+               goto mapped;
+       }
+       if (page_has_buffers(page)) {
+               struct buffer_head *bh, *head;
+               int fully_mapped = 1;
+
+               bh = head = page_buffers(page);
+               do {
+                       if (!buffer_mapped(bh)) {
+                               fully_mapped = 0;
+                               break;
+                       }
+               } while (bh = bh->b_this_page, bh != head);
+
+               if (fully_mapped) {
+                       SetPageMappedToDisk(page);
+                       unlock_page(page);
+                       goto mapped;
+               }
+       }
+       unlock_page(page);
+
+       /*
+        * fill hole blocks
+        */
+       ret = nilfs_transaction_begin(inode->i_sb, &ti, 1);
+       /* never returns -ENOMEM, but may return -ENOSPC */
+       if (unlikely(ret))
+               return VM_FAULT_SIGBUS;
+
+       ret = block_page_mkwrite(vma, vmf, nilfs_get_block);
+       if (unlikely(ret)) {
+               nilfs_transaction_abort(inode->i_sb);
+               return ret;
+       }
+       nilfs_transaction_commit(inode->i_sb);
+
+ mapped:
        SetPageChecked(page);
        wait_on_page_writeback(page);
        return 0;
diff --git a/fs/nilfs2/seglist.h b/fs/nilfs2/seglist.h
new file mode 100644 (file)
index 0000000..d39df91
--- /dev/null
@@ -0,0 +1,85 @@
+/*
+ * seglist.h - expediential structure and routines to handle list of segments
+ *             (would be removed in a future release)
+ *
+ * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Written by Ryusuke Konishi <ryusuke@osrg.net>
+ *
+ */
+#ifndef _NILFS_SEGLIST_H
+#define _NILFS_SEGLIST_H
+
+#include <linux/fs.h>
+#include <linux/buffer_head.h>
+#include <linux/nilfs2_fs.h>
+#include "sufile.h"
+
+struct nilfs_segment_entry {
+       __u64                   segnum;
+
+#define NILFS_SLH_FREED                0x0001  /* The segment was freed provisonally.
+                                          It must be cancelled if
+                                          construction aborted */
+
+       unsigned                flags;
+       struct list_head        list;
+       struct buffer_head     *bh_su;
+       struct nilfs_segment_usage *raw_su;
+};
+
+
+void nilfs_dispose_segment_list(struct list_head *);
+
+static inline struct nilfs_segment_entry *
+nilfs_alloc_segment_entry(__u64 segnum)
+{
+       struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS);
+
+       if (likely(ent)) {
+               ent->segnum = segnum;
+               ent->flags = 0;
+               ent->bh_su = NULL;
+               ent->raw_su = NULL;
+               INIT_LIST_HEAD(&ent->list);
+       }
+       return ent;
+}
+
+static inline int nilfs_open_segment_entry(struct nilfs_segment_entry *ent,
+                                          struct inode *sufile)
+{
+       return nilfs_sufile_get_segment_usage(sufile, ent->segnum,
+                                             &ent->raw_su, &ent->bh_su);
+}
+
+static inline void nilfs_close_segment_entry(struct nilfs_segment_entry *ent,
+                                            struct inode *sufile)
+{
+       if (!ent->bh_su)
+               return;
+       nilfs_sufile_put_segment_usage(sufile, ent->segnum, ent->bh_su);
+       ent->bh_su = NULL;
+       ent->raw_su = NULL;
+}
+
+static inline void nilfs_free_segment_entry(struct nilfs_segment_entry *ent)
+{
+       kfree(ent);
+}
+
+#endif /* _NILFS_SEGLIST_H */
diff --git a/fs/nilfs2/segment.c b/fs/nilfs2/segment.c
new file mode 100644 (file)
index 0000000..2c4c088
--- /dev/null
@@ -0,0 +1,3187 @@
+/*
+ * segment.c - NILFS segment constructor.
+ *
+ * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Written by Ryusuke Konishi <ryusuke@osrg.net>
+ *
+ */
+
+#include <linux/pagemap.h>
+#include <linux/buffer_head.h>
+#include <linux/writeback.h>
+#include <linux/bio.h>
+#include <linux/completion.h>
+#include <linux/blkdev.h>
+#include <linux/backing-dev.h>
+#include <linux/freezer.h>
+#include <linux/kthread.h>
+#include <linux/crc32.h>
+#include <linux/pagevec.h>
+#include "nilfs.h"
+#include "btnode.h"
+#include "page.h"
+#include "segment.h"
+#include "sufile.h"
+#include "cpfile.h"
+#include "ifile.h"
+#include "seglist.h"
+#include "segbuf.h"
+
+
+/*
+ * Segment constructor
+ */
+#define SC_N_INODEVEC  16   /* Size of locally allocated inode vector */
+
+#define SC_MAX_SEGDELTA 64   /* Upper limit of the number of segments
+                               appended in collection retry loop */
+
+/* Construction mode */
+enum {
+       SC_LSEG_SR = 1, /* Make a logical segment having a super root */
+       SC_LSEG_DSYNC,  /* Flush data blocks of a given file and make
+                          a logical segment without a super root */
+       SC_FLUSH_FILE,  /* Flush data files, leads to segment writes without
+                          creating a checkpoint */
+       SC_FLUSH_DAT,   /* Flush DAT file. This also creates segments without
+                          a checkpoint */
+};
+
+/* Stage numbers of dirty block collection */
+enum {
+       NILFS_ST_INIT = 0,
+       NILFS_ST_GC,            /* Collecting dirty blocks for GC */
+       NILFS_ST_FILE,
+       NILFS_ST_SKETCH,
+       NILFS_ST_IFILE,
+       NILFS_ST_CPFILE,
+       NILFS_ST_SUFILE,
+       NILFS_ST_DAT,
+       NILFS_ST_SR,            /* Super root */
+       NILFS_ST_DSYNC,         /* Data sync blocks */
+       NILFS_ST_DONE,
+};
+
+/* State flags of collection */
+#define NILFS_CF_NODE          0x0001  /* Collecting node blocks */
+#define NILFS_CF_IFILE_STARTED 0x0002  /* IFILE stage has started */
+#define NILFS_CF_HISTORY_MASK  (NILFS_CF_IFILE_STARTED)
+
+/* Operations depending on the construction mode and file type */
+struct nilfs_sc_operations {
+       int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
+                           struct inode *);
+       int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
+                           struct inode *);
+       int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
+                           struct inode *);
+       void (*write_data_binfo)(struct nilfs_sc_info *,
+                                struct nilfs_segsum_pointer *,
+                                union nilfs_binfo *);
+       void (*write_node_binfo)(struct nilfs_sc_info *,
+                                struct nilfs_segsum_pointer *,
+                                union nilfs_binfo *);
+};
+
+/*
+ * Other definitions
+ */
+static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
+static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
+static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
+static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
+                              int);
+
+#define nilfs_cnt32_gt(a, b)   \
+       (typecheck(__u32, a) && typecheck(__u32, b) && \
+        ((__s32)(b) - (__s32)(a) < 0))
+#define nilfs_cnt32_ge(a, b)   \
+       (typecheck(__u32, a) && typecheck(__u32, b) && \
+        ((__s32)(a) - (__s32)(b) >= 0))
+#define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
+#define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
+
+/*
+ * Transaction
+ */
+static struct kmem_cache *nilfs_transaction_cachep;
+
+/**
+ * nilfs_init_transaction_cache - create a cache for nilfs_transaction_info
+ *
+ * nilfs_init_transaction_cache() creates a slab cache for the struct
+ * nilfs_transaction_info.
+ *
+ * Return Value: On success, it returns 0. On error, one of the following
+ * negative error code is returned.
+ *
+ * %-ENOMEM - Insufficient memory available.
+ */
+int nilfs_init_transaction_cache(void)
+{
+       nilfs_transaction_cachep =
+               kmem_cache_create("nilfs2_transaction_cache",
+                                 sizeof(struct nilfs_transaction_info),
+                                 0, SLAB_RECLAIM_ACCOUNT, NULL);
+       return (nilfs_transaction_cachep == NULL) ? -ENOMEM : 0;
+}
+
+/**
+ * nilfs_detroy_transaction_cache - destroy the cache for transaction info
+ *
+ * nilfs_destroy_transaction_cache() frees the slab cache for the struct
+ * nilfs_transaction_info.
+ */
+void nilfs_destroy_transaction_cache(void)
+{
+       kmem_cache_destroy(nilfs_transaction_cachep);
+}
+
+static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
+{
+       struct nilfs_transaction_info *cur_ti = current->journal_info;
+       void *save = NULL;
+
+       if (cur_ti) {
+               if (cur_ti->ti_magic == NILFS_TI_MAGIC)
+                       return ++cur_ti->ti_count;
+               else {
+                       /*
+                        * If journal_info field is occupied by other FS,
+                        * we save it and restore on nilfs_transaction_end().
+                        * But this should never happen.
+                        */
+                       printk(KERN_WARNING
+                              "NILFS warning: journal info from a different "
+                              "FS\n");
+                       save = current->journal_info;
+               }
+       }
+       if (!ti) {
+               ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
+               if (!ti)
+                       return -ENOMEM;
+               ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
+       } else {
+               ti->ti_flags = 0;
+       }
+       ti->ti_count = 0;
+       ti->ti_save = save;
+       ti->ti_magic = NILFS_TI_MAGIC;
+       current->journal_info = ti;
+       return 0;
+}
+
+/**
+ * nilfs_transaction_begin - start indivisible file operations.
+ * @sb: super block
+ * @ti: nilfs_transaction_info
+ * @vacancy_check: flags for vacancy rate checks
+ *
+ * nilfs_transaction_begin() acquires a reader/writer semaphore, called
+ * the segment semaphore, to make a segment construction and write tasks
+ * exclusive.  The function is used with nilfs_transaction_end() in pairs.
+ * The region enclosed by these two functions can be nested.  To avoid a
+ * deadlock, the semaphore is only acquired or released in the outermost call.
+ *
+ * This function allocates a nilfs_transaction_info struct to keep context
+ * information on it.  It is initialized and hooked onto the current task in
+ * the outermost call.  If a pre-allocated struct is given to @ti, it is used
+ * instead; othewise a new struct is assigned from a slab.
+ *
+ * When @vacancy_check flag is set, this function will check the amount of
+ * free space, and will wait for the GC to reclaim disk space if low capacity.
+ *
+ * Return Value: On success, 0 is returned. On error, one of the following
+ * negative error code is returned.
+ *
+ * %-ENOMEM - Insufficient memory available.
+ *
+ * %-ERESTARTSYS - Interrupted
+ *
+ * %-ENOSPC - No space left on device
+ */
+int nilfs_transaction_begin(struct super_block *sb,
+                           struct nilfs_transaction_info *ti,
+                           int vacancy_check)
+{
+       struct nilfs_sb_info *sbi;
+       struct the_nilfs *nilfs;
+       int ret = nilfs_prepare_segment_lock(ti);
+
+       if (unlikely(ret < 0))
+               return ret;
+       if (ret > 0)
+               return 0;
+
+       sbi = NILFS_SB(sb);
+       nilfs = sbi->s_nilfs;
+       down_read(&nilfs->ns_segctor_sem);
+       if (vacancy_check && nilfs_near_disk_full(nilfs)) {
+               up_read(&nilfs->ns_segctor_sem);
+               ret = -ENOSPC;
+               goto failed;
+       }
+       return 0;
+
+ failed:
+       ti = current->journal_info;
+       current->journal_info = ti->ti_save;
+       if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
+               kmem_cache_free(nilfs_transaction_cachep, ti);
+       return ret;
+}
+
+/**
+ * nilfs_transaction_end - end indivisible file operations.
+ * @sb: super block
+ * @commit: commit flag (0 for no change)
+ *
+ * nilfs_transaction_end() releases the read semaphore which is
+ * acquired by nilfs_transaction_begin(). Its releasing is only done
+ * in outermost call of this function. If the nilfs_transaction_info
+ * was allocated dynamically, it is given back to a slab cache.
+ */
+int nilfs_transaction_end(struct super_block *sb, int commit)
+{
+       struct nilfs_transaction_info *ti = current->journal_info;
+       struct nilfs_sb_info *sbi;
+       struct nilfs_sc_info *sci;
+       int err = 0;
+
+       BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
+
+       if (commit)
+               ti->ti_flags |= NILFS_TI_COMMIT;
+       if (ti->ti_count > 0) {
+               ti->ti_count--;
+               return 0;
+       }
+       sbi = NILFS_SB(sb);
+       sci = NILFS_SC(sbi);
+       if (sci != NULL) {
+               if (ti->ti_flags & NILFS_TI_COMMIT)
+                       nilfs_segctor_start_timer(sci);
+               if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
+                   sci->sc_watermark)
+                       nilfs_segctor_do_flush(sci, 0);
+       }
+       up_read(&sbi->s_nilfs->ns_segctor_sem);
+       current->journal_info = ti->ti_save;
+
+       if (ti->ti_flags & NILFS_TI_SYNC)
+               err = nilfs_construct_segment(sb);
+       if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
+               kmem_cache_free(nilfs_transaction_cachep, ti);
+       return err;
+}
+
+void nilfs_relax_pressure_in_lock(struct super_block *sb)
+{
+       struct nilfs_sb_info *sbi = NILFS_SB(sb);
+       struct nilfs_sc_info *sci = NILFS_SC(sbi);
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+
+       if (!sci || !sci->sc_flush_request)
+               return;
+
+       set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
+       up_read(&nilfs->ns_segctor_sem);
+
+       down_write(&nilfs->ns_segctor_sem);
+       if (sci->sc_flush_request &&
+           test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
+               struct nilfs_transaction_info *ti = current->journal_info;
+
+               ti->ti_flags |= NILFS_TI_WRITER;
+               nilfs_segctor_do_immediate_flush(sci);
+               ti->ti_flags &= ~NILFS_TI_WRITER;
+       }
+       downgrade_write(&nilfs->ns_segctor_sem);
+}
+
+static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
+                                  struct nilfs_transaction_info *ti,
+                                  int gcflag)
+{
+       struct nilfs_transaction_info *cur_ti = current->journal_info;
+
+       BUG_ON(cur_ti);
+       BUG_ON(!ti);
+       ti->ti_flags = NILFS_TI_WRITER;
+       ti->ti_count = 0;
+       ti->ti_save = cur_ti;
+       ti->ti_magic = NILFS_TI_MAGIC;
+       INIT_LIST_HEAD(&ti->ti_garbage);
+       current->journal_info = ti;
+
+       for (;;) {
+               down_write(&sbi->s_nilfs->ns_segctor_sem);
+               if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
+                       break;
+
+               nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
+
+               up_write(&sbi->s_nilfs->ns_segctor_sem);
+               yield();
+       }
+       if (gcflag)
+               ti->ti_flags |= NILFS_TI_GC;
+}
+
+static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
+{
+       struct nilfs_transaction_info *ti = current->journal_info;
+
+       BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
+       BUG_ON(ti->ti_count > 0);
+
+       up_write(&sbi->s_nilfs->ns_segctor_sem);
+       current->journal_info = ti->ti_save;
+       if (!list_empty(&ti->ti_garbage))
+               nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
+}
+
+static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
+                                           struct nilfs_segsum_pointer *ssp,
+                                           unsigned bytes)
+{
+       struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
+       unsigned blocksize = sci->sc_super->s_blocksize;
+       void *p;
+
+       if (unlikely(ssp->offset + bytes > blocksize)) {
+               ssp->offset = 0;
+               BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
+                                              &segbuf->sb_segsum_buffers));
+               ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
+       }
+       p = ssp->bh->b_data + ssp->offset;
+       ssp->offset += bytes;
+       return p;
+}
+
+/**
+ * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
+ * @sci: nilfs_sc_info
+ */
+static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
+{
+       struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
+       struct buffer_head *sumbh;
+       unsigned sumbytes;
+       unsigned flags = 0;
+       int err;
+
+       if (nilfs_doing_gc())
+               flags = NILFS_SS_GC;
+       err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
+       if (unlikely(err))
+               return err;
+
+       sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
+       sumbytes = segbuf->sb_sum.sumbytes;
+       sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
+       sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
+       sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
+       return 0;
+}
+
+static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
+{
+       sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
+       if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
+               return -E2BIG; /* The current segment is filled up
+                                 (internal code) */
+       sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
+       return nilfs_segctor_reset_segment_buffer(sci);
+}
+
+static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
+{
+       struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
+       int err;
+
+       if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
+               err = nilfs_segctor_feed_segment(sci);
+               if (err)
+                       return err;
+               segbuf = sci->sc_curseg;
+       }
+       err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
+       if (likely(!err))
+               segbuf->sb_sum.flags |= NILFS_SS_SR;
+       return err;
+}
+
+/*
+ * Functions for making segment summary and payloads
+ */
+static int nilfs_segctor_segsum_block_required(
+       struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
+       unsigned binfo_size)
+{
+       unsigned blocksize = sci->sc_super->s_blocksize;
+       /* Size of finfo and binfo is enough small against blocksize */
+
+       return ssp->offset + binfo_size +
+               (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
+               blocksize;
+}
+
+static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
+                                     struct inode *inode)
+{
+       sci->sc_curseg->sb_sum.nfinfo++;
+       sci->sc_binfo_ptr = sci->sc_finfo_ptr;
+       nilfs_segctor_map_segsum_entry(
+               sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
+       /* skip finfo */
+}
+
+static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
+                                   struct inode *inode)
+{
+       struct nilfs_finfo *finfo;
+       struct nilfs_inode_info *ii;
+       struct nilfs_segment_buffer *segbuf;
+
+       if (sci->sc_blk_cnt == 0)
+               return;
+
+       ii = NILFS_I(inode);
+       finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
+                                                sizeof(*finfo));
+       finfo->fi_ino = cpu_to_le64(inode->i_ino);
+       finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
+       finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
+       finfo->fi_cno = cpu_to_le64(ii->i_cno);
+
+       segbuf = sci->sc_curseg;
+       segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
+               sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
+       sci->sc_finfo_ptr = sci->sc_binfo_ptr;
+       sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
+}
+
+static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
+                                       struct buffer_head *bh,
+                                       struct inode *inode,
+                                       unsigned binfo_size)
+{
+       struct nilfs_segment_buffer *segbuf;
+       int required, err = 0;
+
+ retry:
+       segbuf = sci->sc_curseg;
+       required = nilfs_segctor_segsum_block_required(
+               sci, &sci->sc_binfo_ptr, binfo_size);
+       if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
+               nilfs_segctor_end_finfo(sci, inode);
+               err = nilfs_segctor_feed_segment(sci);
+               if (err)
+                       return err;
+               goto retry;
+       }
+       if (unlikely(required)) {
+               err = nilfs_segbuf_extend_segsum(segbuf);
+               if (unlikely(err))
+                       goto failed;
+       }
+       if (sci->sc_blk_cnt == 0)
+               nilfs_segctor_begin_finfo(sci, inode);
+
+       nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
+       /* Substitution to vblocknr is delayed until update_blocknr() */
+       nilfs_segbuf_add_file_buffer(segbuf, bh);
+       sci->sc_blk_cnt++;
+ failed:
+       return err;
+}
+
+static int nilfs_handle_bmap_error(int err, const char *fname,
+                                  struct inode *inode, struct super_block *sb)
+{
+       if (err == -EINVAL) {
+               nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
+                           inode->i_ino);
+               err = -EIO;
+       }
+       return err;
+}
+
+/*
+ * Callback functions that enumerate, mark, and collect dirty blocks
+ */
+static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
+                                  struct buffer_head *bh, struct inode *inode)
+{
+       int err;
+
+       /* BUG_ON(!buffer_dirty(bh)); */
+       /* excluded by scan_dirty_data_buffers() */
+       err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
+       if (unlikely(err < 0))
+               return nilfs_handle_bmap_error(err, __func__, inode,
+                                              sci->sc_super);
+
+       err = nilfs_segctor_add_file_block(sci, bh, inode,
+                                          sizeof(struct nilfs_binfo_v));
+       if (!err)
+               sci->sc_datablk_cnt++;
+       return err;
+}
+
+static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
+                                  struct buffer_head *bh,
+                                  struct inode *inode)
+{
+       int err;
+
+       /* BUG_ON(!buffer_dirty(bh)); */
+       /* excluded by scan_dirty_node_buffers() */
+       err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
+       if (unlikely(err < 0))
+               return nilfs_handle_bmap_error(err, __func__, inode,
+                                              sci->sc_super);
+       return 0;
+}
+
+static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
+                                  struct buffer_head *bh,
+                                  struct inode *inode)
+{
+       BUG_ON(!buffer_dirty(bh));
+       return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
+}
+
+static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
+                                       struct nilfs_segsum_pointer *ssp,
+                                       union nilfs_binfo *binfo)
+{
+       struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
+               sci, ssp, sizeof(*binfo_v));
+       *binfo_v = binfo->bi_v;
+}
+
+static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
+                                       struct nilfs_segsum_pointer *ssp,
+                                       union nilfs_binfo *binfo)
+{
+       __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
+               sci, ssp, sizeof(*vblocknr));
+       *vblocknr = binfo->bi_v.bi_vblocknr;
+}
+
+struct nilfs_sc_operations nilfs_sc_file_ops = {
+       .collect_data = nilfs_collect_file_data,
+       .collect_node = nilfs_collect_file_node,
+       .collect_bmap = nilfs_collect_file_bmap,
+       .write_data_binfo = nilfs_write_file_data_binfo,
+       .write_node_binfo = nilfs_write_file_node_binfo,
+};
+
+static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
+                                 struct buffer_head *bh, struct inode *inode)
+{
+       int err;
+
+       err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
+       if (unlikely(err < 0))
+               return nilfs_handle_bmap_error(err, __func__, inode,
+                                              sci->sc_super);
+
+       err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
+       if (!err)
+               sci->sc_datablk_cnt++;
+       return err;
+}
+
+static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
+                                 struct buffer_head *bh, struct inode *inode)
+{
+       BUG_ON(!buffer_dirty(bh));
+       return nilfs_segctor_add_file_block(sci, bh, inode,
+                                           sizeof(struct nilfs_binfo_dat));
+}
+
+static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
+                                      struct nilfs_segsum_pointer *ssp,
+                                      union nilfs_binfo *binfo)
+{
+       __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
+                                                         sizeof(*blkoff));
+       *blkoff = binfo->bi_dat.bi_blkoff;
+}
+
+static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
+                                      struct nilfs_segsum_pointer *ssp,
+                                      union nilfs_binfo *binfo)
+{
+       struct nilfs_binfo_dat *binfo_dat =
+               nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
+       *binfo_dat = binfo->bi_dat;
+}
+
+struct nilfs_sc_operations nilfs_sc_dat_ops = {
+       .collect_data = nilfs_collect_dat_data,
+       .collect_node = nilfs_collect_file_node,
+       .collect_bmap = nilfs_collect_dat_bmap,
+       .write_data_binfo = nilfs_write_dat_data_binfo,
+       .write_node_binfo = nilfs_write_dat_node_binfo,
+};
+
+struct nilfs_sc_operations nilfs_sc_dsync_ops = {
+       .collect_data = nilfs_collect_file_data,
+       .collect_node = NULL,
+       .collect_bmap = NULL,
+       .write_data_binfo = nilfs_write_file_data_binfo,
+       .write_node_binfo = NULL,
+};
+
+static int nilfs_lookup_dirty_data_buffers(struct inode *inode,
+                                          struct list_head *listp,
+                                          struct nilfs_sc_info *sci)
+{
+       struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
+       struct address_space *mapping = inode->i_mapping;
+       struct pagevec pvec;
+       unsigned i, ndirties = 0, nlimit;
+       pgoff_t index = 0;
+       int err = 0;
+
+       nlimit = sci->sc_segbuf_nblocks -
+               (sci->sc_nblk_this_inc + segbuf->sb_sum.nblocks);
+       pagevec_init(&pvec, 0);
+ repeat:
+       if (!pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
+                               PAGEVEC_SIZE))
+               return 0;
+
+       for (i = 0; i < pagevec_count(&pvec); i++) {
+               struct buffer_head *bh, *head;
+               struct page *page = pvec.pages[i];
+
+               if (mapping->host) {
+                       lock_page(page);
+                       if (!page_has_buffers(page))
+                               create_empty_buffers(page,
+                                                    1 << inode->i_blkbits, 0);
+                       unlock_page(page);
+               }
+
+               bh = head = page_buffers(page);
+               do {
+                       if (buffer_dirty(bh)) {
+                               if (ndirties > nlimit) {
+                                       err = -E2BIG;
+                                       break;
+                               }
+                               get_bh(bh);
+                               list_add_tail(&bh->b_assoc_buffers, listp);
+                               ndirties++;
+                       }
+                       bh = bh->b_this_page;
+               } while (bh != head);
+       }
+       pagevec_release(&pvec);
+       cond_resched();
+
+       if (!err)
+               goto repeat;
+       return err;
+}
+
+static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
+                                           struct list_head *listp)
+{
+       struct nilfs_inode_info *ii = NILFS_I(inode);
+       struct address_space *mapping = &ii->i_btnode_cache;
+       struct pagevec pvec;
+       struct buffer_head *bh, *head;
+       unsigned int i;
+       pgoff_t index = 0;
+
+       pagevec_init(&pvec, 0);
+
+       while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
+                                 PAGEVEC_SIZE)) {
+               for (i = 0; i < pagevec_count(&pvec); i++) {
+                       bh = head = page_buffers(pvec.pages[i]);
+                       do {
+                               if (buffer_dirty(bh)) {
+                                       get_bh(bh);
+                                       list_add_tail(&bh->b_assoc_buffers,
+                                                     listp);
+                               }
+                               bh = bh->b_this_page;
+                       } while (bh != head);
+               }
+               pagevec_release(&pvec);
+               cond_resched();
+       }
+}
+
+static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
+                              struct list_head *head, int force)
+{
+       struct nilfs_inode_info *ii, *n;
+       struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
+       unsigned nv = 0;
+
+       while (!list_empty(head)) {
+               spin_lock(&sbi->s_inode_lock);
+               list_for_each_entry_safe(ii, n, head, i_dirty) {
+                       list_del_init(&ii->i_dirty);
+                       if (force) {
+                               if (unlikely(ii->i_bh)) {
+                                       brelse(ii->i_bh);
+                                       ii->i_bh = NULL;
+                               }
+                       } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
+                               set_bit(NILFS_I_QUEUED, &ii->i_state);
+                               list_add_tail(&ii->i_dirty,
+                                             &sbi->s_dirty_files);
+                               continue;
+                       }
+                       ivec[nv++] = ii;
+                       if (nv == SC_N_INODEVEC)
+                               break;
+               }
+               spin_unlock(&sbi->s_inode_lock);
+
+               for (pii = ivec; nv > 0; pii++, nv--)
+                       iput(&(*pii)->vfs_inode);
+       }
+}
+
+static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
+{
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+       int ret = 0;
+
+       if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
+               ret++;
+       if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
+               ret++;
+       if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
+               ret++;
+       if (ret || nilfs_doing_gc())
+               if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
+                       ret++;
+       return ret;
+}
+
+static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
+{
+       return list_empty(&sci->sc_dirty_files) &&
+               !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
+               list_empty(&sci->sc_cleaning_segments) &&
+               (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
+}
+
+static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
+{
+       struct nilfs_sb_info *sbi = sci->sc_sbi;
+       int ret = 0;
+
+       if (nilfs_test_metadata_dirty(sbi))
+               set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
+
+       spin_lock(&sbi->s_inode_lock);
+       if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
+               ret++;
+
+       spin_unlock(&sbi->s_inode_lock);
+       return ret;
+}
+
+static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
+{
+       struct nilfs_sb_info *sbi = sci->sc_sbi;
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+
+       nilfs_mdt_clear_dirty(sbi->s_ifile);
+       nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
+       nilfs_mdt_clear_dirty(nilfs->ns_sufile);
+       nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
+}
+
+static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
+{
+       struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
+       struct buffer_head *bh_cp;
+       struct nilfs_checkpoint *raw_cp;
+       int err;
+
+       /* XXX: this interface will be changed */
+       err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
+                                         &raw_cp, &bh_cp);
+       if (likely(!err)) {
+               /* The following code is duplicated with cpfile.  But, it is
+                  needed to collect the checkpoint even if it was not newly
+                  created */
+               nilfs_mdt_mark_buffer_dirty(bh_cp);
+               nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
+               nilfs_cpfile_put_checkpoint(
+                       nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
+       } else {
+               BUG_ON(err == -EINVAL || err == -ENOENT);
+       }
+       return err;
+}
+
+static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
+{
+       struct nilfs_sb_info *sbi = sci->sc_sbi;
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+       struct buffer_head *bh_cp;
+       struct nilfs_checkpoint *raw_cp;
+       int err;
+
+       err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
+                                         &raw_cp, &bh_cp);
+       if (unlikely(err)) {
+               BUG_ON(err == -EINVAL || err == -ENOENT);
+               goto failed_ibh;
+       }
+       raw_cp->cp_snapshot_list.ssl_next = 0;
+       raw_cp->cp_snapshot_list.ssl_prev = 0;
+       raw_cp->cp_inodes_count =
+               cpu_to_le64(atomic_read(&sbi->s_inodes_count));
+       raw_cp->cp_blocks_count =
+               cpu_to_le64(atomic_read(&sbi->s_blocks_count));
+       raw_cp->cp_nblk_inc =
+               cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
+       raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
+       raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
+       if (sci->sc_sketch_inode && i_size_read(sci->sc_sketch_inode) > 0)
+               nilfs_checkpoint_set_sketch(raw_cp);
+       nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
+       nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
+       return 0;
+
+ failed_ibh:
+       return err;
+}
+
+static void nilfs_fill_in_file_bmap(struct inode *ifile,
+                                   struct nilfs_inode_info *ii)
+
+{
+       struct buffer_head *ibh;
+       struct nilfs_inode *raw_inode;
+
+       if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
+               ibh = ii->i_bh;
+               BUG_ON(!ibh);
+               raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
+                                                 ibh);
+               nilfs_bmap_write(ii->i_bmap, raw_inode);
+               nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
+       }
+}
+
+static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
+                                           struct inode *ifile)
+{
+       struct nilfs_inode_info *ii;
+
+       list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
+               nilfs_fill_in_file_bmap(ifile, ii);
+               set_bit(NILFS_I_COLLECTED, &ii->i_state);
+       }
+       if (sci->sc_sketch_inode) {
+               ii = NILFS_I(sci->sc_sketch_inode);
+               if (test_bit(NILFS_I_DIRTY, &ii->i_state))
+                       nilfs_fill_in_file_bmap(ifile, ii);
+       }
+}
+
+/*
+ * CRC calculation routines
+ */
+static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
+{
+       struct nilfs_super_root *raw_sr =
+               (struct nilfs_super_root *)bh_sr->b_data;
+       u32 crc;
+
+       BUG_ON(NILFS_SR_BYTES > bh_sr->b_size);
+       crc = crc32_le(seed,
+                      (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
+                      NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
+       raw_sr->sr_sum = cpu_to_le32(crc);
+}
+
+static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
+                                           u32 seed)
+{
+       struct nilfs_segment_buffer *segbuf;
+
+       if (sci->sc_super_root)
+               nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
+
+       list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
+               nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
+               nilfs_segbuf_fill_in_data_crc(segbuf, seed);
+       }
+}
+
+static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
+                                            struct the_nilfs *nilfs)
+{
+       struct buffer_head *bh_sr = sci->sc_super_root;
+       struct nilfs_super_root *raw_sr =
+               (struct nilfs_super_root *)bh_sr->b_data;
+       unsigned isz = nilfs->ns_inode_size;
+
+       raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
+       raw_sr->sr_nongc_ctime
+               = cpu_to_le64(nilfs_doing_gc() ?
+                             nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
+       raw_sr->sr_flags = 0;
+
+       nilfs_mdt_write_inode_direct(
+               nilfs_dat_inode(nilfs), bh_sr, NILFS_SR_DAT_OFFSET(isz));
+       nilfs_mdt_write_inode_direct(
+               nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(isz));
+       nilfs_mdt_write_inode_direct(
+               nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(isz));
+}
+
+static void nilfs_redirty_inodes(struct list_head *head)
+{
+       struct nilfs_inode_info *ii;
+
+       list_for_each_entry(ii, head, i_dirty) {
+               if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
+                       clear_bit(NILFS_I_COLLECTED, &ii->i_state);
+       }
+}
+
+static void nilfs_drop_collected_inodes(struct list_head *head)
+{
+       struct nilfs_inode_info *ii;
+
+       list_for_each_entry(ii, head, i_dirty) {
+               if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
+                       continue;
+
+               clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
+               set_bit(NILFS_I_UPDATED, &ii->i_state);
+       }
+}
+
+static void nilfs_segctor_cancel_free_segments(struct nilfs_sc_info *sci,
+                                              struct inode *sufile)
+
+{
+       struct list_head *head = &sci->sc_cleaning_segments;
+       struct nilfs_segment_entry *ent;
+       int err;
+
+       list_for_each_entry(ent, head, list) {
+               if (!(ent->flags & NILFS_SLH_FREED))
+                       break;
+               err = nilfs_sufile_cancel_free(sufile, ent->segnum);
+               BUG_ON(err);
+
+               ent->flags &= ~NILFS_SLH_FREED;
+       }
+}
+
+static int nilfs_segctor_prepare_free_segments(struct nilfs_sc_info *sci,
+                                              struct inode *sufile)
+{
+       struct list_head *head = &sci->sc_cleaning_segments;
+       struct nilfs_segment_entry *ent;
+       int err;
+
+       list_for_each_entry(ent, head, list) {
+               err = nilfs_sufile_free(sufile, ent->segnum);
+               if (unlikely(err))
+                       return err;
+               ent->flags |= NILFS_SLH_FREED;
+       }
+       return 0;
+}
+
+static void nilfs_segctor_commit_free_segments(struct nilfs_sc_info *sci)
+{
+       nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
+}
+
+static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
+                                      struct inode *inode,
+                                      struct list_head *listp,
+                                      int (*collect)(struct nilfs_sc_info *,
+                                                     struct buffer_head *,
+                                                     struct inode *))
+{
+       struct buffer_head *bh, *n;
+       int err = 0;
+
+       if (collect) {
+               list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
+                       list_del_init(&bh->b_assoc_buffers);
+                       err = collect(sci, bh, inode);
+                       brelse(bh);
+                       if (unlikely(err))
+                               goto dispose_buffers;
+               }
+               return 0;
+       }
+
+ dispose_buffers:
+       while (!list_empty(listp)) {
+               bh = list_entry(listp->next, struct buffer_head,
+                               b_assoc_buffers);
+               list_del_init(&bh->b_assoc_buffers);
+               brelse(bh);
+       }
+       return err;
+}
+
+static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
+                                  struct inode *inode,
+                                  struct nilfs_sc_operations *sc_ops)
+{
+       LIST_HEAD(data_buffers);
+       LIST_HEAD(node_buffers);
+       int err, err2;
+
+       if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
+               err = nilfs_lookup_dirty_data_buffers(inode, &data_buffers,
+                                                     sci);
+               if (err) {
+                       err2 = nilfs_segctor_apply_buffers(
+                               sci, inode, &data_buffers,
+                               err == -E2BIG ? sc_ops->collect_data : NULL);
+                       if (err == -E2BIG)
+                               err = err2;
+                       goto break_or_fail;
+               }
+       }
+       nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
+
+       if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
+               err = nilfs_segctor_apply_buffers(
+                       sci, inode, &data_buffers, sc_ops->collect_data);
+               if (unlikely(err)) {
+                       /* dispose node list */
+                       nilfs_segctor_apply_buffers(
+                               sci, inode, &node_buffers, NULL);
+                       goto break_or_fail;
+               }
+               sci->sc_stage.flags |= NILFS_CF_NODE;
+       }
+       /* Collect node */
+       err = nilfs_segctor_apply_buffers(
+               sci, inode, &node_buffers, sc_ops->collect_node);
+       if (unlikely(err))
+               goto break_or_fail;
+
+       nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
+       err = nilfs_segctor_apply_buffers(
+               sci, inode, &node_buffers, sc_ops->collect_bmap);
+       if (unlikely(err))
+               goto break_or_fail;
+
+       nilfs_segctor_end_finfo(sci, inode);
+       sci->sc_stage.flags &= ~NILFS_CF_NODE;
+
+ break_or_fail:
+       return err;
+}
+
+static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
+                                        struct inode *inode)
+{
+       LIST_HEAD(data_buffers);
+       int err, err2;
+
+       err = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, sci);
+       err2 = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
+                                          (!err || err == -E2BIG) ?
+                                          nilfs_collect_file_data : NULL);
+       if (err == -E2BIG)
+               err = err2;
+       if (!err)
+               nilfs_segctor_end_finfo(sci, inode);
+       return err;
+}
+
+static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
+{
+       struct nilfs_sb_info *sbi = sci->sc_sbi;
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+       struct list_head *head;
+       struct nilfs_inode_info *ii;
+       int err = 0;
+
+       switch (sci->sc_stage.scnt) {
+       case NILFS_ST_INIT:
+               /* Pre-processes */
+               sci->sc_stage.flags = 0;
+
+               if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
+                       sci->sc_nblk_inc = 0;
+                       sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
+                       if (mode == SC_LSEG_DSYNC) {
+                               sci->sc_stage.scnt = NILFS_ST_DSYNC;
+                               goto dsync_mode;
+                       }
+               }
+
+               sci->sc_stage.dirty_file_ptr = NULL;
+               sci->sc_stage.gc_inode_ptr = NULL;
+               if (mode == SC_FLUSH_DAT) {
+                       sci->sc_stage.scnt = NILFS_ST_DAT;
+                       goto dat_stage;
+               }
+               sci->sc_stage.scnt++;  /* Fall through */
+       case NILFS_ST_GC:
+               if (nilfs_doing_gc()) {
+                       head = &sci->sc_gc_inodes;
+                       ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
+                                               head, i_dirty);
+                       list_for_each_entry_continue(ii, head, i_dirty) {
+                               err = nilfs_segctor_scan_file(
+                                       sci, &ii->vfs_inode,
+                                       &nilfs_sc_file_ops);
+                               if (unlikely(err)) {
+                                       sci->sc_stage.gc_inode_ptr = list_entry(
+                                               ii->i_dirty.prev,
+                                               struct nilfs_inode_info,
+                                               i_dirty);
+                                       goto break_or_fail;
+                               }
+                               set_bit(NILFS_I_COLLECTED, &ii->i_state);
+                       }
+                       sci->sc_stage.gc_inode_ptr = NULL;
+               }
+               sci->sc_stage.scnt++;  /* Fall through */
+       case NILFS_ST_FILE:
+               head = &sci->sc_dirty_files;
+               ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
+                                       i_dirty);
+               list_for_each_entry_continue(ii, head, i_dirty) {
+                       clear_bit(NILFS_I_DIRTY, &ii->i_state);
+
+                       err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
+                                                     &nilfs_sc_file_ops);
+                       if (unlikely(err)) {
+                               sci->sc_stage.dirty_file_ptr =
+                                       list_entry(ii->i_dirty.prev,
+                                                  struct nilfs_inode_info,
+                                                  i_dirty);
+                               goto break_or_fail;
+                       }
+                       /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
+                       /* XXX: required ? */
+               }
+               sci->sc_stage.dirty_file_ptr = NULL;
+               if (mode == SC_FLUSH_FILE) {
+                       sci->sc_stage.scnt = NILFS_ST_DONE;
+                       return 0;
+               }
+               sci->sc_stage.scnt++;  /* Fall through */
+       case NILFS_ST_SKETCH:
+               if (mode == SC_LSEG_SR && sci->sc_sketch_inode) {
+                       ii = NILFS_I(sci->sc_sketch_inode);
+                       if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
+                               sci->sc_sketch_inode->i_ctime.tv_sec
+                                       = sci->sc_seg_ctime;
+                               sci->sc_sketch_inode->i_mtime.tv_sec
+                                       = sci->sc_seg_ctime;
+                               err = nilfs_mark_inode_dirty(
+                                       sci->sc_sketch_inode);
+                               if (unlikely(err))
+                                       goto break_or_fail;
+                       }
+                       err = nilfs_segctor_scan_file(sci,
+                                                     sci->sc_sketch_inode,
+                                                     &nilfs_sc_file_ops);
+                       if (unlikely(err))
+                               goto break_or_fail;
+               }
+               sci->sc_stage.scnt++;
+               sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
+               /* Fall through */
+       case NILFS_ST_IFILE:
+               err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
+                                             &nilfs_sc_file_ops);
+               if (unlikely(err))
+                       break;
+               sci->sc_stage.scnt++;
+               /* Creating a checkpoint */
+               err = nilfs_segctor_create_checkpoint(sci);
+               if (unlikely(err))
+                       break;
+               /* Fall through */
+       case NILFS_ST_CPFILE:
+               err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
+                                             &nilfs_sc_file_ops);
+               if (unlikely(err))
+                       break;
+               sci->sc_stage.scnt++;  /* Fall through */
+       case NILFS_ST_SUFILE:
+               err = nilfs_segctor_prepare_free_segments(sci,
+                                                         nilfs->ns_sufile);
+               if (unlikely(err))
+                       break;
+               err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
+                                             &nilfs_sc_file_ops);
+               if (unlikely(err))
+                       break;
+               sci->sc_stage.scnt++;  /* Fall through */
+       case NILFS_ST_DAT:
+ dat_stage:
+               err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
+                                             &nilfs_sc_dat_ops);
+               if (unlikely(err))
+                       break;
+               if (mode == SC_FLUSH_DAT) {
+                       sci->sc_stage.scnt = NILFS_ST_DONE;
+                       return 0;
+               }
+               sci->sc_stage.scnt++;  /* Fall through */
+       case NILFS_ST_SR:
+               if (mode == SC_LSEG_SR) {
+                       /* Appending a super root */
+                       err = nilfs_segctor_add_super_root(sci);
+                       if (unlikely(err))
+                               break;
+               }
+               /* End of a logical segment */
+               sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
+               sci->sc_stage.scnt = NILFS_ST_DONE;
+               return 0;
+       case NILFS_ST_DSYNC:
+ dsync_mode:
+               sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
+               ii = sci->sc_stage.dirty_file_ptr;
+               if (!test_bit(NILFS_I_BUSY, &ii->i_state))
+                       break;
+
+               err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
+               if (unlikely(err))
+                       break;
+               sci->sc_stage.dirty_file_ptr = NULL;
+               sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
+               sci->sc_stage.scnt = NILFS_ST_DONE;
+               return 0;
+       case NILFS_ST_DONE:
+               return 0;
+       default:
+               BUG();
+       }
+
+ break_or_fail:
+       return err;
+}
+
+static int nilfs_segctor_terminate_segment(struct nilfs_sc_info *sci,
+                                          struct nilfs_segment_buffer *segbuf,
+                                          struct inode *sufile)
+{
+       struct nilfs_segment_entry *ent = segbuf->sb_segent;
+       int err;
+
+       err = nilfs_open_segment_entry(ent, sufile);
+       if (unlikely(err))
+               return err;
+       nilfs_mdt_mark_buffer_dirty(ent->bh_su);
+       nilfs_mdt_mark_dirty(sufile);
+       nilfs_close_segment_entry(ent, sufile);
+
+       list_add_tail(&ent->list, &sci->sc_active_segments);
+       segbuf->sb_segent = NULL;
+       return 0;
+}
+
+static int nilfs_touch_segusage(struct inode *sufile, __u64 segnum)
+{
+       struct buffer_head *bh_su;
+       struct nilfs_segment_usage *raw_su;
+       int err;
+
+       err = nilfs_sufile_get_segment_usage(sufile, segnum, &raw_su, &bh_su);
+       if (unlikely(err))
+               return err;
+       nilfs_mdt_mark_buffer_dirty(bh_su);
+       nilfs_mdt_mark_dirty(sufile);
+       nilfs_sufile_put_segment_usage(sufile, segnum, bh_su);
+       return 0;
+}
+
+static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
+                                           struct the_nilfs *nilfs)
+{
+       struct nilfs_segment_buffer *segbuf, *n;
+       struct inode *sufile = nilfs->ns_sufile;
+       __u64 nextnum;
+       int err;
+
+       if (list_empty(&sci->sc_segbufs)) {
+               segbuf = nilfs_segbuf_new(sci->sc_super);
+               if (unlikely(!segbuf))
+                       return -ENOMEM;
+               list_add(&segbuf->sb_list, &sci->sc_segbufs);
+       } else
+               segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
+
+       err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
+                              nilfs->ns_pseg_offset, nilfs);
+       if (unlikely(err))
+               return err;
+
+       if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
+               err = nilfs_segctor_terminate_segment(sci, segbuf, sufile);
+               if (unlikely(err))
+                       return err;
+
+               nilfs_shift_to_next_segment(nilfs);
+               err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
+       }
+       sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
+
+       err = nilfs_touch_segusage(sufile, segbuf->sb_segnum);
+       if (unlikely(err))
+               return err;
+
+       if (nilfs->ns_segnum == nilfs->ns_nextnum) {
+               /* Start from the head of a new full segment */
+               err = nilfs_sufile_alloc(sufile, &nextnum);
+               if (unlikely(err))
+                       return err;
+       } else
+               nextnum = nilfs->ns_nextnum;
+
+       segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
+       nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
+
+       /* truncating segment buffers */
+       list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
+                                         sb_list) {
+               list_del_init(&segbuf->sb_list);
+               nilfs_segbuf_free(segbuf);
+       }
+       return err;
+}
+
+static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
+                                        struct the_nilfs *nilfs, int nadd)
+{
+       struct nilfs_segment_buffer *segbuf, *prev, *n;
+       struct inode *sufile = nilfs->ns_sufile;
+       __u64 nextnextnum;
+       LIST_HEAD(list);
+       int err, ret, i;
+
+       prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
+       /*
+        * Since the segment specified with nextnum might be allocated during
+        * the previous construction, the buffer including its segusage may
+        * not be dirty.  The following call ensures that the buffer is dirty
+        * and will pin the buffer on memory until the sufile is written.
+        */
+       err = nilfs_touch_segusage(sufile, prev->sb_nextnum);
+       if (unlikely(err))
+               return err;
+
+       for (i = 0; i < nadd; i++) {
+               /* extend segment info */
+               err = -ENOMEM;
+               segbuf = nilfs_segbuf_new(sci->sc_super);
+               if (unlikely(!segbuf))
+                       goto failed;
+
+               /* map this buffer to region of segment on-disk */
+               err = nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
+               if (unlikely(err))
+                       goto failed_segbuf;
+
+               sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
+
+               /* allocate the next next full segment */
+               err = nilfs_sufile_alloc(sufile, &nextnextnum);
+               if (unlikely(err))
+                       goto failed_segbuf;
+
+               segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
+               nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
+
+               list_add_tail(&segbuf->sb_list, &list);
+               prev = segbuf;
+       }
+       list_splice(&list, sci->sc_segbufs.prev);
+       return 0;
+
+ failed_segbuf:
+       nilfs_segbuf_free(segbuf);
+ failed:
+       list_for_each_entry_safe(segbuf, n, &list, sb_list) {
+               ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
+               BUG_ON(ret);
+               list_del_init(&segbuf->sb_list);
+               nilfs_segbuf_free(segbuf);
+       }
+       return err;
+}
+
+static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
+                                                  struct the_nilfs *nilfs)
+{
+       struct nilfs_segment_buffer *segbuf;
+       int ret, done = 0;
+
+       segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
+       if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
+               ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
+               BUG_ON(ret);
+       }
+       if (segbuf->sb_io_error) {
+               /* Case 1: The first segment failed */
+               if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
+                       /* Case 1a:  Partial segment appended into an existing
+                          segment */
+                       nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
+                                               segbuf->sb_fseg_end);
+               else /* Case 1b:  New full segment */
+                       set_nilfs_discontinued(nilfs);
+               done++;
+       }
+
+       list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
+               ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
+               BUG_ON(ret);
+               if (!done && segbuf->sb_io_error) {
+                       if (segbuf->sb_segnum != nilfs->ns_nextnum)
+                               /* Case 2: extended segment (!= next) failed */
+                               nilfs_sufile_set_error(nilfs->ns_sufile,
+                                                      segbuf->sb_segnum);
+                       done++;
+               }
+       }
+}
+
+static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info *sci)
+{
+       struct nilfs_segment_buffer *segbuf;
+
+       list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list)
+               nilfs_segbuf_clear(segbuf);
+       sci->sc_super_root = NULL;
+}
+
+static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info *sci)
+{
+       struct nilfs_segment_buffer *segbuf;
+
+       while (!list_empty(&sci->sc_segbufs)) {
+               segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
+               list_del_init(&segbuf->sb_list);
+               nilfs_segbuf_free(segbuf);
+       }
+       /* sci->sc_curseg = NULL; */
+}
+
+static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
+                                          struct the_nilfs *nilfs, int err)
+{
+       if (unlikely(err)) {
+               nilfs_segctor_free_incomplete_segments(sci, nilfs);
+               nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
+       }
+       nilfs_segctor_clear_segment_buffers(sci);
+}
+
+static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
+                                         struct inode *sufile)
+{
+       struct nilfs_segment_buffer *segbuf;
+       struct buffer_head *bh_su;
+       struct nilfs_segment_usage *raw_su;
+       unsigned long live_blocks;
+       int ret;
+
+       list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
+               ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
+                                                    &raw_su, &bh_su);
+               BUG_ON(ret); /* always succeed because bh_su is dirty */
+               live_blocks = segbuf->sb_sum.nblocks +
+                       (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
+               raw_su->su_lastmod = cpu_to_le64(sci->sc_seg_ctime);
+               raw_su->su_nblocks = cpu_to_le32(live_blocks);
+               nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
+                                              bh_su);
+       }
+}
+
+static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
+                                         struct inode *sufile)
+{
+       struct nilfs_segment_buffer *segbuf;
+       struct buffer_head *bh_su;
+       struct nilfs_segment_usage *raw_su;
+       int ret;
+
+       segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
+       ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
+                                            &raw_su, &bh_su);
+       BUG_ON(ret); /* always succeed because bh_su is dirty */
+       raw_su->su_nblocks = cpu_to_le32(segbuf->sb_pseg_start -
+                                        segbuf->sb_fseg_start);
+       nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum, bh_su);
+
+       list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
+               ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
+                                                    &raw_su, &bh_su);
+               BUG_ON(ret); /* always succeed */
+               raw_su->su_nblocks = 0;
+               nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
+                                              bh_su);
+       }
+}
+
+static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
+                                           struct nilfs_segment_buffer *last,
+                                           struct inode *sufile)
+{
+       struct nilfs_segment_buffer *segbuf = last, *n;
+       int ret;
+
+       list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
+                                         sb_list) {
+               list_del_init(&segbuf->sb_list);
+               sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
+               ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
+               BUG_ON(ret);
+               nilfs_segbuf_free(segbuf);
+       }
+}
+
+
+static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
+                                struct the_nilfs *nilfs, int mode)
+{
+       struct nilfs_cstage prev_stage = sci->sc_stage;
+       int err, nadd = 1;
+
+       /* Collection retry loop */
+       for (;;) {
+               sci->sc_super_root = NULL;
+               sci->sc_nblk_this_inc = 0;
+               sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
+
+               err = nilfs_segctor_reset_segment_buffer(sci);
+               if (unlikely(err))
+                       goto failed;
+
+               err = nilfs_segctor_collect_blocks(sci, mode);
+               sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
+               if (!err)
+                       break;
+
+               if (unlikely(err != -E2BIG))
+                       goto failed;
+
+               /* The current segment is filled up */
+               if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
+                       break;
+
+               nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
+               nilfs_segctor_clear_segment_buffers(sci);
+
+               err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
+               if (unlikely(err))
+                       return err;
+
+               nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
+               sci->sc_stage = prev_stage;
+       }
+       nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
+       return 0;
+
+ failed:
+       return err;
+}
+
+static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
+                                     struct buffer_head *new_bh)
+{
+       BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
+
+       list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
+       /* The caller must release old_bh */
+}
+
+static int
+nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
+                                    struct nilfs_segment_buffer *segbuf,
+                                    int mode)
+{
+       struct inode *inode = NULL;
+       sector_t blocknr;
+       unsigned long nfinfo = segbuf->sb_sum.nfinfo;
+       unsigned long nblocks = 0, ndatablk = 0;
+       struct nilfs_sc_operations *sc_op = NULL;
+       struct nilfs_segsum_pointer ssp;
+       struct nilfs_finfo *finfo = NULL;
+       union nilfs_binfo binfo;
+       struct buffer_head *bh, *bh_org;
+       ino_t ino = 0;
+       int err = 0;
+
+       if (!nfinfo)
+               goto out;
+
+       blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
+       ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
+       ssp.offset = sizeof(struct nilfs_segment_summary);
+
+       list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
+               if (bh == sci->sc_super_root)
+                       break;
+               if (!finfo) {
+                       finfo = nilfs_segctor_map_segsum_entry(
+                               sci, &ssp, sizeof(*finfo));
+                       ino = le64_to_cpu(finfo->fi_ino);
+                       nblocks = le32_to_cpu(finfo->fi_nblocks);
+                       ndatablk = le32_to_cpu(finfo->fi_ndatablk);
+
+                       if (buffer_nilfs_node(bh))
+                               inode = NILFS_BTNC_I(bh->b_page->mapping);
+                       else
+                               inode = NILFS_AS_I(bh->b_page->mapping);
+
+                       if (mode == SC_LSEG_DSYNC)
+                               sc_op = &nilfs_sc_dsync_ops;
+                       else if (ino == NILFS_DAT_INO)
+                               sc_op = &nilfs_sc_dat_ops;
+                       else /* file blocks */
+                               sc_op = &nilfs_sc_file_ops;
+               }
+               bh_org = bh;
+               get_bh(bh_org);
+               err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
+                                       &binfo);
+               if (bh != bh_org)
+                       nilfs_list_replace_buffer(bh_org, bh);
+               brelse(bh_org);
+               if (unlikely(err))
+                       goto failed_bmap;
+
+               if (ndatablk > 0)
+                       sc_op->write_data_binfo(sci, &ssp, &binfo);
+               else
+                       sc_op->write_node_binfo(sci, &ssp, &binfo);
+
+               blocknr++;
+               if (--nblocks == 0) {
+                       finfo = NULL;
+                       if (--nfinfo == 0)
+                               break;
+               } else if (ndatablk > 0)
+                       ndatablk--;
+       }
+ out:
+       return 0;
+
+ failed_bmap:
+       err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
+       return err;
+}
+
+static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
+{
+       struct nilfs_segment_buffer *segbuf;
+       int err;
+
+       list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
+               err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
+               if (unlikely(err))
+                       return err;
+               nilfs_segbuf_fill_in_segsum(segbuf);
+       }
+       return 0;
+}
+
+static int
+nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
+{
+       struct page *clone_page;
+       struct buffer_head *bh, *head, *bh2;
+       void *kaddr;
+
+       bh = head = page_buffers(page);
+
+       clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
+       if (unlikely(!clone_page))
+               return -ENOMEM;
+
+       bh2 = page_buffers(clone_page);
+       kaddr = kmap_atomic(page, KM_USER0);
+       do {
+               if (list_empty(&bh->b_assoc_buffers))
+                       continue;
+               get_bh(bh2);
+               page_cache_get(clone_page); /* for each bh */
+               memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
+               bh2->b_blocknr = bh->b_blocknr;
+               list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
+               list_add_tail(&bh->b_assoc_buffers, out);
+       } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
+       kunmap_atomic(kaddr, KM_USER0);
+
+       if (!TestSetPageWriteback(clone_page))
+               inc_zone_page_state(clone_page, NR_WRITEBACK);
+       unlock_page(clone_page);
+
+       return 0;
+}
+
+static int nilfs_test_page_to_be_frozen(struct page *page)
+{
+       struct address_space *mapping = page->mapping;
+
+       if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
+               return 0;
+
+       if (page_mapped(page)) {
+               ClearPageChecked(page);
+               return 1;
+       }
+       return PageChecked(page);
+}
+
+static int nilfs_begin_page_io(struct page *page, struct list_head *out)
+{
+       if (!page || PageWriteback(page))
+               /* For split b-tree node pages, this function may be called
+                  twice.  We ignore the 2nd or later calls by this check. */
+               return 0;
+
+       lock_page(page);
+       clear_page_dirty_for_io(page);
+       set_page_writeback(page);
+       unlock_page(page);
+
+       if (nilfs_test_page_to_be_frozen(page)) {
+               int err = nilfs_copy_replace_page_buffers(page, out);
+               if (unlikely(err))
+                       return err;
+       }
+       return 0;
+}
+
+static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
+                                      struct page **failed_page)
+{
+       struct nilfs_segment_buffer *segbuf;
+       struct page *bd_page = NULL, *fs_page = NULL;
+       struct list_head *list = &sci->sc_copied_buffers;
+       int err;
+
+       *failed_page = NULL;
+       list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
+               struct buffer_head *bh;
+
+               list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
+                                   b_assoc_buffers) {
+                       if (bh->b_page != bd_page) {
+                               if (bd_page) {
+                                       lock_page(bd_page);
+                                       clear_page_dirty_for_io(bd_page);
+                                       set_page_writeback(bd_page);
+                                       unlock_page(bd_page);
+                               }
+                               bd_page = bh->b_page;
+                       }
+               }
+
+               list_for_each_entry(bh, &segbuf->sb_payload_buffers,
+                                   b_assoc_buffers) {
+                       if (bh == sci->sc_super_root) {
+                               if (bh->b_page != bd_page) {
+                                       lock_page(bd_page);
+                                       clear_page_dirty_for_io(bd_page);
+                                       set_page_writeback(bd_page);
+                                       unlock_page(bd_page);
+                                       bd_page = bh->b_page;
+                               }
+                               break;
+                       }
+                       if (bh->b_page != fs_page) {
+                               err = nilfs_begin_page_io(fs_page, list);
+                               if (unlikely(err)) {
+                                       *failed_page = fs_page;
+                                       goto out;
+                               }
+                               fs_page = bh->b_page;
+                       }
+               }
+       }
+       if (bd_page) {
+               lock_page(bd_page);
+               clear_page_dirty_for_io(bd_page);
+               set_page_writeback(bd_page);
+               unlock_page(bd_page);
+       }
+       err = nilfs_begin_page_io(fs_page, list);
+       if (unlikely(err))
+               *failed_page = fs_page;
+ out:
+       return err;
+}
+
+static int nilfs_segctor_write(struct nilfs_sc_info *sci,
+                              struct backing_dev_info *bdi)
+{
+       struct nilfs_segment_buffer *segbuf;
+       struct nilfs_write_info wi;
+       int err, res;
+
+       wi.sb = sci->sc_super;
+       wi.bh_sr = sci->sc_super_root;
+       wi.bdi = bdi;
+
+       list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
+               nilfs_segbuf_prepare_write(segbuf, &wi);
+               err = nilfs_segbuf_write(segbuf, &wi);
+
+               res = nilfs_segbuf_wait(segbuf, &wi);
+               err = unlikely(err) ? : res;
+               if (unlikely(err))
+                       return err;
+       }
+       return 0;
+}
+
+static int nilfs_page_has_uncleared_buffer(struct page *page)
+{
+       struct buffer_head *head, *bh;
+
+       head = bh = page_buffers(page);
+       do {
+               if (buffer_dirty(bh) && !list_empty(&bh->b_assoc_buffers))
+                       return 1;
+               bh = bh->b_this_page;
+       } while (bh != head);
+       return 0;
+}
+
+static void __nilfs_end_page_io(struct page *page, int err)
+{
+       /* BUG_ON(err > 0); */
+       if (!err) {
+               if (!nilfs_page_buffers_clean(page))
+                       __set_page_dirty_nobuffers(page);
+               ClearPageError(page);
+       } else {
+               __set_page_dirty_nobuffers(page);
+               SetPageError(page);
+       }
+
+       if (buffer_nilfs_allocated(page_buffers(page))) {
+               if (TestClearPageWriteback(page))
+                       dec_zone_page_state(page, NR_WRITEBACK);
+       } else
+               end_page_writeback(page);
+}
+
+static void nilfs_end_page_io(struct page *page, int err)
+{
+       if (!page)
+               return;
+
+       if (buffer_nilfs_node(page_buffers(page)) &&
+           nilfs_page_has_uncleared_buffer(page))
+               /* For b-tree node pages, this function may be called twice
+                  or more because they might be split in a segment.
+                  This check assures that cleanup has been done for all
+                  buffers in a split btnode page. */
+               return;
+
+       __nilfs_end_page_io(page, err);
+}
+
+static void nilfs_clear_copied_buffers(struct list_head *list, int err)
+{
+       struct buffer_head *bh, *head;
+       struct page *page;
+
+       while (!list_empty(list)) {
+               bh = list_entry(list->next, struct buffer_head,
+                               b_assoc_buffers);
+               page = bh->b_page;
+               page_cache_get(page);
+               head = bh = page_buffers(page);
+               do {
+                       if (!list_empty(&bh->b_assoc_buffers)) {
+                               list_del_init(&bh->b_assoc_buffers);
+                               if (!err) {
+                                       set_buffer_uptodate(bh);
+                                       clear_buffer_dirty(bh);
+                                       clear_buffer_nilfs_volatile(bh);
+                               }
+                               brelse(bh); /* for b_assoc_buffers */
+                       }
+               } while ((bh = bh->b_this_page) != head);
+
+               __nilfs_end_page_io(page, err);
+               page_cache_release(page);
+       }
+}
+
+static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
+                                     struct page *failed_page, int err)
+{
+       struct nilfs_segment_buffer *segbuf;
+       struct page *bd_page = NULL, *fs_page = NULL;
+
+       list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
+               struct buffer_head *bh;
+
+               list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
+                                   b_assoc_buffers) {
+                       if (bh->b_page != bd_page) {
+                               if (bd_page)
+                                       end_page_writeback(bd_page);
+                               bd_page = bh->b_page;
+                       }
+               }
+
+               list_for_each_entry(bh, &segbuf->sb_payload_buffers,
+                                   b_assoc_buffers) {
+                       if (bh == sci->sc_super_root) {
+                               if (bh->b_page != bd_page) {
+                                       end_page_writeback(bd_page);
+                                       bd_page = bh->b_page;
+                               }
+                               break;
+                       }
+                       if (bh->b_page != fs_page) {
+                               nilfs_end_page_io(fs_page, err);
+                               if (unlikely(fs_page == failed_page))
+                                       goto done;
+                               fs_page = bh->b_page;
+                       }
+               }
+       }
+       if (bd_page)
+               end_page_writeback(bd_page);
+
+       nilfs_end_page_io(fs_page, err);
+ done:
+       nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
+}
+
+static void nilfs_set_next_segment(struct the_nilfs *nilfs,
+                                  struct nilfs_segment_buffer *segbuf)
+{
+       nilfs->ns_segnum = segbuf->sb_segnum;
+       nilfs->ns_nextnum = segbuf->sb_nextnum;
+       nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
+               + segbuf->sb_sum.nblocks;
+       nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
+       nilfs->ns_ctime = segbuf->sb_sum.ctime;
+}
+
+static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
+{
+       struct nilfs_segment_buffer *segbuf;
+       struct page *bd_page = NULL, *fs_page = NULL;
+       struct nilfs_sb_info *sbi = sci->sc_sbi;
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+       int update_sr = (sci->sc_super_root != NULL);
+
+       list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
+               struct buffer_head *bh;
+
+               list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
+                                   b_assoc_buffers) {
+                       set_buffer_uptodate(bh);
+                       clear_buffer_dirty(bh);
+                       if (bh->b_page != bd_page) {
+                               if (bd_page)
+                                       end_page_writeback(bd_page);
+                               bd_page = bh->b_page;
+                       }
+               }
+               /*
+                * We assume that the buffers which belong to the same page
+                * continue over the buffer list.
+                * Under this assumption, the last BHs of pages is
+                * identifiable by the discontinuity of bh->b_page
+                * (page != fs_page).
+                *
+                * For B-tree node blocks, however, this assumption is not
+                * guaranteed.  The cleanup code of B-tree node pages needs
+                * special care.
+                */
+               list_for_each_entry(bh, &segbuf->sb_payload_buffers,
+                                   b_assoc_buffers) {
+                       set_buffer_uptodate(bh);
+                       clear_buffer_dirty(bh);
+                       clear_buffer_nilfs_volatile(bh);
+                       if (bh == sci->sc_super_root) {
+                               if (bh->b_page != bd_page) {
+                                       end_page_writeback(bd_page);
+                                       bd_page = bh->b_page;
+                               }
+                               break;
+                       }
+                       if (bh->b_page != fs_page) {
+                               nilfs_end_page_io(fs_page, 0);
+                               fs_page = bh->b_page;
+                       }
+               }
+
+               if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
+                       if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
+                               set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
+                               sci->sc_lseg_stime = jiffies;
+                       }
+                       if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
+                               clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
+               }
+       }
+       /*
+        * Since pages may continue over multiple segment buffers,
+        * end of the last page must be checked outside of the loop.
+        */
+       if (bd_page)
+               end_page_writeback(bd_page);
+
+       nilfs_end_page_io(fs_page, 0);
+
+       nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
+
+       nilfs_drop_collected_inodes(&sci->sc_dirty_files);
+
+       if (nilfs_doing_gc()) {
+               nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
+               if (update_sr)
+                       nilfs_commit_gcdat_inode(nilfs);
+       } else {
+               nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
+               set_nilfs_cond_nongc_write(nilfs);
+               wake_up(&nilfs->ns_cleanerd_wq);
+       }
+
+       sci->sc_nblk_inc += sci->sc_nblk_this_inc;
+
+       segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
+       nilfs_set_next_segment(nilfs, segbuf);
+
+       if (update_sr) {
+               nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
+                                      segbuf->sb_sum.seg_seq, nilfs->ns_cno);
+
+               clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
+               set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
+       } else
+               clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
+}
+
+static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
+                                       struct nilfs_sb_info *sbi)
+{
+       struct nilfs_inode_info *ii, *n;
+       __u64 cno = sbi->s_nilfs->ns_cno;
+
+       spin_lock(&sbi->s_inode_lock);
+ retry:
+       list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
+               if (!ii->i_bh) {
+                       struct buffer_head *ibh;
+                       int err;
+
+                       spin_unlock(&sbi->s_inode_lock);
+                       err = nilfs_ifile_get_inode_block(
+                               sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
+                       if (unlikely(err)) {
+                               nilfs_warning(sbi->s_super, __func__,
+                                             "failed to get inode block.\n");
+                               return err;
+                       }
+                       nilfs_mdt_mark_buffer_dirty(ibh);
+                       nilfs_mdt_mark_dirty(sbi->s_ifile);
+                       spin_lock(&sbi->s_inode_lock);
+                       if (likely(!ii->i_bh))
+                               ii->i_bh = ibh;
+                       else
+                               brelse(ibh);
+                       goto retry;
+               }
+               ii->i_cno = cno;
+
+               clear_bit(NILFS_I_QUEUED, &ii->i_state);
+               set_bit(NILFS_I_BUSY, &ii->i_state);
+               list_del(&ii->i_dirty);
+               list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
+       }
+       spin_unlock(&sbi->s_inode_lock);
+
+       NILFS_I(sbi->s_ifile)->i_cno = cno;
+
+       return 0;
+}
+
+static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
+                                         struct nilfs_sb_info *sbi)
+{
+       struct nilfs_transaction_info *ti = current->journal_info;
+       struct nilfs_inode_info *ii, *n;
+       __u64 cno = sbi->s_nilfs->ns_cno;
+
+       spin_lock(&sbi->s_inode_lock);
+       list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
+               if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
+                   test_bit(NILFS_I_DIRTY, &ii->i_state)) {
+                       /* The current checkpoint number (=nilfs->ns_cno) is
+                          changed between check-in and check-out only if the
+                          super root is written out.  So, we can update i_cno
+                          for the inodes that remain in the dirty list. */
+                       ii->i_cno = cno;
+                       continue;
+               }
+               clear_bit(NILFS_I_BUSY, &ii->i_state);
+               brelse(ii->i_bh);
+               ii->i_bh = NULL;
+               list_del(&ii->i_dirty);
+               list_add_tail(&ii->i_dirty, &ti->ti_garbage);
+       }
+       spin_unlock(&sbi->s_inode_lock);
+}
+
+/*
+ * Nasty routines to manipulate active flags on sufile.
+ * These would be removed in a future release.
+ */
+static void nilfs_segctor_reactivate_segments(struct nilfs_sc_info *sci,
+                                             struct the_nilfs *nilfs)
+{
+       struct nilfs_segment_buffer *segbuf, *last;
+       struct nilfs_segment_entry *ent, *n;
+       struct inode *sufile = nilfs->ns_sufile;
+       struct list_head *head;
+
+       last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
+       nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
+               ent = segbuf->sb_segent;
+               if (!ent)
+                       break; /* ignore unmapped segments (should check it?)*/
+               nilfs_segment_usage_set_active(ent->raw_su);
+               nilfs_close_segment_entry(ent, sufile);
+       }
+
+       head = &sci->sc_active_segments;
+       list_for_each_entry_safe(ent, n, head, list) {
+               nilfs_segment_usage_set_active(ent->raw_su);
+               nilfs_close_segment_entry(ent, sufile);
+       }
+
+       down_write(&nilfs->ns_sem);
+       head = &nilfs->ns_used_segments;
+       list_for_each_entry(ent, head, list) {
+               nilfs_segment_usage_set_volatile_active(ent->raw_su);
+       }
+       up_write(&nilfs->ns_sem);
+}
+
+static int nilfs_segctor_deactivate_segments(struct nilfs_sc_info *sci,
+                                            struct the_nilfs *nilfs)
+{
+       struct nilfs_segment_buffer *segbuf, *last;
+       struct nilfs_segment_entry *ent;
+       struct inode *sufile = nilfs->ns_sufile;
+       struct list_head *head;
+       int err;
+
+       last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
+       nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
+               /*
+                * Deactivate ongoing full segments.  The last segment is kept
+                * active because it is a start point of recovery, and is not
+                * relocatable until the super block points to a newer
+                * checkpoint.
+                */
+               ent = segbuf->sb_segent;
+               if (!ent)
+                       break; /* ignore unmapped segments (should check it?)*/
+               err = nilfs_open_segment_entry(ent, sufile);
+               if (unlikely(err))
+                       goto failed;
+               nilfs_segment_usage_clear_active(ent->raw_su);
+               BUG_ON(!buffer_dirty(ent->bh_su));
+       }
+
+       head = &sci->sc_active_segments;
+       list_for_each_entry(ent, head, list) {
+               err = nilfs_open_segment_entry(ent, sufile);
+               if (unlikely(err))
+                       goto failed;
+               nilfs_segment_usage_clear_active(ent->raw_su);
+               BUG_ON(!buffer_dirty(ent->bh_su));
+       }
+
+       down_write(&nilfs->ns_sem);
+       head = &nilfs->ns_used_segments;
+       list_for_each_entry(ent, head, list) {
+               /* clear volatile active for segments of older generations */
+               nilfs_segment_usage_clear_volatile_active(ent->raw_su);
+       }
+       up_write(&nilfs->ns_sem);
+       return 0;
+
+ failed:
+       nilfs_segctor_reactivate_segments(sci, nilfs);
+       return err;
+}
+
+static void nilfs_segctor_bead_completed_segments(struct nilfs_sc_info *sci)
+{
+       struct nilfs_segment_buffer *segbuf, *last;
+       struct nilfs_segment_entry *ent;
+
+       /* move each segbuf->sb_segent to the list of used active segments */
+       last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
+       nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
+               ent = segbuf->sb_segent;
+               if (!ent)
+                       break; /* ignore unmapped segments (should check it?)*/
+               list_add_tail(&ent->list, &sci->sc_active_segments);
+               segbuf->sb_segent = NULL;
+       }
+}
+
+static void
+__nilfs_segctor_commit_deactivate_segments(struct nilfs_sc_info *sci,
+                                          struct the_nilfs *nilfs)
+
+{
+       struct nilfs_segment_entry *ent;
+
+       list_splice_init(&sci->sc_active_segments,
+                        nilfs->ns_used_segments.prev);
+
+       list_for_each_entry(ent, &nilfs->ns_used_segments, list) {
+               nilfs_segment_usage_set_volatile_active(ent->raw_su);
+               /* These segments are kept open */
+       }
+}
+
+/*
+ * Main procedure of segment constructor
+ */
+static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
+{
+       struct nilfs_sb_info *sbi = sci->sc_sbi;
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+       struct page *failed_page;
+       int err, has_sr = 0;
+
+       sci->sc_stage.scnt = NILFS_ST_INIT;
+
+       err = nilfs_segctor_check_in_files(sci, sbi);
+       if (unlikely(err))
+               goto out;
+
+       if (nilfs_test_metadata_dirty(sbi))
+               set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
+
+       if (nilfs_segctor_clean(sci))
+               goto out;
+
+       do {
+               sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
+
+               err = nilfs_segctor_begin_construction(sci, nilfs);
+               if (unlikely(err))
+                       goto out;
+
+               /* Update time stamp */
+               sci->sc_seg_ctime = get_seconds();
+
+               err = nilfs_segctor_collect(sci, nilfs, mode);
+               if (unlikely(err))
+                       goto failed;
+
+               has_sr = (sci->sc_super_root != NULL);
+
+               /* Avoid empty segment */
+               if (sci->sc_stage.scnt == NILFS_ST_DONE &&
+                   NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
+                       BUG_ON(mode == SC_LSEG_SR);
+                       nilfs_segctor_end_construction(sci, nilfs, 1);
+                       goto out;
+               }
+
+               err = nilfs_segctor_assign(sci, mode);
+               if (unlikely(err))
+                       goto failed;
+
+               if (has_sr) {
+                       err = nilfs_segctor_deactivate_segments(sci, nilfs);
+                       if (unlikely(err))
+                               goto failed;
+               }
+               if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
+                       nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
+
+               if (has_sr) {
+                       err = nilfs_segctor_fill_in_checkpoint(sci);
+                       if (unlikely(err))
+                               goto failed_to_make_up;
+
+                       nilfs_segctor_fill_in_super_root(sci, nilfs);
+               }
+               nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
+
+               /* Write partial segments */
+               err = nilfs_segctor_prepare_write(sci, &failed_page);
+               if (unlikely(err))
+                       goto failed_to_write;
+
+               nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
+
+               err = nilfs_segctor_write(sci, nilfs->ns_bdi);
+               if (unlikely(err))
+                       goto failed_to_write;
+
+               nilfs_segctor_complete_write(sci);
+
+               /* Commit segments */
+               nilfs_segctor_bead_completed_segments(sci);
+               if (has_sr) {
+                       down_write(&nilfs->ns_sem);
+                       nilfs_update_last_segment(sbi, 1);
+                       __nilfs_segctor_commit_deactivate_segments(sci, nilfs);
+                       up_write(&nilfs->ns_sem);
+                       nilfs_segctor_commit_free_segments(sci);
+                       nilfs_segctor_clear_metadata_dirty(sci);
+               }
+
+               nilfs_segctor_end_construction(sci, nilfs, 0);
+
+       } while (sci->sc_stage.scnt != NILFS_ST_DONE);
+
+       /* Clearing sketch data */
+       if (has_sr && sci->sc_sketch_inode) {
+               if (i_size_read(sci->sc_sketch_inode) == 0)
+                       clear_bit(NILFS_I_DIRTY,
+                                 &NILFS_I(sci->sc_sketch_inode)->i_state);
+               i_size_write(sci->sc_sketch_inode, 0);
+       }
+ out:
+       nilfs_segctor_destroy_segment_buffers(sci);
+       nilfs_segctor_check_out_files(sci, sbi);
+       return err;
+
+ failed_to_write:
+       nilfs_segctor_abort_write(sci, failed_page, err);
+       nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
+
+ failed_to_make_up:
+       if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
+               nilfs_redirty_inodes(&sci->sc_dirty_files);
+       if (has_sr)
+               nilfs_segctor_reactivate_segments(sci, nilfs);
+
+ failed:
+       if (nilfs_doing_gc())
+               nilfs_redirty_inodes(&sci->sc_gc_inodes);
+       nilfs_segctor_end_construction(sci, nilfs, err);
+       goto out;
+}
+
+/**
+ * nilfs_secgtor_start_timer - set timer of background write
+ * @sci: nilfs_sc_info
+ *
+ * If the timer has already been set, it ignores the new request.
+ * This function MUST be called within a section locking the segment
+ * semaphore.
+ */
+static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
+{
+       spin_lock(&sci->sc_state_lock);
+       if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
+               sci->sc_timer->expires = jiffies + sci->sc_interval;
+               add_timer(sci->sc_timer);
+               sci->sc_state |= NILFS_SEGCTOR_COMMIT;
+       }
+       spin_unlock(&sci->sc_state_lock);
+}
+
+static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
+{
+       spin_lock(&sci->sc_state_lock);
+       if (!(sci->sc_flush_request & (1 << bn))) {
+               unsigned long prev_req = sci->sc_flush_request;
+
+               sci->sc_flush_request |= (1 << bn);
+               if (!prev_req)
+                       wake_up(&sci->sc_wait_daemon);
+       }
+       spin_unlock(&sci->sc_state_lock);
+}
+
+/**
+ * nilfs_flush_segment - trigger a segment construction for resource control
+ * @sb: super block
+ * @ino: inode number of the file to be flushed out.
+ */
+void nilfs_flush_segment(struct super_block *sb, ino_t ino)
+{
+       struct nilfs_sb_info *sbi = NILFS_SB(sb);
+       struct nilfs_sc_info *sci = NILFS_SC(sbi);
+
+       if (!sci || nilfs_doing_construction())
+               return;
+       nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
+                                       /* assign bit 0 to data files */
+}
+
+int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info *sci,
+                                          __u64 *segnum, size_t nsegs)
+{
+       struct nilfs_segment_entry *ent;
+       struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
+       struct inode *sufile = nilfs->ns_sufile;
+       LIST_HEAD(list);
+       __u64 *pnum;
+       const char *flag_name;
+       size_t i;
+       int err, err2 = 0;
+
+       for (pnum = segnum, i = 0; i < nsegs; pnum++, i++) {
+               ent = nilfs_alloc_segment_entry(*pnum);
+               if (unlikely(!ent)) {
+                       err = -ENOMEM;
+                       goto failed;
+               }
+               list_add_tail(&ent->list, &list);
+
+               err = nilfs_open_segment_entry(ent, sufile);
+               if (unlikely(err))
+                       goto failed;
+
+               if (unlikely(le32_to_cpu(ent->raw_su->su_flags) !=
+                            (1UL << NILFS_SEGMENT_USAGE_DIRTY))) {
+                       if (nilfs_segment_usage_clean(ent->raw_su))
+                               flag_name = "clean";
+                       else if (nilfs_segment_usage_active(ent->raw_su))
+                               flag_name = "active";
+                       else if (nilfs_segment_usage_volatile_active(
+                                        ent->raw_su))
+                               flag_name = "volatile active";
+                       else if (!nilfs_segment_usage_dirty(ent->raw_su))
+                               flag_name = "non-dirty";
+                       else
+                               flag_name = "erroneous";
+
+                       printk(KERN_ERR
+                              "NILFS: %s segment is requested to be cleaned "
+                              "(segnum=%llu)\n",
+                              flag_name, (unsigned long long)ent->segnum);
+                       err2 = -EINVAL;
+               }
+               nilfs_close_segment_entry(ent, sufile);
+       }
+       if (unlikely(err2)) {
+               err = err2;
+               goto failed;
+       }
+       list_splice(&list, sci->sc_cleaning_segments.prev);
+       return 0;
+
+ failed:
+       nilfs_dispose_segment_list(&list);
+       return err;
+}
+
+void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info *sci)
+{
+       nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
+}
+
+struct nilfs_segctor_wait_request {
+       wait_queue_t    wq;
+       __u32           seq;
+       int             err;
+       atomic_t        done;
+};
+
+static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
+{
+       struct nilfs_segctor_wait_request wait_req;
+       int err = 0;
+
+       spin_lock(&sci->sc_state_lock);
+       init_wait(&wait_req.wq);
+       wait_req.err = 0;
+       atomic_set(&wait_req.done, 0);
+       wait_req.seq = ++sci->sc_seq_request;
+       spin_unlock(&sci->sc_state_lock);
+
+       init_waitqueue_entry(&wait_req.wq, current);
+       add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
+       set_current_state(TASK_INTERRUPTIBLE);
+       wake_up(&sci->sc_wait_daemon);
+
+       for (;;) {
+               if (atomic_read(&wait_req.done)) {
+                       err = wait_req.err;
+                       break;
+               }
+               if (!signal_pending(current)) {
+                       schedule();
+                       continue;
+               }
+               err = -ERESTARTSYS;
+               break;
+       }
+       finish_wait(&sci->sc_wait_request, &wait_req.wq);
+       return err;
+}
+
+static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
+{
+       struct nilfs_segctor_wait_request *wrq, *n;
+       unsigned long flags;
+
+       spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
+       list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
+                                wq.task_list) {
+               if (!atomic_read(&wrq->done) &&
+                   nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
+                       wrq->err = err;
+                       atomic_set(&wrq->done, 1);
+               }
+               if (atomic_read(&wrq->done)) {
+                       wrq->wq.func(&wrq->wq,
+                                    TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
+                                    0, NULL);
+               }
+       }
+       spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
+}
+
+/**
+ * nilfs_construct_segment - construct a logical segment
+ * @sb: super block
+ *
+ * Return Value: On success, 0 is retured. On errors, one of the following
+ * negative error code is returned.
+ *
+ * %-EROFS - Read only filesystem.
+ *
+ * %-EIO - I/O error
+ *
+ * %-ENOSPC - No space left on device (only in a panic state).
+ *
+ * %-ERESTARTSYS - Interrupted.
+ *
+ * %-ENOMEM - Insufficient memory available.
+ */
+int nilfs_construct_segment(struct super_block *sb)
+{
+       struct nilfs_sb_info *sbi = NILFS_SB(sb);
+       struct nilfs_sc_info *sci = NILFS_SC(sbi);
+       struct nilfs_transaction_info *ti;
+       int err;
+
+       if (!sci)
+               return -EROFS;
+
+       /* A call inside transactions causes a deadlock. */
+       BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
+
+       err = nilfs_segctor_sync(sci);
+       return err;
+}
+
+/**
+ * nilfs_construct_dsync_segment - construct a data-only logical segment
+ * @sb: super block
+ * @inode: the inode whose data blocks should be written out
+ *
+ * Return Value: On success, 0 is retured. On errors, one of the following
+ * negative error code is returned.
+ *
+ * %-EROFS - Read only filesystem.
+ *
+ * %-EIO - I/O error
+ *
+ * %-ENOSPC - No space left on device (only in a panic state).
+ *
+ * %-ERESTARTSYS - Interrupted.
+ *
+ * %-ENOMEM - Insufficient memory available.
+ */
+int nilfs_construct_dsync_segment(struct super_block *sb,
+                                 struct inode *inode)
+{
+       struct nilfs_sb_info *sbi = NILFS_SB(sb);
+       struct nilfs_sc_info *sci = NILFS_SC(sbi);
+       struct nilfs_inode_info *ii;
+       struct nilfs_transaction_info ti;
+       int err = 0;
+
+       if (!sci)
+               return -EROFS;
+
+       nilfs_transaction_lock(sbi, &ti, 0);
+
+       ii = NILFS_I(inode);
+       if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
+           nilfs_test_opt(sbi, STRICT_ORDER) ||
+           test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
+           nilfs_discontinued(sbi->s_nilfs)) {
+               nilfs_transaction_unlock(sbi);
+               err = nilfs_segctor_sync(sci);
+               return err;
+       }
+
+       spin_lock(&sbi->s_inode_lock);
+       if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
+           !test_bit(NILFS_I_BUSY, &ii->i_state)) {
+               spin_unlock(&sbi->s_inode_lock);
+               nilfs_transaction_unlock(sbi);
+               return 0;
+       }
+       spin_unlock(&sbi->s_inode_lock);
+       sci->sc_stage.dirty_file_ptr = ii;
+
+       err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
+
+       nilfs_transaction_unlock(sbi);
+       return err;
+}
+
+struct nilfs_segctor_req {
+       int mode;
+       __u32 seq_accepted;
+       int sc_err;  /* construction failure */
+       int sb_err;  /* super block writeback failure */
+};
+
+#define FLUSH_FILE_BIT (0x1) /* data file only */
+#define FLUSH_DAT_BIT  (1 << NILFS_DAT_INO) /* DAT only */
+
+static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
+                                struct nilfs_segctor_req *req)
+{
+       BUG_ON(!sci);
+
+       req->sc_err = req->sb_err = 0;
+       spin_lock(&sci->sc_state_lock);
+       req->seq_accepted = sci->sc_seq_request;
+       spin_unlock(&sci->sc_state_lock);
+
+       if (sci->sc_timer)
+               del_timer_sync(sci->sc_timer);
+}
+
+static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
+                                struct nilfs_segctor_req *req)
+{
+       /* Clear requests (even when the construction failed) */
+       spin_lock(&sci->sc_state_lock);
+
+       sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
+
+       if (req->mode == SC_LSEG_SR) {
+               sci->sc_seq_done = req->seq_accepted;
+               nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
+               sci->sc_flush_request = 0;
+       } else if (req->mode == SC_FLUSH_FILE)
+               sci->sc_flush_request &= ~FLUSH_FILE_BIT;
+       else if (req->mode == SC_FLUSH_DAT)
+               sci->sc_flush_request &= ~FLUSH_DAT_BIT;
+
+       spin_unlock(&sci->sc_state_lock);
+}
+
+static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
+                                  struct nilfs_segctor_req *req)
+{
+       struct nilfs_sb_info *sbi = sci->sc_sbi;
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+       int err = 0;
+
+       if (nilfs_discontinued(nilfs))
+               req->mode = SC_LSEG_SR;
+       if (!nilfs_segctor_confirm(sci)) {
+               err = nilfs_segctor_do_construct(sci, req->mode);
+               req->sc_err = err;
+       }
+       if (likely(!err)) {
+               if (req->mode != SC_FLUSH_DAT)
+                       atomic_set(&nilfs->ns_ndirtyblks, 0);
+               if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
+                   nilfs_discontinued(nilfs)) {
+                       down_write(&nilfs->ns_sem);
+                       req->sb_err = nilfs_commit_super(sbi);
+                       up_write(&nilfs->ns_sem);
+               }
+       }
+       return err;
+}
+
+static void nilfs_construction_timeout(unsigned long data)
+{
+       struct task_struct *p = (struct task_struct *)data;
+       wake_up_process(p);
+}
+
+static void
+nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
+{
+       struct nilfs_inode_info *ii, *n;
+
+       list_for_each_entry_safe(ii, n, head, i_dirty) {
+               if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
+                       continue;
+               hlist_del_init(&ii->vfs_inode.i_hash);
+               list_del_init(&ii->i_dirty);
+               nilfs_clear_gcinode(&ii->vfs_inode);
+       }
+}
+
+int nilfs_clean_segments(struct super_block *sb, void __user *argp)
+{
+       struct nilfs_sb_info *sbi = NILFS_SB(sb);
+       struct nilfs_sc_info *sci = NILFS_SC(sbi);
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+       struct nilfs_transaction_info ti;
+       struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
+       int err;
+
+       if (unlikely(!sci))
+               return -EROFS;
+
+       nilfs_transaction_lock(sbi, &ti, 1);
+
+       err = nilfs_init_gcdat_inode(nilfs);
+       if (unlikely(err))
+               goto out_unlock;
+       err = nilfs_ioctl_prepare_clean_segments(nilfs, argp);
+       if (unlikely(err))
+               goto out_unlock;
+
+       list_splice_init(&nilfs->ns_gc_inodes, sci->sc_gc_inodes.prev);
+
+       for (;;) {
+               nilfs_segctor_accept(sci, &req);
+               err = nilfs_segctor_construct(sci, &req);
+               nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
+               nilfs_segctor_notify(sci, &req);
+
+               if (likely(!err))
+                       break;
+
+               nilfs_warning(sb, __func__,
+                             "segment construction failed. (err=%d)", err);
+               set_current_state(TASK_INTERRUPTIBLE);
+               schedule_timeout(sci->sc_interval);
+       }
+
+ out_unlock:
+       nilfs_clear_gcdat_inode(nilfs);
+       nilfs_transaction_unlock(sbi);
+       return err;
+}
+
+static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
+{
+       struct nilfs_sb_info *sbi = sci->sc_sbi;
+       struct nilfs_transaction_info ti;
+       struct nilfs_segctor_req req = { .mode = mode };
+
+       nilfs_transaction_lock(sbi, &ti, 0);
+
+       nilfs_segctor_accept(sci, &req);
+       nilfs_segctor_construct(sci, &req);
+       nilfs_segctor_notify(sci, &req);
+
+       /*
+        * Unclosed segment should be retried.  We do this using sc_timer.
+        * Timeout of sc_timer will invoke complete construction which leads
+        * to close the current logical segment.
+        */
+       if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
+               nilfs_segctor_start_timer(sci);
+
+       nilfs_transaction_unlock(sbi);
+}
+
+static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
+{
+       int mode = 0;
+       int err;
+
+       spin_lock(&sci->sc_state_lock);
+       mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
+               SC_FLUSH_DAT : SC_FLUSH_FILE;
+       spin_unlock(&sci->sc_state_lock);
+
+       if (mode) {
+               err = nilfs_segctor_do_construct(sci, mode);
+
+               spin_lock(&sci->sc_state_lock);
+               sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
+                       ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
+               spin_unlock(&sci->sc_state_lock);
+       }
+       clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
+}
+
+static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
+{
+       if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
+           time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
+               if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
+                       return SC_FLUSH_FILE;
+               else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
+                       return SC_FLUSH_DAT;
+       }
+       return SC_LSEG_SR;
+}
+
+/**
+ * nilfs_segctor_thread - main loop of the segment constructor thread.
+ * @arg: pointer to a struct nilfs_sc_info.
+ *
+ * nilfs_segctor_thread() initializes a timer and serves as a daemon
+ * to execute segment constructions.
+ */
+static int nilfs_segctor_thread(void *arg)
+{
+       struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
+       struct timer_list timer;
+       int timeout = 0;
+
+       init_timer(&timer);
+       timer.data = (unsigned long)current;
+       timer.function = nilfs_construction_timeout;
+       sci->sc_timer = &timer;
+
+       /* start sync. */
+       sci->sc_task = current;
+       wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
+       printk(KERN_INFO
+              "segctord starting. Construction interval = %lu seconds, "
+              "CP frequency < %lu seconds\n",
+              sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
+
+       spin_lock(&sci->sc_state_lock);
+ loop:
+       for (;;) {
+               int mode;
+
+               if (sci->sc_state & NILFS_SEGCTOR_QUIT)
+                       goto end_thread;
+
+               if (timeout || sci->sc_seq_request != sci->sc_seq_done)
+                       mode = SC_LSEG_SR;
+               else if (!sci->sc_flush_request)
+                       break;
+               else
+                       mode = nilfs_segctor_flush_mode(sci);
+
+               spin_unlock(&sci->sc_state_lock);
+               nilfs_segctor_thread_construct(sci, mode);
+               spin_lock(&sci->sc_state_lock);
+               timeout = 0;
+       }
+
+
+       if (freezing(current)) {
+               spin_unlock(&sci->sc_state_lock);
+               refrigerator();
+               spin_lock(&sci->sc_state_lock);
+       } else {
+               DEFINE_WAIT(wait);
+               int should_sleep = 1;
+
+               prepare_to_wait(&sci->sc_wait_daemon, &wait,
+                               TASK_INTERRUPTIBLE);
+
+               if (sci->sc_seq_request != sci->sc_seq_done)
+                       should_sleep = 0;
+               else if (sci->sc_flush_request)
+                       should_sleep = 0;
+               else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
+                       should_sleep = time_before(jiffies,
+                                                  sci->sc_timer->expires);
+
+               if (should_sleep) {
+                       spin_unlock(&sci->sc_state_lock);
+                       schedule();
+                       spin_lock(&sci->sc_state_lock);
+               }
+               finish_wait(&sci->sc_wait_daemon, &wait);
+               timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
+                          time_after_eq(jiffies, sci->sc_timer->expires));
+       }
+       goto loop;
+
+ end_thread:
+       spin_unlock(&sci->sc_state_lock);
+       del_timer_sync(sci->sc_timer);
+       sci->sc_timer = NULL;
+
+       /* end sync. */
+       sci->sc_task = NULL;
+       wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
+       return 0;
+}
+
+static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
+{
+       struct task_struct *t;
+
+       t = kthread_run(nilfs_segctor_thread, sci, "segctord");
+       if (IS_ERR(t)) {
+               int err = PTR_ERR(t);
+
+               printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
+                      err);
+               return err;
+       }
+       wait_event(sci->sc_wait_task, sci->sc_task != NULL);
+       return 0;
+}
+
+static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
+{
+       sci->sc_state |= NILFS_SEGCTOR_QUIT;
+
+       while (sci->sc_task) {
+               wake_up(&sci->sc_wait_daemon);
+               spin_unlock(&sci->sc_state_lock);
+               wait_event(sci->sc_wait_task, sci->sc_task == NULL);
+               spin_lock(&sci->sc_state_lock);
+       }
+}
+
+static int nilfs_segctor_init(struct nilfs_sc_info *sci,
+                             struct nilfs_recovery_info *ri)
+{
+       int err;
+       struct inode *inode = nilfs_iget(sci->sc_super, NILFS_SKETCH_INO);
+
+       sci->sc_sketch_inode = IS_ERR(inode) ? NULL : inode;
+       if (sci->sc_sketch_inode)
+               i_size_write(sci->sc_sketch_inode, 0);
+
+       sci->sc_seq_done = sci->sc_seq_request;
+       if (ri)
+               list_splice_init(&ri->ri_used_segments,
+                                sci->sc_active_segments.prev);
+
+       err = nilfs_segctor_start_thread(sci);
+       if (err) {
+               if (ri)
+                       list_splice_init(&sci->sc_active_segments,
+                                        ri->ri_used_segments.prev);
+               if (sci->sc_sketch_inode) {
+                       iput(sci->sc_sketch_inode);
+                       sci->sc_sketch_inode = NULL;
+               }
+       }
+       return err;
+}
+
+/*
+ * Setup & clean-up functions
+ */
+static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
+{
+       struct nilfs_sc_info *sci;
+
+       sci = kzalloc(sizeof(*sci), GFP_KERNEL);
+       if (!sci)
+               return NULL;
+
+       sci->sc_sbi = sbi;
+       sci->sc_super = sbi->s_super;
+
+       init_waitqueue_head(&sci->sc_wait_request);
+       init_waitqueue_head(&sci->sc_wait_daemon);
+       init_waitqueue_head(&sci->sc_wait_task);
+       spin_lock_init(&sci->sc_state_lock);
+       INIT_LIST_HEAD(&sci->sc_dirty_files);
+       INIT_LIST_HEAD(&sci->sc_segbufs);
+       INIT_LIST_HEAD(&sci->sc_gc_inodes);
+       INIT_LIST_HEAD(&sci->sc_active_segments);
+       INIT_LIST_HEAD(&sci->sc_cleaning_segments);
+       INIT_LIST_HEAD(&sci->sc_copied_buffers);
+
+       sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
+       sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
+       sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
+
+       if (sbi->s_interval)
+               sci->sc_interval = sbi->s_interval;
+       if (sbi->s_watermark)
+               sci->sc_watermark = sbi->s_watermark;
+       return sci;
+}
+
+static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
+{
+       int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
+
+       /* The segctord thread was stopped and its timer was removed.
+          But some tasks remain. */
+       do {
+               struct nilfs_sb_info *sbi = sci->sc_sbi;
+               struct nilfs_transaction_info ti;
+               struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
+
+               nilfs_transaction_lock(sbi, &ti, 0);
+               nilfs_segctor_accept(sci, &req);
+               ret = nilfs_segctor_construct(sci, &req);
+               nilfs_segctor_notify(sci, &req);
+               nilfs_transaction_unlock(sbi);
+
+       } while (ret && retrycount-- > 0);
+}
+
+/**
+ * nilfs_segctor_destroy - destroy the segment constructor.
+ * @sci: nilfs_sc_info
+ *
+ * nilfs_segctor_destroy() kills the segctord thread and frees
+ * the nilfs_sc_info struct.
+ * Caller must hold the segment semaphore.
+ */
+static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
+{
+       struct nilfs_sb_info *sbi = sci->sc_sbi;
+       int flag;
+
+       up_write(&sbi->s_nilfs->ns_segctor_sem);
+
+       spin_lock(&sci->sc_state_lock);
+       nilfs_segctor_kill_thread(sci);
+       flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
+               || sci->sc_seq_request != sci->sc_seq_done);
+       spin_unlock(&sci->sc_state_lock);
+
+       if (flag || nilfs_segctor_confirm(sci))
+               nilfs_segctor_write_out(sci);
+
+       BUG_ON(!list_empty(&sci->sc_copied_buffers));
+
+       if (!list_empty(&sci->sc_dirty_files)) {
+               nilfs_warning(sbi->s_super, __func__,
+                             "dirty file(s) after the final construction\n");
+               nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
+       }
+       if (!list_empty(&sci->sc_active_segments))
+               nilfs_dispose_segment_list(&sci->sc_active_segments);
+
+       if (!list_empty(&sci->sc_cleaning_segments))
+               nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
+
+       BUG_ON(!list_empty(&sci->sc_segbufs));
+
+       if (sci->sc_sketch_inode) {
+               iput(sci->sc_sketch_inode);
+               sci->sc_sketch_inode = NULL;
+       }
+       down_write(&sbi->s_nilfs->ns_segctor_sem);
+
+       kfree(sci);
+}
+
+/**
+ * nilfs_attach_segment_constructor - attach a segment constructor
+ * @sbi: nilfs_sb_info
+ * @ri: nilfs_recovery_info
+ *
+ * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
+ * initilizes it, and starts the segment constructor.
+ *
+ * Return Value: On success, 0 is returned. On error, one of the following
+ * negative error code is returned.
+ *
+ * %-ENOMEM - Insufficient memory available.
+ */
+int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
+                                    struct nilfs_recovery_info *ri)
+{
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+       int err;
+
+       /* Each field of nilfs_segctor is cleared through the initialization
+          of super-block info */
+       sbi->s_sc_info = nilfs_segctor_new(sbi);
+       if (!sbi->s_sc_info)
+               return -ENOMEM;
+
+       nilfs_attach_writer(nilfs, sbi);
+       err = nilfs_segctor_init(NILFS_SC(sbi), ri);
+       if (err) {
+               nilfs_detach_writer(nilfs, sbi);
+               kfree(sbi->s_sc_info);
+               sbi->s_sc_info = NULL;
+       }
+       return err;
+}
+
+/**
+ * nilfs_detach_segment_constructor - destroy the segment constructor
+ * @sbi: nilfs_sb_info
+ *
+ * nilfs_detach_segment_constructor() kills the segment constructor daemon,
+ * frees the struct nilfs_sc_info, and destroy the dirty file list.
+ */
+void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
+{
+       struct the_nilfs *nilfs = sbi->s_nilfs;
+       LIST_HEAD(garbage_list);
+
+       down_write(&nilfs->ns_segctor_sem);
+       if (NILFS_SC(sbi)) {
+               nilfs_segctor_destroy(NILFS_SC(sbi));
+               sbi->s_sc_info = NULL;
+       }
+
+       /* Force to free the list of dirty files */
+       spin_lock(&sbi->s_inode_lock);
+       if (!list_empty(&sbi->s_dirty_files)) {
+               list_splice_init(&sbi->s_dirty_files, &garbage_list);
+               nilfs_warning(sbi->s_super, __func__,
+                             "Non empty dirty list after the last "
+                             "segment construction\n");
+       }
+       spin_unlock(&sbi->s_inode_lock);
+       up_write(&nilfs->ns_segctor_sem);
+
+       nilfs_dispose_list(sbi, &garbage_list, 1);
+       nilfs_detach_writer(nilfs, sbi);
+}
diff --git a/fs/nilfs2/segment.h b/fs/nilfs2/segment.h
new file mode 100644 (file)
index 0000000..615654b
--- /dev/null
@@ -0,0 +1,246 @@
+/*
+ * segment.h - NILFS Segment constructor prototypes and definitions
+ *
+ * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Written by Ryusuke Konishi <ryusuke@osrg.net>
+ *
+ */
+#ifndef _NILFS_SEGMENT_H
+#define _NILFS_SEGMENT_H
+
+#include <linux/types.h>
+#include <linux/fs.h>
+#include <linux/buffer_head.h>
+#include <linux/nilfs2_fs.h>
+#include "sb.h"
+
+/**
+ * struct nilfs_recovery_info - Recovery infomation
+ * @ri_need_recovery: Recovery status
+ * @ri_super_root: Block number of the last super root
+ * @ri_ri_cno: Number of the last checkpoint
+ * @ri_lsegs_start: Region for roll-forwarding (start block number)
+ * @ri_lsegs_end: Region for roll-forwarding (end block number)
+ * @ri_lseg_start_seq: Sequence value of the segment at ri_lsegs_start
+ * @ri_used_segments: List of segments to be mark active
+ * @ri_pseg_start: Block number of the last partial segment
+ * @ri_seq: Sequence number on the last partial segment
+ * @ri_segnum: Segment number on the last partial segment
+ * @ri_nextnum: Next segment number on the last partial segment
+ */
+struct nilfs_recovery_info {
+       int                     ri_need_recovery;
+       sector_t                ri_super_root;
+       __u64                   ri_cno;
+
+       sector_t                ri_lsegs_start;
+       sector_t                ri_lsegs_end;
+       u64                     ri_lsegs_start_seq;
+       struct list_head        ri_used_segments;
+       sector_t                ri_pseg_start;
+       u64                     ri_seq;
+       __u64                   ri_segnum;
+       __u64                   ri_nextnum;
+};
+
+/* ri_need_recovery */
+#define NILFS_RECOVERY_SR_UPDATED       1  /* The super root was updated */
+#define NILFS_RECOVERY_ROLLFORWARD_DONE         2  /* Rollforward was carried out */
+
+/**
+ * struct nilfs_cstage - Context of collection stage
+ * @scnt: Stage count
+ * @flags: State flags
+ * @dirty_file_ptr: Pointer on dirty_files list, or inode of a target file
+ * @gc_inode_ptr: Pointer on the list of gc-inodes
+ */
+struct nilfs_cstage {
+       int                     scnt;
+       unsigned                flags;
+       struct nilfs_inode_info *dirty_file_ptr;
+       struct nilfs_inode_info *gc_inode_ptr;
+};
+
+struct nilfs_segment_buffer;
+
+struct nilfs_segsum_pointer {
+       struct buffer_head     *bh;
+       unsigned                offset; /* offset in bytes */
+};
+
+/**
+ * struct nilfs_sc_info - Segment constructor information
+ * @sc_super: Back pointer to super_block struct
+ * @sc_sbi: Back pointer to nilfs_sb_info struct
+ * @sc_nblk_inc: Block count of current generation
+ * @sc_dirty_files: List of files to be written
+ * @sc_gc_inodes: List of GC inodes having blocks to be written
+ * @sc_active_segments: List of active segments that were already written out
+ * @sc_cleaning_segments: List of segments to be freed through construction
+ * @sc_copied_buffers: List of copied buffers (buffer heads) to freeze data
+ * @sc_segbufs: List of segment buffers
+ * @sc_segbuf_nblocks: Number of available blocks in segment buffers.
+ * @sc_curseg: Current segment buffer
+ * @sc_super_root: Pointer to the super root buffer
+ * @sc_stage: Collection stage
+ * @sc_finfo_ptr: pointer to the current finfo struct in the segment summary
+ * @sc_binfo_ptr: pointer to the current binfo struct in the segment summary
+ * @sc_blk_cnt:        Block count of a file
+ * @sc_datablk_cnt: Data block count of a file
+ * @sc_nblk_this_inc: Number of blocks included in the current logical segment
+ * @sc_seg_ctime: Creation time
+ * @sc_flags: Internal flags
+ * @sc_sketch_inode: Inode of the sketch file
+ * @sc_state_lock: spinlock for sc_state and so on
+ * @sc_state: Segctord state flags
+ * @sc_flush_request: inode bitmap of metadata files to be flushed
+ * @sc_wait_request: Client request queue
+ * @sc_wait_daemon: Daemon wait queue
+ * @sc_wait_task: Start/end wait queue to control segctord task
+ * @sc_seq_request: Request counter
+ * @sc_seq_done: Completion counter
+ * @sc_sync: Request of explicit sync operation
+ * @sc_interval: Timeout value of background construction
+ * @sc_mjcp_freq: Frequency of creating checkpoints
+ * @sc_lseg_stime: Start time of the latest logical segment
+ * @sc_watermark: Watermark for the number of dirty buffers
+ * @sc_timer: Timer for segctord
+ * @sc_task: current thread of segctord
+ */
+struct nilfs_sc_info {
+       struct super_block     *sc_super;
+       struct nilfs_sb_info   *sc_sbi;
+
+       unsigned long           sc_nblk_inc;
+
+       struct list_head        sc_dirty_files;
+       struct list_head        sc_gc_inodes;
+       struct list_head        sc_active_segments;
+       struct list_head        sc_cleaning_segments;
+       struct list_head        sc_copied_buffers;
+
+       /* Segment buffers */
+       struct list_head        sc_segbufs;
+       unsigned long           sc_segbuf_nblocks;
+       struct nilfs_segment_buffer *sc_curseg;
+       struct buffer_head     *sc_super_root;
+
+       struct nilfs_cstage     sc_stage;
+
+       struct nilfs_segsum_pointer sc_finfo_ptr;
+       struct nilfs_segsum_pointer sc_binfo_ptr;
+       unsigned long           sc_blk_cnt;
+       unsigned long           sc_datablk_cnt;
+       unsigned long           sc_nblk_this_inc;
+       time_t                  sc_seg_ctime;
+
+       unsigned long           sc_flags;
+
+       /*
+        * Pointer to an inode of the sketch.
+        * This pointer is kept only while it contains data.
+        * We protect it with a semaphore of the segment constructor.
+        */
+       struct inode           *sc_sketch_inode;
+
+       spinlock_t              sc_state_lock;
+       unsigned long           sc_state;
+       unsigned long           sc_flush_request;
+
+       wait_queue_head_t       sc_wait_request;
+       wait_queue_head_t       sc_wait_daemon;
+       wait_queue_head_t       sc_wait_task;
+
+       __u32                   sc_seq_request;
+       __u32                   sc_seq_done;
+
+       int                     sc_sync;
+       unsigned long           sc_interval;
+       unsigned long           sc_mjcp_freq;
+       unsigned long           sc_lseg_stime;  /* in 1/HZ seconds */
+       unsigned long           sc_watermark;
+
+       struct timer_list      *sc_timer;
+       struct task_struct     *sc_task;
+};
+
+/* sc_flags */
+enum {
+       NILFS_SC_DIRTY,         /* One or more dirty meta-data blocks exist */
+       NILFS_SC_UNCLOSED,      /* Logical segment is not closed */
+       NILFS_SC_SUPER_ROOT,    /* The latest segment has a super root */
+       NILFS_SC_PRIOR_FLUSH,   /* Requesting immediate flush without making a
+                                  checkpoint */
+};
+
+/* sc_state */
+#define NILFS_SEGCTOR_QUIT         0x0001  /* segctord is being destroyed */
+#define NILFS_SEGCTOR_COMMIT       0x0004  /* committed transaction exists */
+
+/*
+ * Constant parameters
+ */
+#define NILFS_SC_CLEANUP_RETRY     3  /* Retry count of construction when
+                                         destroying segctord */
+
+/*
+ * Default values of timeout, in seconds.
+ */
+#define NILFS_SC_DEFAULT_TIMEOUT    5   /* Timeout value of dirty blocks.
+                                          It triggers construction of a
+                                          logical segment with a super root */
+#define NILFS_SC_DEFAULT_SR_FREQ    30  /* Maximum frequency of super root
+                                          creation */
+#define NILFS_SC_DEFAULT_SB_FREQ    30  /* Minimum interval of periodical
+                                          update of superblock (reserved) */
+
+/*
+ * The default threshold amount of data, in block counts.
+ */
+#define NILFS_SC_DEFAULT_WATERMARK  3600
+
+
+/* segment.c */
+extern int nilfs_init_transaction_cache(void);
+extern void nilfs_destroy_transaction_cache(void);
+extern void nilfs_relax_pressure_in_lock(struct super_block *);
+
+extern int nilfs_construct_segment(struct super_block *);
+extern int nilfs_construct_dsync_segment(struct super_block *,
+                                        struct inode *);
+extern void nilfs_flush_segment(struct super_block *, ino_t);
+extern int nilfs_clean_segments(struct super_block *, void __user *);
+
+extern int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info *,
+                                                 __u64 *, size_t);
+extern void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info *);
+
+extern int nilfs_attach_segment_constructor(struct nilfs_sb_info *,
+                                           struct nilfs_recovery_info *);
+extern void nilfs_detach_segment_constructor(struct nilfs_sb_info *);
+
+/* recovery.c */
+extern int nilfs_read_super_root_block(struct super_block *, sector_t,
+                                      struct buffer_head **, int);
+extern int nilfs_search_super_root(struct the_nilfs *, struct nilfs_sb_info *,
+                                  struct nilfs_recovery_info *);
+extern int nilfs_recover_logical_segments(struct the_nilfs *,
+                                         struct nilfs_sb_info *,
+                                         struct nilfs_recovery_info *);
+
+#endif /* _NILFS_SEGMENT_H */