nilfs2: B-tree node cache
authorRyusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Tue, 7 Apr 2009 02:01:25 +0000 (19:01 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 7 Apr 2009 15:31:13 +0000 (08:31 -0700)
This adds routines for B-tree node buffers.

Signed-off-by: Seiji Kihara <kihara.seiji@lab.ntt.co.jp>
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/btnode.c [new file with mode: 0644]
fs/nilfs2/btnode.h [new file with mode: 0644]

diff --git a/fs/nilfs2/btnode.c b/fs/nilfs2/btnode.c
new file mode 100644 (file)
index 0000000..4cc07b2
--- /dev/null
@@ -0,0 +1,316 @@
+/*
+ * btnode.c - NILFS B-tree node cache
+ *
+ * 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
+ *
+ * This file was originally written by Seiji Kihara <kihara@osrg.net>
+ * and fully revised by Ryusuke Konishi <ryusuke@osrg.net> for
+ * stabilization and simplification.
+ *
+ */
+
+#include <linux/types.h>
+#include <linux/buffer_head.h>
+#include <linux/mm.h>
+#include <linux/backing-dev.h>
+#include "nilfs.h"
+#include "mdt.h"
+#include "dat.h"
+#include "page.h"
+#include "btnode.h"
+
+
+void nilfs_btnode_cache_init_once(struct address_space *btnc)
+{
+       INIT_RADIX_TREE(&btnc->page_tree, GFP_ATOMIC);
+       spin_lock_init(&btnc->tree_lock);
+       INIT_LIST_HEAD(&btnc->private_list);
+       spin_lock_init(&btnc->private_lock);
+
+       spin_lock_init(&btnc->i_mmap_lock);
+       INIT_RAW_PRIO_TREE_ROOT(&btnc->i_mmap);
+       INIT_LIST_HEAD(&btnc->i_mmap_nonlinear);
+}
+
+static struct address_space_operations def_btnode_aops;
+
+void nilfs_btnode_cache_init(struct address_space *btnc)
+{
+       btnc->host = NULL;  /* can safely set to host inode ? */
+       btnc->flags = 0;
+       mapping_set_gfp_mask(btnc, GFP_NOFS);
+       btnc->assoc_mapping = NULL;
+       btnc->backing_dev_info = &default_backing_dev_info;
+       btnc->a_ops = &def_btnode_aops;
+}
+
+void nilfs_btnode_cache_clear(struct address_space *btnc)
+{
+       invalidate_mapping_pages(btnc, 0, -1);
+       truncate_inode_pages(btnc, 0);
+}
+
+int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
+                             sector_t pblocknr, struct buffer_head **pbh,
+                             int newblk)
+{
+       struct buffer_head *bh;
+       struct inode *inode = NILFS_BTNC_I(btnc);
+       int err;
+
+       bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
+       if (unlikely(!bh))
+               return -ENOMEM;
+
+       err = -EEXIST; /* internal code */
+       if (newblk) {
+               if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
+                            buffer_dirty(bh))) {
+                       brelse(bh);
+                       BUG();
+               }
+               bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
+               bh->b_blocknr = blocknr;
+               set_buffer_mapped(bh);
+               set_buffer_uptodate(bh);
+               goto found;
+       }
+
+       if (buffer_uptodate(bh) || buffer_dirty(bh))
+               goto found;
+
+       if (pblocknr == 0) {
+               pblocknr = blocknr;
+               if (inode->i_ino != NILFS_DAT_INO) {
+                       struct inode *dat =
+                               nilfs_dat_inode(NILFS_I_NILFS(inode));
+
+                       /* blocknr is a virtual block number */
+                       err = nilfs_dat_translate(dat, blocknr, &pblocknr);
+                       if (unlikely(err)) {
+                               brelse(bh);
+                               goto out_locked;
+                       }
+               }
+       }
+       lock_buffer(bh);
+       if (buffer_uptodate(bh)) {
+               unlock_buffer(bh);
+               err = -EEXIST; /* internal code */
+               goto found;
+       }
+       set_buffer_mapped(bh);
+       bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
+       bh->b_blocknr = pblocknr; /* set block address for read */
+       bh->b_end_io = end_buffer_read_sync;
+       get_bh(bh);
+       submit_bh(READ, bh);
+       bh->b_blocknr = blocknr; /* set back to the given block address */
+       err = 0;
+found:
+       *pbh = bh;
+
+out_locked:
+       unlock_page(bh->b_page);
+       page_cache_release(bh->b_page);
+       return err;
+}
+
+int nilfs_btnode_get(struct address_space *btnc, __u64 blocknr,
+                    sector_t pblocknr, struct buffer_head **pbh, int newblk)
+{
+       struct buffer_head *bh;
+       int err;
+
+       err = nilfs_btnode_submit_block(btnc, blocknr, pblocknr, pbh, newblk);
+       if (err == -EEXIST) /* internal code (cache hit) */
+               return 0;
+       if (unlikely(err))
+               return err;
+
+       bh = *pbh;
+       wait_on_buffer(bh);
+       if (!buffer_uptodate(bh)) {
+               brelse(bh);
+               return -EIO;
+       }
+       return 0;
+}
+
+/**
+ * nilfs_btnode_delete - delete B-tree node buffer
+ * @bh: buffer to be deleted
+ *
+ * nilfs_btnode_delete() invalidates the specified buffer and delete the page
+ * including the buffer if the page gets unbusy.
+ */
+void nilfs_btnode_delete(struct buffer_head *bh)
+{
+       struct address_space *mapping;
+       struct page *page = bh->b_page;
+       pgoff_t index = page_index(page);
+       int still_dirty;
+
+       page_cache_get(page);
+       lock_page(page);
+       wait_on_page_writeback(page);
+
+       nilfs_forget_buffer(bh);
+       still_dirty = PageDirty(page);
+       mapping = page->mapping;
+       unlock_page(page);
+       page_cache_release(page);
+
+       if (!still_dirty && mapping)
+               invalidate_inode_pages2_range(mapping, index, index);
+}
+
+/**
+ * nilfs_btnode_prepare_change_key
+ *  prepare to move contents of the block for old key to one of new key.
+ *  the old buffer will not be removed, but might be reused for new buffer.
+ *  it might return -ENOMEM because of memory allocation errors,
+ *  and might return -EIO because of disk read errors.
+ */
+int nilfs_btnode_prepare_change_key(struct address_space *btnc,
+                                   struct nilfs_btnode_chkey_ctxt *ctxt)
+{
+       struct buffer_head *obh, *nbh;
+       struct inode *inode = NILFS_BTNC_I(btnc);
+       __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
+       int err;
+
+       if (oldkey == newkey)
+               return 0;
+
+       obh = ctxt->bh;
+       ctxt->newbh = NULL;
+
+       if (inode->i_blkbits == PAGE_CACHE_SHIFT) {
+               lock_page(obh->b_page);
+               /*
+                * We cannot call radix_tree_preload for the kernels older
+                * than 2.6.23, because it is not exported for modules.
+                */
+               err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
+               if (err)
+                       goto failed_unlock;
+               /* BUG_ON(oldkey != obh->b_page->index); */
+               if (unlikely(oldkey != obh->b_page->index))
+                       NILFS_PAGE_BUG(obh->b_page,
+                                      "invalid oldkey %lld (newkey=%lld)",
+                                      (unsigned long long)oldkey,
+                                      (unsigned long long)newkey);
+
+retry:
+               spin_lock_irq(&btnc->tree_lock);
+               err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
+               spin_unlock_irq(&btnc->tree_lock);
+               /*
+                * Note: page->index will not change to newkey until
+                * nilfs_btnode_commit_change_key() will be called.
+                * To protect the page in intermediate state, the page lock
+                * is held.
+                */
+               radix_tree_preload_end();
+               if (!err)
+                       return 0;
+               else if (err != -EEXIST)
+                       goto failed_unlock;
+
+               err = invalidate_inode_pages2_range(btnc, newkey, newkey);
+               if (!err)
+                       goto retry;
+               /* fallback to copy mode */
+               unlock_page(obh->b_page);
+       }
+
+       err = nilfs_btnode_get(btnc, newkey, 0, &nbh, 1);
+       if (likely(!err)) {
+               BUG_ON(nbh == obh);
+               ctxt->newbh = nbh;
+       }
+       return err;
+
+ failed_unlock:
+       unlock_page(obh->b_page);
+       return err;
+}
+
+/**
+ * nilfs_btnode_commit_change_key
+ *  commit the change_key operation prepared by prepare_change_key().
+ */
+void nilfs_btnode_commit_change_key(struct address_space *btnc,
+                                   struct nilfs_btnode_chkey_ctxt *ctxt)
+{
+       struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
+       __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
+       struct page *opage;
+
+       if (oldkey == newkey)
+               return;
+
+       if (nbh == NULL) {      /* blocksize == pagesize */
+               opage = obh->b_page;
+               if (unlikely(oldkey != opage->index))
+                       NILFS_PAGE_BUG(opage,
+                                      "invalid oldkey %lld (newkey=%lld)",
+                                      (unsigned long long)oldkey,
+                                      (unsigned long long)newkey);
+               if (!test_set_buffer_dirty(obh) && TestSetPageDirty(opage))
+                       BUG();
+
+               spin_lock_irq(&btnc->tree_lock);
+               radix_tree_delete(&btnc->page_tree, oldkey);
+               radix_tree_tag_set(&btnc->page_tree, newkey,
+                                  PAGECACHE_TAG_DIRTY);
+               spin_unlock_irq(&btnc->tree_lock);
+
+               opage->index = obh->b_blocknr = newkey;
+               unlock_page(opage);
+       } else {
+               nilfs_copy_buffer(nbh, obh);
+               nilfs_btnode_mark_dirty(nbh);
+
+               nbh->b_blocknr = newkey;
+               ctxt->bh = nbh;
+               nilfs_btnode_delete(obh); /* will decrement bh->b_count */
+       }
+}
+
+/**
+ * nilfs_btnode_abort_change_key
+ *  abort the change_key operation prepared by prepare_change_key().
+ */
+void nilfs_btnode_abort_change_key(struct address_space *btnc,
+                                  struct nilfs_btnode_chkey_ctxt *ctxt)
+{
+       struct buffer_head *nbh = ctxt->newbh;
+       __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
+
+       if (oldkey == newkey)
+               return;
+
+       if (nbh == NULL) {      /* blocksize == pagesize */
+               spin_lock_irq(&btnc->tree_lock);
+               radix_tree_delete(&btnc->page_tree, newkey);
+               spin_unlock_irq(&btnc->tree_lock);
+               unlock_page(ctxt->bh->b_page);
+       } else
+               brelse(nbh);
+}
diff --git a/fs/nilfs2/btnode.h b/fs/nilfs2/btnode.h
new file mode 100644 (file)
index 0000000..35faa86
--- /dev/null
@@ -0,0 +1,58 @@
+/*
+ * btnode.h - NILFS B-tree node cache
+ *
+ * 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 Seiji Kihara <kihara@osrg.net>
+ * Revised by Ryusuke Konishi <ryusuke@osrg.net>
+ */
+
+#ifndef _NILFS_BTNODE_H
+#define _NILFS_BTNODE_H
+
+#include <linux/types.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/backing-dev.h>
+
+
+struct nilfs_btnode_chkey_ctxt {
+       __u64 oldkey;
+       __u64 newkey;
+       struct buffer_head *bh;
+       struct buffer_head *newbh;
+};
+
+void nilfs_btnode_cache_init_once(struct address_space *);
+void nilfs_btnode_cache_init(struct address_space *);
+void nilfs_btnode_cache_clear(struct address_space *);
+int nilfs_btnode_submit_block(struct address_space *, __u64, sector_t,
+                             struct buffer_head **, int);
+int nilfs_btnode_get(struct address_space *, __u64, sector_t,
+                    struct buffer_head **, int);
+void nilfs_btnode_delete(struct buffer_head *);
+int nilfs_btnode_prepare_change_key(struct address_space *,
+                                   struct nilfs_btnode_chkey_ctxt *);
+void nilfs_btnode_commit_change_key(struct address_space *,
+                                   struct nilfs_btnode_chkey_ctxt *);
+void nilfs_btnode_abort_change_key(struct address_space *,
+                                  struct nilfs_btnode_chkey_ctxt *);
+
+#define nilfs_btnode_mark_dirty(bh)    nilfs_mark_buffer_dirty(bh)
+
+
+#endif /* _NILFS_BTNODE_H */