GFS2: Wait for unlock completion on umount
[linux-2.6.git] / fs / gfs2 / meta_io.c
index da49973..6f68a5f 100644 (file)
@@ -1,10 +1,10 @@
 /*
  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
- * Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
+ * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
  *
  * This copyrighted material is made available to anyone wishing to use,
  * modify, copy, or redistribute it subject to the terms and conditions
- * of the GNU General Public License v.2.
+ * of the GNU General Public License version 2.
  */
 
 #include <linux/sched.h>
 #include <linux/writeback.h>
 #include <linux/swap.h>
 #include <linux/delay.h>
+#include <linux/bio.h>
 #include <linux/gfs2_ondisk.h>
-#include <asm/semaphore.h>
 
 #include "gfs2.h"
-#include "lm_interface.h"
 #include "incore.h"
 #include "glock.h"
 #include "glops.h"
 #include "trans.h"
 #include "util.h"
 
-#define buffer_busy(bh) \
-((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock) | (1ul << BH_Pinned)))
-#define buffer_in_io(bh) \
-((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock)))
-
-static int aspace_get_block(struct inode *inode, sector_t lblock,
-                           struct buffer_head *bh_result, int create)
+static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc)
 {
-       gfs2_assert_warn(inode->i_sb->s_fs_info, 0);
-       return -EOPNOTSUPP;
-}
-
-static int gfs2_aspace_writepage(struct page *page,
-                                struct writeback_control *wbc)
-{
-       return block_write_full_page(page, aspace_get_block, wbc);
-}
-
-/**
- * stuck_releasepage - We're stuck in gfs2_releasepage().  Print stuff out.
- * @bh: the buffer we're stuck on
- *
- */
-
-static void stuck_releasepage(struct buffer_head *bh)
-{
-       struct inode *inode = bh->b_page->mapping->host;
-       struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
-       struct gfs2_bufdata *bd = bh->b_private;
-       struct gfs2_glock *gl;
-
-       fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
-       fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
-               (uint64_t)bh->b_blocknr, atomic_read(&bh->b_count));
-       fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
-       fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
-
-       if (!bd)
-               return;
-
-       gl = bd->bd_gl;
-
-       fs_warn(sdp, "gl = (%u, %llu)\n", 
-               gl->gl_name.ln_type, gl->gl_name.ln_number);
-
-       fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
-               (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
-               (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
-
-       if (gl->gl_ops == &gfs2_inode_glops) {
-               struct gfs2_inode *ip = gl->gl_object;
-               unsigned int x;
-
-               if (!ip)
-                       return;
-
-               fs_warn(sdp, "ip = %llu %llu\n",
-                       ip->i_num.no_formal_ino, ip->i_num.no_addr);
-               fs_warn(sdp, "ip->i_count = %d, ip->i_vnode = %s\n",
-                       atomic_read(&ip->i_count),
-                       (ip->i_vnode) ? "!NULL" : "NULL");
-
-               for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
-                       fs_warn(sdp, "ip->i_cache[%u] = %s\n",
-                               x, (ip->i_cache[x]) ? "!NULL" : "NULL");
-       }
-}
-
-/**
- * gfs2_aspace_releasepage - free the metadata associated with a page
- * @page: the page that's being released
- * @gfp_mask: passed from Linux VFS, ignored by us
- *
- * Call try_to_free_buffers() if the buffers in this page can be
- * released.
- *
- * Returns: 0
- */
-
-static int gfs2_aspace_releasepage(struct page *page, gfp_t gfp_mask)
-{
-       struct inode *aspace = page->mapping->host;
-       struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
+       int err;
        struct buffer_head *bh, *head;
-       struct gfs2_bufdata *bd;
-       unsigned long t;
+       int nr_underway = 0;
+       int write_op = (1 << BIO_RW_META) | ((wbc->sync_mode == WB_SYNC_ALL ?
+                       WRITE_SYNC_PLUG : WRITE));
 
-       if (!page_has_buffers(page))
-               goto out;
-
-       head = bh = page_buffers(page);
-       do {
-               t = jiffies;
-
-               while (atomic_read(&bh->b_count)) {
-                       if (atomic_read(&aspace->i_writecount)) {
-                               if (time_after_eq(jiffies, t +
-                                   gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
-                                       stuck_releasepage(bh);
-                                       t = jiffies;
-                               }
+       BUG_ON(!PageLocked(page));
+       BUG_ON(!page_has_buffers(page));
 
-                               yield();
-                               continue;
-                       }
+       head = page_buffers(page);
+       bh = head;
 
-                       return 0;
+       do {
+               if (!buffer_mapped(bh))
+                       continue;
+               /*
+                * If it's a fully non-blocking write attempt and we cannot
+                * lock the buffer then redirty the page.  Note that this can
+                * potentially cause a busy-wait loop from pdflush and kswapd
+                * activity, but those code paths have their own higher-level
+                * throttling.
+                */
+               if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
+                       lock_buffer(bh);
+               } else if (!trylock_buffer(bh)) {
+                       redirty_page_for_writepage(wbc, page);
+                       continue;
+               }
+               if (test_clear_buffer_dirty(bh)) {
+                       mark_buffer_async_write(bh);
+               } else {
+                       unlock_buffer(bh);
                }
+       } while ((bh = bh->b_this_page) != head);
 
-               gfs2_assert_warn(sdp, !buffer_pinned(bh));
+       /*
+        * The page and its buffers are protected by PageWriteback(), so we can
+        * drop the bh refcounts early.
+        */
+       BUG_ON(PageWriteback(page));
+       set_page_writeback(page);
 
-               bd = bh->b_private;
-               if (bd) {
-                       gfs2_assert_warn(sdp, bd->bd_bh == bh);
-                       gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
-                       gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list));
-                       gfs2_assert_warn(sdp, !bd->bd_ail);
-                       kmem_cache_free(gfs2_bufdata_cachep, bd);
-                       bh->b_private = NULL;
+       do {
+               struct buffer_head *next = bh->b_this_page;
+               if (buffer_async_write(bh)) {
+                       submit_bh(write_op, bh);
+                       nr_underway++;
                }
+               bh = next;
+       } while (bh != head);
+       unlock_page(page);
 
-               bh = bh->b_this_page;
-       }
-       while (bh != head);
+       err = 0;
+       if (nr_underway == 0)
+               end_page_writeback(page);
 
- out:
-       return try_to_free_buffers(page);
+       return err;
 }
 
-static struct address_space_operations aspace_aops = {
+static const struct address_space_operations aspace_aops = {
        .writepage = gfs2_aspace_writepage,
-       .releasepage = gfs2_aspace_releasepage,
+       .releasepage = gfs2_releasepage,
+       .sync_page = block_sync_page,
 };
 
 /**
@@ -183,13 +115,15 @@ static struct address_space_operations aspace_aops = {
 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
 {
        struct inode *aspace;
+       struct gfs2_inode *ip;
 
        aspace = new_inode(sdp->sd_vfs);
        if (aspace) {
-               mapping_set_gfp_mask(aspace->i_mapping, GFP_KERNEL);
+               mapping_set_gfp_mask(aspace->i_mapping, GFP_NOFS);
                aspace->i_mapping->a_ops = &aspace_aops;
-               aspace->i_size = ~0ULL;
-               aspace->u.generic_ip = NULL;
+               aspace->i_size = MAX_LFS_FILESIZE;
+               ip = GFS2_I(aspace);
+               clear_bit(GIF_USER, &ip->i_flags);
                insert_inode_hash(aspace);
        }
        return aspace;
@@ -202,217 +136,36 @@ void gfs2_aspace_put(struct inode *aspace)
 }
 
 /**
- * gfs2_ail1_start_one - Start I/O on a part of the AIL
- * @sdp: the filesystem
- * @tr: the part of the AIL
- *
- */
-
-void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
-{
-       struct gfs2_bufdata *bd, *s;
-       struct buffer_head *bh;
-       int retry;
-
-       BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
-
-       do {
-               retry = 0;
-
-               list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
-                                                bd_ail_st_list) {
-                       bh = bd->bd_bh;
-
-                       gfs2_assert(sdp, bd->bd_ail == ai);
-
-                       if (!buffer_busy(bh)) {
-                               if (!buffer_uptodate(bh))
-                                       gfs2_io_error_bh(sdp, bh);
-                               list_move(&bd->bd_ail_st_list,
-                                         &ai->ai_ail2_list);
-                               continue;
-                       }
-
-                       if (!buffer_dirty(bh))
-                               continue;
-
-                       list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
-
-                       gfs2_log_unlock(sdp);
-                       wait_on_buffer(bh);
-                       ll_rw_block(WRITE, 1, &bh);
-                       gfs2_log_lock(sdp);
-
-                       retry = 1;
-                       break;
-               }
-       } while (retry);
-}
-
-/**
- * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
- * @sdp: the filesystem
- * @ai: the AIL entry
- *
- */
-
-int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
-{
-       struct gfs2_bufdata *bd, *s;
-       struct buffer_head *bh;
-
-       list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
-                                        bd_ail_st_list) {
-               bh = bd->bd_bh;
-
-               gfs2_assert(sdp, bd->bd_ail == ai);
-
-               if (buffer_busy(bh)) {
-                       if (flags & DIO_ALL)
-                               continue;
-                       else
-                               break;
-               }
-
-               if (!buffer_uptodate(bh))
-                       gfs2_io_error_bh(sdp, bh);
-
-               list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
-       }
-
-       return list_empty(&ai->ai_ail1_list);
-}
-
-/**
- * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
- * @sdp: the filesystem
- * @ai: the AIL entry
- *
- */
-
-void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
-{
-       struct list_head *head = &ai->ai_ail2_list;
-       struct gfs2_bufdata *bd;
-
-       while (!list_empty(head)) {
-               bd = list_entry(head->prev, struct gfs2_bufdata,
-                               bd_ail_st_list);
-               gfs2_assert(sdp, bd->bd_ail == ai);
-               bd->bd_ail = NULL;
-               list_del(&bd->bd_ail_st_list);
-               list_del(&bd->bd_ail_gl_list);
-               atomic_dec(&bd->bd_gl->gl_ail_count);
-               brelse(bd->bd_bh);
-       }
-}
-
-/**
- * ail_empty_gl - remove all buffers for a given lock from the AIL
- * @gl: the glock
- *
- * None of the buffers should be dirty, locked, or pinned.
- */
-
-void gfs2_ail_empty_gl(struct gfs2_glock *gl)
-{
-       struct gfs2_sbd *sdp = gl->gl_sbd;
-       unsigned int blocks;
-       struct list_head *head = &gl->gl_ail_list;
-       struct gfs2_bufdata *bd;
-       struct buffer_head *bh;
-       uint64_t blkno;
-       int error;
-
-       blocks = atomic_read(&gl->gl_ail_count);
-       if (!blocks)
-               return;
-
-       error = gfs2_trans_begin(sdp, 0, blocks);
-       if (gfs2_assert_withdraw(sdp, !error))
-               return;
-
-       gfs2_log_lock(sdp);
-       while (!list_empty(head)) {
-               bd = list_entry(head->next, struct gfs2_bufdata,
-                               bd_ail_gl_list);
-               bh = bd->bd_bh;
-               blkno = bh->b_blocknr;
-               gfs2_assert_withdraw(sdp, !buffer_busy(bh));
-
-               bd->bd_ail = NULL;
-               list_del(&bd->bd_ail_st_list);
-               list_del(&bd->bd_ail_gl_list);
-               atomic_dec(&gl->gl_ail_count);
-               brelse(bh);
-               gfs2_log_unlock(sdp);
-
-               gfs2_trans_add_revoke(sdp, blkno);
-
-               gfs2_log_lock(sdp);
-       }
-       gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
-       gfs2_log_unlock(sdp);
-
-       gfs2_trans_end(sdp);
-       gfs2_log_flush(sdp, NULL);
-}
-
-/**
- * gfs2_meta_inval - Invalidate all buffers associated with a glock
- * @gl: the glock
- *
- */
-
-void gfs2_meta_inval(struct gfs2_glock *gl)
-{
-       struct gfs2_sbd *sdp = gl->gl_sbd;
-       struct inode *aspace = gl->gl_aspace;
-       struct address_space *mapping = gl->gl_aspace->i_mapping;
-
-       gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
-
-       atomic_inc(&aspace->i_writecount);
-       truncate_inode_pages(mapping, 0);
-       atomic_dec(&aspace->i_writecount);
-
-       gfs2_assert_withdraw(sdp, !mapping->nrpages);
-}
-
-/**
  * gfs2_meta_sync - Sync all buffers associated with a glock
  * @gl: The glock
- * @flags: DIO_START | DIO_WAIT
  *
  */
 
-void gfs2_meta_sync(struct gfs2_glock *gl, int flags)
+void gfs2_meta_sync(struct gfs2_glock *gl)
 {
        struct address_space *mapping = gl->gl_aspace->i_mapping;
-       int error = 0;
+       int error;
 
-       if (flags & DIO_START)
-               filemap_fdatawrite(mapping);
-       if (!error && (flags & DIO_WAIT))
-               error = filemap_fdatawait(mapping);
+       filemap_fdatawrite(mapping);
+       error = filemap_fdatawait(mapping);
 
        if (error)
                gfs2_io_error(gl->gl_sbd);
 }
 
 /**
- * getbuf - Get a buffer with a given address space
- * @sdp: the filesystem
- * @aspace: the address space
+ * gfs2_getbuf - Get a buffer with a given address space
+ * @gl: the glock
  * @blkno: the block number (filesystem scope)
  * @create: 1 if the buffer should be created
  *
  * Returns: the buffer
  */
 
-static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
-                                 uint64_t blkno, int create)
+struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create)
 {
+       struct address_space *mapping = gl->gl_aspace->i_mapping;
+       struct gfs2_sbd *sdp = gl->gl_sbd;
        struct page *page;
        struct buffer_head *bh;
        unsigned int shift;
@@ -425,13 +178,13 @@ static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
 
        if (create) {
                for (;;) {
-                       page = grab_cache_page(aspace->i_mapping, index);
+                       page = grab_cache_page(mapping, index);
                        if (page)
                                break;
                        yield();
                }
        } else {
-               page = find_lock_page(aspace->i_mapping, index);
+               page = find_lock_page(mapping, index);
                if (!page)
                        return NULL;
        }
@@ -474,10 +227,10 @@ static void meta_prep_new(struct buffer_head *bh)
  * Returns: The buffer
  */
 
-struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, uint64_t blkno)
+struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
 {
        struct buffer_head *bh;
-       bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
+       bh = gfs2_getbuf(gl, blkno, CREATE);
        meta_prep_new(bh);
        return bh;
 }
@@ -486,57 +239,69 @@ struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, uint64_t blkno)
  * gfs2_meta_read - Read a block from disk
  * @gl: The glock covering the block
  * @blkno: The block number
- * @flags: flags to gfs2_dreread()
+ * @flags: flags
  * @bhp: the place where the buffer is returned (NULL on failure)
  *
  * Returns: errno
  */
 
-int gfs2_meta_read(struct gfs2_glock *gl, uint64_t blkno, int flags,
+int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
                   struct buffer_head **bhp)
 {
-       int error;
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+       struct buffer_head *bh;
 
-       *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
-       error = gfs2_meta_reread(gl->gl_sbd, *bhp, flags);
-       if (error)
-               brelse(*bhp);
+       if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
+               return -EIO;
 
-       return error;
+       *bhp = bh = gfs2_getbuf(gl, blkno, CREATE);
+
+       lock_buffer(bh);
+       if (buffer_uptodate(bh)) {
+               unlock_buffer(bh);
+               return 0;
+       }
+       bh->b_end_io = end_buffer_read_sync;
+       get_bh(bh);
+       submit_bh(READ_SYNC | (1 << BIO_RW_META), bh);
+       if (!(flags & DIO_WAIT))
+               return 0;
+
+       wait_on_buffer(bh);
+       if (unlikely(!buffer_uptodate(bh))) {
+               struct gfs2_trans *tr = current->journal_info;
+               if (tr && tr->tr_touched)
+                       gfs2_io_error_bh(sdp, bh);
+               brelse(bh);
+               return -EIO;
+       }
+
+       return 0;
 }
 
 /**
- * gfs2_meta_reread - Reread a block from disk
+ * gfs2_meta_wait - Reread a block from disk
  * @sdp: the filesystem
- * @bh: The block to read
- * @flags: Flags that control the read
+ * @bh: The block to wait for
  *
  * Returns: errno
  */
 
-int gfs2_meta_reread(struct gfs2_sbd *sdp, struct buffer_head *bh, int flags)
+int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
 {
        if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
                return -EIO;
 
-       if (flags & DIO_FORCE)
-               clear_buffer_uptodate(bh);
-
-       if ((flags & DIO_START) && !buffer_uptodate(bh))
-               ll_rw_block(READ, 1, &bh);
-
-       if (flags & DIO_WAIT) {
-               wait_on_buffer(bh);
+       wait_on_buffer(bh);
 
-               if (!buffer_uptodate(bh)) {
-                       struct gfs2_trans *tr = current->journal_info;
-                       if (tr && tr->tr_touched)
-                               gfs2_io_error_bh(sdp, bh);
-                       return -EIO;
-               }
-               if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
-                       return -EIO;
+       if (!buffer_uptodate(bh)) {
+               struct gfs2_trans *tr = current->journal_info;
+               if (tr && tr->tr_touched)
+                       gfs2_io_error_bh(sdp, bh);
+               return -EIO;
        }
+       if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
+               return -EIO;
 
        return 0;
 }
@@ -562,93 +327,50 @@ void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
                return;
        }
 
-       bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL),
-       memset(bd, 0, sizeof(struct gfs2_bufdata));
-
+       bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
        bd->bd_bh = bh;
        bd->bd_gl = gl;
 
        INIT_LIST_HEAD(&bd->bd_list_tr);
-       if (meta) {
+       if (meta)
                lops_init_le(&bd->bd_le, &gfs2_buf_lops);
-       } else {
+       else
                lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
-               get_bh(bh);
-       }
        bh->b_private = bd;
 
        if (meta)
                unlock_page(bh->b_page);
 }
 
-/**
- * gfs2_pin - Pin a buffer in memory
- * @sdp: the filesystem the buffer belongs to
- * @bh: The buffer to be pinned
- *
- */
-
-void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
-{
-       struct gfs2_bufdata *bd = bh->b_private;
-
-       gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
-
-       if (test_set_buffer_pinned(bh))
-               gfs2_assert_withdraw(sdp, 0);
-
-       wait_on_buffer(bh);
-
-       /* If this buffer is in the AIL and it has already been written
-          to in-place disk block, remove it from the AIL. */
-
-       gfs2_log_lock(sdp);
-       if (bd->bd_ail && !buffer_in_io(bh))
-               list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
-       gfs2_log_unlock(sdp);
-
-       clear_buffer_dirty(bh);
-       wait_on_buffer(bh);
-
-       if (!buffer_uptodate(bh))
-               gfs2_io_error_bh(sdp, bh);
-
-       get_bh(bh);
-}
-
-/**
- * gfs2_unpin - Unpin a buffer
- * @sdp: the filesystem the buffer belongs to
- * @bh: The buffer to unpin
- * @ai:
- *
- */
-
-void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
-               struct gfs2_ail *ai)
+void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta)
 {
+       struct gfs2_sbd *sdp = GFS2_SB(bh->b_page->mapping->host);
        struct gfs2_bufdata *bd = bh->b_private;
-
-       gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
-
-       if (!buffer_pinned(bh))
-               gfs2_assert_withdraw(sdp, 0);
-
-       mark_buffer_dirty(bh);
-       clear_buffer_pinned(bh);
-
-       gfs2_log_lock(sdp);
-       if (bd->bd_ail) {
-               list_del(&bd->bd_ail_st_list);
+       if (test_clear_buffer_pinned(bh)) {
+               list_del_init(&bd->bd_le.le_list);
+               if (meta) {
+                       gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
+                       sdp->sd_log_num_buf--;
+                       tr->tr_num_buf_rm++;
+               } else {
+                       gfs2_assert_warn(sdp, sdp->sd_log_num_databuf);
+                       sdp->sd_log_num_databuf--;
+                       tr->tr_num_databuf_rm++;
+               }
+               tr->tr_touched = 1;
                brelse(bh);
-       } else {
-               struct gfs2_glock *gl = bd->bd_gl;
-               list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
-               atomic_inc(&gl->gl_ail_count);
        }
-       bd->bd_ail = ai;
-       list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
-       gfs2_log_unlock(sdp);
+       if (bd) {
+               if (bd->bd_ail) {
+                       gfs2_remove_from_ail(bd);
+                       bh->b_private = NULL;
+                       bd->bd_bh = NULL;
+                       bd->bd_blkno = bh->b_blocknr;
+                       gfs2_trans_add_revoke(sdp, bd);
+               }
+       }
+       clear_buffer_dirty(bh);
+       clear_buffer_uptodate(bh);
 }
 
 /**
@@ -659,47 +381,19 @@ void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  *
  */
 
-void gfs2_meta_wipe(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
+void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
 {
-       struct gfs2_sbd *sdp = ip->i_sbd;
-       struct inode *aspace = ip->i_gl->gl_aspace;
+       struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
        struct buffer_head *bh;
 
        while (blen) {
-               bh = getbuf(sdp, aspace, bstart, NO_CREATE);
+               bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE);
                if (bh) {
-                       struct gfs2_bufdata *bd = bh->b_private;
-
-                       if (test_clear_buffer_pinned(bh)) {
-                               struct gfs2_trans *tr = current->journal_info;
-                               gfs2_log_lock(sdp);
-                               list_del_init(&bd->bd_le.le_list);
-                               gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
-                               sdp->sd_log_num_buf--;
-                               gfs2_log_unlock(sdp);
-                               tr->tr_num_buf_rm++;
-                               brelse(bh);
-                       }
-                       if (bd) {
-                               gfs2_log_lock(sdp);
-                               if (bd->bd_ail) {
-                                       uint64_t blkno = bh->b_blocknr;
-                                       bd->bd_ail = NULL;
-                                       list_del(&bd->bd_ail_st_list);
-                                       list_del(&bd->bd_ail_gl_list);
-                                       atomic_dec(&bd->bd_gl->gl_ail_count);
-                                       brelse(bh);
-                                       gfs2_log_unlock(sdp);
-                                       gfs2_trans_add_revoke(sdp, blkno);
-                               } else
-                                       gfs2_log_unlock(sdp);
-                       }
-
                        lock_buffer(bh);
-                       clear_buffer_dirty(bh);
-                       clear_buffer_uptodate(bh);
+                       gfs2_log_lock(sdp);
+                       gfs2_remove_from_journal(bh, current->journal_info, 1);
+                       gfs2_log_unlock(sdp);
                        unlock_buffer(bh);
-
                        brelse(bh);
                }
 
@@ -709,32 +403,6 @@ void gfs2_meta_wipe(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
 }
 
 /**
- * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
- * @ip: The GFS2 inode
- *
- * This releases buffers that are in the most-recently-used array of
- * blocks used for indirect block addressing for this inode.
- */
-
-void gfs2_meta_cache_flush(struct gfs2_inode *ip)
-{
-       struct buffer_head **bh_slot;
-       unsigned int x;
-
-       spin_lock(&ip->i_spin);
-
-       for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) {
-               bh_slot = &ip->i_cache[x];
-               if (!*bh_slot)
-                       break;
-               brelse(*bh_slot);
-               *bh_slot = NULL;
-       }
-
-       spin_unlock(&ip->i_spin);
-}
-
-/**
  * gfs2_meta_indirect_buffer - Get a metadata buffer
  * @ip: The GFS2 inode
  * @height: The level of this buf in the metadata (indir addr) tree (if any)
@@ -742,75 +410,33 @@ void gfs2_meta_cache_flush(struct gfs2_inode *ip)
  * @new: Non-zero if we may create a new buffer
  * @bhp: the buffer is returned here
  *
- * Try to use the gfs2_inode's MRU metadata tree cache.
- *
  * Returns: errno
  */
 
-int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, uint64_t num,
+int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
                              int new, struct buffer_head **bhp)
 {
-       struct buffer_head *bh, **bh_slot = ip->i_cache + height;
-       int error;
-
-       spin_lock(&ip->i_spin);
-       bh = *bh_slot;
-       if (bh) {
-               if (bh->b_blocknr == num)
-                       get_bh(bh);
-               else
-                       bh = NULL;
-       }
-       spin_unlock(&ip->i_spin);
-
-       if (bh) {
-               if (new)
-                       meta_prep_new(bh);
-               else {
-                       error = gfs2_meta_reread(ip->i_sbd, bh,
-                                                DIO_START | DIO_WAIT);
-                       if (error) {
-                               brelse(bh);
-                               return error;
-                       }
-               }
-       } else {
-               if (new)
-                       bh = gfs2_meta_new(ip->i_gl, num);
-               else {
-                       error = gfs2_meta_read(ip->i_gl, num,
-                                              DIO_START | DIO_WAIT, &bh);
-                       if (error)
-                               return error;
-               }
-
-               spin_lock(&ip->i_spin);
-               if (*bh_slot != bh) {
-                       brelse(*bh_slot);
-                       *bh_slot = bh;
-                       get_bh(bh);
-               }
-               spin_unlock(&ip->i_spin);
-       }
+       struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
+       struct gfs2_glock *gl = ip->i_gl;
+       struct buffer_head *bh;
+       int ret = 0;
 
        if (new) {
-               if (gfs2_assert_warn(ip->i_sbd, height)) {
-                       brelse(bh);
-                       return -EIO;
-               }
+               BUG_ON(height == 0);
+               bh = gfs2_meta_new(gl, num);
                gfs2_trans_add_bh(ip->i_gl, bh, 1);
                gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
                gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
-
-       } else if (gfs2_metatype_check(ip->i_sbd, bh,
-                            (height) ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)) {
-               brelse(bh);
-               return -EIO;
+       } else {
+               u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
+               ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh);
+               if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
+                       brelse(bh);
+                       ret = -EIO;
+               }
        }
-
        *bhp = bh;
-
-       return 0;
+       return ret;
 }
 
 /**
@@ -819,71 +445,47 @@ int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, uint64_t num,
  * @dblock: the starting disk block
  * @extlen: the number of blocks in the extent
  *
+ * returns: the first buffer in the extent
  */
 
-void gfs2_meta_ra(struct gfs2_glock *gl, uint64_t dblock, uint32_t extlen)
+struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
 {
        struct gfs2_sbd *sdp = gl->gl_sbd;
-       struct inode *aspace = gl->gl_aspace;
        struct buffer_head *first_bh, *bh;
-       uint32_t max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
+       u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
                          sdp->sd_sb.sb_bsize_shift;
-       int error;
 
-       if (!extlen || !max_ra)
-               return;
+       BUG_ON(!extlen);
+
+       if (max_ra < 1)
+               max_ra = 1;
        if (extlen > max_ra)
                extlen = max_ra;
 
-       first_bh = getbuf(sdp, aspace, dblock, CREATE);
+       first_bh = gfs2_getbuf(gl, dblock, CREATE);
 
        if (buffer_uptodate(first_bh))
                goto out;
-       if (!buffer_locked(first_bh)) {
-               error = gfs2_meta_reread(sdp, first_bh, DIO_START);
-               if (error)
-                       goto out;
-       }
+       if (!buffer_locked(first_bh))
+               ll_rw_block(READ_SYNC | (1 << BIO_RW_META), 1, &first_bh);
 
        dblock++;
        extlen--;
 
        while (extlen) {
-               bh = getbuf(sdp, aspace, dblock, CREATE);
-
-               if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
-                       error = gfs2_meta_reread(sdp, bh, DIO_START);
-                       brelse(bh);
-                       if (error)
-                               goto out;
-               } else
-                       brelse(bh);
+               bh = gfs2_getbuf(gl, dblock, CREATE);
 
+               if (!buffer_uptodate(bh) && !buffer_locked(bh))
+                       ll_rw_block(READA, 1, &bh);
+               brelse(bh);
                dblock++;
                extlen--;
-
-               if (buffer_uptodate(first_bh))
-                       break;
+               if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
+                       goto out;
        }
 
- out:
-       brelse(first_bh);
-}
-
-/**
- * gfs2_meta_syncfs - sync all the buffers in a filesystem
- * @sdp: the filesystem
- *
- */
-
-void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
-{
-       gfs2_log_flush(sdp, NULL);
-       for (;;) {
-               gfs2_ail1_start(sdp, DIO_ALL);
-               if (gfs2_ail1_empty(sdp, DIO_ALL))
-                       break;
-               msleep(10);
-       }
+       wait_on_buffer(first_bh);
+out:
+       return first_bh;
 }