Merge master.kernel.org:/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw
[pandora-kernel.git] / fs / gfs2 / lops.c
index d501e82..4d7f94d 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) 2004-2006 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/spinlock.h>
 #include <linux/completion.h>
 #include <linux/buffer_head.h>
-#include <asm/semaphore.h>
+#include <linux/gfs2_ondisk.h>
+#include <linux/lm_interface.h>
 
 #include "gfs2.h"
+#include "incore.h"
 #include "glock.h"
 #include "log.h"
 #include "lops.h"
 #include "recovery.h"
 #include "rgrp.h"
 #include "trans.h"
+#include "util.h"
 
 static void glock_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 {
        struct gfs2_glock *gl;
+       struct gfs2_trans *tr = current->journal_info;
 
-       get_transaction->tr_touched = 1;
+       tr->tr_touched = 1;
 
        if (!list_empty(&le->le_list))
                return;
@@ -68,7 +72,7 @@ static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
        if (!list_empty(&bd->bd_list_tr))
                return;
 
-       tr = get_transaction;
+       tr = current->journal_info;
        tr->tr_touched = 1;
        tr->tr_num_buf++;
        list_add(&bd->bd_list_tr, &tr->tr_list_buf);
@@ -79,7 +83,7 @@ static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
        gfs2_trans_add_gl(bd->bd_gl);
 
        gfs2_meta_check(sdp, bd->bd_bh);
-       gfs2_meta_pin(sdp, bd->bd_bh);
+       gfs2_pin(sdp, bd->bd_bh);
 
        gfs2_log_lock(sdp);
        sdp->sd_log_num_buf++;
@@ -114,7 +118,7 @@ static void buf_lo_before_commit(struct gfs2_sbd *sdp)
        unsigned n;
        __be64 *ptr;
 
-       offset += (sizeof(__be64) - 1);
+       offset += sizeof(__be64) - 1;
        offset &= ~(sizeof(__be64) - 1);
        limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
        /* for 4k blocks, limit = 503 */
@@ -125,11 +129,12 @@ static void buf_lo_before_commit(struct gfs2_sbd *sdp)
                if (total > limit)
                        num = limit;
                bh = gfs2_log_get_buf(sdp);
+               sdp->sd_log_num_hdrs++;
                ld = (struct gfs2_log_descriptor *)bh->b_data;
                ptr = (__be64 *)(bh->b_data + offset);
                ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
-               ld->ld_header.mh_type = cpu_to_be16(GFS2_METATYPE_LD);
-               ld->ld_header.mh_format = cpu_to_be16(GFS2_FORMAT_LD);
+               ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
+               ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
                ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_METADATA);
                ld->ld_length = cpu_to_be32(num + 1);
                ld->ld_data1 = cpu_to_be32(num);
@@ -137,7 +142,8 @@ static void buf_lo_before_commit(struct gfs2_sbd *sdp)
                memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 
                n = 0;
-               list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf, bd_le.le_list) {
+               list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
+                                            bd_le.le_list) {
                        *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
                        if (++n >= num)
                                break;
@@ -147,7 +153,8 @@ static void buf_lo_before_commit(struct gfs2_sbd *sdp)
                ll_rw_block(WRITE, 1, &bh);
 
                n = 0;
-               list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf, bd_le.le_list) {
+               list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
+                                            bd_le.le_list) {
                        bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
                        set_buffer_dirty(bh);
                        ll_rw_block(WRITE, 1, &bh);
@@ -169,15 +176,15 @@ static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
                list_del_init(&bd->bd_le.le_list);
                sdp->sd_log_num_buf--;
 
-               gfs2_meta_unpin(sdp, bd->bd_bh, ai);
+               gfs2_unpin(sdp, bd->bd_bh, ai);
        }
        gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 }
 
 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
-                              struct gfs2_log_header *head, int pass)
+                              struct gfs2_log_header_host *head, int pass)
 {
-       struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
+       struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 
        if (pass != 0)
                return;
@@ -190,11 +197,12 @@ static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
                                struct gfs2_log_descriptor *ld, __be64 *ptr,
                                int pass)
 {
-       struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
-       struct gfs2_glock *gl = jd->jd_inode->i_gl;
+       struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
+       struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
+       struct gfs2_glock *gl = ip->i_gl;
        unsigned int blks = be32_to_cpu(ld->ld_data1);
        struct buffer_head *bh_log, *bh_ip;
-       uint64_t blkno;
+       u64 blkno;
        int error = 0;
 
        if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
@@ -211,8 +219,8 @@ static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
                        continue;
 
                error = gfs2_replay_read_block(jd, start, &bh_log);
-                if (error)
-                        return error;
+               if (error)
+                       return error;
 
                bh_ip = gfs2_meta_new(gl, blkno);
                memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
@@ -236,16 +244,17 @@ static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 
 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 {
-       struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
+       struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
+       struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 
        if (error) {
-               gfs2_meta_sync(jd->jd_inode->i_gl, DIO_START | DIO_WAIT);
+               gfs2_meta_sync(ip->i_gl);
                return;
        }
        if (pass != 1)
                return;
 
-       gfs2_meta_sync(jd->jd_inode->i_gl, DIO_START | DIO_WAIT);
+       gfs2_meta_sync(ip->i_gl);
 
        fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
                jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
@@ -255,7 +264,7 @@ static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 {
        struct gfs2_trans *tr;
 
-       tr = get_transaction;
+       tr = current->journal_info;
        tr->tr_touched = 1;
        tr->tr_num_revoke++;
 
@@ -280,10 +289,11 @@ static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
        bh = gfs2_log_get_buf(sdp);
        ld = (struct gfs2_log_descriptor *)bh->b_data;
        ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
-       ld->ld_header.mh_type = cpu_to_be16(GFS2_METATYPE_LD);
-       ld->ld_header.mh_format = cpu_to_be16(GFS2_FORMAT_LD);
+       ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
+       ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
        ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE);
-       ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(uint64_t)));
+       ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
+                                                   sizeof(u64)));
        ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
        ld->ld_data2 = cpu_to_be32(0);
        memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
@@ -291,25 +301,25 @@ static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 
        while (!list_empty(head)) {
                rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list);
-               list_del(&rv->rv_le.le_list);
+               list_del_init(&rv->rv_le.le_list);
                sdp->sd_log_num_revoke--;
 
-               if (offset + sizeof(uint64_t) > sdp->sd_sb.sb_bsize) {
+               if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
                        set_buffer_dirty(bh);
                        ll_rw_block(WRITE, 1, &bh);
 
                        bh = gfs2_log_get_buf(sdp);
                        mh = (struct gfs2_meta_header *)bh->b_data;
                        mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
-                       mh->mh_type = cpu_to_be16(GFS2_METATYPE_LB);
-                       mh->mh_format = cpu_to_be16(GFS2_FORMAT_LB);
+                       mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
+                       mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
                        offset = sizeof(struct gfs2_meta_header);
                }
 
                *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno);
                kfree(rv);
 
-               offset += sizeof(uint64_t);
+               offset += sizeof(u64);
        }
        gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 
@@ -318,9 +328,9 @@ static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 }
 
 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
-                                 struct gfs2_log_header *head, int pass)
+                                 struct gfs2_log_header_host *head, int pass)
 {
-       struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
+       struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 
        if (pass != 0)
                return;
@@ -333,12 +343,12 @@ static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
                                   struct gfs2_log_descriptor *ld, __be64 *ptr,
                                   int pass)
 {
-       struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
+       struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
        unsigned int blks = be32_to_cpu(ld->ld_length);
        unsigned int revokes = be32_to_cpu(ld->ld_data1);
        struct buffer_head *bh;
        unsigned int offset;
-       uint64_t blkno;
+       u64 blkno;
        int first = 1;
        int error;
 
@@ -355,7 +365,7 @@ static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
                if (!first)
                        gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 
-               while (offset + sizeof(uint64_t) <= sdp->sd_sb.sb_bsize) {
+               while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
                        blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 
                        error = gfs2_revoke_add(sdp, blkno, start);
@@ -366,7 +376,7 @@ static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 
                        if (!--revokes)
                                break;
-                       offset += sizeof(uint64_t);
+                       offset += sizeof(u64);
                }
 
                brelse(bh);
@@ -379,7 +389,7 @@ static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 
 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 {
-       struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
+       struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 
        if (error) {
                gfs2_revoke_clean(sdp);
@@ -397,8 +407,9 @@ static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 {
        struct gfs2_rgrpd *rgd;
+       struct gfs2_trans *tr = current->journal_info;
 
-       get_transaction->tr_touched = 1;
+       tr->tr_touched = 1;
 
        if (!list_empty(&le->le_list))
                return;
@@ -409,7 +420,7 @@ static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
        gfs2_log_lock(sdp);
        sdp->sd_log_num_rg++;
        list_add(&le->le_list, &sdp->sd_log_le_rg);
-       gfs2_log_unlock(sdp);   
+       gfs2_log_unlock(sdp);
 }
 
 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
@@ -428,70 +439,336 @@ static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
        gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
 }
 
+/**
+ * databuf_lo_add - Add a databuf to the transaction.
+ *
+ * This is used in two distinct cases:
+ * i) In ordered write mode
+ *    We put the data buffer on a list so that we can ensure that its
+ *    synced to disk at the right time
+ * ii) In journaled data mode
+ *    We need to journal the data block in the same way as metadata in
+ *    the functions above. The difference is that here we have a tag
+ *    which is two __be64's being the block number (as per meta data)
+ *    and a flag which says whether the data block needs escaping or
+ *    not. This means we need a new log entry for each 251 or so data
+ *    blocks, which isn't an enormous overhead but twice as much as
+ *    for normal metadata blocks.
+ */
 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 {
-       get_transaction->tr_touched = 1;
+       struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
+       struct gfs2_trans *tr = current->journal_info;
+       struct address_space *mapping = bd->bd_bh->b_page->mapping;
+       struct gfs2_inode *ip = GFS2_I(mapping->host);
 
+       tr->tr_touched = 1;
+       if (list_empty(&bd->bd_list_tr) &&
+           (ip->i_di.di_flags & GFS2_DIF_JDATA)) {
+               tr->tr_num_buf++;
+               list_add(&bd->bd_list_tr, &tr->tr_list_buf);
+               gfs2_pin(sdp, bd->bd_bh);
+               tr->tr_num_buf_new++;
+       }
+       gfs2_trans_add_gl(bd->bd_gl);
        gfs2_log_lock(sdp);
-       sdp->sd_log_num_databuf++;
-       list_add(&le->le_list, &sdp->sd_log_le_databuf);
+       if (list_empty(&le->le_list)) {
+               if (ip->i_di.di_flags & GFS2_DIF_JDATA)
+                       sdp->sd_log_num_jdata++;
+               sdp->sd_log_num_databuf++;
+               list_add(&le->le_list, &sdp->sd_log_le_databuf);
+       }
        gfs2_log_unlock(sdp);
 }
 
+static int gfs2_check_magic(struct buffer_head *bh)
+{
+       struct page *page = bh->b_page;
+       void *kaddr;
+       __be32 *ptr;
+       int rv = 0;
+
+       kaddr = kmap_atomic(page, KM_USER0);
+       ptr = kaddr + bh_offset(bh);
+       if (*ptr == cpu_to_be32(GFS2_MAGIC))
+               rv = 1;
+       kunmap_atomic(kaddr, KM_USER0);
+
+       return rv;
+}
+
+/**
+ * databuf_lo_before_commit - Scan the data buffers, writing as we go
+ *
+ * Here we scan through the lists of buffers and make the assumption
+ * that any buffer thats been pinned is being journaled, and that
+ * any unpinned buffer is an ordered write data buffer and therefore
+ * will be written back rather than journaled.
+ */
 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 {
-       struct list_head *head = &sdp->sd_log_le_databuf;
        LIST_HEAD(started);
-       struct gfs2_databuf *db;
-       struct buffer_head *bh;
+       struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt;
+       struct buffer_head *bh = NULL,*bh1 = NULL;
+       unsigned int offset = sizeof(struct gfs2_log_descriptor);
+       struct gfs2_log_descriptor *ld;
+       unsigned int limit;
+       unsigned int total_dbuf = sdp->sd_log_num_databuf;
+       unsigned int total_jdata = sdp->sd_log_num_jdata;
+       unsigned int num, n;
+       __be64 *ptr = NULL;
 
-       while (!list_empty(head)) {
-               db = list_entry(head->prev, struct gfs2_databuf, db_le.le_list);
-               list_move(&db->db_le.le_list, &started);
+       offset += 2*sizeof(__be64) - 1;
+       offset &= ~(2*sizeof(__be64) - 1);
+       limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
 
-               gfs2_log_lock(sdp);
-               bh = db->db_bh;
+       /*
+        * Start writing ordered buffers, write journaled buffers
+        * into the log along with a header
+        */
+       gfs2_log_lock(sdp);
+       bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf,
+                                      bd_le.le_list);
+       while(total_dbuf) {
+               num = total_jdata;
+               if (num > limit)
+                       num = limit;
+               n = 0;
+               list_for_each_entry_safe_continue(bd1, bdt,
+                                                 &sdp->sd_log_le_databuf,
+                                                 bd_le.le_list) {
+                       /* store off the buffer head in a local ptr since
+                        * gfs2_bufdata might change when we drop the log lock
+                        */
+                       bh1 = bd1->bd_bh;
+
+                       /* An ordered write buffer */
+                       if (bh1 && !buffer_pinned(bh1)) {
+                               list_move(&bd1->bd_le.le_list, &started);
+                               if (bd1 == bd2) {
+                                       bd2 = NULL;
+                                       bd2 = list_prepare_entry(bd2,
+                                                       &sdp->sd_log_le_databuf,
+                                                       bd_le.le_list);
+                               }
+                               total_dbuf--;
+                               if (bh1) {
+                                       if (buffer_dirty(bh1)) {
+                                               get_bh(bh1);
+
+                                               gfs2_log_unlock(sdp);
+
+                                               ll_rw_block(SWRITE, 1, &bh1);
+                                               brelse(bh1);
+
+                                               gfs2_log_lock(sdp);
+                                       }
+                                       continue;
+                               }
+                               continue;
+                       } else if (bh1) { /* A journaled buffer */
+                               int magic;
+                               gfs2_log_unlock(sdp);
+                               if (!bh) {
+                                       bh = gfs2_log_get_buf(sdp);
+                                       sdp->sd_log_num_hdrs++;
+                                       ld = (struct gfs2_log_descriptor *)
+                                            bh->b_data;
+                                       ptr = (__be64 *)(bh->b_data + offset);
+                                       ld->ld_header.mh_magic =
+                                               cpu_to_be32(GFS2_MAGIC);
+                                       ld->ld_header.mh_type =
+                                               cpu_to_be32(GFS2_METATYPE_LD);
+                                       ld->ld_header.mh_format =
+                                               cpu_to_be32(GFS2_FORMAT_LD);
+                                       ld->ld_type =
+                                               cpu_to_be32(GFS2_LOG_DESC_JDATA);
+                                       ld->ld_length = cpu_to_be32(num + 1);
+                                       ld->ld_data1 = cpu_to_be32(num);
+                                       ld->ld_data2 = cpu_to_be32(0);
+                                       memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
+                               }
+                               magic = gfs2_check_magic(bh1);
+                               *ptr++ = cpu_to_be64(bh1->b_blocknr);
+                               *ptr++ = cpu_to_be64((__u64)magic);
+                               clear_buffer_escaped(bh1);
+                               if (unlikely(magic != 0))
+                                       set_buffer_escaped(bh1);
+                               gfs2_log_lock(sdp);
+                               if (n++ > num)
+                                       break;
+                       } else if (!bh1) {
+                               total_dbuf--;
+                               sdp->sd_log_num_databuf--;
+                               list_del_init(&bd1->bd_le.le_list);
+                               if (bd1 == bd2) {
+                                       bd2 = NULL;
+                                       bd2 = list_prepare_entry(bd2,
+                                               &sdp->sd_log_le_databuf,
+                                               bd_le.le_list);
+                                }
+                               kmem_cache_free(gfs2_bufdata_cachep, bd1);
+                       }
+               }
+               gfs2_log_unlock(sdp);
                if (bh) {
-                       get_bh(bh);
+                       set_buffer_dirty(bh);
+                       ll_rw_block(WRITE, 1, &bh);
+                       bh = NULL;
+               }
+               n = 0;
+               gfs2_log_lock(sdp);
+               list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf,
+                                            bd_le.le_list) {
+                       if (!bd2->bd_bh)
+                               continue;
+                       /* copy buffer if it needs escaping */
                        gfs2_log_unlock(sdp);
-                       if (buffer_dirty(bh)) {
-                               wait_on_buffer(bh);
-                               ll_rw_block(WRITE, 1, &bh);
+                       if (unlikely(buffer_escaped(bd2->bd_bh))) {
+                               void *kaddr;
+                               struct page *page = bd2->bd_bh->b_page;
+                               bh = gfs2_log_get_buf(sdp);
+                               kaddr = kmap_atomic(page, KM_USER0);
+                               memcpy(bh->b_data,
+                                      kaddr + bh_offset(bd2->bd_bh),
+                                      sdp->sd_sb.sb_bsize);
+                               kunmap_atomic(kaddr, KM_USER0);
+                               *(__be32 *)bh->b_data = 0;
+                       } else {
+                               bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
                        }
-                       brelse(bh);
-               } else
-                       gfs2_log_unlock(sdp);
+                       set_buffer_dirty(bh);
+                       ll_rw_block(WRITE, 1, &bh);
+                       gfs2_log_lock(sdp);
+                       if (++n >= num)
+                               break;
+               }
+               bh = NULL;
+               total_dbuf -= num;
+               total_jdata -= num;
        }
+       gfs2_log_unlock(sdp);
 
+       /* Wait on all ordered buffers */
        while (!list_empty(&started)) {
-               db = list_entry(started.next, struct gfs2_databuf,
-                               db_le.le_list);
-               list_del(&db->db_le.le_list);
-               sdp->sd_log_num_databuf--;
-
                gfs2_log_lock(sdp);
-               bh = db->db_bh;
+               bd1 = list_entry(started.next, struct gfs2_bufdata,
+                                bd_le.le_list);
+               list_del_init(&bd1->bd_le.le_list);
+               sdp->sd_log_num_databuf--;
+               bh = bd1->bd_bh;
                if (bh) {
-                       set_v2db(bh, NULL);
+                       bh->b_private = NULL;
+                       get_bh(bh);
                        gfs2_log_unlock(sdp);
                        wait_on_buffer(bh);
                        brelse(bh);
                } else
                        gfs2_log_unlock(sdp);
 
-               kfree(db);
+               kmem_cache_free(gfs2_bufdata_cachep, bd1);
+       }
+
+       /* We've removed all the ordered write bufs here, so only jdata left */
+       gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata);
+}
+
+static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
+                                   struct gfs2_log_descriptor *ld,
+                                   __be64 *ptr, int pass)
+{
+       struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
+       struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
+       struct gfs2_glock *gl = ip->i_gl;
+       unsigned int blks = be32_to_cpu(ld->ld_data1);
+       struct buffer_head *bh_log, *bh_ip;
+       u64 blkno;
+       u64 esc;
+       int error = 0;
+
+       if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
+               return 0;
+
+       gfs2_replay_incr_blk(sdp, &start);
+       for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
+               blkno = be64_to_cpu(*ptr++);
+               esc = be64_to_cpu(*ptr++);
+
+               sdp->sd_found_blocks++;
+
+               if (gfs2_revoke_check(sdp, blkno, start))
+                       continue;
+
+               error = gfs2_replay_read_block(jd, start, &bh_log);
+               if (error)
+                       return error;
+
+               bh_ip = gfs2_meta_new(gl, blkno);
+               memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
+
+               /* Unescape */
+               if (esc) {
+                       __be32 *eptr = (__be32 *)bh_ip->b_data;
+                       *eptr = cpu_to_be32(GFS2_MAGIC);
+               }
+               mark_buffer_dirty(bh_ip);
+
+               brelse(bh_log);
+               brelse(bh_ip);
+               if (error)
+                       break;
+
+               sdp->sd_replayed_blocks++;
+       }
+
+       return error;
+}
+
+/* FIXME: sort out accounting for log blocks etc. */
+
+static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
+{
+       struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
+       struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
+
+       if (error) {
+               gfs2_meta_sync(ip->i_gl);
+               return;
        }
+       if (pass != 1)
+               return;
+
+       /* data sync? */
+       gfs2_meta_sync(ip->i_gl);
 
+       fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
+               jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
+}
+
+static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
+{
+       struct list_head *head = &sdp->sd_log_le_databuf;
+       struct gfs2_bufdata *bd;
+
+       while (!list_empty(head)) {
+               bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
+               list_del_init(&bd->bd_le.le_list);
+               sdp->sd_log_num_databuf--;
+               sdp->sd_log_num_jdata--;
+               gfs2_unpin(sdp, bd->bd_bh, ai);
+       }
        gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
+       gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata);
 }
 
-struct gfs2_log_operations gfs2_glock_lops = {
+
+const struct gfs2_log_operations gfs2_glock_lops = {
        .lo_add = glock_lo_add,
        .lo_after_commit = glock_lo_after_commit,
-       .lo_name = "glock"
+       .lo_name = "glock",
 };
 
-struct gfs2_log_operations gfs2_buf_lops = {
+const struct gfs2_log_operations gfs2_buf_lops = {
        .lo_add = buf_lo_add,
        .lo_incore_commit = buf_lo_incore_commit,
        .lo_before_commit = buf_lo_before_commit,
@@ -499,36 +776,40 @@ struct gfs2_log_operations gfs2_buf_lops = {
        .lo_before_scan = buf_lo_before_scan,
        .lo_scan_elements = buf_lo_scan_elements,
        .lo_after_scan = buf_lo_after_scan,
-       .lo_name = "buf"
+       .lo_name = "buf",
 };
 
-struct gfs2_log_operations gfs2_revoke_lops = {
+const struct gfs2_log_operations gfs2_revoke_lops = {
        .lo_add = revoke_lo_add,
        .lo_before_commit = revoke_lo_before_commit,
        .lo_before_scan = revoke_lo_before_scan,
        .lo_scan_elements = revoke_lo_scan_elements,
        .lo_after_scan = revoke_lo_after_scan,
-       .lo_name = "revoke"
+       .lo_name = "revoke",
 };
 
-struct gfs2_log_operations gfs2_rg_lops = {
+const struct gfs2_log_operations gfs2_rg_lops = {
        .lo_add = rg_lo_add,
        .lo_after_commit = rg_lo_after_commit,
-       .lo_name = "rg"
+       .lo_name = "rg",
 };
 
-struct gfs2_log_operations gfs2_databuf_lops = {
+const struct gfs2_log_operations gfs2_databuf_lops = {
        .lo_add = databuf_lo_add,
+       .lo_incore_commit = buf_lo_incore_commit,
        .lo_before_commit = databuf_lo_before_commit,
-       .lo_name = "databuf"
+       .lo_after_commit = databuf_lo_after_commit,
+       .lo_scan_elements = databuf_lo_scan_elements,
+       .lo_after_scan = databuf_lo_after_scan,
+       .lo_name = "databuf",
 };
 
-struct gfs2_log_operations *gfs2_log_ops[] = {
+const struct gfs2_log_operations *gfs2_log_ops[] = {
        &gfs2_glock_lops,
        &gfs2_buf_lops,
        &gfs2_revoke_lops,
        &gfs2_rg_lops,
        &gfs2_databuf_lops,
-       NULL
+       NULL,
 };