[XFS] Avoid replaying inode buffer initialisation log items if on-disk version is...
authorLachlan McIlroy <lachlan@sgi.com>
Fri, 14 Sep 2007 05:23:04 +0000 (15:23 +1000)
committerTim Shimmin <tes@chook.melbourne.sgi.com>
Tue, 18 Sep 2007 10:16:00 +0000 (20:16 +1000)
SGI-PV: 969656
SGI-Modid: xfs-linux-melb:xfs-kern:29676a

Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
fs/xfs/xfs_buf_item.h
fs/xfs/xfs_log_recover.c
fs/xfs/xfs_trans_buf.c

index d7e13614306635bfae1f31bfbd1728e0ff3c1e36..fa25b7dcc6c3515b3e4c7c5f217b91cbca77e4ae 100644 (file)
@@ -52,6 +52,11 @@ typedef struct xfs_buf_log_format_t {
 #define        XFS_BLI_UDQUOT_BUF      0x4
 #define XFS_BLI_PDQUOT_BUF     0x8
 #define        XFS_BLI_GDQUOT_BUF      0x10
+/*
+ * This flag indicates that the buffer contains newly allocated
+ * inodes.
+ */
+#define        XFS_BLI_INODE_NEW_BUF   0x20
 
 #define        XFS_BLI_CHUNK           128
 #define        XFS_BLI_SHIFT           7
index 8ae6e8e5f3db70e1a5c0ee276361f0907fe3f793..dacb19739cc2830ae561a811321362e0fbf2bd6d 100644 (file)
@@ -1874,6 +1874,7 @@ xlog_recover_do_inode_buffer(
 /*ARGSUSED*/
 STATIC void
 xlog_recover_do_reg_buffer(
+       xfs_mount_t             *mp,
        xlog_recover_item_t     *item,
        xfs_buf_t               *bp,
        xfs_buf_log_format_t    *buf_f)
@@ -1884,6 +1885,50 @@ xlog_recover_do_reg_buffer(
        unsigned int            *data_map = NULL;
        unsigned int            map_size = 0;
        int                     error;
+       int                     stale_buf = 1;
+
+       /*
+        * Scan through the on-disk inode buffer and attempt to
+        * determine if it has been written to since it was logged.
+        *
+        * - If any of the magic numbers are incorrect then the buffer is stale
+        * - If any of the modes are non-zero then the buffer is not stale
+        * - If all of the modes are zero and at least one of the generation
+        *   counts is non-zero then the buffer is stale
+        *
+        * If the end result is a stale buffer then the log buffer is replayed
+        * otherwise it is skipped.
+        *
+        * This heuristic is not perfect.  It can be improved by scanning the
+        * entire inode chunk for evidence that any of the inode clusters have
+        * been updated.  To fix this problem completely we will need a major
+        * architectural change to the logging system.
+        */
+       if (buf_f->blf_flags & XFS_BLI_INODE_NEW_BUF) {
+               xfs_dinode_t    *dip;
+               int             inodes_per_buf;
+               int             mode_count = 0;
+               int             gen_count = 0;
+
+               stale_buf = 0;
+               inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
+               for (i = 0; i < inodes_per_buf; i++) {
+                       dip = (xfs_dinode_t *)xfs_buf_offset(bp,
+                               i * mp->m_sb.sb_inodesize);
+                       if (be16_to_cpu(dip->di_core.di_magic) !=
+                                       XFS_DINODE_MAGIC) {
+                               stale_buf = 1;
+                               break;
+                       }
+                       if (be16_to_cpu(dip->di_core.di_mode))
+                               mode_count++;
+                       if (be16_to_cpu(dip->di_core.di_gen))
+                               gen_count++;
+               }
+
+               if (!mode_count && gen_count)
+                       stale_buf = 1;
+       }
 
        switch (buf_f->blf_type) {
        case XFS_LI_BUF:
@@ -1917,7 +1962,7 @@ xlog_recover_do_reg_buffer(
                                               -1, 0, XFS_QMOPT_DOWARN,
                                               "dquot_buf_recover");
                }
-               if (!error)
+               if (!error && stale_buf)
                        memcpy(xfs_buf_offset(bp,
                                (uint)bit << XFS_BLI_SHIFT),    /* dest */
                                item->ri_buf[i].i_addr,         /* source */
@@ -2089,7 +2134,7 @@ xlog_recover_do_dquot_buffer(
        if (log->l_quotaoffs_flag & type)
                return;
 
-       xlog_recover_do_reg_buffer(item, bp, buf_f);
+       xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
 }
 
 /*
@@ -2190,7 +2235,7 @@ xlog_recover_do_buffer_trans(
                  (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
                xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
        } else {
-               xlog_recover_do_reg_buffer(item, bp, buf_f);
+               xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
        }
        if (error)
                return XFS_ERROR(error);
index 60b6b898022bcb1be98f96f2a21615e1bff5e131..95fff6872a2fcd32c17cdff342bebe062ded95e7 100644 (file)
@@ -966,6 +966,7 @@ xfs_trans_inode_alloc_buf(
        ASSERT(atomic_read(&bip->bli_refcount) > 0);
 
        bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
+       bip->bli_format.blf_flags |= XFS_BLI_INODE_NEW_BUF;
 }