diff --git a/lustre/kernel_patches/patches/snapfs_core-2.4.20.patch b/lustre/kernel_patches/patches/snapfs_core-2.4.20.patch
index 3cc6d9b5c5364d4b515e4c0bccf7cbc843a014f4..60b27d0f728cbc3b7f3945dda747cfd638588597 100644
--- a/lustre/kernel_patches/patches/snapfs_core-2.4.20.patch
+++ b/lustre/kernel_patches/patches/snapfs_core-2.4.20.patch
@@ -2,8 +2,8 @@
Index: linux-2.4.20-8/fs/ext3/snap.c
===================================================================
--- linux-2.4.20-8.orig/fs/ext3/snap.c 2003-01-30 18:24:37.000000000 +0800
-+++ linux-2.4.20-8/fs/ext3/snap.c 2004-01-19 23:24:39.000000000 +0800
-@@ -0,0 +1,2583 @@
++++ linux-2.4.20-8/fs/ext3/snap.c 2004-01-27 00:07:10.000000000 +0800
+@@ -0,0 +1,2577 @@
+/* fs/ext3/snap.c
+ *
+ * Copyright (c) 2002 Cluster File Systems, Inc. <info@clusterfs.com>
@@ -46,6 +46,7 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+#define EXT3_MAX_SNAP_DATA (sizeof(struct snap_ea))
+#define EXT3_SNAP_INDEX EXT3_XATTR_INDEX_LUSTRE
+
++#define EXT3_SNAP_DEBUG
+#ifdef EXT3_SNAP_DEBUG
+ #define snap_debug(f, a...) \
+ do { \
@@ -227,9 +228,10 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+/*SET FLAGS*/
+extern int ext3_bmap(struct address_space *mapping, long block);
+extern int ext3_load_inode_bitmap (struct super_block * sb, unsigned int block_group);
++extern int ext3_block_truncate_page(handle_t *handle, struct address_space *mapping,
++ loff_t from);
+/* helper functions to manipulate field 'parent' in snap_ea */
-+//static inline int
-+static int
++static inline int
+set_parent_ino(struct snap_ea *pea, int size, int index, ino_t val)
+{
+ char * p = (char*) pea;
@@ -241,26 +243,6 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+
+ return 0;
+}
-+static inline ino_t
-+get_parent_ino(struct snap_ea *pea, int size, int index)
-+{
-+ char * p = (char*)pea;
-+ int offset;
-+
-+ offset = sizeof(ino_t)*2 + (size - sizeof(ino_t)*2)/2;
-+ offset += sizeof(ino_t) * index;
-+ return *(ino_t*)(p+offset);
-+}
-+static inline void snap_double_lock(struct inode *i1, struct inode *i2)
-+{
-+ double_down(&i1->i_sem, &i2->i_sem);
-+}
-+
-+static inline void snap_double_unlock(struct inode *i1, struct inode *i2)
-+{
-+ double_up(&i1->i_sem, &i2->i_sem);
-+}
-+
+/* ext3_iterate_cowed_inode:
+ * iterate all the cowed inode with the same index and
+ * run the associate function @repeat
@@ -515,7 +497,8 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ if (inode->i_ino == le32_to_cpu(SB_FIRST_COWED_INO(inode->i_sb))) {
+ SB_FIRST_COWED_INO(inode->i_sb) = cpu_to_le32(next_ino);
+ EXT3_I(inode)->i_flags &= ~EXT3_SNAP_PRI_FLAG;
-+
++ if (next_ino == 0)
++ SB_LAST_COWED_INO(inode->i_sb) = 0;
+ } else {
+ if (!prev_ino)
+ goto err_exit;
@@ -673,7 +656,6 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ * Is it right in snapfs EXT3, check it later?
+ */
+ inlist = 0;
-+ // ea_size = SNAP_EA_SIZE_FROM_INDEX(index);
+ } else if (err < 0 || err > EXT3_MAX_SNAP_DATA) {
+ goto out_unlock;
+ }
@@ -804,8 +786,20 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+{
+ return (EXT3_I(inode)->i_file_acl != 0);
+}
++/* XXX This function has a very bad effect to
++ * the performance of filesystem,
++ * will find another way to fix it
++ */
++static void fs_flushinval_pages(handle_t *handle, struct inode* inode)
++{
++ if (inode->i_blocks > 0 && inode->i_mapping) {
++ fsync_inode_data_buffers(inode);
++ // ext3_block_truncate_page(handle, inode->i_mapping, inode->i_size);
++ truncate_inode_pages(inode->i_mapping, 0);
++ }
++}
+
-+/* ext3_migrate_data:
++/* ext3_migrate_data2:
+ * MOVE all the data blocks from inode src to inode dst as well as
+ * COPY all attributes(meta data) from inode src to inode dst.
+ * For extended attributes(EA), we COPY all the EAs but skip the Snap EA from src to dst.
@@ -826,6 +820,8 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ if (dst->i_ino == src->i_ino)
+ return 0;
+
++ fs_flushinval_pages(handle, src);
++
+ ext3_copy_meta(handle, dst, src);
+
+ snap_debug("migrating data blocks from %lu to %lu\n", src->i_ino, dst->i_ino);
@@ -857,7 +853,6 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ ext3_mark_inode_dirty(handle, src);
+ ext3_mark_inode_dirty(handle, dst);
+
-+ truncate_inode_pages(src->i_mapping, 0);
+
+ return SNAP_ERROR(err);
+}
@@ -1261,9 +1256,6 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ ext3_journal_stop(handle, pri);
+}
+
-+
-+
-+
+static handle_t * ext3_copy_data(handle_t *handle, struct inode *dst,
+ struct inode *src, int *has_orphan)
+{
@@ -1287,7 +1279,9 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ snap_debug("create_indirect:fail to extend "
+ "journal, restart trans\n");
+ loopfail( 3 );
-+ if( !*has_orphan ){
++ if(!*has_orphan) {
++ snap_debug("add orphan ino %lu nlink %d to orphan list \n",
++ dst->i_ino, dst->i_nlink);
+#ifdef EXT3_ENABLE_SNAP_ORPHAN
+ add_snap_orphan(handle, dst, src);
+#else
@@ -1487,9 +1481,11 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ * or ext3_rmdir will report errors "bad dir, no data blocks"
+ */
+ if( S_ISDIR(pri->i_mode)) {
-+ handle = ext3_copy_data( handle, pri, ind, &has_orphan );
-+ if( !handle )
++ handle = ext3_copy_data(handle, pri, ind, &has_orphan);
++ if(!handle) {
++ err = -EINVAL;
+ goto exit_unlock;
++ }
+ }
+
+ pri->u.ext3_i.i_flags |= EXT3_DEL_FL;
@@ -1526,17 +1522,15 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ * it has been cowed and has ea )
+ */
+ if( S_ISLNK(ind->i_mode) &&
-+ (( ind->i_blocks == 0) || (ext3_has_ea(ind) && ind->i_blocks == bpib )) ){
-+ snap_debug("ino %lu is fast symlink\n",
-+ pri->i_ino);
-+ memcpy(EXT3_I(pri)->i_data,
-+ EXT3_I(ind)->i_data,
-+ sizeof(EXT3_I(ind)->i_data));
++ ((ind->i_blocks == 0) || (ext3_has_ea(ind) && ind->i_blocks == bpib))) {
++ snap_debug("ino %lu is fast symlink\n", pri->i_ino);
++ memcpy(EXT3_I(pri)->i_data, EXT3_I(ind)->i_data,
++ sizeof(EXT3_I(ind)->i_data));
+ pri->i_size = ind->i_size;
+ }
+ else {
+ handle = ext3_copy_data(handle, pri, ind, &has_orphan);
-+ if( !handle )
++ if (!handle)
+ goto exit_unlock;
+ }
+ }
@@ -1564,7 +1558,9 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ pri->i_sb->s_dirt = 1;
+ unlock_super(pri->i_sb);
+ }
-+ if(has_orphan) {
++ if (has_orphan) {
++ snap_debug("del %lu nlink %d from orphan list\n",
++ ind->i_ino, ind->i_nlink);
+#ifdef EXT3_ENABLE_SNAP_ORPHAN
+ remove_snap_orphan(handle, ind);
+#else
@@ -1581,9 +1577,19 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ up(&ind->i_sem);
+ ind->i_nlink = 0;
+exit:
++ if (has_orphan) {
++ snap_debug("del %lu nlink %d from orphan list\n",
++ ind->i_ino, ind->i_nlink);
++#ifdef EXT3_ENABLE_SNAP_ORPHAN
++ remove_snap_orphan(handle, ind);
++#else
++ ext3_orphan_del(handle, ind);
++#endif
++ }
+ iput(ind);
+ ext3_journal_stop(handle, pri);
-+ snap_debug("exiting with error %d\n", err);
++ if (err)
++ snap_err("exiting with error %d\n", err);
+ return NULL;
+}
+
@@ -1934,19 +1940,22 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ }
+ /* if it's block level cow, first copy the blocks back */
+ if (EXT3_HAS_COMPAT_FEATURE(pri->i_sb, EXT3_FEATURE_COMPAT_BLOCKCOW) &&
-+ S_ISREG(pri->i_mode)) {
++ S_ISREG(pri->i_mode)) {
+
+ int blocks;
-+ if( !next_ind ) next_ind = pri;
++ if (!next_ind) {
++ next_ind = pri;
++ down(&ind->i_sem);
++ } else {
++ double_down(&next_ind->i_sem, &ind->i_sem);
++ }
+ blocks = (next_ind->i_size + next_ind->i_sb->s_blocksize-1)
+ >> next_ind->i_sb->s_blocksize_bits;
-+
+#define FAST_MIGRATE_BLOCK
+#ifdef FAST_MIGRATE_BLOCK
+ snap_debug("migrate block back from ino %lu to %lu\n",
+ ind->i_ino, next_ind->i_ino);
+
-+ snap_double_lock(next_ind, ind);
+ for(i = 0; i < blocks; i++) {
+ if( ext3_bmap(next_ind->i_mapping, i) )
+ continue;
@@ -1959,25 +1968,7 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ next_ind->i_blocks = calculate_i_blocks( next_ind, blocks);
+ ext3_mark_inode_dirty(handle, next_ind);
+
-+ snap_double_unlock(next_ind, ind);
-+
-+#if 0
-+ snap_double_lock(pri, ind);
-+
-+ for(i = 0; i < blocks; i++) {
-+ if( ext3_bmap(pri, i) ) continue;
-+ if( !ext3_bmap(ind, i) ) continue;
-+ ext3_migrate_block( pri, ind, i) ;
-+ }
-+ /* Now re-compute the i_blocks */
-+ /* XXX shall we take care of ind here? probably not */
-+ pri->i_blocks = calculate_i_blocks( pri, blocks);
-+ mark_inode_dirty(pri);
-+
-+ double_unlock(pri, ind);
-+#endif
+#else
-+ snap_double_lock(next_ind, ind);
+ for (i = 0; i < blocks; i++) {
+ if (ext3_bmap(next_ind->i_mapping, i))
+ continue;
@@ -1985,10 +1976,14 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ break;
+ }
+ ext3_mark_inode_dirty(handle, next_ind);
-+ double_unlock(next_ind, ind);
+#endif
-+ }
++ if (next_ind == pri)
++ up(&ind->i_sem);
++ else
++ double_up(&next_ind->i_sem, &ind->i_sem);
+
++ }
++
+ snap_debug("delete indirect ino %lu\n", ind->i_ino);
+ snap_debug("iput ind %lu, ref count = %d\n",
+ ind->i_ino, atomic_read(&ind->i_count));
@@ -2093,23 +2088,22 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+ /* XXX: check this, ext3_new_inode, the first arg should be "dir" */
+ tmp = ext3_new_inode(handle, pri, (int)pri->i_mode, 0);
+ if(tmp) {
-+ snap_double_lock(pri, tmp);
++ double_down(&pri->i_sem, &tmp->i_sem);
+ ext3_migrate_data(handle, tmp, pri);
-+ snap_double_unlock(pri, tmp);
++ double_up(&pri->i_sem, &tmp->i_sem);
+
+ tmp->i_nlink = 0;
+ iput(tmp);
+ }
+ else
+ snap_err("restore_indirect, new_inode err\n");
-+
-+ snap_double_lock(pri, ind);
++
++ double_down(&pri->i_sem, &ind->i_sem);
+ ext3_migrate_data(handle, pri, ind);
+ /* clear the cow flag for pri because ind has it */
+ pri->u.ext3_i.i_flags &= ~EXT3_COW_FL;
+ ext3_mark_inode_dirty(handle, pri);
-+ snap_double_unlock(pri, ind);
-+
++ double_up(&pri->i_sem, &ind->i_sem);
+ iput(ind);
+
+// ext3_destroy_indirect(pri, index);
@@ -2589,8 +2583,8 @@ Index: linux-2.4.20-8/fs/ext3/snap.c
+
Index: linux-2.4.20-8/fs/ext3/Makefile
===================================================================
---- linux-2.4.20-8.orig/fs/ext3/Makefile 2004-01-05 10:54:03.000000000 +0800
-+++ linux-2.4.20-8/fs/ext3/Makefile 2004-01-13 00:10:14.000000000 +0800
+--- linux-2.4.20-8.orig/fs/ext3/Makefile 2004-01-19 22:06:25.000000000 +0800
++++ linux-2.4.20-8/fs/ext3/Makefile 2004-01-19 22:06:25.000000000 +0800
@@ -13,7 +13,7 @@
obj-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o iopen.o \
@@ -2602,8 +2596,8 @@ Index: linux-2.4.20-8/fs/ext3/Makefile
export-objs += xattr.o
Index: linux-2.4.20-8/fs/ext3/inode.c
===================================================================
---- linux-2.4.20-8.orig/fs/ext3/inode.c 2004-01-05 10:54:03.000000000 +0800
-+++ linux-2.4.20-8/fs/ext3/inode.c 2004-01-18 01:48:19.000000000 +0800
+--- linux-2.4.20-8.orig/fs/ext3/inode.c 2004-01-19 22:06:24.000000000 +0800
++++ linux-2.4.20-8/fs/ext3/inode.c 2004-01-26 01:12:48.000000000 +0800
@@ -1191,7 +1191,7 @@
* So, if we see any bmap calls here on a modified, data-journaled file,
* take extra steps to flush any blocks which might be in the cache.
@@ -2613,10 +2607,19 @@ Index: linux-2.4.20-8/fs/ext3/inode.c
{
struct inode *inode = mapping->host;
journal_t *journal;
+@@ -1403,7 +1403,7 @@
+ * This required during truncate. We need to physically zero the tail end
+ * of that block so it doesn't yield old data if the file is later grown.
+ */
+-static int ext3_block_truncate_page(handle_t *handle,
++int ext3_block_truncate_page(handle_t *handle,
+ struct address_space *mapping, loff_t from)
+ {
+ unsigned long index = from >> PAGE_CACHE_SHIFT;
Index: linux-2.4.20-8/fs/ext3/ialloc.c
===================================================================
---- linux-2.4.20-8.orig/fs/ext3/ialloc.c 2004-01-05 10:54:03.000000000 +0800
-+++ linux-2.4.20-8/fs/ext3/ialloc.c 2004-01-16 20:48:29.000000000 +0800
+--- linux-2.4.20-8.orig/fs/ext3/ialloc.c 2004-01-19 22:06:24.000000000 +0800
++++ linux-2.4.20-8/fs/ext3/ialloc.c 2004-01-19 22:06:25.000000000 +0800
@@ -160,6 +160,13 @@
return retval;
}
@@ -2633,8 +2636,8 @@ Index: linux-2.4.20-8/fs/ext3/ialloc.c
* that have access to it, and as such there are no
Index: linux-2.4.20-8/fs/ext3/super.c
===================================================================
---- linux-2.4.20-8.orig/fs/ext3/super.c 2004-01-05 10:54:03.000000000 +0800
-+++ linux-2.4.20-8/fs/ext3/super.c 2004-01-18 01:40:10.000000000 +0800
+--- linux-2.4.20-8.orig/fs/ext3/super.c 2004-01-19 22:06:24.000000000 +0800
++++ linux-2.4.20-8/fs/ext3/super.c 2004-01-19 22:06:25.000000000 +0800
@@ -1324,6 +1324,13 @@
sbi->s_mount_state = le16_to_cpu(es->s_state);
sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
@@ -2649,10 +2652,24 @@ Index: linux-2.4.20-8/fs/ext3/super.c
for (i=0; i < 4; i++)
sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
sbi->s_def_hash_version = es->s_def_hash_version;
+Index: linux-2.4.20-8/fs/ext3/ext3-exports.c
+===================================================================
+--- linux-2.4.20-8.orig/fs/ext3/ext3-exports.c 2004-01-19 22:06:19.000000000 +0800
++++ linux-2.4.20-8/fs/ext3/ext3-exports.c 2004-01-26 01:13:53.000000000 +0800
+@@ -21,6 +21,9 @@
+ EXPORT_SYMBOL(ext3_xattr_set);
+ EXPORT_SYMBOL(ext3_prep_san_write);
+ EXPORT_SYMBOL(ext3_map_inode_page);
++EXPORT_SYMBOL(ext3_orphan_add);
++EXPORT_SYMBOL(ext3_orphan_del);
++EXPORT_SYMBOL(ext3_block_truncate_page)
+
+ EXPORT_SYMBOL(ext3_abort);
+ EXPORT_SYMBOL(ext3_decode_error);
Index: linux-2.4.20-8/include/linux/snap.h
===================================================================
--- linux-2.4.20-8.orig/include/linux/snap.h 2003-01-30 18:24:37.000000000 +0800
-+++ linux-2.4.20-8/include/linux/snap.h 2004-01-13 00:10:14.000000000 +0800
++++ linux-2.4.20-8/include/linux/snap.h 2004-01-19 22:11:26.000000000 +0800
@@ -0,0 +1,266 @@
+/*
+ * Copyright (c) 2002 Cluster File Systems, Inc. <info@clusterfs.com>
@@ -2922,8 +2939,8 @@ Index: linux-2.4.20-8/include/linux/snap.h
+#endif
Index: linux-2.4.20-8/include/linux/ext3_fs.h
===================================================================
---- linux-2.4.20-8.orig/include/linux/ext3_fs.h 2004-01-05 10:54:03.000000000 +0800
-+++ linux-2.4.20-8/include/linux/ext3_fs.h 2004-01-16 20:45:09.000000000 +0800
+--- linux-2.4.20-8.orig/include/linux/ext3_fs.h 2004-01-19 22:06:24.000000000 +0800
++++ linux-2.4.20-8/include/linux/ext3_fs.h 2004-01-19 22:11:15.000000000 +0800
@@ -183,7 +183,13 @@
#define EXT3_INDEX_FL 0x00001000 /* hash-indexed directory */
#define EXT3_IMAGIC_FL 0x00002000 /* AFS directory */
@@ -2995,8 +3012,8 @@ Index: linux-2.4.20-8/include/linux/ext3_fs.h
EXT3_FEATURE_INCOMPAT_RECOVER)
Index: linux-2.4.20-8/include/linux/ext3_fs_sb.h
===================================================================
---- linux-2.4.20-8.orig/include/linux/ext3_fs_sb.h 2004-01-05 10:54:00.000000000 +0800
-+++ linux-2.4.20-8/include/linux/ext3_fs_sb.h 2004-01-13 00:10:14.000000000 +0800
+--- linux-2.4.20-8.orig/include/linux/ext3_fs_sb.h 2004-01-19 22:06:18.000000000 +0800
++++ linux-2.4.20-8/include/linux/ext3_fs_sb.h 2004-01-19 22:10:06.000000000 +0800
@@ -86,6 +86,13 @@
wait_queue_head_t s_delete_thread_queue;
wait_queue_head_t s_delete_waiter_queue;
@@ -3013,8 +3030,8 @@ Index: linux-2.4.20-8/include/linux/ext3_fs_sb.h
#endif /* _LINUX_EXT3_FS_SB */
Index: linux-2.4.20-8/include/linux/ext3_jbd.h
===================================================================
---- linux-2.4.20-8.orig/include/linux/ext3_jbd.h 2004-01-05 10:53:59.000000000 +0800
-+++ linux-2.4.20-8/include/linux/ext3_jbd.h 2004-01-13 00:10:14.000000000 +0800
+--- linux-2.4.20-8.orig/include/linux/ext3_jbd.h 2004-01-19 22:06:15.000000000 +0800
++++ linux-2.4.20-8/include/linux/ext3_jbd.h 2004-01-19 22:11:15.000000000 +0800
@@ -71,6 +71,33 @@
#define EXT3_INDEX_EXTRA_TRANS_BLOCKS 8
@@ -3052,13 +3069,14 @@ Index: linux-2.4.20-8/include/linux/ext3_jbd.h
%diffstat
fs/ext3/Makefile | 2
+ fs/ext3/ext3-exports.c | 3
fs/ext3/ialloc.c | 7
- fs/ext3/inode.c | 2
- fs/ext3/snap.c | 2583 +++++++++++++++++++++++++++++++++++++++++++++
+ fs/ext3/inode.c | 4
+ fs/ext3/snap.c | 2577 +++++++++++++++++++++++++++++++++++++++++++++
fs/ext3/super.c | 7
include/linux/ext3_fs.h | 38
include/linux/ext3_fs_sb.h | 7
include/linux/ext3_jbd.h | 27
include/linux/snap.h | 266 ++++
- 9 files changed, 2934 insertions(+), 5 deletions(-)
+ 10 files changed, 2932 insertions(+), 6 deletions(-)