Whamcloud - gitweb
LU-8871 kernel: kernel upgrade [SLES12 SP2 4.4.21-84] 57/24257/12
authorBob Glossman <bob.glossman@intel.com>
Thu, 8 Dec 2016 16:49:51 +0000 (08:49 -0800)
committerOleg Drokin <oleg.drokin@intel.com>
Wed, 18 Jan 2017 19:39:09 +0000 (19:39 +0000)
Major linux version upgrade, from linux 3.12 to linux 4.4.
All new ldiskfs and kernel patches.
All new target and config files.

Test-Parameters: clientdistro=sles12sp2 testgroup=review-ldiskfs \
  mdsdistro=sles12sp2 ossdistro=sles12sp2 mdsfilesystemtype=ldiskfs \
  mdtfilesystemtype=ldiskfs ostfilesystemtype=ldiskfs

Signed-off-by: Bob Glossman <bob.glossman@intel.com>
Change-Id: I354197d5426ebde3da6c814bfb7f97f0f04113eb
Reviewed-on: https://review.whamcloud.com/24257
Tested-by: Jenkins
Reviewed-by: Yang Sheng <yang.sheng@intel.com>
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Minh Diep <minh.diep@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
38 files changed:
config/lustre-build-ldiskfs.m4
contrib/lbuild/funcs.sh
contrib/lbuild/lbuild
contrib/lbuild/lbuild-sles12
ldiskfs/kernel_patches/patches/sles12sp2/ext4-corrupted-inode-block-bitmaps-handling-patches.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-data-in-dirent.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-disable-mb-cache.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-give-warning-with-dir-htree-growing.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-hash-indexed-dir-dotdot-update.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-inode-version.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-kill-dx-root.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-large-dir.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-large-eas.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-lookup-dotdot.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-max-dir-size.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-mballoc-extra-checks.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-misc.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-mmp-brelse.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-osd-iop-common.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-pdirop.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-prealloc.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4-print-inum-in-htree-warning.patch [new file with mode: 0644]
ldiskfs/kernel_patches/patches/sles12sp2/ext4_s_max_ext_tree_depth.patch [new file with mode: 0644]
ldiskfs/kernel_patches/series/ldiskfs-4.4-sles12sp2.series [new file with mode: 0644]
lustre/ChangeLog
lustre/autoconf/lustre-core.m4
lustre/include/lustre_compat.h
lustre/kernel_patches/kernel_configs/kernel-4.4-4.4-sles12-x86_64.config [new file with mode: 0644]
lustre/kernel_patches/patches/blkdev_tunables-3.8-sles12.patch [new file with mode: 0644]
lustre/kernel_patches/patches/dev_read_only-3.8.patch [new file with mode: 0644]
lustre/kernel_patches/patches/raid5-mmp-unplug-dev-sles12sp2.patch [new file with mode: 0644]
lustre/kernel_patches/series/4.4-sles12.series [new file with mode: 0644]
lustre/kernel_patches/targets/4.4-sles12.target.in [new file with mode: 0644]
lustre/kernel_patches/which_patch
lustre/osd-ldiskfs/osd_io.c
lustre/osd-ldiskfs/osd_quota.c
lustre/scripts/lustre_rmmod
lustre/utils/mount_utils_ldiskfs.c

index 7de591f..12dbd63 100644 (file)
@@ -24,6 +24,7 @@ AS_IF([test x$RHEL_KERNEL = xyes], [
        6[0-3]) LDISKFS_SERIES="2.6-rhel6.series"       ;;
        esac
 ], [test x$SUSE_KERNEL = xyes], [
+       AS_VERSION_COMPARE([$LINUXRELEASE],[4.4.0],[
        AS_VERSION_COMPARE([$LINUXRELEASE],[3.12.0],[
        AS_VERSION_COMPARE([$LINUXRELEASE],[3.0.0],[
        AS_VERSION_COMPARE([$LINUXRELEASE],[2.6.32], [],
@@ -44,7 +45,8 @@ AS_IF([test x$RHEL_KERNEL = xyes], [
                *) LDISKFS_SERIES="3.12-sles12.series"
                        ;;
                esac
-       ])
+       ])],[LDISKFS_SERIES="4.4-sles12sp2.series"],
+           [LDISKFS_SERIES="4.4-sles12sp2.series"])
 ])
 ])
 AS_IF([test -z "$LDISKFS_SERIES"],
@@ -143,6 +145,27 @@ ext4_bread, [
 ]) # LB_EXT4_BREAD_4ARGS
 
 #
+# LB_EXT4_HAVE_INFO_DQUOT
+#
+# in linux 4.4 i_dqout is in ext4_inode_info, not in struct inode
+#
+AC_DEFUN([LB_EXT4_HAVE_INFO_DQUOT], [
+LB_CHECK_COMPILE([if i_dquot is in ext4_inode_info],
+ext4_info_dquot, [
+       #include <linux/fs.h>
+       #include <linux/quota.h>
+       #include "$EXT4_SRC_DIR/ext4.h"
+],[
+       struct ext4_inode_info in;
+       struct dquot *dq;
+
+       dq = in.i_dquot[0];
+],[
+       AC_DEFINE(HAVE_EXT4_INFO_DQUOT, 1, [i_dquot is in ext4_inode_info])
+])
+]) # LB_EXT4_HAVE_INFO_DQUOT
+
+#
 # LDISKFS_AC_PATCH_PROGRAM
 #
 # Determine which program should be used to apply the patches to
@@ -222,10 +245,12 @@ AS_IF([test x$enable_ldiskfs != xno],[
        LB_EXT4_JOURNAL_START_3ARGS
        LB_LDISKFS_MAP_BLOCKS
        LB_EXT4_BREAD_4ARGS
+       LB_EXT4_HAVE_INFO_DQUOT
        AC_DEFINE(CONFIG_LDISKFS_FS_POSIX_ACL, 1, [posix acls for ldiskfs])
        AC_DEFINE(CONFIG_LDISKFS_FS_SECURITY, 1, [fs security for ldiskfs])
        AC_DEFINE(CONFIG_LDISKFS_FS_XATTR, 1, [extened attributes for ldiskfs])
        AC_DEFINE(CONFIG_LDISKFS_FS_RW, 1, [enable rw access for ldiskfs])
+       AC_DEFINE(CONFIG_LDISKFS_FS_ENCRYPTION, 1, [enable encryption for ldiskfs])
        AC_SUBST(LDISKFS_SUBDIR, ldiskfs)
        AC_DEFINE(HAVE_LDISKFS_OSD, 1, Enable ldiskfs osd)
 ])
index e57e5bf..b3e033f 100644 (file)
@@ -138,12 +138,18 @@ autodetect_distro() {
             "RedHatEnterpriseServer" | "ScientificSL" | "CentOS")
                 name="rhel"
                 ;;
-            "SUSE LINUX")
-                name="sles"
-               PATCHLEVEL=$(sed -n -e 's/^PATCHLEVEL = //p' /etc/SuSE-release)
-               if [ "$PATCHLEVEL" -ne "0" ]; then
-                       version="${version}.$PATCHLEVEL"
-               fi
+           "SUSE LINUX" | "SUSE")
+               name="sles"
+               case "$version" in
+               *.*)    # $version already has patchlevel
+                       ;;
+               *)      # add patchlevel
+                       PATCHLEVEL=$(sed -n -e 's/^PATCHLEVEL = //p' /etc/SuSE-release)
+                       if [ "$PATCHLEVEL" -ne "0" ]; then
+                               version="${version}.$PATCHLEVEL"
+                       fi
+                       ;;
+               esac
                ;;
             "Fedora")
                 name="fc"
index 668d909..f7c1028 100755 (executable)
@@ -282,7 +282,7 @@ check_options() {
     fi
 
     case $TARGET in
-        3.12-sles12)
+        3.12-sles12 | 4.4-sles12)
             CANONICAL_TARGET="sles12"
             ;;
         3.10-rhel7)
index 1a02945..57febcc 100644 (file)
@@ -13,7 +13,15 @@ source ${LBUILD_DIR}/lbuild-sles
 
 kernel_srpm_location() {
 
-    local site="http://mgmt/cobbler/repo_mirror/updates-sles12.1-src"
+    local site
+    case $lnxmaj in
+    3.12)
+       site="http://mgmt/cobbler/repo_mirror/updates-sles12.1-src"
+       ;;
+    4.4)
+       site="http://mgmt/cobbler/repo_mirror/updates-sles12.2-src"
+       ;;
+    esac
     echo "$site/src/"
 
 }
@@ -25,12 +33,24 @@ edit_specs() {
     (cd $TOPDIR/SOURCES; ./mkspec --release ${lnxrel})
     for spec in $RPMSMPTYPE source syms; do
         #cp $TOPDIR/SOURCES/kernel-$spec.spec{,.orig}
+       case $lnxmaj in
+       4.4)
+        sed -i -e "s/^\(Release: *\).*$/\1${lnxrel}_${buildid}/" \
+               -e '/^%setup /a\
+cp  %_sourcedir/linux-4.4-lustre.patch %_builddir/%{name}-%{version}/lustre.patch\
+! grep -q lustre.patch %_sourcedir/series.conf && echo -e "\\n\\tlustre.patch" >> %_sourcedir/series.conf' \
+          SOURCES/kernel-${spec}.spec || \
+            fatal 1 "Error while editing SOURCES/kernel-${spec}.spec"
+       ;;
+       3.12)
         sed -i -e "s/^\(Release: *\).*$/\1${lnxrel}_${buildid}/" \
                -e '/^%setup /a\
 cp  %_sourcedir/linux-3.12-lustre.patch %_builddir/%{name}-%{version}/lustre.patch\
 ! grep -q lustre.patch %_sourcedir/series.conf && echo -e "\\n\\tlustre.patch" >> %_sourcedir/series.conf' \
           SOURCES/kernel-${spec}.spec || \
             fatal 1 "Error while editing SOURCES/kernel-${spec}.spec"
+       ;;
+       esac
 
         if $KERNEL_LUSTRE_NAMING; then
             # these are all of the changes needed because we change the package names
@@ -112,7 +132,15 @@ find_linux_rpm-sles12() {
     local wanted_kernel="$2"
     local pathtorpms=${3:-"$KERNELRPMSBASE/$lnxmaj/$DISTROMAJ/$TARGET_ARCH"}
 
-    local site="http://mgmt/cobbler/repo_mirror/updates-sles12.1-x86_64"
+    local site
+    case $lnxmaj in
+    3.12)
+       site="http://mgmt/cobbler/repo_mirror/updates-sles12.1-x86_64"
+       ;;
+    4.4)
+       site="http://mgmt/cobbler/repo_mirror/updates-sles12.2-x86_64"
+       ;;
+    esac
     if [ ! -f $pathtorpms/kernel-default-devel-$wanted_kernel.$(resolve_arch $TARGET_ARCH).rpm ]; then
         fetch_url "$site/x86_64/kernel-default-devel-$wanted_kernel.$(resolve_arch $TARGET_ARCH).rpm" $pathtorpms
     fi
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-corrupted-inode-block-bitmaps-handling-patches.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-corrupted-inode-block-bitmaps-handling-patches.patch
new file mode 100644 (file)
index 0000000..a1fcea9
--- /dev/null
@@ -0,0 +1,442 @@
+Since we could skip corrupt block groups, this patch
+use ext4_warning() intead of ext4_error() to make FS not
+emount RO in default, also fix a leftover from upstream
+commit 163a203ddb36c36d4a1c942
+---
+diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
+index e069155..692b5e4 100644
+--- a/fs/ext4/balloc.c
++++ b/fs/ext4/balloc.c
+@@ -185,25 +185,17 @@ static int ext4_init_block_bitmap(struct super_block *sb,
+       struct ext4_sb_info *sbi = EXT4_SB(sb);
+       ext4_fsblk_t start, tmp;
+       int flex_bg = 0;
+-      struct ext4_group_info *grp;
+       J_ASSERT_BH(bh, buffer_locked(bh));
+       /* If checksum is bad mark all blocks used to prevent allocation
+        * essentially implementing a per-group read-only flag. */
+       if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
+-              grp = ext4_get_group_info(sb, block_group);
+-              if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
+-                      percpu_counter_sub(&sbi->s_freeclusters_counter,
+-                                         grp->bb_free);
+-              set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
+-              if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
+-                      int count;
+-                      count = ext4_free_inodes_count(sb, gdp);
+-                      percpu_counter_sub(&sbi->s_freeinodes_counter,
+-                                         count);
+-              }
+-              set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
++              ext4_corrupted_block_group(sb, block_group,
++                              EXT4_GROUP_INFO_BBITMAP_CORRUPT |
++                              EXT4_GROUP_INFO_IBITMAP_CORRUPT,
++                              "Checksum bad for group %u",
++                              block_group);
+               return -EFSBADCRC;
+       }
+       memset(bh->b_data, 0, sb->s_blocksize);
+@@ -367,7 +359,6 @@ static void ext4_validate_block_bitmap(struct super_block *sb,
+ {
+       ext4_fsblk_t    blk;
+       struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
+-      struct ext4_sb_info *sbi = EXT4_SB(sb);
+       if (buffer_verified(bh))
+               return 0;
+@@ -377,22 +367,19 @@ static void ext4_validate_block_bitmap(struct super_block *sb,
+       if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
+                       desc, bh))) {
+               ext4_unlock_group(sb, block_group);
+-              ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
+-              if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
+-                      percpu_counter_sub(&sbi->s_freeclusters_counter,
+-                                         grp->bb_free);
+-              set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
++              ext4_corrupted_block_group(sb, block_group,
++                              EXT4_GROUP_INFO_BBITMAP_CORRUPT,
++                              "bg %u: bad block bitmap checksum",
++                              block_group);
+               return -EFSBADCRC;
+       }
+       blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
+       if (unlikely(blk != 0)) {
+               ext4_unlock_group(sb, block_group);
+-              ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
+-                         block_group, blk);
+-              if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
+-                      percpu_counter_sub(&sbi->s_freeclusters_counter,
+-                                         grp->bb_free);
+-              set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
++              ext4_corrupted_block_group(sb, block_group,
++                              EXT4_GROUP_INFO_BBITMAP_CORRUPT,
++                              "bg %u: block %llu: invalid block bitmap",
++                              block_group, blk);
+               return -EFSCORRUPTED;
+       }
+       set_buffer_verified(bh);
+@@ -445,8 +432,6 @@ ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
+               ext4_unlock_group(sb, block_group);
+               unlock_buffer(bh);
+               if (err) {
+-                      ext4_error(sb, "Failed to init block bitmap for group "
+-                                 "%u: %d", block_group, err);
+                       goto out;
+               }
+               goto verify;
+diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
+index 3c41773..63a63b6 100644
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -91,6 +91,17 @@ typedef __u32 ext4_lblk_t;
+ /* data type for block group number */
+ typedef unsigned int ext4_group_t;
++void __ext4_corrupted_block_group(struct super_block *sb,
++                                ext4_group_t group, unsigned int flags,
++                                const char *function, unsigned int line);
++
++#define ext4_corrupted_block_group(sb, group, flags, fmt, ...)                \
++      do {                                                            \
++              __ext4_warning(sb, __func__, __LINE__, fmt,             \
++                              ##__VA_ARGS__);                         \
++              __ext4_corrupted_block_group(sb, group, flags,          \
++                                      __func__, __LINE__);            \
++      } while (0)
+ /*
+  * Flags used in mballoc's allocation_context flags field.
+  *
+@@ -2673,7 +2684,11 @@ struct ext4_group_info {
+ #define EXT4_GROUP_INFO_NEED_INIT_BIT         0
+ #define EXT4_GROUP_INFO_WAS_TRIMMED_BIT               1
+ #define EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT   2
++#define EXT4_GROUP_INFO_BBITMAP_CORRUPT               \
++      (1 << EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT)
+ #define EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT   3
++#define EXT4_GROUP_INFO_IBITMAP_CORRUPT               \
++      (1 << EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT)
+ #define EXT4_MB_GRP_NEED_INIT(grp)    \
+       (test_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &((grp)->bb_state)))
+diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
+index fc65310..92bcc8d 100644
+--- a/fs/ext4/ialloc.c
++++ b/fs/ext4/ialloc.c
+@@ -70,25 +70,15 @@ static unsigned ext4_init_inode_bitmap(struct super_block *sb,
+                                      ext4_group_t block_group,
+                                      struct ext4_group_desc *gdp)
+ {
+-      struct ext4_group_info *grp;
+-      struct ext4_sb_info *sbi = EXT4_SB(sb);
+       J_ASSERT_BH(bh, buffer_locked(bh));
+       /* If checksum is bad mark all blocks and inodes use to prevent
+        * allocation, essentially implementing a per-group read-only flag. */
+       if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
+-              grp = ext4_get_group_info(sb, block_group);
+-              if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
+-                      percpu_counter_sub(&sbi->s_freeclusters_counter,
+-                                         grp->bb_free);
+-              set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
+-              if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
+-                      int count;
+-                      count = ext4_free_inodes_count(sb, gdp);
+-                      percpu_counter_sub(&sbi->s_freeinodes_counter,
+-                                         count);
+-              }
+-              set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
++              ext4_corrupted_block_group(sb, block_group,
++                              EXT4_GROUP_INFO_BBITMAP_CORRUPT |
++                              EXT4_GROUP_INFO_IBITMAP_CORRUPT,
++                              "Checksum bad for group %u", block_group);
+               return -EFSBADCRC;
+       }
+@@ -193,8 +180,6 @@ verify:
+               ext4_unlock_group(sb, block_group);
+               unlock_buffer(bh);
+               if (err) {
+-                      ext4_error(sb, "Failed to init inode bitmap for group "
+-                                 "%u: %d", block_group, err);
+                       goto out;
+               }
+               return bh;
+@@ -337,14 +318,9 @@ out:
+               if (!fatal)
+                       fatal = err;
+       } else {
+-              ext4_error(sb, "bit already cleared for inode %lu", ino);
+-              if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
+-                      int count;
+-                      count = ext4_free_inodes_count(sb, gdp);
+-                      percpu_counter_sub(&sbi->s_freeinodes_counter,
+-                                         count);
+-              }
+-              set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
++              ext4_corrupted_block_group(sb, block_group,
++                              EXT4_GROUP_INFO_IBITMAP_CORRUPT,
++                              "bit already cleared for inode %lu", ino);
+       }
+ error_return:
+diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
+index 7282d07..e6805e6 100644
+--- a/fs/ext4/mballoc.c
++++ b/fs/ext4/mballoc.c
+@@ -752,10 +752,18 @@ int ext4_mb_generate_buddy(struct super_block *sb,
+       if (free != grp->bb_free) {
+               struct ext4_group_desc *gdp;
+               gdp = ext4_get_group_desc(sb, group, NULL);
+-              ext4_error(sb, "group %lu: %u blocks in bitmap, %u in bb, "
+-                      "%u in gd, %lu pa's\n", (long unsigned int)group,
+-                      free, grp->bb_free, ext4_free_group_clusters(sb, gdp),
+-                      grp->bb_prealloc_nr);
++
++              ext4_corrupted_block_group(sb, group,
++                              EXT4_GROUP_INFO_BBITMAP_CORRUPT,
++                              "group %lu: %u blocks in bitmap, %u in bb, %u in gd, %lu pa's block bitmap corrupt",
++                              (unsigned long int)group, free, grp->bb_free,
++                              ext4_free_group_clusters(sb, gdp),
++                              grp->bb_prealloc_nr);
++              /*
++               * If we intend to continue, we consider group descriptor
++               * corrupt and update bb_free using bitmap value
++               */
++              grp->bb_free = free;
+               return -EIO;
+       }
+       mb_set_largest_free_order(sb, grp);
+@@ -1101,7 +1109,7 @@ ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
+       int block;
+       int pnum;
+       int poff;
+-      struct page *page;
++      struct page *page = NULL;
+       int ret;
+       struct ext4_group_info *grp;
+       struct ext4_sb_info *sbi = EXT4_SB(sb);
+@@ -1127,7 +1135,7 @@ ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
+                */
+               ret = ext4_mb_init_group(sb, group);
+               if (ret)
+-                      return ret;
++                      goto err;
+       }
+       /*
+@@ -1227,6 +1235,7 @@ err:
+               page_cache_release(e4b->bd_buddy_page);
+       e4b->bd_buddy = NULL;
+       e4b->bd_bitmap = NULL;
++      ext4_warning(sb, "Error loading buddy information for %u", group);
+       return ret;
+ }
+@@ -3599,9 +3608,11 @@ int ext4_mb_check_ondisk_bitmap(struct super_block *sb, void *bitmap,
+       }
+       if (free != ext4_free_group_clusters(sb, gdp)) {
+-              ext4_error(sb, "on-disk bitmap for group %d"
+-                      "corrupted: %u blocks free in bitmap, %u - in gd\n",
+-                      group, free, ext4_free_group_clusters(sb, gdp));
++              ext4_corrupted_block_group(sb, group,
++                              EXT4_GROUP_INFO_BBITMAP_CORRUPT,
++                              "on-disk bitmap for group %d corrupted: %u blocks free in bitmap, %u - in gd\n",
++                              group, free,
++                              ext4_free_group_clusters(sb, gdp));
+               return -EIO;
+       }
+       return 0;
+@@ -3962,16 +3973,8 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
+       /* "free < pa->pa_free" means we maybe double alloc the same blocks,
+        * otherwise maybe leave some free blocks unavailable, no need to BUG.*/
+       if ((free > pa->pa_free && !pa->pa_error) || (free < pa->pa_free)) {
+-              ext4_error(sb, "pa free mismatch: [pa %p] "
+-                              "[phy %lu] [logic %lu] [len %u] [free %u] "
+-                              "[error %u] [inode %lu] [freed %u]", pa,
+-                              (unsigned long)pa->pa_pstart,
+-                              (unsigned long)pa->pa_lstart,
+-                              (unsigned)pa->pa_len, (unsigned)pa->pa_free,
+-                              (unsigned)pa->pa_error, pa->pa_inode->i_ino,
+-                              free);
+               ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u",
+-                                      free, pa->pa_free);
++                                    free, pa->pa_free);
+               /*
+                * pa is already deleted so we use the value obtained
+                * from the bitmap and continue.
+@@ -4031,14 +4034,11 @@ ext4_mb_discard_group_preallocations(struct super_block *sb,
+       bitmap_bh = ext4_read_block_bitmap(sb, group);
+       if (IS_ERR(bitmap_bh)) {
+               err = PTR_ERR(bitmap_bh);
+-              ext4_error(sb, "Error %d reading block bitmap for %u",
+-                         err, group);
+               return 0;
+       }
+       err = ext4_mb_load_buddy(sb, group, &e4b);
+       if (err) {
+-              ext4_error(sb, "Error loading buddy information for %u", group);
+               put_bh(bitmap_bh);
+               return 0;
+       }
+@@ -4198,17 +4198,12 @@ repeat:
+               group = ext4_get_group_number(sb, pa->pa_pstart);
+               err = ext4_mb_load_buddy(sb, group, &e4b);
+-              if (err) {
+-                      ext4_error(sb, "Error loading buddy information for %u",
+-                                      group);
++              if (err)
+                       return;
+-              }
+               bitmap_bh = ext4_read_block_bitmap(sb, group);
+               if (IS_ERR(bitmap_bh)) {
+                       err = PTR_ERR(bitmap_bh);
+-                      ext4_error(sb, "Error %d reading block bitmap for %u",
+-                                      err, group);
+                       ext4_mb_unload_buddy(&e4b);
+                       continue;
+               }
+@@ -4467,11 +4462,8 @@ ext4_mb_discard_lg_preallocations(struct super_block *sb,
+       list_for_each_entry_safe(pa, tmp, &discard_list, u.pa_tmp_list) {
+               group = ext4_get_group_number(sb, pa->pa_pstart);
+-              if (ext4_mb_load_buddy(sb, group, &e4b)) {
+-                      ext4_error(sb, "Error loading buddy information for %u",
+-                                      group);
++              if (ext4_mb_load_buddy(sb, group, &e4b))
+                       continue;
+-              }
+               ext4_lock_group(sb, group);
+               list_del(&pa->pa_group_list);
+               ext4_get_group_info(sb, group)->bb_prealloc_nr--;
+@@ -4742,17 +4734,18 @@ errout:
+                        * been updated or not when fail case. So can
+                        * not revert pa_free back, just mark pa_error*/
+                       pa->pa_error++;
+-                      ext4_error(sb,
+-                              "Updating bitmap error: [err %d] "
+-                              "[pa %p] [phy %lu] [logic %lu] "
+-                              "[len %u] [free %u] [error %u] "
+-                              "[inode %lu]", *errp, pa,
+-                              (unsigned long)pa->pa_pstart,
+-                              (unsigned long)pa->pa_lstart,
+-                              (unsigned)pa->pa_len,
+-                              (unsigned)pa->pa_free,
+-                              (unsigned)pa->pa_error,
+-                              pa->pa_inode ? pa->pa_inode->i_ino : 0);
++                      ext4_corrupted_block_group(sb, 0, 0,
++                                      "Updating bitmap error: [err %d] "
++                                      "[pa %p] [phy %lu] [logic %lu] "
++                                      "[len %u] [free %u] [error %u] "
++                                      "[inode %lu]", *errp, pa,
++                                      (unsigned long)pa->pa_pstart,
++                                      (unsigned long)pa->pa_lstart,
++                                      (unsigned)pa->pa_len,
++                                      (unsigned)pa->pa_free,
++                                      (unsigned)pa->pa_error,
++                                      pa->pa_inode ?
++                                      pa->pa_inode->i_ino : 0);
+               }
+       }
+       ext4_mb_release_context(ac);
+@@ -5037,7 +5030,7 @@ do_more:
+       err = ext4_mb_load_buddy(sb, block_group, &e4b);
+       if (err)
+-              goto error_return;
++              goto error_brelse;
+       if ((flags & EXT4_FREE_BLOCKS_METADATA) && ext4_handle_valid(handle)) {
+               struct ext4_free_data *new_entry;
+@@ -5119,8 +5112,9 @@ do_more:
+               goto do_more;
+       }
+ error_return:
+-      brelse(bitmap_bh);
+       ext4_std_error(sb, err);
++error_brelse:
++      brelse(bitmap_bh);
+       return;
+ }
+@@ -5216,7 +5210,7 @@ int ext4_group_add_blocks(handle_t *handle, struct super_block *sb,
+       err = ext4_mb_load_buddy(sb, block_group, &e4b);
+       if (err)
+-              goto error_return;
++              goto error_brelse;
+       /*
+        * need to update group_info->bb_free and bitmap
+@@ -5253,8 +5247,9 @@ int ext4_group_add_blocks(handle_t *handle, struct super_block *sb,
+               err = ret;
+ error_return:
+-      brelse(bitmap_bh);
+       ext4_std_error(sb, err);
++error_brelse:
++      brelse(bitmap_bh);
+       return err;
+ }
+@@ -5329,11 +5324,9 @@ ext4_trim_all_free(struct super_block *sb, ext4_group_t group,
+       trace_ext4_trim_all_free(sb, group, start, max);
+       ret = ext4_mb_load_buddy(sb, group, &e4b);
+-      if (ret) {
+-              ext4_error(sb, "Error in loading buddy "
+-                              "information for %u", group);
++      if (ret)
+               return ret;
+-      }
++
+       bitmap = e4b.bd_bitmap;
+       ext4_lock_group(sb, group);
+diff --git a/fs/ext4/super.c b/fs/ext4/super.c
+index c625960..0de22f2 100644
+--- a/fs/ext4/super.c
++++ b/fs/ext4/super.c
+@@ -633,6 +633,37 @@ void __ext4_warning(struct super_block *sb, const char *function,
+       va_end(args);
+ }
++void __ext4_corrupted_block_group(struct super_block *sb, ext4_group_t group,
++                                unsigned int flags, const char *function,
++                                unsigned int line)
++{
++      struct ext4_sb_info *sbi = EXT4_SB(sb);
++      struct ext4_group_info *grp = ext4_get_group_info(sb, group);
++      struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
++
++      if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT &&
++          !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) {
++              percpu_counter_sub(&sbi->s_freeclusters_counter,
++                                      grp->bb_free);
++              set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
++                      &grp->bb_state);
++      }
++
++      if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT &&
++          !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
++              if (gdp) {
++                      int count;
++
++                      count = ext4_free_inodes_count(sb, gdp);
++                      percpu_counter_sub(&sbi->s_freeinodes_counter,
++                                         count);
++              }
++              set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
++                      &grp->bb_state);
++      }
++      save_error_info(sb, function, line);
++}
++
+ void __ext4_grp_locked_error(const char *function, unsigned int line,
+                            struct super_block *sb, ext4_group_t grp,
+                            unsigned long ino, ext4_fsblk_t block,
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-data-in-dirent.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-data-in-dirent.patch
new file mode 100644 (file)
index 0000000..5116a6e
--- /dev/null
@@ -0,0 +1,802 @@
+this patch implements feature which allows ext4 fs users (e.g. Lustre)
+to store data in ext4 dirent.
+data is stored in ext4 dirent after file-name, this space is accounted
+in de->rec_len. flag EXT4_DIRENT_LUFID added to d_type if extra data
+is present.
+
+make use of dentry->d_fsdata to pass fid to ext4. so no
+changes in ext4_add_entry() interface required.
+
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/dir.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/dir.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/dir.c
+@@ -71,11 +71,11 @@ int __ext4_check_dir_entry(const char *f
+       const int rlen = ext4_rec_len_from_disk(de->rec_len,
+                                               dir->i_sb->s_blocksize);
+-      if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
++      if (unlikely(rlen < __EXT4_DIR_REC_LEN(1)))
+               error_msg = "rec_len is smaller than minimal";
+       else if (unlikely(rlen % 4 != 0))
+               error_msg = "rec_len % 4 != 0";
+-      else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
++      else if (unlikely(rlen < EXT4_DIR_REC_LEN(de)))
+               error_msg = "rec_len is too small for name_len";
+       else if (unlikely(((char *) de - buf) + rlen > size))
+               error_msg = "directory entry across range";
+@@ -208,7 +208,7 @@ revalidate:
+                                * failure will be detected in the
+                                * dirent test below. */
+                               if (ext4_rec_len_from_disk(de->rec_len,
+-                                      sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
++                                  sb->s_blocksize) < __EXT4_DIR_REC_LEN(1))
+                                       break;
+                               i += ext4_rec_len_from_disk(de->rec_len,
+                                                           sb->s_blocksize);
+@@ -438,12 +438,17 @@ int ext4_htree_store_dirent(struct file
+       struct fname *fname, *new_fn;
+       struct dir_private_info *info;
+       int len;
++      int extra_data = 0;
+       info = dir_file->private_data;
+       p = &info->root.rb_node;
+       /* Create and allocate the fname structure */
+-      len = sizeof(struct fname) + ent_name->len + 1;
++      if (dirent->file_type & EXT4_DIRENT_LUFID)
++              extra_data = ext4_get_dirent_data_len(dirent);
++
++      len = sizeof(struct fname) + ent_name->len + extra_data + 1;
++
+       new_fn = kzalloc(len, GFP_KERNEL);
+       if (!new_fn)
+               return -ENOMEM;
+@@ -452,7 +457,7 @@ int ext4_htree_store_dirent(struct file
+       new_fn->inode = le32_to_cpu(dirent->inode);
+       new_fn->name_len = dirent->name_len;
+       new_fn->file_type = dirent->file_type;
+-      memcpy(new_fn->name, ent_name->name, ent_name->len);
++      memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data);
+       new_fn->name[ent_name->len] = 0;
+       while (*p) {
+@@ -652,7 +457,7 @@ int ext4_htree_store_dirent(struct file
+               if (ldiskfs_check_dir_entry(dir, NULL, de, bh,
+                                        buf, buf_size, offset))
+                       return -EFSCORRUPTED;
+-              nlen = EXT4_DIR_REC_LEN(de->name_len);
++              nlen = EXT4_DIR_REC_LEN(de);
+               rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
+               de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
+               offset += rlen;
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
+@@ -952,6 +952,7 @@ struct ext4_inode_info {
+ #define EXT4_MOUNT_POSIX_ACL          0x08000 /* POSIX Access Control Lists */
+ #define EXT4_MOUNT_NO_AUTO_DA_ALLOC   0x10000 /* No auto delalloc mapping */
+ #define EXT4_MOUNT_BARRIER            0x20000 /* Use block barriers */
++#define EXT4_MOUNT_DIRDATA            0x40000 /* Data in directory entries*/
+ #define EXT4_MOUNT_QUOTA              0x80000 /* Some quota option set */
+ #define EXT4_MOUNT_USRQUOTA           0x100000 /* "old" user quota */
+ #define EXT4_MOUNT_GRPQUOTA           0x200000 /* "old" group quota */
+@@ -1534,6 +1535,7 @@ static inline void ext4_clear_state_flag
+                                        EXT4_FEATURE_INCOMPAT_64BIT| \
+                                        EXT4_FEATURE_INCOMPAT_FLEX_BG| \
+                                        EXT4_FEATURE_INCOMPAT_MMP | \
++                                       EXT4_FEATURE_INCOMPAT_DIRDATA| \
+                                        EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
+                                        EXT4_FEATURE_INCOMPAT_ENCRYPT | \
+                                        EXT4_FEATURE_INCOMPAT_CSUM_SEED)
+@@ -1640,6 +1642,43 @@ struct ext4_dir_entry_tail {
+ #define EXT4_FT_SYMLINK               7
+ #define EXT4_FT_MAX           8
++#define EXT4_FT_MASK          0xf
++
++#if EXT4_FT_MAX > EXT4_FT_MASK
++#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
++#endif
++
++/*
++ * d_type has 4 unused bits, so it can hold four types data. these different
++ * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
++ * stored, in flag order, after file-name in ext4 dirent.
++*/
++/*
++ * this flag is added to d_type if ext4 dirent has extra data after
++ * filename. this data length is variable and length is stored in first byte
++ * of data. data start after filename NUL byte.
++ * This is used by Lustre FS.
++  */
++#define EXT4_DIRENT_LUFID             0x10
++
++#define EXT4_LUFID_MAGIC    0xAD200907UL
++struct ext4_dentry_param {
++      __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
++      char   edp_len;         /* size of edp_data in bytes */
++      char   edp_data[0];     /* packed array of data */
++} __packed;
++
++static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
++                                                struct ext4_dentry_param *p)
++
++{
++      if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_DIRDATA))
++              return NULL;
++      if (p && p->edp_magic == EXT4_LUFID_MAGIC)
++              return &p->edp_len;
++      else
++              return NULL;
++}
+ #define EXT4_FT_DIR_CSUM      0xDE
+@@ -1650,8 +1689,11 @@ struct ext4_dir_entry_tail {
+  */
+ #define EXT4_DIR_PAD                  4
+ #define EXT4_DIR_ROUND                        (EXT4_DIR_PAD - 1)
+-#define EXT4_DIR_REC_LEN(name_len)    (((name_len) + 8 + EXT4_DIR_ROUND) & \
++#define __EXT4_DIR_REC_LEN(name_len)  (((name_len) + 8 + EXT4_DIR_ROUND) & \
+                                        ~EXT4_DIR_ROUND)
++#define EXT4_DIR_REC_LEN(de)          (__EXT4_DIR_REC_LEN((de)->name_len +\
++                                      ext4_get_dirent_data_len(de)))
++
+ #define EXT4_MAX_REC_LEN              ((1<<16)-1)
+ /*
+@@ -1987,12 +2029,12 @@ extern int ext4_find_dest_de(struct inod
+                            struct buffer_head *bh,
+                            void *buf, int buf_size,
+                            struct ext4_filename *fname,
+-                           struct ext4_dir_entry_2 **dest_de);
++                           struct ext4_dir_entry_2 **dest_de, int *dlen);
+ int ext4_insert_dentry(struct inode *dir,
+                      struct inode *inode,
+                      struct ext4_dir_entry_2 *de,
+                      int buf_size,
+-                     struct ext4_filename *fname);
++                     struct ext4_filename *fname, void *data);
+ static inline void ext4_update_dx_flag(struct inode *inode)
+ {
+       if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
+@@ -2004,10 +2046,17 @@ static unsigned char ext4_filetype_table
+ static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
+ {
++      int fl_index = filetype & EXT4_FT_MASK;
++
+-      if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
++      if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX)
+               return DT_UNKNOWN;
+-      return ext4_filetype_table[filetype];
++      if (!test_opt(sb, DIRDATA))
++              return ext4_filetype_table[fl_index];
++
++      return (ext4_filetype_table[fl_index]) |
++              (filetype & EXT4_DIRENT_LUFID);
++
+ }
+ /* fsync.c */
+@@ -2157,6 +2206,8 @@ extern struct buffer_head * ext4_find_en
+ extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
+                            struct ext4_dir_entry_2 *de_del,
+                            struct buffer_head *bh);
++extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
++                             struct inode *inode, const void *, const void *);
+ extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
+                               __u32 start_minor_hash, __u32 *next_hash);
+ extern int ext4_search_dir(struct buffer_head *bh,
+@@ -2761,6 +2810,28 @@ extern struct mutex ext4__aio_mutex[EXT4
+ extern int ext4_resize_begin(struct super_block *sb);
+ extern void ext4_resize_end(struct super_block *sb);
++/*
++ * Compute the total directory entry data length.
++ * This includes the filename and an implicit NUL terminator (always present),
++ * and optional extensions.  Each extension has a bit set in the high 4 bits of
++ * de->file_type, and the extension length is the first byte in each entry.
++ */
++static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
++{
++      char *len = de->name + de->name_len + 1 /* NUL terminator */;
++      int dlen = 0;
++      __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
++
++      while (extra_data_flags) {
++              if (extra_data_flags & 1) {
++                      dlen += *len + (dlen == 0);
++                      len += *len;
++              }
++              extra_data_flags >>= 1;
++      }
++      return dlen;
++}
++
+ #endif        /* __KERNEL__ */
+ #endif        /* _EXT4_H */
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/namei.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c
+@@ -239,7 +239,8 @@ static unsigned dx_get_count(struct dx_e
+ static unsigned dx_get_limit(struct dx_entry *entries);
+ static void dx_set_count(struct dx_entry *entries, unsigned value);
+ static void dx_set_limit(struct dx_entry *entries, unsigned value);
+-static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
++static inline unsigned dx_root_limit(struct inode *dir,
++              struct ext4_dir_entry_2 *dot_de, unsigned infosize);
+ static unsigned dx_node_limit(struct inode *dir);
+ static struct dx_frame *dx_probe(const struct qstr *d_name,
+                                struct inode *dir,
+@@ -504,11 +505,12 @@ ext4_next_entry(struct ext4_dir_entry_2
+ */
+ struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
+ {
+-      /* get dotdot first */
+-      de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
++      BUG_ON(de->name_len != 1);
++      /* get dotdot first */
++      de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
+-      /* dx root info is after dotdot entry */
+-      de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
++      /* dx root info is after dotdot entry */
++      de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
+       return (struct dx_root_info *)de;
+ }
+@@ -553,10 +555,16 @@ static inline void dx_set_limit(struct d
+       ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
+ }
+-static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
++static inline unsigned dx_root_limit(struct inode *dir,
++              struct ext4_dir_entry_2 *dot_de, unsigned infosize)
+ {
+-      unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
+-              EXT4_DIR_REC_LEN(2) - infosize;
++      struct ext4_dir_entry_2 *dotdot_de;
++      unsigned entry_space;
++
++      BUG_ON(dot_de->name_len != 1);
++      dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
++      entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(dot_de) -
++                       EXT4_DIR_REC_LEN(dotdot_de) - infosize;
+       if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
+                                      EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
+@@ -566,7 +574,7 @@ static inline unsigned dx_root_limit(str
+ static inline unsigned dx_node_limit(struct inode *dir)
+ {
+-      unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
++      unsigned entry_space = dir->i_sb->s_blocksize - __EXT4_DIR_REC_LEN(0);
+       if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
+                                      EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
+@@ -617,7 +625,7 @@ static struct stats dx_show_leaf(struct
+                               printk(":%x.%u ", h.hash,
+                                      (unsigned) ((char *) de - base));
+                       }
+-                      space += EXT4_DIR_REC_LEN(de->name_len);
++                      space += EXT4_DIR_REC_LEN(de);
+                       names++;
+               }
+               de = ext4_next_entry(de, size);
+@@ -723,11 +731,14 @@ dx_probe(const struct qstr *d_name, stru
+       entries = (struct dx_entry *) (((char *)info) + info->info_length);
+-      if (dx_get_limit(entries) != dx_root_limit(dir,
+-                                                 info->info_length)) {
++      if (dx_get_limit(entries) !=
++          dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
++                        info->info_length)) {
+               ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
+                                  dx_get_limit(entries),
+-                                 dx_root_limit(dir, info->info_length));
++                                 dx_root_limit(dir,
++                                        (struct ext4_dir_entry_2 *)frame->bh->b_data,
++                                        info->info_length));
+               goto fail;
+       }
+
+@@ -916,7 +925,7 @@ static int htree_dirblock_to_tree(struct
+       de = (struct ext4_dir_entry_2 *) bh->b_data;
+       top = (struct ext4_dir_entry_2 *) ((char *) de +
+                                          dir->i_sb->s_blocksize -
+-                                         EXT4_DIR_REC_LEN(0));
++                                         __EXT4_DIR_REC_LEN(0));
+ #ifdef CONFIG_EXT4_FS_ENCRYPTION
+       /* Check if the directory is encrypted */
+       if (ext4_encrypted_inode(dir)) {
+@@ -1508,7 +1517,7 @@ dx_move_dirents(char *from, char *to, st
+       while (count--) {
+               struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
+                                               (from + (map->offs<<2));
+-              rec_len = EXT4_DIR_REC_LEN(de->name_len);
++              rec_len = EXT4_DIR_REC_LEN(de);
+               memcpy (to, de, rec_len);
+               ((struct ext4_dir_entry_2 *) to)->rec_len =
+                               ext4_rec_len_to_disk(rec_len, blocksize);
+@@ -1532,7 +1541,7 @@ static struct ext4_dir_entry_2* dx_pack_
+       while ((char*)de < base + blocksize) {
+               next = ext4_next_entry(de, blocksize);
+               if (de->inode && de->name_len) {
+-                      rec_len = EXT4_DIR_REC_LEN(de->name_len);
++                      rec_len = EXT4_DIR_REC_LEN(de);
+                       if (de > to)
+                               memmove(to, de, rec_len);
+                       to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
+@@ -1664,15 +1673,17 @@ int ext4_find_dest_de(struct inode *dir,
+                     struct buffer_head *bh,
+                     void *buf, int buf_size,
+                     struct ext4_filename *fname,
+-                    struct ext4_dir_entry_2 **dest_de)
++                    struct ext4_dir_entry_2 **dest_de, int *dlen)
+ {
+       struct ext4_dir_entry_2 *de;
+-      unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
++      unsigned short reclen = __EXT4_DIR_REC_LEN(fname_len(fname)) +
++                                                (dlen ? *dlen : 0);
+       int nlen, rlen;
+       unsigned int offset = 0;
+       char *top;
+       int res;
++      dlen ? *dlen = 0 : 0; /* default set to 0 */
+       de = (struct ext4_dir_entry_2 *)buf;
+       top = buf + buf_size - reclen;
+       while ((char *) de <= top) {
+@@ -1680,10 +1690,26 @@ int ext4_find_dest_de(struct inode *dir,
+                       res = -EEXIST;
+                       goto return_result;
+               }
+-              nlen = EXT4_DIR_REC_LEN(de->name_len);
++              nlen = EXT4_DIR_REC_LEN(de);
+               rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
+               if ((de->inode ? rlen - nlen : rlen) >= reclen)
+                       break;
++              /* Then for dotdot entries, check for the smaller space
++               * required for just the entry, no FID */
++              if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
++                      if ((de->inode ? rlen - nlen : rlen) >=
++                          __EXT4_DIR_REC_LEN(fname_len(fname))) {
++                              /* set dlen=1 to indicate not
++                               * enough space store fid */
++                              dlen ? *dlen = 1 : 0;
++                              break;
++                      }
++                      /* The new ".." entry must be written over the
++                       * previous ".." entry, which is the first
++                       * entry traversed by this scan. If it doesn't
++                       * fit, something is badly wrong, so -EIO. */
++                      return -EIO;
++              }
+               de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
+               offset += rlen;
+       }
+@@ -1697,12 +1723,12 @@ int ext4_find_dest_de(struct inode *dir,
+ void ext4_insert_dentry(struct inode *inode,
+                       struct ext4_dir_entry_2 *de,
+                      int buf_size,
+-                     struct ext4_filename *fname)
++                     struct ext4_filename *fname, void *data)
+ {
+       int nlen, rlen;
+-      nlen = EXT4_DIR_REC_LEN(de->name_len);
++      nlen = EXT4_DIR_REC_LEN(de);
+       rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
+       if (de->inode) {
+               struct ext4_dir_entry_2 *de1 =
+@@ -1716,6 +1742,11 @@ void ext4_insert_dentry(struct inode *in
+       ext4_set_de_type(inode->i_sb, de, inode->i_mode);
+       de->name_len = fname_len(fname);
+       memcpy(de->name, fname_name(fname), fname_len(fname));
++      if (data) {
++              de->name[fname_len(fname)] = 0;
++              memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
++              de->file_type |= EXT4_DIRENT_LUFID;
++      }
+       return 0;
+ }
+ /*
+@@ -1734,18 +1765,23 @@ static int add_dirent_to_buf(handle_t *h
+ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
+                            struct inode *dir,
+                            struct inode *inode, struct ext4_dir_entry_2 *de,
+-                           struct buffer_head *bh)
++                           struct buffer_head *bh, struct dentry *dentry)
+ {
+       unsigned int    blocksize = dir->i_sb->s_blocksize;
+       int             csum_size = 0;
+-      int             err;
++      int             err, dlen = 0;
++      unsigned char   *data;
++      data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
++                                              dentry->d_fsdata);
+       if (ext4_has_metadata_csum(inode->i_sb))
+               csum_size = sizeof(struct ext4_dir_entry_tail);
+       if (!de) {
++              if (data)
++                      dlen = (*data) + 1;
+               err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
+-                                      blocksize - csum_size, fname, &de);
++                                      blocksize - csum_size, fname, &de, &dlen);
+               if (err)
+                       return err;
+       }
+@@ -1755,7 +1791,10 @@ static int add_dirent_to_buf(handle_t *h
+       /* By now the buffer is marked for journaling. Due to crypto operations,
+        * the following function call may fail */
+-      err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
++      /* If writing the short form of "dotdot", don't add the data section */
++      if (dlen == 1)
++              data = NULL;
++      err = ext4_insert_dentry(dir, inode, de, blocksize, fname, data);
+       if (err < 0)
+               return err;
+@@ -1866,7 +1905,8 @@ static int make_indexed_dir(handle_t *ha
+       dx_set_block(entries, 1);
+       dx_set_count(entries, 1);
+-      dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
++      dx_set_limit(entries, dx_root_limit(dir,
++                                       dot_de, sizeof(*dx_info)));
+       /* Initialize as for dx_probe */
+       hinfo.hash_version = dx_info->hash_version;
+@@ -1876,7 +2476,7 @@ static int make_indexed_dir(handle_t *ha
+       }
+       dx_release(frames);
+-      retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
++      retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh, dentry);
+       brelse(bh);
+       return retval;
+ out_frames:
+@@ -1909,6 +1949,8 @@ static int ext4_update_dotdot(handle_t *
+       struct buffer_head * dir_block;
+       struct ext4_dir_entry_2 * de;
+       int len, journal = 0, err = 0;
++      int dlen = 0;
++      char *data;
+       if (IS_ERR(handle))
+               return PTR_ERR(handle);
+@@ -1924,19 +1966,24 @@ static int ext4_update_dotdot(handle_t *
+       /* the first item must be "." */
+       assert(de->name_len == 1 && de->name[0] == '.');
+       len = le16_to_cpu(de->rec_len);
+-      assert(len >= EXT4_DIR_REC_LEN(1));
+-      if (len > EXT4_DIR_REC_LEN(1)) {
++      assert(len >= __EXT4_DIR_REC_LEN(1));
++      if (len > __EXT4_DIR_REC_LEN(1)) {
+               BUFFER_TRACE(dir_block, "get_write_access");
+               err = ext4_journal_get_write_access(handle, dir_block);
+               if (err)
+                       goto out_journal;
+               journal = 1;
+-              de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
++              de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
+       }
+-      len -= EXT4_DIR_REC_LEN(1);
+-      assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
++      len -= EXT4_DIR_REC_LEN(de);
++      data = ext4_dentry_get_data(dir->i_sb,
++                      (struct ext4_dentry_param *)dentry->d_fsdata);
++      if (data)
++              dlen = *data + 1;
++      assert(len == 0 || len >= __EXT4_DIR_REC_LEN(2 + dlen));
++
+       de = (struct ext4_dir_entry_2 *)
+                       ((char *) de + le16_to_cpu(de->rec_len));
+       if (!journal) {
+@@ -1950,10 +1997,15 @@ static int ext4_update_dotdot(handle_t *
+       if (len > 0)
+               de->rec_len = cpu_to_le16(len);
+       else
+-              assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
++              assert(le16_to_cpu(de->rec_len) >= __EXT4_DIR_REC_LEN(2));
+       de->name_len = 2;
+       strcpy(de->name, "..");
+-      ext4_set_de_type(dir->i_sb, de, S_IFDIR);
++      if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
++              de->name[2] = 0;
++              memcpy(&de->name[2 + 1], data, *data);
++              ext4_set_de_type(dir->i_sb, de, S_IFDIR);
++              de->file_type |= EXT4_DIRENT_LUFID;
++      }
+ out_journal:
+       if (journal) {
+@@ -2039,7 +2639,7 @@ int __ext4_add_entry(handle_t *handle
+                       goto out;
+               }
+               retval = add_dirent_to_buf(handle, &fname, dir, inode,
+-                                         NULL, bh);
++                                         NULL, bh, dentry);
+               if (retval != -ENOSPC)
+                       goto out;
+@@ -2067,7 +2667,7 @@ int __ext4_add_entry(handle_t *handle
+               initialize_dirent_tail(t, blocksize);
+       }
+-      retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
++      retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh, dentry);
+ out:
+       ext4_fname_free_filename(&fname);
+       brelse(bh);
+@@ -2156,7 +2756,7 @@ again:
+               goto cleanup;
+       }
+-      err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
++      err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh, dentry);
+       if (err != -ENOSPC)
+               goto cleanup;
+@@ -2213,7 +2913,7 @@ again:
+               err = PTR_ERR(de);
+               goto cleanup;
+       }
+-      err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
++      err = add_dirent_to_buf(handle, fname, dir, inode, de, bh, dentry);
+       goto cleanup;
+ journal_error:
+@@ -2428,30 +2480,61 @@ retry:
+       return err;
+ }
++struct tp_block {
++      struct inode *inode;
++      void *data1;
++      void *data2;
++};
++
+ struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
+                         struct ext4_dir_entry_2 *de,
+                         int blocksize, int csum_size,
+                         unsigned int parent_ino, int dotdot_real_len)
+ {
++      void *data1 = NULL, *data2 = NULL;
++      int dot_reclen = 0;
++
++      if (dotdot_real_len == 10) {
++              struct tp_block *tpb = (struct tp_block *)inode;
++              data1 = tpb->data1;
++              data2 = tpb->data2;
++              inode = tpb->inode;
++              dotdot_real_len = 0;
++      }
+       de->inode = cpu_to_le32(inode->i_ino);
+       de->name_len = 1;
+-      de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
+-                                         blocksize);
+       strcpy(de->name, ".");
+       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
++      /* get packed fid data*/
++      data1 = ext4_dentry_get_data(inode->i_sb,
++                              (struct ext4_dentry_param *) data1);
++      if (data1) {
++              de->name[1] = 0;
++              memcpy(&de->name[2], data1, *(char *) data1);
++              de->file_type |= EXT4_DIRENT_LUFID;
++      }
++      de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
++      dot_reclen = cpu_to_le16(de->rec_len);
+       de = ext4_next_entry(de, blocksize);
+       de->inode = cpu_to_le32(parent_ino);
+       de->name_len = 2;
++      strcpy(de->name, "..");
++      ext4_set_de_type(inode->i_sb, de, S_IFDIR);
++      data2 = ext4_dentry_get_data(inode->i_sb,
++                      (struct ext4_dentry_param *) data2);
++      if (data2) {
++              de->name[2] = 0;
++              memcpy(&de->name[3], data2, *(char *) data2);
++              de->file_type |= EXT4_DIRENT_LUFID;
++      }
+       if (!dotdot_real_len)
+               de->rec_len = ext4_rec_len_to_disk(blocksize -
+-                                      (csum_size + EXT4_DIR_REC_LEN(1)),
++                                      (csum_size + dot_reclen),
+                                       blocksize);
+       else
+               de->rec_len = ext4_rec_len_to_disk(
+-                              EXT4_DIR_REC_LEN(de->name_len), blocksize);
++                              EXT4_DIR_REC_LEN(de), blocksize);
+-      strcpy(de->name, "..");
+-      ext4_set_de_type(inode->i_sb, de, S_IFDIR);
+       return ext4_next_entry(de, blocksize);
+ }
+@@ -2457,8 +2540,10 @@ struct ext4_dir_entry_2 *ext4_init_dot_d
+ }
+ static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
+-                           struct inode *inode)
++                           struct inode *inode,
++                           const void *data1, const void *data2)
+ {
++      struct tp_block param;
+       struct buffer_head *dir_block = NULL;
+       struct ext4_dir_entry_2 *de;
+       struct ext4_dir_entry_tail *t;
+@@ -2488,7 +2573,11 @@ static int ext4_init_new_dir(handle_t *h
+       if (err)
+               goto out;
+       de = (struct ext4_dir_entry_2 *)dir_block->b_data;
+-      ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
++      param.inode = inode;
++      param.data1 = (void *)data1;
++      param.data2 = (void *)data2;
++      ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
++                           csum_size, dir->i_ino, 10);
+       set_nlink(inode, 2);
+       if (csum_size) {
+               t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
+@@ -2402,6 +2426,29 @@ out:
+       return err;
+ }
++/* Initialize @inode as a subdirectory of @dir, and add the
++ * "." and ".." entries into the first directory block. */
++int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
++                      struct inode *inode,
++                      const void *data1, const void *data2)
++{
++      int rc;
++
++      if (IS_ERR(handle))
++              return PTR_ERR(handle);
++
++      if (IS_DIRSYNC(dir))
++              ext4_handle_sync(handle);
++
++      inode->i_op = &ext4_dir_inode_operations;
++      inode->i_fop = &ext4_dir_operations;
++      rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
++      if (!rc)
++              rc = ext4_mark_inode_dirty(handle, inode);
++      return rc;
++}
++EXPORT_SYMBOL(ext4_add_dot_dotdot);
++
+ static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
+ {
+       handle_t *handle;
+@@ -2546,7 +2636,7 @@ retry:
+       inode->i_op = &ext4_dir_inode_operations;
+       inode->i_fop = &ext4_dir_operations;
+-      err = ext4_init_new_dir(handle, dir, inode);
++      err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
+       if (err)
+               goto out_clear_inode;
+       err = ext4_mark_inode_dirty(handle, inode);
+@@ -2598,7 +2688,7 @@ static int empty_dir(struct inode *inode
+       }
+       sb = inode->i_sb;
+-      if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
++      if (inode->i_size < __EXT4_DIR_REC_LEN(1) + __EXT4_DIR_REC_LEN(2)) {
+               EXT4_ERROR_INODE(inode, "invalid size");
+               return 1;
+       }
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inline.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inline.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inline.c
+@@ -988,7 +998,7 @@ static int ext4_add_dirent_to_inline(han
+       err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
+-                              inline_size, fname, &de);
++                              inline_size, fname, &de, NULL);
+       if (err)
+               return err;
+@@ -998,7 +998,7 @@ static int ext4_add_dirent_to_inline(han
+       err = ext4_journal_get_write_access(handle, iloc->bh);
+       if (err)
+               return err;
+-      ext4_insert_dentry(dir, inode, de, inline_size, fname);
++      ext4_insert_dentry(dir, inode, de, inline_size, fname, NULL);
+       ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
+@@ -1078,7 +1078,7 @@ static int ext4_update_inline_dir(handle
+       int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
+       int new_size = get_max_inline_xattr_value_size(dir, iloc);
+-      if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
++      if (new_size - old_size <= __EXT4_DIR_REC_LEN(1))
+               return -ENOSPC;
+       ret = ext4_update_inline_data(handle, dir,
+@@ -1348,7 +1348,7 @@ int htree_inlinedir_to_tree(struct file
+                       fake.name_len = 1;
+                       strcpy(fake.name, ".");
+                       fake.rec_len = ext4_rec_len_to_disk(
+-                                              EXT4_DIR_REC_LEN(fake.name_len),
++                                              EXT4_DIR_REC_LEN(&fake),
+                                               inline_size);
+                       ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
+                       de = &fake;
+@@ -1358,7 +1358,7 @@ int htree_inlinedir_to_tree(struct file
+                       fake.name_len = 2;
+                       strcpy(fake.name, "..");
+                       fake.rec_len = ext4_rec_len_to_disk(
+-                                              EXT4_DIR_REC_LEN(fake.name_len),
++                                              EXT4_DIR_REC_LEN(&fake),
+                                               inline_size);
+                       ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
+                       de = &fake;
+@@ -1455,8 +1455,8 @@ int ext4_read_inline_dir(struct file *fi
+        * So we will use extra_offset and extra_size to indicate them
+        * during the inline dir iteration.
+        */
+-      dotdot_offset = EXT4_DIR_REC_LEN(1);
+-      dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
++      dotdot_offset = __EXT4_DIR_REC_LEN(1);
++      dotdot_size = dotdot_offset + __EXT4_DIR_REC_LEN(2);
+       extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
+       extra_size = extra_offset + inline_size;
+@@ -1493,7 +1493,7 @@ revalidate:
+                        * failure will be detected in the
+                        * dirent test below. */
+                       if (ext4_rec_len_from_disk(de->rec_len, extra_size)
+-                              < EXT4_DIR_REC_LEN(1))
++                              < __EXT4_DIR_REC_LEN(1))
+                               break;
+                               i += ext4_rec_len_from_disk(de->rec_len,
+                                                           extra_size);
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/super.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c
+@@ -1151,7 +1151,7 @@ enum {
+       Opt_data_err_abort, Opt_data_err_ignore,
+       Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
+       Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
+-      Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
++      Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
+       Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
+       Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
+       Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
+@@ -1216,6 +1216,7 @@ static const match_table_t tokens = {
+       {Opt_stripe, "stripe=%u"},
+       {Opt_delalloc, "delalloc"},
+       {Opt_nodelalloc, "nodelalloc"},
++      {Opt_dirdata, "dirdata"},
+       {Opt_removed, "mblk_io_submit"},
+       {Opt_removed, "nomblk_io_submit"},
+       {Opt_block_validity, "block_validity"},
+@@ -1424,6 +1425,7 @@ static const struct mount_opts {
+       {Opt_usrjquota, 0, MOPT_Q},
+       {Opt_grpjquota, 0, MOPT_Q},
+       {Opt_offusrjquota, 0, MOPT_Q},
++      {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
+       {Opt_offgrpjquota, 0, MOPT_Q},
+       {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
+       {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-disable-mb-cache.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-disable-mb-cache.patch
new file mode 100644 (file)
index 0000000..bc95109
--- /dev/null
@@ -0,0 +1,154 @@
+mbcache provides absolutely no value for Lustre xattrs (because
+they are unique and cannot be shared between files) and as we can
+see it has a noticable overhead in some cases. In the past there
+was a CONFIG_MBCACHE option that would allow it to be disabled,
+but this was removed in newer kernels, so we will need to patch
+ldiskfs to fix this.
+
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
+@@ -944,6 +944,7 @@ struct ext4_inode_info {
+ /*
+  * Mount flags set via mount options or defaults
+  */
++#define EXT4_MOUNT_NO_MBCACHE         0x00001 /* Disable mbcache */
+ #define EXT4_MOUNT_GRPID              0x00004 /* Create files with directory's group */
+ #define EXT4_MOUNT_DEBUG              0x00008 /* Some debugging messages */
+ #define EXT4_MOUNT_ERRORS_CONT                0x00010 /* Continue on errors */
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/super.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c
+@@ -1157,6 +1157,7 @@ enum {
+       Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
+       Opt_inode_readahead_blks, Opt_journal_ioprio,
+       Opt_dioread_nolock, Opt_dioread_lock,
++      Opt_no_mbcache,
+       Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
+       Opt_max_dir_size_kb,
+ };
+@@ -1231,6 +1232,7 @@ static const match_table_t tokens = {
+       {Opt_discard, "discard"},
+       {Opt_nodiscard, "nodiscard"},
+       {Opt_init_itable, "init_itable=%u"},
++      {Opt_no_mbcache, "no_mbcache"},
+       {Opt_init_itable, "init_itable"},
+       {Opt_noinit_itable, "noinit_itable"},
+       {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
+@@ -1390,6 +1392,7 @@ static const struct mount_opts {
+       {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
+       {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
+       {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
++      {Opt_no_mbcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
+       {Opt_commit, 0, MOPT_GTE0},
+       {Opt_max_batch_time, 0, MOPT_GTE0},
+       {Opt_min_batch_time, 0, MOPT_GTE0},
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/xattr.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.c
+@@ -80,7 +80,7 @@
+ # define ea_bdebug(bh, fmt, ...)      no_printk(fmt, ##__VA_ARGS__)
+ #endif
+-static void ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
++static void _ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
+ static struct buffer_head *ext4_xattr_cache_find(struct inode *,
+                                                struct ext4_xattr_header *,
+                                                struct mb_cache_entry **);
+@@ -401,7 +401,8 @@ bad_block:
+               error = -EFSCORRUPTED;
+               goto cleanup;
+       }
+-      ext4_xattr_cache_insert(ext4_mb_cache, bh);
++      if (!test_opt(inode->i_sb, NO_MBCACHE))
++              _ext4_xattr_cache_insert(ext4_mb_cache, bh);
+       entry = BFIRST(bh);
+       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
+                                     inode);
+@@ -565,7 +566,8 @@ ext4_xattr_block_list(struct dentry *den
+               error = -EFSCORRUPTED;
+               goto cleanup;
+       }
+-      ext4_xattr_cache_insert(ext4_mb_cache, bh);
++      if (!test_opt(inode->i_sb, NO_MBCACHE))
++              _ext4_xattr_cache_insert(ext4_mb_cache, bh);
+       error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
+ cleanup:
+@@ -678,7 +680,9 @@ ext4_xattr_release_block(handle_t *handl
+                * This must happen under buffer lock for
+                * ext4_xattr_block_set() to reliably detect freed block
+                */
+-              mb_cache_entry_delete_block(ext4_mb_cache, hash, bh->b_blocknr);
++              if (!test_opt(inode->i_sb, NO_MBCACHE))
++                      mb_cache_entry_delete_block(ext4_mb_cache,
++                                                  hash, bh->b_blocknr);
+               get_bh(bh);
+               unlock_buffer(bh);
+               ext4_free_blocks(handle, inode, bh, 0, 1,
+@@ -690,9 +694,10 @@ ext4_xattr_release_block(handle_t *handl
+               if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
+-                      struct mb_cache_entry *ce;
++                      struct mb_cache_entry *ce = NULL;
+-                      ce = mb_cache_entry_get(ext4_mb_cache, hash,
++                      if (!test_opt(inode->i_sb, NO_MBCACHE))
++                              ce = mb_cache_entry_get(ext4_mb_cache, hash,
+                                               bh->b_blocknr);
+                       if (ce) {
+                               ce->e_reusable = 1;
+@@ -1107,7 +1112,8 @@ ext4_xattr_block_set(handle_t *handle, s
+                        * ext4_xattr_block_set() to reliably detect modified
+                        * block
+                        */
+-                      mb_cache_entry_delete_block(ext4_mb_cache, hash,
++                      if (!test_opt(inode->i_sb, NO_MBCACHE))
++                              mb_cache_entry_delete_block(ext4_mb_cache, hash,
+                                                   bs->bh->b_blocknr);
+                       ea_bdebug(bs->bh, "modifying in-place");
+                       error = ext4_xattr_set_entry(i, s, handle, inode);
+@@ -1115,8 +1121,9 @@ ext4_xattr_block_set(handle_t *handle, s
+                               if (!IS_LAST_ENTRY(s->first))
+                                       ext4_xattr_rehash(header(s->base),
+                                                         s->here);
+-                              ext4_xattr_cache_insert(ext4_mb_cache,
+-                                      bs->bh);
++                              if (!test_opt(inode->i_sb, NO_MBCACHE))
++                                      _ext4_xattr_cache_insert(ext4_mb_cache,
++                                                              bs->bh);
+                       }
+                       unlock_buffer(bs->bh);
+                       if (error == -EFSCORRUPTED)
+@@ -1277,7 +1284,8 @@ getblk_failed:
+                       memcpy(new_bh->b_data, s->base, new_bh->b_size);
+                       set_buffer_uptodate(new_bh);
+                       unlock_buffer(new_bh);
+-                      ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
++                      if (!test_opt(inode->i_sb, NO_MBCACHE))
++                              _ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
+                       error = ext4_handle_dirty_xattr_block(handle,
+                                                             inode, new_bh);
+                       if (error)
+@@ -2068,7 +2076,7 @@ ext4_xattr_inode_array_free(struct inode
+  * Returns 0, or a negative error number on failure.
+  */
+ static void
+-ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
++_ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
+ {
+       struct ext4_xattr_header *header = BHDR(bh);
+       __u32 hash = le32_to_cpu(header->h_hash);
+@@ -2140,6 +2148,8 @@ ext4_xattr_cache_find(struct inode *inod
+       struct mb_cache_entry *ce;
+       struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
++      if (test_opt(inode->i_sb, NO_MBCACHE))
++              return NULL;
+       if (!header->h_hash)
+               return NULL;  /* never share */
+       ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-give-warning-with-dir-htree-growing.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-give-warning-with-dir-htree-growing.patch
new file mode 100644 (file)
index 0000000..83a3f3c
--- /dev/null
@@ -0,0 +1,168 @@
+diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
+index 3c41773..157438f 100644
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -1286,6 +1286,7 @@ struct ext4_sb_info {
+       unsigned long s_mb_prealloc_table_size;
+       unsigned int s_mb_group_prealloc;
+       unsigned int s_max_dir_size_kb;
++      unsigned long s_warning_dir_size;
+       /* where last allocation was done - for stream allocation */
+       unsigned long s_mb_last_group;
+       unsigned long s_mb_last_start;
+diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
+index 7e9e04a..094d54f 100644
+--- a/fs/ext4/namei.c
++++ b/fs/ext4/namei.c
+@@ -687,12 +687,20 @@ struct ext4_dir_lock_data {
+ #define ext4_htree_lock_data(l)       ((struct ext4_dir_lock_data *)(l)->lk_private)
+ #define ext4_find_entry(dir, name, dirent, inline) \
+                       __ext4_find_entry(dir, name, dirent, inline, NULL)
+-#define ext4_add_entry(handle, dentry, inode) \
+-                      __ext4_add_entry(handle, dentry, inode, NULL)
+ /* NB: ext4_lblk_t is 32 bits so we use high bits to identify invalid blk */
+ #define EXT4_HTREE_NODE_CHANGED       (0xcafeULL << 32)
++inline int ext4_add_entry(handle_t *handle, struct dentry *dentry,
++                        struct inode *inode)
++{
++      int ret = __ext4_add_entry(handle, dentry, inode, NULL);
++
++      if (ret == -ENOBUFS)
++              ret = 0;
++      return ret;
++}
++
+ static void ext4_htree_event_cb(void *target, void *event)
+ {
+       u64 *block = (u64 *)target;
+@@ -2479,6 +2487,54 @@ int __ext4_add_entry(handle_t *handle, struct dentry *dentry,
+ }
+ EXPORT_SYMBOL(__ext4_add_entry);
++static unsigned long __ext4_max_dir_size(struct dx_frame *frames,
++                             struct dx_frame *frame, struct inode *dir)
++{
++      unsigned long max_dir_size;
++
++      if (EXT4_SB(dir->i_sb)->s_max_dir_size_kb) {
++              max_dir_size = EXT4_SB(dir->i_sb)->s_max_dir_size_kb << 10;
++      } else {
++              max_dir_size = EXT4_BLOCK_SIZE(dir->i_sb);
++              while (frame >= frames) {
++                      max_dir_size *= dx_get_limit(frame->entries);
++                      if (frame == frames)
++                              break;
++                      frame--;
++              }
++              /* use 75% of max dir size in average */
++              max_dir_size = max_dir_size / 4 * 3;
++      }
++      return max_dir_size;
++}
++
++/*
++ * With hash tree growing, it is easy to hit ENOSPC, but it is hard
++ * to predict when it will happen. let's give administrators warning
++ * when reaching 3/5 and 2/3 of limit
++ */
++static inline bool dir_size_in_warning_range(struct dx_frame *frames,
++                                           struct dx_frame *frame,
++                                           struct inode *dir)
++{
++      unsigned long size1, size2;
++      struct super_block *sb = dir->i_sb;
++
++      if (unlikely(!EXT4_SB(sb)->s_warning_dir_size))
++              EXT4_SB(sb)->s_warning_dir_size =
++                      __ext4_max_dir_size(frames, frame, dir);
++
++      size1 = EXT4_SB(sb)->s_warning_dir_size / 16 * 10;
++      size1 = size1 & ~(EXT4_BLOCK_SIZE(sb) - 1);
++      size2 = EXT4_SB(sb)->s_warning_dir_size / 16 * 11;
++      size2 = size2 & ~(EXT4_BLOCK_SIZE(sb) - 1);
++      if (in_range(dir->i_size, size1, EXT4_BLOCK_SIZE(sb)) ||
++          in_range(dir->i_size, size2, EXT4_BLOCK_SIZE(sb)))
++              return true;
++
++      return false;
++}
++
+ /*
+  * Returns 0 for success, or a negative error value
+  */
+@@ -2494,6 +2550,7 @@ static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
+       struct ext4_dir_entry_2 *de;
+       int restart;
+       int err;
++      bool ret_warn = false;
+ again:
+       restart = 0;
+@@ -2517,6 +2574,11 @@ again:
+       /* Block full, should compress but for now just split */
+       dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
+                      dx_get_count(entries), dx_get_limit(entries)));
++
++      if (frame - frames + 1 >= ext4_dir_htree_level(sb) ||
++          EXT4_SB(sb)->s_warning_dir_size)
++              ret_warn = dir_size_in_warning_range(frames, frame, dir);
++
+       /* Need to split index? */
+       if (dx_get_count(entries) == dx_get_limit(entries)) {
+               ext4_lblk_t newblock;
+@@ -2548,7 +2610,7 @@ again:
+                                        dir->i_ino, current->comm, levels,
+                                        ext4_dir_htree_level(sb));
+                       if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
+-                              ext4_warning(sb, "Large directory feature is"
++                              ext4_warning(sb, "Large directory feature is "
+                                                "not enabled on this "
+                                                "filesystem");
+                       }
+@@ -2674,6 +2736,8 @@ cleanup:
+        * repeat dx_probe() to find out valid htree-path */
+       if (restart && err == 0)
+               goto again;
++      if (err == 0 && ret_warn)
++              err = -ENOBUFS;
+       return err;
+ }
+diff --git a/fs/ext4/super.c b/fs/ext4/super.c
+index c625960..1914379 100644
+--- a/fs/ext4/super.c
++++ b/fs/ext4/super.c
+@@ -1546,6 +1546,8 @@ static int handle_mount_opt(struct super_block *sb, char *opt, int token,
+               sbi->s_li_wait_mult = arg;
+       } else if (token == Opt_max_dir_size_kb) {
+               sbi->s_max_dir_size_kb = arg;
++              /* reset s_warning_dir_size and make it re-calculated */
++              sbi->s_warning_dir_size = 0;
+       } else if (token == Opt_stripe) {
+               sbi->s_stripe = arg;
+       } else if (token == Opt_resuid) {
+diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
+index c625960..1914379 100644
+--- a/fs/ext4/sysfs.c
++++ b/fs/ext4/sysfs.c
+@@ -2657,6 +2659,7 @@ EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
+ EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
+ EXT4_RW_ATTR_SBI_UI(max_dir_size, s_max_dir_size_kb);
+ EXT4_RW_ATTR_SBI_UI(max_dir_size_kb, s_max_dir_size_kb);
++EXT4_RW_ATTR_SBI_UI(warning_dir_size, s_warning_dir_size);
+ EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
+ EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
+ EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
+@@ -2686,6 +2689,7 @@ static struct attribute *ext4_attrs[] = {
+       ATTR_LIST(inode_goal),
+       ATTR_LIST(max_dir_size),
+       ATTR_LIST(max_dir_size_kb),
++      ATTR_LIST(warning_dir_size),
+       ATTR_LIST(mb_stats),
+       ATTR_LIST(mb_max_to_scan),
+       ATTR_LIST(mb_min_to_scan),
+-- 
+1.8.3.1
+
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-hash-indexed-dir-dotdot-update.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-hash-indexed-dir-dotdot-update.patch
new file mode 100644 (file)
index 0000000..d781202
--- /dev/null
@@ -0,0 +1,89 @@
+Index: linux-3.10.0-123.9.3.el7.x86_64/fs/ext4/namei.c
+===================================================================
+--- linux-3.10.0-123.9.3.el7.x86_64.orig/fs/ext4/namei.c
++++ linux-3.10.0-123.9.3.el7.x86_64/fs/ext4/namei.c
+@@ -1894,6 +1894,74 @@ static int make_indexed_dir(handle_t *ha
+       return retval;
+ }
++/* update ".." for hash-indexed directory, split the item "." if necessary */
++static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
++                            struct inode *inode)
++{
++      struct inode *dir = dentry->d_parent->d_inode;
++      struct buffer_head *dir_block;
++      struct ext4_dir_entry_2 *de;
++      int len, journal = 0, err = 0;
++
++      if (IS_ERR(handle))
++              return PTR_ERR(handle);
++
++      if (IS_DIRSYNC(dir))
++              handle->h_sync = 1;
++
++      dir_block = ext4_bread(handle, dir, 0, 0);
++      if (IS_ERR(dir_block)) {
++              err = PTR_ERR(dir_block);
++              goto out;
++      }
++
++      de = (struct ext4_dir_entry_2 *)dir_block->b_data;
++      /* the first item must be "." */
++      assert(de->name_len == 1 && de->name[0] == '.');
++      len = le16_to_cpu(de->rec_len);
++      assert(len >= EXT4_DIR_REC_LEN(1));
++      if (len > EXT4_DIR_REC_LEN(1)) {
++              BUFFER_TRACE(dir_block, "get_write_access");
++              err = ext4_journal_get_write_access(handle, dir_block);
++              if (err)
++                      goto out_journal;
++
++              journal = 1;
++              de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
++      }
++
++      len -= EXT4_DIR_REC_LEN(1);
++      assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
++      de = (struct ext4_dir_entry_2 *)
++                      ((char *) de + le16_to_cpu(de->rec_len));
++      if (!journal) {
++              BUFFER_TRACE(dir_block, "get_write_access");
++              err = ext4_journal_get_write_access(handle, dir_block);
++              if (err)
++                      goto out_journal;
++      }
++
++      de->inode = cpu_to_le32(inode->i_ino);
++      if (len > 0)
++              de->rec_len = cpu_to_le16(len);
++      else
++              assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
++      de->name_len = 2;
++      strcpy(de->name, "..");
++      ext4_set_de_type(dir->i_sb, de, S_IFDIR);
++
++out_journal:
++      if (journal) {
++              BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
++              err = ext4_handle_dirty_dirent_node(handle, dir, dir_block);
++              ext4_mark_inode_dirty(handle, dir);
++      }
++      brelse(dir_block);
++
++out:
++      return err;
++}
++
+ /*
+  *    ext4_add_entry()
+  *
+@@ -1938,6 +2004,9 @@ int ext4_add_entry(handle_t *handle, str
+       }
+       if (is_dx(dir)) {
++              if (dentry->d_name.len == 2 &&
++                  memcmp(dentry->d_name.name, "..", 2) == 0)
++                      return ext4_update_dotdot(handle, dentry, inode);
+               retval = ext4_dx_add_entry(handle, &fname, dentry, inode);
+               if (!retval || (retval != ERR_BAD_DX_DIR))
+                       return retval;
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-inode-version.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-inode-version.patch
new file mode 100644 (file)
index 0000000..ababc82
--- /dev/null
@@ -0,0 +1,56 @@
+Index: linux-stage/fs/ext4/inode.c
+===================================================================
+--- linux-stage.orig/fs/ext4/inode.c
++++ linux-stage/fs/ext4/inode.c
+@@ -4549,10 +4549,10 @@ struct inode *ext4_iget(struct super_blo
+       EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
+       if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
+-              inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
++              ei->i_fs_version = le32_to_cpu(raw_inode->i_disk_version);
+               if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
+                       if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
+-                              inode->i_version |=
++                              ei->i_fs_version |=
+                   (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
+               }
+       }
+@@ -4837,11 +4837,11 @@ static int ext4_do_update_inode(handle_t
+       }
+       if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
+-              raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
++              raw_inode->i_disk_version = cpu_to_le32(ei->i_fs_version);
+               if (ei->i_extra_isize) {
+                       if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
+                               raw_inode->i_version_hi =
+-                                      cpu_to_le32(inode->i_version >> 32);
++                                      cpu_to_le32(ei->i_fs_version >> 32);
+                       raw_inode->i_extra_isize =
+                               cpu_to_le16(ei->i_extra_isize);
+               }
+Index: linux-stage/fs/ext4/ialloc.c
+===================================================================
+--- linux-stage.orig/fs/ext4/ialloc.c
++++ linux-stage/fs/ext4/ialloc.c
+@@ -1045,6 +1045,7 @@ got:
+       ei->i_dtime = 0;
+       ei->i_block_group = group;
+       ei->i_last_alloc_group = ~0;
++      ei->i_fs_version = 0;
+       ext4_set_inode_flags(inode);
+       if (IS_DIRSYNC(inode))
+Index: linux-stage/fs/ext4/ext4.h
+===================================================================
+--- linux-stage.orig/fs/ext4/ext4.h
++++ linux-stage/fs/ext4/ext4.h
+@@ -1022,6 +1022,8 @@ struct ext4_inode_info {
+       struct dquot *i_dquot[MAXQUOTAS];
+ #endif
++      __u64 i_fs_version;
++
+       /* Precomputed uuid+inum+igen checksum for seeding inode checksums */
+       __u32 i_csum_seed;
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-kill-dx-root.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-kill-dx-root.patch
new file mode 100644 (file)
index 0000000..231f76b
--- /dev/null
@@ -0,0 +1,222 @@
+removes static definition of dx_root struct. so that "." and ".." dirent can
+have extra data. This patch does not change any functionality but is required for
+ext4_data_in_dirent patch.
+Index: linux-3.10.0-123.9.3.el7.x86_64/fs/ext4/namei.c
+===================================================================
+--- linux-3.10.0-123.9.3.el7.x86_64.orig/fs/ext4/namei.c
++++ linux-3.10.0-123.9.3.el7.x86_64/fs/ext4/namei.c
+@@ -193,22 +193,13 @@ struct dx_entry
+  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
+  */
+-struct dx_root
++struct dx_root_info
+ {
+-      struct fake_dirent dot;
+-      char dot_name[4];
+-      struct fake_dirent dotdot;
+-      char dotdot_name[4];
+-      struct dx_root_info
+-      {
+-              __le32 reserved_zero;
+-              u8 hash_version;
+-              u8 info_length; /* 8 */
+-              u8 indirect_levels;
+-              u8 unused_flags;
+-      }
+-      info;
+-      struct dx_entry entries[0];
++      __le32 reserved_zero;
++      u8 hash_version;
++      u8 info_length; /* 8 */
++      u8 indirect_levels;
++      u8 unused_flags;
+ };
+ struct dx_node
+@@ -511,6 +502,16 @@ ext4_next_entry(struct ext4_dir_entry_2
+  * Future: use high four bits of block for coalesce-on-delete flags
+  * Mask them off for now.
+  */
++struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
++{
++      /* get dotdot first */
++      de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
++
++      /* dx root info is after dotdot entry */
++      de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
++
++      return (struct dx_root_info *)de;
++}
+ static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
+ {
+@@ -673,7 +674,7 @@ dx_probe(const struct qstr *d_name, stru
+ {
+       unsigned count, indirect;
+       struct dx_entry *at, *entries, *p, *q, *m;
+-      struct dx_root *root;
++      struct dx_root_info *info;
+       struct dx_frame *frame = frame_in;
+       u32 hash;
+@@ -684,17 +685,17 @@ dx_probe(const struct qstr *d_name, stru
+       if (IS_ERR(frame->bh))
+               return (struct dx_frame *) frame->bh;
+-      root = (struct dx_root *) frame->bh->b_data;
+-      if (root->info.hash_version != DX_HASH_TEA &&
+-          root->info.hash_version != DX_HASH_HALF_MD4 &&
+-          root->info.hash_version != DX_HASH_LEGACY) {
++      info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data);
++      if (info->hash_version != DX_HASH_TEA &&
++          info->hash_version != DX_HASH_HALF_MD4 &&
++          info->hash_version != DX_HASH_LEGACY) {
+               ext4_warning_inode(dir, "Unrecognised inode hash code %u for directory "
+-                                 "%lu", root->info.hash_version, dir->i_ino);
++                                 "%lu", info->hash_version, dir->i_ino);
+               goto fail;
+       }
+       if (fname)
+               hinfo = &fname->hinfo;
+-      hinfo->hash_version = root->info.hash_version;
++      hinfo->hash_version = info->hash_version;
+       if (hinfo->hash_version <= DX_HASH_TEA)
+               hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
+       hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
+@@ -702,27 +704,26 @@ dx_probe(const struct qstr *d_name, stru
+               ext4fs_dirhash(d_name->name, d_name->len, hinfo);
+       hash = hinfo->hash;
+-      if (root->info.unused_flags & 1) {
++      if (info->unused_flags & 1) {
+               ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
+-                                 root->info.unused_flags);
++                                 info->unused_flags);
+               goto fail;
+       }
+-      indirect = root->info.indirect_levels;
++      indirect = info->indirect_levels;
+       if (indirect > 1) {
+               ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
+-                                 root->info.indirect_levels);
++                                 info->indirect_levels);
+               goto fail;
+       }
+-      entries = (struct dx_entry *)(((char *)&root->info) +
+-                                    root->info.info_length);
++      entries = (struct dx_entry *)(((char *)info) + info->info_length);
+       if (dx_get_limit(entries) != dx_root_limit(dir,
+-                                                 root->info.info_length)) {
++                                                 info->info_length)) {
+               ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
+                                  dx_get_limit(entries),
+-                                 dx_root_limit(dir, root->info.info_length));
++                                 dx_root_limit(dir, info->info_length));
+               goto fail;
+       }
+@@ -1787,10 +1790,9 @@ static int make_indexed_dir(handle_t *ha
+       const char      *name = dentry->d_name.name;
+       int             namelen = dentry->d_name.len;
+       struct buffer_head *bh2;
+-      struct dx_root  *root;
+       struct dx_frame frames[2], *frame;
+       struct dx_entry *entries;
+-      struct ext4_dir_entry_2 *de, *de2;
++      struct ext4_dir_entry_2 *de, *de2, *dot_de, *dotdot_de;
+       struct ext4_dir_entry_tail *t;
+       char            *data1, *top;
+       unsigned        len;
+@@ -1798,7 +1800,7 @@ static int make_indexed_dir(handle_t *ha
+       unsigned        blocksize;
+       struct dx_hash_info hinfo;
+       ext4_lblk_t  block;
+-      struct fake_dirent *fde;
++      struct dx_root_info *dx_info;
+       int csum_size = 0;
+       if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
+@@ -1813,18 +1815,19 @@ static int make_indexed_dir(handle_t *ha
+               brelse(bh);
+               return retval;
+       }
+-      root = (struct dx_root *) bh->b_data;
++
++      dot_de = (struct ext4_dir_entry_2 *)bh->b_data;
++      dotdot_de = ext4_next_entry(dot_de, blocksize);
+       /* The 0th block becomes the root, move the dirents out */
+-      fde = &root->dotdot;
+-      de = (struct ext4_dir_entry_2 *)((char *)fde +
+-              ext4_rec_len_from_disk(fde->rec_len, blocksize));
+-      if ((char *) de >= (((char *) root) + blocksize)) {
++      de = (struct ext4_dir_entry_2 *)((char *)dotdot_de +
++              ext4_rec_len_from_disk(dotdot_de->rec_len, blocksize));
++      if ((char *)de >= (((char *)dot_de) + blocksize)) {
+               EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
+               brelse(bh);
+               return -EFSCORRUPTED;
+       }
+-      len = ((char *) root) + (blocksize - csum_size) - (char *) de;
++      len = ((char *)dot_de) + (blocksize - csum_size) - (char *)de;
+       /* Allocate new block for the 0th block's dirents */
+       bh2 = ext4_append(handle, dir, &block);
+@@ -1850,19 +1853,24 @@ static int make_indexed_dir(handle_t *ha
+       }
+       /* Initialize the root; the dot dirents already exist */
+-      de = (struct ext4_dir_entry_2 *) (&root->dotdot);
+-      de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
+-                                         blocksize);
+-      memset (&root->info, 0, sizeof(root->info));
+-      root->info.info_length = sizeof(root->info);
+-      root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
+-      entries = root->entries;
++      dotdot_de->rec_len =
++              ext4_rec_len_to_disk(blocksize - le16_to_cpu(dot_de->rec_len),
++                                   blocksize);
++
++      /* initialize hashing info */
++      dx_info = dx_get_dx_info(dot_de);
++      memset(dx_info, 0, sizeof(*dx_info));
++      dx_info->info_length = sizeof(*dx_info);
++      dx_info->hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
++
++      entries = (void *)dx_info + sizeof(*dx_info);
++
+       dx_set_block(entries, 1);
+       dx_set_count(entries, 1);
+-      dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
++      dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
+       /* Initialize as for dx_probe */
+-      fname->hinfo.hash_version = root->info.hash_version;
++      fname->hinfo.hash_version = dx_info->hash_version;
+       if (fname->hinfo.hash_version <= DX_HASH_TEA)
+               hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
+       hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
+@@ -2152,6 +2159,7 @@ static int ext4_dx_add_entry(handle_t *h
+                               goto journal_error;
+                       brelse (bh2);
+               } else {
++                      struct dx_root_info *info;
+                       dxtrace(printk(KERN_DEBUG
+                                      "Creating second level index...\n"));
+                       memcpy((char *) entries2, (char *) entries,
+@@ -2161,7 +2169,9 @@ static int ext4_dx_add_entry(handle_t *h
+                       /* Set up root */
+                       dx_set_count(entries, 1);
+                       dx_set_block(entries + 0, newblock);
+-                      ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
++                      info = dx_get_dx_info((struct ext4_dir_entry_2 *)
++                                            frames[0].bh->b_data);
++                      info->indirect_levels = 1;
+                       /* Add new access path frame */
+                       frame = frames + 1;
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-large-dir.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-large-dir.patch
new file mode 100644 (file)
index 0000000..019daca
--- /dev/null
@@ -0,0 +1,337 @@
+This INCOMPAT_LARGEDIR feature allows larger directories
+to be created in ldiskfs, both with directory sizes over
+2GB and and a maximum htree depth of 3 instead of the
+current limit of 2. These features are needed in order
+to exceed the current limit of approximately 10M entries
+in a single directory.
+
+Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/ext4.h
++++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/ext4.h
+@@ -1585,6 +1585,7 @@ static inline void ext4_clear_state_flag
+                                        EXT4_FEATURE_INCOMPAT_MMP |    \
+                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
+                                        EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
++                                       EXT4_FEATURE_INCOMPAT_LARGEDIR | \
+                                        EXT4_FEATURE_INCOMPAT_ENCRYPT | \
+                                        EXT4_FEATURE_INCOMPAT_CSUM_SEED)
+ #define EXT4_FEATURE_RO_COMPAT_SUPP   (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
+@@ -1999,6 +2000,9 @@ struct mmpd_data {
+ # define NORET_TYPE   /**/
+ # define ATTRIB_NORET __attribute__((noreturn))
+ # define NORET_AND    noreturn,
++/* htree levels for ext4 */
++#define EXT4_HTREE_LEVEL_COMPAT 2
++#define EXT4_HTREE_LEVEL      3
+ struct ext4_xattr_ino_array {
+       unsigned int xia_count;         /* # of used item in the array */
+@@ -2472,13 +2476,16 @@ static inline void ext4_r_blocks_count_s
+       es->s_r_blocks_count_hi = cpu_to_le32(blk >> 32);
+ }
+-static inline loff_t ext4_isize(struct ext4_inode *raw_inode)
++static inline loff_t ext4_isize(struct super_block *sb,
++                              struct ext4_inode *raw_inode)
+ {
+-      if (S_ISREG(le16_to_cpu(raw_inode->i_mode)))
++      if (S_ISREG(le16_to_cpu(raw_inode->i_mode)) ||
++          (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_LARGEDIR) &&
++          S_ISDIR(le16_to_cpu(raw_inode->i_mode))))
+               return ((loff_t)le32_to_cpu(raw_inode->i_size_high) << 32) |
+                       le32_to_cpu(raw_inode->i_size_lo);
+-      else
+-              return (loff_t) le32_to_cpu(raw_inode->i_size_lo);
++
++      return (loff_t)le32_to_cpu(raw_inode->i_size_lo);
+ }
+ static inline void ext4_isize_set(struct ext4_inode *raw_inode, loff_t i_size)
+Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/namei.c
+===================================================================
+--- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/namei.c
++++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/namei.c
+@@ -513,7 +513,14 @@ struct dx_root_info * dx_get_dx_info(str
+ static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
+ {
+-      return le32_to_cpu(entry->block) & 0x00ffffff;
++      return le32_to_cpu(entry->block) & 0x0fffffff;
++}
++
++static inline int
++ext4_dir_htree_level(struct super_block *sb)
++{
++      return EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_LARGEDIR) ?
++              EXT4_HTREE_LEVEL : EXT4_HTREE_LEVEL_COMPAT;
+ }
+ static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
+@@ -681,6 +688,7 @@ dx_probe(const struct qstr *d_name, stru
+       struct dx_frame *frame = frame_in;
+       u32 hash;
++      memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
+       frame->bh = ext4_read_dirblock(dir, 0, INDEX);
+       if (IS_ERR(frame->bh))
+               return (struct dx_frame *) frame->bh;
+@@ -714,9 +721,13 @@ dx_probe(const struct qstr *d_name, stru
+       }
+       indirect = info->indirect_levels;
+-      if (indirect > 1) {
+-              ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
+-                                 info->indirect_levels);
++      if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
++              ext4_warning_inode(dir, "htree depth: %#06x exceed max depth %u",
++                                 indirect, ext4_dir_htree_level(dir->i_sb));
++              if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
++                      ext4_warning(dir->i_sb, "Enable large directory "
++                                              "feature to access it");
++              }
+               goto fail;
+       }
+@@ -812,12 +826,20 @@ fail:
+ static void dx_release (struct dx_frame *frames)
+ {
++      int i;
++      struct dx_root_info *info;
++
+       if (frames[0].bh == NULL)
+               return;
+-      if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
+-              brelse(frames[1].bh);
+-      brelse(frames[0].bh);
++      for (i = 0, info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data);
++           i <= info->indirect_levels;
++           i++) {
++              if (frames[i].bh == NULL)
++                      break;
++              brelse(frames[i].bh);
++              frames[i].bh = NULL;
++      }
+ }
+ /*
+@@ -960,7 +979,7 @@ int ext4_htree_fill_tree(struct file *di
+ {
+       struct dx_hash_info hinfo;
+       struct ext4_dir_entry_2 *de;
+-      struct dx_frame frames[2], *frame;
++      struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+       struct inode *dir;
+       ext4_lblk_t block;
+       int count = 0;
+@@ -1376,7 +1395,7 @@ static struct buffer_head * ext4_dx_find
+       struct dx_hash_info     hinfo;
+ {
+       struct super_block * sb = dir->i_sb;
+-      struct dx_frame frames[2], *frame;
++      struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+       const struct qstr *d_name = fname->usr_fname;
+       struct buffer_head *bh;
+       ext4_lblk_t block;
+@@ -1832,7 +1851,7 @@ static int make_indexed_dir(handle_t *ha
+       const char      *name = dentry->d_name.name;
+       int             namelen = dentry->d_name.len;
+       struct buffer_head *bh2;
+-      struct dx_frame frames[2], *frame;
++      struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+       struct dx_entry *entries;
+       struct ext4_dir_entry_2 *de, *de2, *dot_de, *dotdot_de;
+       struct ext4_dir_entry_tail *t;
+@@ -2117,14 +2136,17 @@ static int ext4_add_entry(handle_t *hand
+ static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
+                            struct inode *inode)
+ {
+-      struct dx_frame frames[2], *frame;
++      struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+       struct dx_entry *entries, *at;
+       struct buffer_head *bh;
+       struct inode *dir = d_inode(dentry->d_parent);
+       struct super_block *sb = dir->i_sb;
+       struct ext4_dir_entry_2 *de;
++      int restart;
+       int err;
++again:
++      restart = 0;
+       frame = dx_probe(fname, dir, NULL, frames);
+       if (IS_ERR(frame))
+               return PTR_ERR(frame);
+@@ -2138,33 +2160,48 @@ static int ext4_dx_add_entry(handle_t *h
+               goto cleanup;
+       }
+-      BUFFER_TRACE(bh, "get_write_access");
+-      err = ext4_journal_get_write_access(handle, bh);
+-      if (err)
+-              goto journal_error;
+-
+       err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh, dentry);
+       if (err != -ENOSPC)
+               goto cleanup;
++      err = 0;
+       /* Block full, should compress but for now just split */
+       dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
+                      dx_get_count(entries), dx_get_limit(entries)));
+       /* Need to split index? */
+       if (dx_get_count(entries) == dx_get_limit(entries)) {
+               ext4_lblk_t newblock;
+-              unsigned icount = dx_get_count(entries);
+-              int levels = frame - frames;
++              int levels = frame - frames + 1;
++              unsigned icount;
++              int add_level = 1;
+               struct dx_entry *entries2;
+               struct dx_node *node2;
+               struct buffer_head *bh2;
+-              if (levels && (dx_get_count(frames->entries) ==
+-                             dx_get_limit(frames->entries))) {
+-                      ext4_warning_inode(dir, "Directory index full!");
++              while (frame > frames) {
++                      if (dx_get_count((frame - 1)->entries) <
++                          dx_get_limit((frame - 1)->entries)) {
++                              add_level = 0;
++                              break;
++                      }
++                      frame--; /* split higher index block */
++                      at = frame->at;
++                      entries = frame->entries;
++                      restart = 1;
++              }
++              if (add_level && levels == ext4_dir_htree_level(sb)) {
++                      ext4_warning(sb, "inode %lu: comm %s: index %u: reach max htree level %u",
++                                       dir->i_ino, current->comm, levels,
++                                       ext4_dir_htree_level(sb));
++                      if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
++                              ext4_warning(sb, "Large directory feature is"
++                                               "not enabled on this "
++                                               "filesystem");
++                      }
+                       err = -ENOSPC;
+                       goto cleanup;
+               }
++              icount = dx_get_count(entries);
+               bh2 = ext4_append(handle, dir, &newblock);
+               if (IS_ERR(bh2)) {
+                       err = PTR_ERR(bh2);
+@@ -2179,7 +2216,7 @@ static int ext4_dx_add_entry(handle_t *h
+               err = ext4_journal_get_write_access(handle, frame->bh);
+               if (err)
+                       goto journal_error;
+-              if (levels) {
++              if (!add_level) {
+                       unsigned icount1 = icount/2, icount2 = icount - icount1;
+                       unsigned hash2 = dx_get_hash(entries + icount1);
+                       dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
+@@ -2187,7 +2224,7 @@ static int ext4_dx_add_entry(handle_t *h
+                       BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
+                       err = ext4_journal_get_write_access(handle,
+-                                                           frames[0].bh);
++                                                          (frame - 1)->bh);
+                       if (err)
+                               goto journal_error;
+@@ -2203,19 +2240,25 @@ static int ext4_dx_add_entry(handle_t *h
+                               frame->entries = entries = entries2;
+                               swap(frame->bh, bh2);
+                       }
+-                      dx_insert_block(frames + 0, hash2, newblock);
+-                      dxtrace(dx_show_index("node", frames[1].entries));
++                      dx_insert_block(frame - 1, hash2, newblock);
++                      dxtrace(dx_show_index("node", frame->entries));
+                       dxtrace(dx_show_index("node",
+-                             ((struct dx_node *) bh2->b_data)->entries));
++                             ((struct dx_node *)bh2->b_data)->entries));
+                       err = ext4_handle_dirty_dx_node(handle, dir, bh2);
+                       if (err)
+                               goto journal_error;
+                       brelse (bh2);
++                      ext4_handle_dirty_dirent_node(handle, dir,
++                                                    (frame - 1)->bh);
++                      if (restart) {
++                              ext4_handle_dirty_dirent_node(handle, dir,
++                                                            frame->bh);
++                              goto cleanup;
++                      }
+               } else {
+                       struct dx_root_info *info;
+-                      dxtrace(printk(KERN_DEBUG
+-                                     "Creating second level index...\n"));
+-                      memcpy((char *) entries2, (char *) entries,
++
++                      memcpy((char *)entries2, (char *)entries,
+                              icount * sizeof(struct dx_entry));
+                       dx_set_limit(entries2, dx_node_limit(dir));
+@@ -2224,21 +2267,14 @@ static int ext4_dx_add_entry(handle_t *h
+                       dx_set_block(entries + 0, newblock);
+                       info = dx_get_dx_info((struct ext4_dir_entry_2*)
+                                             frames[0].bh->b_data);
+-                      info->indirect_levels = 1;
+-
+-                      /* Add new access path frame */
+-                      frame = frames + 1;
+-                      frame->at = at = at - entries + entries2;
+-                      frame->entries = entries = entries2;
+-                      frame->bh = bh2;
+-                      err = ext4_journal_get_write_access(handle,
+-                                                           frame->bh);
+-                      if (err)
+-                              goto journal_error;
+-              }
+-              err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
+-              if (err) {
+-                      ext4_std_error(inode->i_sb, err);
++                      info->indirect_levels += 1;
++                      dxtrace(printk(KERN_DEBUG
++                                     "Creating %d level index...\n",
++                                     info->indirect_levels));
++                      ext4_handle_dirty_dirent_node(handle, dir, frame->bh);
++                      ext4_handle_dirty_dirent_node(handle, dir, bh2);
++                      brelse(bh2);
++                      restart = 1;
+                       goto cleanup;
+               }
+       }
+@@ -2253,6 +2289,10 @@ journal_error:
+ cleanup:
+       brelse(bh);
+       dx_release(frames);
++      /* @restart is true means htree-path has been changed, we need to
++       * repeat dx_probe() to find out valid htree-path */
++      if (restart && err == 0)
++              goto again;
+       return err;
+ }
+Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/inode.c
+===================================================================
+--- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/inode.c
++++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/inode.c
+@@ -4056,7 +4056,7 @@ struct inode *ext4_iget(struct super_blo
+       if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
+               ei->i_file_acl |=
+                       ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
+-      inode->i_size = ext4_isize(raw_inode);
++      inode->i_size = ext4_isize(sb, raw_inode);
+       ei->i_disksize = inode->i_size;
+ #ifdef CONFIG_QUOTA
+       ei->i_reserved_quota = 0;
+@@ -4306,7 +4306,7 @@ static int ext4_do_update_inode(handle_t
+               raw_inode->i_file_acl_high =
+                       cpu_to_le16(ei->i_file_acl >> 32);
+       raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
+-      if (ei->i_disksize != ext4_isize(raw_inode)) {
++      if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
+               ext4_isize_set(raw_inode, ei->i_disksize);
+               need_datasync = 1;
+       }
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-large-eas.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-large-eas.patch
new file mode 100644 (file)
index 0000000..962512e
--- /dev/null
@@ -0,0 +1,1086 @@
+This patch implements the large EA support in ext4. If the size of
+an EA value is larger than the blocksize, then the EA value would
+not be saved in the external EA block, instead it would be saved
+in an external EA inode. So, the patch also helps support a larger
+number of EAs.
+
+Index: linux-stage/fs/ext4/ext4.h
+===================================================================
+--- linux-stage.orig/fs/ext4/ext4.h
++++ linux-stage/fs/ext4/ext4.h
+@@ -1579,6 +1579,7 @@ static inline void ext4_clear_state_flag
+                                        EXT4_FEATURE_INCOMPAT_EXTENTS| \
+                                        EXT4_FEATURE_INCOMPAT_64BIT| \
+                                        EXT4_FEATURE_INCOMPAT_FLEX_BG| \
++                                       EXT4_FEATURE_INCOMPAT_EA_INODE| \
+                                        EXT4_FEATURE_INCOMPAT_MMP | \
+                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
+                                        EXT4_FEATURE_INCOMPAT_INLINE_DATA)
+@@ -1979,6 +1980,12 @@ struct mmpd_data {
+ #define EXT4_MMP_MAX_CHECK_INTERVAL   300UL
+ /*
++ * Maximum size of xattr attributes for FEATURE_INCOMPAT_EA_INODE 1Mb
++ * This limit is arbitrary, but is reasonable for the xattr API.
++ */
++#define EXT4_XATTR_MAX_LARGE_EA_SIZE    (1024 * 1024)
++
++/*
+  * Function prototypes
+  */
+@@ -1990,6 +1997,10 @@ struct mmpd_data {
+ # define ATTRIB_NORET __attribute__((noreturn))
+ # define NORET_AND    noreturn,
++struct ext4_xattr_ino_array {
++      unsigned int xia_count;         /* # of used item in the array */
++      unsigned int xia_inodes[0];
++};
+ /* bitmap.c */
+ extern unsigned int ext4_count_free(char *bitmap, unsigned numchars);
+ void ext4_inode_bitmap_csum_set(struct super_block *sb, ext4_group_t group,
+@@ -2194,6 +2205,7 @@ extern void ext4_set_inode_flags(struct
+ extern void ext4_get_inode_flags(struct ext4_inode_info *);
+ extern int ext4_alloc_da_blocks(struct inode *inode);
+ extern void ext4_set_aops(struct inode *inode);
++extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int chunk);
+ extern int ext4_writepage_trans_blocks(struct inode *);
+ extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks);
+ extern int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
+Index: linux-stage/fs/ext4/inode.c
+===================================================================
+--- linux-stage.orig/fs/ext4/inode.c
++++ linux-stage/fs/ext4/inode.c
+@@ -134,8 +134,6 @@ static void ext4_invalidatepage(struct p
+                               unsigned int length);
+ static int __ext4_journalled_writepage(struct page *page, unsigned int len);
+ static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
+-static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
+-                                int pextents);
+ /*
+  * Test whether an inode is a fast symlink.
+@@ -184,6 +182,8 @@ void ext4_evict_inode(struct inode *inod
+ {
+       handle_t *handle;
+       int err;
++      int extra_credits = 3;
++      struct ext4_xattr_ino_array *lea_ino_array = NULL;
+       trace_ext4_evict_inode(inode);
+@@ -236,8 +236,8 @@ void ext4_evict_inode(struct inode *inod
+        * protection against it
+        */
+       sb_start_intwrite(inode->i_sb);
+-      handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
+-                                  ext4_blocks_for_truncate(inode)+3);
++
++      handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, extra_credits);
+       if (IS_ERR(handle)) {
+               ext4_std_error(inode->i_sb, PTR_ERR(handle));
+               /*
+@@ -249,9 +249,36 @@ void ext4_evict_inode(struct inode *inod
+               sb_end_intwrite(inode->i_sb);
+               goto no_delete;
+       }
+-
+       if (IS_SYNC(inode))
+               ext4_handle_sync(handle);
++
++      /*
++       * Delete xattr inode before deleting the main inode.
++       */
++      err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
++      if (err) {
++              ext4_warning(inode->i_sb,
++                           "couldn't delete inode's xattr (err %d)", err);
++              goto stop_handle;
++      }
++
++      if (!IS_NOQUOTA(inode))
++              extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
++
++      if (!ext4_handle_has_enough_credits(handle,
++                      ext4_blocks_for_truncate(inode) + extra_credits)) {
++              err = ext4_journal_extend(handle,
++                      ext4_blocks_for_truncate(inode) + extra_credits);
++              if (err > 0)
++                      err = ext4_journal_restart(handle,
++                      ext4_blocks_for_truncate(inode) + extra_credits);
++              if (err != 0) {
++                      ext4_warning(inode->i_sb,
++                                   "couldn't extend journal (err %d)", err);
++                      goto stop_handle;
++              }
++      }
++
+       inode->i_size = 0;
+       err = ext4_mark_inode_dirty(handle, inode);
+       if (err) {
+@@ -269,10 +296,10 @@ void ext4_evict_inode(struct inode *inod
+        * enough credits left in the handle to remove the inode from
+        * the orphan list and set the dtime field.
+        */
+-      if (!ext4_handle_has_enough_credits(handle, 3)) {
+-              err = ext4_journal_extend(handle, 3);
++      if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
++              err = ext4_journal_extend(handle, extra_credits);
+               if (err > 0)
+-                      err = ext4_journal_restart(handle, 3);
++                      err = ext4_journal_restart(handle, extra_credits);
+               if (err != 0) {
+                       ext4_warning(inode->i_sb,
+                                    "couldn't extend journal (err %d)", err);
+@@ -306,8 +333,12 @@ void ext4_evict_inode(struct inode *inod
+               ext4_clear_inode(inode);
+       else
+               ext4_free_inode(handle, inode);
++
+       ext4_journal_stop(handle);
+       sb_end_intwrite(inode->i_sb);
++
++      if (lea_ino_array != NULL)
++              ext4_xattr_inode_array_free(inode, lea_ino_array);
+       return;
+ no_delete:
+       ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
+@@ -4681,7 +4712,7 @@ static int ext4_index_trans_blocks(struc
+  *
+  * Also account for superblock, inode, quota and xattr blocks
+  */
+-static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
++int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
+                                 int pextents)
+ {
+       ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
+Index: linux-stage/fs/ext4/xattr.c
+===================================================================
+--- linux-stage.orig/fs/ext4/xattr.c
++++ linux-stage/fs/ext4/xattr.c
+@@ -201,6 +201,7 @@ ext4_xattr_check_names(struct ext4_xattr
+       while (!IS_LAST_ENTRY(entry)) {
+               if (entry->e_value_size != 0 &&
++                  entry->e_value_inum == 0 &&
+                   (value_start + le16_to_cpu(entry->e_value_offs) <
+                    (void *)e + sizeof(__u32) ||
+                    value_start + le16_to_cpu(entry->e_value_offs) +
+@@ -233,19 +234,26 @@ ext4_xattr_check_block(struct inode *ino
+ }
+ static inline int
+-ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
++ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
++                     struct inode *inode)
+ {
+       size_t value_size = le32_to_cpu(entry->e_value_size);
+-      if (entry->e_value_block != 0 || value_size > size ||
++      if (!entry->e_value_inum &&
+           le16_to_cpu(entry->e_value_offs) + value_size > size)
+               return -EFSCORRUPTED;
++      if (entry->e_value_inum &&
++          (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
++           le32_to_cpu(entry->e_value_inum) >
++           le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
++              return -EFSCORRUPTED;
+       return 0;
+ }
+ static int
+ ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
+-                    const char *name, size_t size, int sorted)
++                    const char *name, size_t size, int sorted,
++                    struct inode *inode)
+ {
+       struct ext4_xattr_entry *entry;
+       size_t name_len;
+@@ -265,11 +273,104 @@ ext4_xattr_find_entry(struct ext4_xattr_
+                       break;
+       }
+       *pentry = entry;
+-      if (!cmp && ext4_xattr_check_entry(entry, size))
++      if (!cmp && ext4_xattr_check_entry(entry, size, inode))
+               return -EFSCORRUPTED;
+       return cmp ? -ENODATA : 0;
+ }
++/*
++ * Read the EA value from an inode.
++ */
++static int
++ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
++{
++      unsigned long block = 0;
++      struct buffer_head *bh = NULL;
++      int blocksize;
++      size_t csize, ret_size = 0;
++
++      if (*size == 0)
++              return 0;
++
++      blocksize = ea_inode->i_sb->s_blocksize;
++
++      while (ret_size < *size) {
++              csize = (*size - ret_size) > blocksize ? blocksize :
++                                                      *size - ret_size;
++              bh = ext4_bread(NULL, ea_inode, block, 0);
++              if (IS_ERR(bh)) {
++                      *size = ret_size;
++                      return PTR_ERR(bh);
++              }
++              memcpy(buf, bh->b_data, csize);
++              brelse(bh);
++
++              buf += csize;
++              block += 1;
++              ret_size += csize;
++      }
++
++      *size = ret_size;
++
++      return 0;
++}
++
++struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, int *err)
++{
++      struct inode *ea_inode = NULL;
++
++      ea_inode = ext4_iget(parent->i_sb, ea_ino);
++      if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
++              int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
++              ext4_error(parent->i_sb, "error while reading EA inode %lu "
++                         "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
++              *err = rc != 0 ? rc : -EIO;
++              return NULL;
++      }
++
++      if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
++          ea_inode->i_generation != parent->i_generation) {
++              ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
++                         "to parent invalid.", ea_ino);
++              *err = -EINVAL;
++              goto error;
++      }
++
++      if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
++              ext4_error(parent->i_sb, "EA inode %lu does not have "
++                         "EXT4_EA_INODE_FL flag set.\n", ea_ino);
++              *err = -EINVAL;
++              goto error;
++      }
++
++      *err = 0;
++      return ea_inode;
++
++error:
++      iput(ea_inode);
++      return NULL;
++}
++
++/*
++ * Read the value from the EA inode.
++ */
++static int
++ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino, void *buffer,
++                   size_t *size)
++{
++      struct inode *ea_inode = NULL;
++      int err;
++
++      ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
++      if (err)
++              return err;
++
++      err = ext4_xattr_inode_read(ea_inode, buffer, size);
++      iput(ea_inode);
++
++      return err;
++}
++
+ static int
+ ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
+                    void *buffer, size_t buffer_size)
+@@ -301,7 +401,8 @@ bad_block:
+       }
+       ext4_xattr_cache_insert(bh);
+       entry = BFIRST(bh);
+-      error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
++      error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
++                                    inode);
+       if (error == -EFSCORRUPTED)
+               goto bad_block;
+       if (error)
+@@ -311,8 +412,16 @@ bad_block:
+               error = -ERANGE;
+               if (size > buffer_size)
+                       goto cleanup;
+-              memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
+-                     size);
++              if (entry->e_value_inum) {
++                      error = ext4_xattr_inode_get(inode,
++                                           le32_to_cpu(entry->e_value_inum),
++                                           buffer, &size);
++                      if (error)
++                              goto cleanup;
++              } else {
++                      memcpy(buffer, bh->b_data +
++                             le16_to_cpu(entry->e_value_offs), size);
++              }
+       }
+       error = size;
+@@ -346,7 +455,7 @@ ext4_xattr_ibody_get(struct inode *inode
+       if (error)
+               goto cleanup;
+       error = ext4_xattr_find_entry(&entry, name_index, name,
+-                                    end - (void *)entry, 0);
++                                    end - (void *)entry, 0, inode);
+       if (error)
+               goto cleanup;
+       size = le32_to_cpu(entry->e_value_size);
+@@ -354,8 +463,16 @@ ext4_xattr_ibody_get(struct inode *inode
+               error = -ERANGE;
+               if (size > buffer_size)
+                       goto cleanup;
+-              memcpy(buffer, (void *)IFIRST(header) +
+-                     le16_to_cpu(entry->e_value_offs), size);
++              if (entry->e_value_inum) {
++                      error = ext4_xattr_inode_get(inode,
++                                           le32_to_cpu(entry->e_value_inum),
++                                           buffer, &size);
++                      if (error)
++                              goto cleanup;
++              } else {
++                      memcpy(buffer, (void *)IFIRST(header) +
++                             le16_to_cpu(entry->e_value_offs), size);
++              }
+       }
+       error = size;
+@@ -600,7 +717,7 @@ static size_t ext4_xattr_free_space(stru
+                                   size_t *min_offs, void *base, int *total)
+ {
+       for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
+-              if (!last->e_value_block && last->e_value_size) {
++              if (!last->e_value_inum && last->e_value_size) {
+                       size_t offs = le16_to_cpu(last->e_value_offs);
+                       if (offs < *min_offs)
+                               *min_offs = offs;
+@@ -611,16 +728,176 @@ static size_t ext4_xattr_free_space(stru
+       return (*min_offs - ((void *)last - base) - sizeof(__u32));
+ }
++/*
++ * Write the value of the EA in an inode.
++ */
+ static int
+-ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
++ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
++                     const void *buf, int bufsize)
++{
++      struct buffer_head *bh = NULL;
++      unsigned long block = 0;
++      unsigned blocksize = ea_inode->i_sb->s_blocksize;
++      unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
++      int csize, wsize = 0;
++      int ret = 0;
++      int retries = 0;
++
++retry:
++      while (ret >= 0 && ret < max_blocks) {
++              struct ext4_map_blocks map;
++              map.m_lblk = block += ret;
++              map.m_len = max_blocks -= ret;
++
++              ret = ext4_map_blocks(handle, ea_inode, &map,
++                                    EXT4_GET_BLOCKS_CREATE);
++              if (ret <= 0) {
++                      ext4_mark_inode_dirty(handle, ea_inode);
++                      if (ret == -ENOSPC &&
++                          ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
++                              ret = 0;
++                              goto retry;
++                      }
++                      break;
++              }
++      }
++
++      if (ret < 0)
++              return ret;
++
++      block = 0;
++      while (wsize < bufsize) {
++              if (bh != NULL)
++                      brelse(bh);
++              csize = (bufsize - wsize) > blocksize ? blocksize :
++                                                              bufsize - wsize;
++              bh = ext4_getblk(handle, ea_inode, block, 0);
++              if (IS_ERR(bh)) {
++                      ret = PTR_ERR(bh);
++                      goto out;
++              }
++              ret = ext4_journal_get_write_access(handle, bh);
++              if (ret)
++                      goto out;
++
++              memcpy(bh->b_data, buf, csize);
++              set_buffer_uptodate(bh);
++              ext4_handle_dirty_metadata(handle, ea_inode, bh);
++
++              buf += csize;
++              wsize += csize;
++              block += 1;
++      }
++
++      mutex_lock(&ea_inode->i_mutex);
++      i_size_write(ea_inode, wsize);
++      ext4_update_i_disksize(ea_inode, wsize);
++      mutex_unlock(&ea_inode->i_mutex);
++
++      ext4_mark_inode_dirty(handle, ea_inode);
++
++out:
++      brelse(bh);
++
++      return ret;
++}
++
++/*
++ * Create an inode to store the value of a large EA.
++ */
++static struct inode *
++ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
++{
++      struct inode *ea_inode = NULL;
++
++      /*
++       * Let the next inode be the goal, so we try and allocate the EA inode
++       * in the same group, or nearby one.
++       */
++      ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
++                                S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
++
++      if (!IS_ERR(ea_inode)) {
++              ea_inode->i_op = &ext4_file_inode_operations;
++              ea_inode->i_fop = &ext4_file_operations;
++              ext4_set_aops(ea_inode);
++              ea_inode->i_generation = inode->i_generation;
++              EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
++
++              /*
++               * A back-pointer from EA inode to parent inode will be useful
++               * for e2fsck.
++               */
++              EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
++              unlock_new_inode(ea_inode);
++      }
++
++      return ea_inode;
++}
++
++/*
++ * Unlink the inode storing the value of the EA.
++ */
++int
++ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
++{
++      struct inode *ea_inode = NULL;
++      int err;
++
++      ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
++      if (err)
++              return err;
++
++      clear_nlink(ea_inode);
++      iput(ea_inode);
++
++      return 0;
++}
++
++/*
++ * Add value of the EA in an inode.
++ */
++static int
++ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
++                   const void *value, size_t value_len)
++{
++      struct inode *ea_inode = NULL;
++      int err;
++
++      /* Create an inode for the EA value */
++      ea_inode = ext4_xattr_inode_create(handle, inode);
++      if (IS_ERR(ea_inode))
++              return -1;
++
++      err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
++      if (err)
++              clear_nlink(ea_inode);
++      else
++              *ea_ino = ea_inode->i_ino;
++
++      iput(ea_inode);
++
++      return err;
++}
++
++static int
++ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
++                   handle_t *handle, struct inode *inode)
+ {
+       struct ext4_xattr_entry *last;
+       size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
++      int in_inode = i->in_inode;
++
++      if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
++               EXT4_FEATURE_INCOMPAT_EA_INODE) &&
++          (EXT4_XATTR_SIZE(i->value_len) >
++           EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
++              in_inode = 1;
+       /* Compute min_offs and last. */
+       last = s->first;
+       for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
+-              if (!last->e_value_block && last->e_value_size) {
++              if (!last->e_value_inum && last->e_value_size) {
+                       size_t offs = le16_to_cpu(last->e_value_offs);
+                       if (offs < min_offs)
+                               min_offs = offs;
+@@ -628,15 +903,20 @@ ext4_xattr_set_entry(struct ext4_xattr_i
+       }
+       free = min_offs - ((void *)last - s->base) - sizeof(__u32);
+       if (!s->not_found) {
+-              if (!s->here->e_value_block && s->here->e_value_size) {
++              if (!in_inode &&
++                  !s->here->e_value_inum && s->here->e_value_size) {
+                       size_t size = le32_to_cpu(s->here->e_value_size);
+                       free += EXT4_XATTR_SIZE(size);
+               }
+               free += EXT4_XATTR_LEN(name_len);
+       }
+       if (i->value) {
+-              if (free < EXT4_XATTR_LEN(name_len) +
+-                         EXT4_XATTR_SIZE(i->value_len))
++              size_t value_len = EXT4_XATTR_SIZE(i->value_len);
++
++              if (in_inode)
++                      value_len = 0;
++
++              if (free < EXT4_XATTR_LEN(name_len) + value_len)
+                       return -ENOSPC;
+       }
+@@ -651,7 +931,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
+               s->here->e_name_len = name_len;
+               memcpy(s->here->e_name, i->name, name_len);
+       } else {
+-              if (!s->here->e_value_block && s->here->e_value_size) {
++              if (!s->here->e_value_inum && s->here->e_value_size &&
++                  s->here->e_value_offs > 0) {
+                       void *first_val = s->base + min_offs;
+                       size_t offs = le16_to_cpu(s->here->e_value_offs);
+                       void *val = s->base + offs;
+@@ -685,13 +966,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
+                       last = s->first;
+                       while (!IS_LAST_ENTRY(last)) {
+                               size_t o = le16_to_cpu(last->e_value_offs);
+-                              if (!last->e_value_block &&
++                              if (!last->e_value_inum &&
+                                   last->e_value_size && o < offs)
+                                       last->e_value_offs =
+                                               cpu_to_le16(o + size);
+                               last = EXT4_XATTR_NEXT(last);
+                       }
+               }
++              if (s->here->e_value_inum) {
++                      ext4_xattr_inode_unlink(inode,
++                                      le32_to_cpu(s->here->e_value_inum));
++                      s->here->e_value_inum = 0;
++              }
+               if (!i->value) {
+                       /* Remove the old name. */
+                       size_t size = EXT4_XATTR_LEN(name_len);
+@@ -705,10 +990,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
+       if (i->value) {
+               /* Insert the new value. */
+               s->here->e_value_size = cpu_to_le32(i->value_len);
+-              if (i->value_len) {
++              if (in_inode) {
++                      unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
++                      ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
++                                           i->value_len);
++                      s->here->e_value_inum = cpu_to_le32(ea_ino);
++                      s->here->e_value_offs = 0;
++              } else if (i->value_len) {
+                       size_t size = EXT4_XATTR_SIZE(i->value_len);
+                       void *val = s->base + min_offs - size;
+                       s->here->e_value_offs = cpu_to_le16(min_offs - size);
++                      s->here->e_value_inum = 0;
+                       if (i->value == EXT4_ZERO_XATTR_VALUE) {
+                               memset(val, 0, size);
+                       } else {
+@@ -758,7 +1050,7 @@ ext4_xattr_block_find(struct inode *inod
+               bs->s.end = bs->bh->b_data + bs->bh->b_size;
+               bs->s.here = bs->s.first;
+               error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
+-                                            i->name, bs->bh->b_size, 1);
++                                           i->name, bs->bh->b_size, 1, inode);
+               if (error && error != -ENODATA)
+                       goto cleanup;
+               bs->s.not_found = error;
+@@ -782,8 +1074,6 @@ ext4_xattr_block_set(handle_t *handle, s
+ #define header(x) ((struct ext4_xattr_header *)(x))
+-      if (i->value && i->value_len > sb->s_blocksize)
+-              return -ENOSPC;
+       if (s->base) {
+               ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
+                                       bs->bh->b_blocknr);
+@@ -799,7 +1089,7 @@ ext4_xattr_block_set(handle_t *handle, s
+                               ce = NULL;
+                       }
+                       ea_bdebug(bs->bh, "modifying in-place");
+-                      error = ext4_xattr_set_entry(i, s);
++                      error = ext4_xattr_set_entry(i, s, handle, inode);
+                       if (!error) {
+                               if (!IS_LAST_ENTRY(s->first))
+                                       ext4_xattr_rehash(header(s->base),
+@@ -850,7 +1140,7 @@ ext4_xattr_block_set(handle_t *handle, s
+               s->end = s->base + sb->s_blocksize;
+       }
+-      error = ext4_xattr_set_entry(i, s);
++      error = ext4_xattr_set_entry(i, s, handle, inode);
+       if (error == -EFSCORRUPTED)
+               goto bad_block;
+       if (error)
+@@ -1000,7 +1290,7 @@ int ext4_xattr_ibody_find(struct inode *
+               /* Find the named attribute. */
+               error = ext4_xattr_find_entry(&is->s.here, i->name_index,
+                                             i->name, is->s.end -
+-                                            (void *)is->s.base, 0);
++                                            (void *)is->s.base, 0, inode);
+               if (error && error != -ENODATA)
+                       return error;
+               is->s.not_found = error;
+@@ -1018,7 +1308,7 @@ int ext4_xattr_ibody_inline_set(handle_t
+       if (EXT4_I(inode)->i_extra_isize == 0)
+               return -ENOSPC;
+-      error = ext4_xattr_set_entry(i, s);
++      error = ext4_xattr_set_entry(i, s, handle, inode);
+       if (error) {
+               if (error == -ENOSPC &&
+                   ext4_has_inline_data(inode)) {
+@@ -1030,7 +1320,7 @@ int ext4_xattr_ibody_inline_set(handle_t
+                       error = ext4_xattr_ibody_find(inode, i, is);
+                       if (error)
+                               return error;
+-                      error = ext4_xattr_set_entry(i, s);
++                      error = ext4_xattr_set_entry(i, s, handle, inode);
+               }
+               if (error)
+                       return error;
+@@ -1056,7 +1346,7 @@ static int ext4_xattr_ibody_set(handle_t
+       if (EXT4_I(inode)->i_extra_isize == 0)
+               return -ENOSPC;
+-      error = ext4_xattr_set_entry(i, s);
++      error = ext4_xattr_set_entry(i, s, handle, inode);
+       if (error)
+               return error;
+       header = IHDR(inode, ext4_raw_inode(&is->iloc));
+@@ -1092,7 +1382,7 @@ ext4_xattr_set_handle(handle_t *handle, 
+               .name = name,
+               .value = value,
+               .value_len = value_len,
+-
++              .in_inode = 0,
+       };
+       struct ext4_xattr_ibody_find is = {
+               .s = { .not_found = -ENODATA, },
+@@ -1157,6 +1447,15 @@ ext4_xattr_set_handle(handle_t *handle, 
+                                       goto cleanup;
+                       }
+                       error = ext4_xattr_block_set(handle, inode, &i, &bs);
++                      if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
++                                      EXT4_FEATURE_INCOMPAT_EA_INODE) &&
++                          error == -ENOSPC) {
++                              /* xattr not fit to block, store at external
++                               * inode */
++                              i.in_inode = 1;
++                              error = ext4_xattr_ibody_set(handle, inode,
++                                                           &i, &is);
++                      }
+                       if (error)
+                               goto cleanup;
+                       if (!is.s.not_found) {
+@@ -1203,9 +1502,22 @@ ext4_xattr_set(struct inode *inode, int 
+              const void *value, size_t value_len, int flags)
+ {
+       handle_t *handle;
++      struct super_block *sb = inode->i_sb;
+       int error, retries = 0;
+       int credits = ext4_jbd2_credits_xattr(inode);
++      if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
++          EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
++              int nrblocks = (value_len + sb->s_blocksize - 1) >>
++                                      sb->s_blocksize_bits;
++
++              /* For new inode */
++              credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
++
++              /* For data blocks of EA inode */
++              credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
++      }
++
+ retry:
+       handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
+       if (IS_ERR(handle)) {
+@@ -1217,7 +1529,7 @@ retry:
+                                             value, value_len, flags);
+               error2 = ext4_journal_stop(handle);
+               if (error == -ENOSPC &&
+-                  ext4_should_retry_alloc(inode->i_sb, &retries))
++                  ext4_should_retry_alloc(sb, &retries))
+                       goto retry;
+               if (error == 0)
+                       error = error2;
+@@ -1239,7 +1551,7 @@ static void ext4_xattr_shift_entries(str
+       /* Adjust the value offsets of the entries */
+       for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
+-              if (!last->e_value_block && last->e_value_size) {
++              if (!last->e_value_inum && last->e_value_size) {
+                       new_offs = le16_to_cpu(last->e_value_offs) +
+                                                       value_offs_shift;
+                       BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
+@@ -1477,21 +1789,135 @@ cleanup:
+ }
++#define EIA_INCR 16 /* must be 2^n */
++#define EIA_MASK (EIA_INCR - 1)
++/* Add the large xattr @ino into @lea_ino_array for later deletion.
++ * If @lea_ino_array is new or full it will be grown and the old
++ * contents copied over.
++ */
++static int
++ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
++{
++      if (*lea_ino_array == NULL) {
++              /*
++               * Start with 15 inodes, so it fits into a power-of-two size.
++               * If *lea_ino_array is NULL, this is essentially offsetof()
++               */
++              (*lea_ino_array) =
++                      kmalloc(offsetof(struct ext4_xattr_ino_array,
++                                       xia_inodes[EIA_MASK]),
++                              GFP_NOFS);
++              if (*lea_ino_array == NULL)
++                      return -ENOMEM;
++              (*lea_ino_array)->xia_count = 0;
++      } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
++              /* expand the array once all 15 + n * 16 slots are full */
++              struct ext4_xattr_ino_array *new_array = NULL;
++              int count = (*lea_ino_array)->xia_count;
++
++              /* if new_array is NULL, this is essentially offsetof() */
++              new_array = kmalloc(
++                              offsetof(struct ext4_xattr_ino_array,
++                                       xia_inodes[count + EIA_INCR]),
++                              GFP_NOFS);
++              if (new_array == NULL)
++                      return -ENOMEM;
++              memcpy(new_array, *lea_ino_array,
++                     offsetof(struct ext4_xattr_ino_array,
++                              xia_inodes[count]));
++              kfree(*lea_ino_array);
++              *lea_ino_array = new_array;
++      }
++      (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
++      return 0;
++}
++
++/**
++ * Add xattr inode to orphan list
++ */
++static int
++ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
++                      int credits, struct ext4_xattr_ino_array *lea_ino_array)
++{
++      struct inode *ea_inode = NULL;
++      int idx = 0, error = 0;
++
++      if (lea_ino_array == NULL)
++              return 0;
++
++      for (; idx < lea_ino_array->xia_count; ++idx) {
++              if (!ext4_handle_has_enough_credits(handle, credits)) {
++                      error = ext4_journal_extend(handle, credits);
++                      if (error > 0)
++                              error = ext4_journal_restart(handle, credits);
++
++                      if (error != 0) {
++                              ext4_warning(inode->i_sb,
++                                      "couldn't extend journal "
++                                      "(err %d)", error);
++                              return error;
++                      }
++              }
++              ea_inode = ext4_xattr_inode_iget(inode,
++                              lea_ino_array->xia_inodes[idx], &error);
++              if (error)
++                      continue;
++              ext4_orphan_add(handle, ea_inode);
++              /* the inode's i_count will be released by caller */
++      }
++
++      return 0;
++}
+ /*
+  * ext4_xattr_delete_inode()
+  *
+- * Free extended attribute resources associated with this inode. This
++ * Free extended attribute resources associated with this inode. Traverse
++ * all entries and unlink any xattr inodes associated with this inode. This
+  * is called immediately before an inode is freed. We have exclusive
+- * access to the inode.
++ * access to the inode. If an orphan inode is deleted it will also delete any
++ * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
++ * to ensure they belong to the parent inode and were not deleted already.
+  */
+-void
+-ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
++int
++ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
++                      struct ext4_xattr_ino_array **lea_ino_array)
+ {
+       struct buffer_head *bh = NULL;
++      struct ext4_xattr_ibody_header *header;
++      struct ext4_inode *raw_inode;
++      struct ext4_iloc iloc;
++      struct ext4_xattr_entry *entry;
++      int credits = 3, error = 0;
+-      if (!EXT4_I(inode)->i_file_acl)
++      if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
++              goto delete_external_ea;
++
++      error = ext4_get_inode_loc(inode, &iloc);
++      if (error)
+               goto cleanup;
++      raw_inode = ext4_raw_inode(&iloc);
++      header = IHDR(inode, raw_inode);
++      for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
++           entry = EXT4_XATTR_NEXT(entry)) {
++              if (!entry->e_value_inum)
++                      continue;
++              if (ext4_expand_ino_array(lea_ino_array,
++                                        entry->e_value_inum) != 0) {
++                      brelse(iloc.bh);
++                      goto cleanup;
++              }
++              entry->e_value_inum = 0;
++      }
++      brelse(iloc.bh);
++
++delete_external_ea:
++      if (!EXT4_I(inode)->i_file_acl) {
++              /* add xattr inode to orphan list */
++              ext4_xattr_inode_orphan_add(handle, inode, credits,
++                                              *lea_ino_array);
++              goto cleanup;
++      }
+       bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
+       if (!bh) {
+               EXT4_ERROR_INODE(inode, "block %llu read error",
+@@ -1504,11 +1930,69 @@ ext4_xattr_delete_inode(handle_t *handle
+                                EXT4_I(inode)->i_file_acl);
+               goto cleanup;
+       }
++
++      for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
++           entry = EXT4_XATTR_NEXT(entry)) {
++              if (!entry->e_value_inum)
++                      continue;
++              if (ext4_expand_ino_array(lea_ino_array,
++                                        entry->e_value_inum) != 0)
++                      goto cleanup;
++              entry->e_value_inum = 0;
++      }
++
++      /* add xattr inode to orphan list */
++      error = ext4_xattr_inode_orphan_add(handle, inode, credits,
++                                      *lea_ino_array);
++      if (error != 0)
++              goto cleanup;
++
++      if (!IS_NOQUOTA(inode))
++              credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
++
++      if (!ext4_handle_has_enough_credits(handle, credits)) {
++              error = ext4_journal_extend(handle, credits);
++              if (error > 0)
++                      error = ext4_journal_restart(handle, credits);
++              if (error != 0) {
++                      ext4_warning(inode->i_sb,
++                              "couldn't extend journal (err %d)", error);
++                      goto cleanup;
++              }
++      }
++
+       ext4_xattr_release_block(handle, inode, bh);
+       EXT4_I(inode)->i_file_acl = 0;
+ cleanup:
+       brelse(bh);
++
++      return error;
++}
++
++void
++ext4_xattr_inode_array_free(struct inode *inode,
++                          struct ext4_xattr_ino_array *lea_ino_array)
++{
++      struct inode    *ea_inode = NULL;
++      int             idx = 0;
++      int             err;
++
++      if (lea_ino_array == NULL)
++              return;
++
++      for (; idx < lea_ino_array->xia_count; ++idx) {
++              ea_inode = ext4_xattr_inode_iget(inode,
++                              lea_ino_array->xia_inodes[idx], &err);
++              if (err)
++                      continue;
++              /* for inode's i_count get from ext4_xattr_delete_inode */
++              if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
++                      iput(ea_inode);
++              clear_nlink(ea_inode);
++              iput(ea_inode);
++      }
++      kfree(lea_ino_array);
+ }
+ /*
+@@ -1578,10 +2062,9 @@ ext4_xattr_cmp(struct ext4_xattr_header 
+                   entry1->e_name_index != entry2->e_name_index ||
+                   entry1->e_name_len != entry2->e_name_len ||
+                   entry1->e_value_size != entry2->e_value_size ||
++                  entry1->e_value_inum != entry2->e_value_inum ||
+                   memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
+                       return 1;
+-              if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
+-                      return -EFSCORRUPTED;
+               if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
+                          (char *)header2 + le16_to_cpu(entry2->e_value_offs),
+                          le32_to_cpu(entry1->e_value_size)))
+@@ -1665,7 +2148,7 @@ static inline void ext4_xattr_hash_entry
+                      *name++;
+       }
+-      if (entry->e_value_block == 0 && entry->e_value_size != 0) {
++      if (!entry->e_value_inum && entry->e_value_size) {
+               __le32 *value = (__le32 *)((char *)header +
+                       le16_to_cpu(entry->e_value_offs));
+               for (n = (le32_to_cpu(entry->e_value_size) +
+Index: linux-stage/fs/ext4/xattr.h
+===================================================================
+--- linux-stage.orig/fs/ext4/xattr.h
++++ linux-stage/fs/ext4/xattr.h
+@@ -42,7 +42,7 @@ struct ext4_xattr_entry {
+       __u8    e_name_len;     /* length of name */
+       __u8    e_name_index;   /* attribute name index */
+       __le16  e_value_offs;   /* offset in disk block of value */
+-      __le32  e_value_block;  /* disk block attribute is stored on (n/i) */
++      __le32  e_value_inum;   /* inode in which the value is stored */
+       __le32  e_value_size;   /* size of attribute value */
+       __le32  e_hash;         /* hash value of name and value */
+       char    e_name[0];      /* attribute name */
+@@ -67,6 +67,26 @@ struct ext4_xattr_entry {
+               EXT4_I(inode)->i_extra_isize))
+ #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
++/*
++ * Link EA inode back to parent one using i_mtime field.
++ * Extra integer type conversion added to ignore higher
++ * bits in i_mtime.tv_sec which might be set by ext4_get()
++ */
++#define EXT4_XATTR_INODE_SET_PARENT(inode, inum)      \
++do {                                                  \
++      (inode)->i_mtime.tv_sec = inum;                 \
++} while(0)
++
++#define EXT4_XATTR_INODE_GET_PARENT(inode)            \
++((__u32)(inode)->i_mtime.tv_sec)
++
++/*
++ * The minimum size of EA value when you start storing it in an external inode
++ * size of block - size of header - size of 1 entry - 4 null bytes
++*/
++#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b)                                       \
++      ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
++
+ #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
+ #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
+ #define BFIRST(bh) ENTRY(BHDR(bh)+1)
+@@ -75,10 +84,11 @@ struct ext4_xattr_entry {
+ #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
+ struct ext4_xattr_info {
+-      int name_index;
+       const char *name;
+       const void *value;
+       size_t value_len;
++      int name_index;
++      int in_inode;
+ };
+ struct ext4_xattr_search {
+@@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
+ extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
+ extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
+-extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
++extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
++                                         int *err);
++extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
++extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
++                                 struct ext4_xattr_ino_array **array);
++extern void ext4_xattr_inode_array_free(struct inode *inode,
++                                      struct ext4_xattr_ino_array *array);
+ extern void ext4_xattr_put_super(struct super_block *);
+ extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
+Index: linux-stage/fs/ext4/ialloc.c
+===================================================================
+--- linux-stage.orig/fs/ext4/ialloc.c
++++ linux-stage/fs/ext4/ialloc.c
+@@ -269,7 +269,6 @@ void ext4_free_inode(handle_t *handle, s
+        * as writing the quota to disk may need the lock as well.
+        */
+       dquot_initialize(inode);
+-      ext4_xattr_delete_inode(handle, inode);
+       dquot_free_inode(inode);
+       dquot_drop(inode);
+Index: linux-stage/fs/ext4/inline.c
+===================================================================
+--- linux-stage.orig/fs/ext4/inline.c
++++ linux-stage/fs/ext4/inline.c
+@@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
+       /* Compute min_offs. */
+       for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
+-              if (!entry->e_value_block && entry->e_value_size) {
++              if (!entry->e_value_inum && entry->e_value_size) {
+                       size_t offs = le16_to_cpu(entry->e_value_offs);
+                       if (offs < min_offs)
+                               min_offs = offs;
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-lookup-dotdot.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-lookup-dotdot.patch
new file mode 100644 (file)
index 0000000..c6f0371
--- /dev/null
@@ -0,0 +1,38 @@
+Index: linux-3.10.9-200.fc17.x86_64/fs/ext4/namei.c
+===================================================================
+--- linux-3.10.9-200.fc17.x86_64.orig/fs/ext4/namei.c
++++ linux-3.10.9-200.fc17.x86_64/fs/ext4/namei.c
+@@ -1438,6 +1438,33 @@ static struct dentry *ext4_lookup(struct
+                       return ERR_PTR(-EIO);
+               }
+       }
++      /* ".." shouldn't go into dcache to preserve dcache hierarchy
++       * otherwise we'll get parent being a child of actual child.
++       * see bug 10458 for details -bzzz */
++      if (inode && (dentry->d_name.name[0] == '.' &&
++                    (dentry->d_name.len == 1 || (dentry->d_name.len == 2 &&
++                                           dentry->d_name.name[1] == '.')))) {
++              struct dentry *goal = NULL;
++
++              /* first, look for an existing dentry - any one is good */
++              goal = d_find_any_alias(inode);
++              if (goal == NULL) {
++                      spin_lock(&dentry->d_lock);
++                      /* there is no alias, we need to make current dentry:
++                       *  a) inaccessible for __d_lookup()
++                       *  b) inaccessible for iopen */
++                      J_ASSERT(hlist_unhashed(&dentry->d_u.d_alias));
++                      dentry->d_flags |= DCACHE_NFSFS_RENAMED;
++                      /* this is d_instantiate() ... */
++                      hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
++                      dentry->d_inode = inode;
++                      spin_unlock(&dentry->d_lock);
++              }
++              if (goal)
++                      iput(inode);
++              return goal;
++      }
++
+       return d_splice_alias(inode, dentry);
+ }
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-max-dir-size.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-max-dir-size.patch
new file mode 100644 (file)
index 0000000..38ff250
--- /dev/null
@@ -0,0 +1,44 @@
+Add a proc interface for max_dir_size.
+
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/sysfs.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/sysfs.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/sysfs.c
+@@ -2557,6 +2562,8 @@ EXT4_RW_ATTR(reserved_clusters);
+ EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
+                inode_readahead_blks_store, s_inode_readahead_blks);
+ EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
++EXT4_RW_ATTR_SBI_UI(max_dir_size, s_max_dir_size_kb);
++EXT4_RW_ATTR_SBI_UI(max_dir_size_kb, s_max_dir_size_kb);
+ EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
+ EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
+ EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
+@@ -2581,6 +2588,8 @@ static struct attribute *ext4_attrs[] =
+       ATTR_LIST(reserved_clusters),
+       ATTR_LIST(inode_readahead_blks),
+       ATTR_LIST(inode_goal),
++      ATTR_LIST(max_dir_size),
++      ATTR_LIST(max_dir_size_kb),
+       ATTR_LIST(mb_stats),
+       ATTR_LIST(mb_max_to_scan),
+       ATTR_LIST(mb_min_to_scan),
+@@ -2676,7 +2476,9 @@ static ssize_t sbi_ui_show(struct ext4_a
+               if (!ptr)
+                       return 0;
+               return snprintf(buf, PAGE_SIZE, "%u\n",
+-                              *((unsigned int *) ptr));
++                              strcmp("max_dir_size", a->attr.name) ?
++                              *((unsigned int *) ptr) :
++                              (*((unsigned int *) ptr)) << 10);
+       case attr_pointer_atomic:
+               if (!ptr)
+                       return 0;
+@@ -2691,6 +2494,8 @@ static ssize_t sbi_ui_store(struct ext4_
+               ret = kstrtoul(skip_spaces(buf), 0, &t);
+               if (ret)
+                       return ret;
++              if (strcmp("max_dir_size", a->attr.name) == 0)
++                      t >>= 10;
+               *((unsigned int *) ptr) = t;
+               return len;
+       case attr_inode_readahead:
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-mballoc-extra-checks.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-mballoc-extra-checks.patch
new file mode 100644 (file)
index 0000000..a0dda46
--- /dev/null
@@ -0,0 +1,316 @@
+Index: linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/ext4.h
++++ linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h
+@@ -2391,6 +2391,7 @@ struct ext4_group_info {
+       ext4_grpblk_t   bb_fragments;   /* nr of freespace fragments */
+       ext4_grpblk_t   bb_largest_free_order;/* order of largest frag in BG */
+       struct          list_head bb_prealloc_list;
++      unsigned long   bb_prealloc_nr;
+ #ifdef DOUBLE_CHECK
+       void            *bb_bitmap;
+ #endif
+Index: linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.c
+===================================================================
+--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/mballoc.c
++++ linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.c
+@@ -362,7 +362,7 @@ static const char *ext4_groupinfo_slab_n
+       "ext4_groupinfo_64k", "ext4_groupinfo_128k"
+ };
+-static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
++static int ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
+                                       ext4_group_t group);
+ static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
+                                               ext4_group_t group);
+@@ -718,7 +718,7 @@ mb_set_largest_free_order(struct super_b
+ }
+ static noinline_for_stack
+-void ext4_mb_generate_buddy(struct super_block *sb,
++int ext4_mb_generate_buddy(struct super_block *sb,
+                               void *buddy, void *bitmap, ext4_group_t group)
+ {
+       struct ext4_group_info *grp = ext4_get_group_info(sb, group);
+@@ -750,19 +750,13 @@ void ext4_mb_generate_buddy(struct super
+       grp->bb_fragments = fragments;
+       if (free != grp->bb_free) {
+-              ext4_grp_locked_error(sb, group, 0, 0,
+-                                    "block bitmap and bg descriptor "
+-                                    "inconsistent: %u vs %u free clusters",
+-                                    free, grp->bb_free);
+-              /*
+-               * If we intend to continue, we consider group descriptor
+-               * corrupt and update bb_free using bitmap value
+-               */
+-              grp->bb_free = free;
+-              if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
+-                      percpu_counter_sub(&sbi->s_freeclusters_counter,
+-                                         grp->bb_free);
+-              set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
++              struct ext4_group_desc *gdp;
++              gdp = ext4_get_group_desc(sb, group, NULL);
++              ext4_error(sb, "group %lu: %u blocks in bitmap, %u in bb, "
++                      "%u in gd, %lu pa's\n", (long unsigned int)group,
++                      free, grp->bb_free, ext4_free_group_clusters(sb, gdp),
++                      grp->bb_prealloc_nr);
++              return -EIO;
+       }
+       mb_set_largest_free_order(sb, grp);
+@@ -768,6 +767,8 @@ void ext4_mb_generate_buddy(struct super
+       EXT4_SB(sb)->s_mb_buddies_generated++;
+       EXT4_SB(sb)->s_mb_generation_time += period;
+       spin_unlock(&EXT4_SB(sb)->s_bal_lock);
++
++      return 0;
+ }
+ static void mb_regenerate_buddy(struct ext4_buddy *e4b)
+@@ -883,7 +884,7 @@ static int ext4_mb_init_cache(struct pag
+       }
+       first_block = page->index * blocks_per_page;
+-      for (i = 0; i < blocks_per_page; i++) {
++      for (i = 0; i < blocks_per_page && err == 0; i++) {
+               group = (first_block + i) >> 1;
+               if (group >= ngroups)
+                       break;
+@@ -922,7 +923,7 @@ static int ext4_mb_init_cache(struct pag
+                       ext4_lock_group(sb, group);
+                       /* init the buddy */
+                       memset(data, 0xff, blocksize);
+-                      ext4_mb_generate_buddy(sb, data, incore, group);
++                      err = ext4_mb_generate_buddy(sb, data, incore, group);
+                       ext4_unlock_group(sb, group);
+                       incore = NULL;
+               } else {
+@@ -937,7 +938,7 @@ static int ext4_mb_init_cache(struct pag
+                       memcpy(data, bitmap, blocksize);
+                       /* mark all preallocated blks used in in-core bitmap */
+-                      ext4_mb_generate_from_pa(sb, data, group);
++                      err = ext4_mb_generate_from_pa(sb, data, group);
+                       ext4_mb_generate_from_freelist(sb, data, group);
+                       ext4_unlock_group(sb, group);
+@@ -947,7 +948,8 @@ static int ext4_mb_init_cache(struct pag
+                       incore = data;
+               }
+       }
+-      SetPageUptodate(page);
++      if (likely(err == 0))
++              SetPageUptodate(page);
+ out:
+       if (bh) {
+@@ -2224,9 +2226,11 @@ static void *ext4_mb_seq_groups_next(str
+ static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
+ {
+       struct super_block *sb = seq->private;
++      struct ext4_group_desc *gdp;
+       ext4_group_t group = (ext4_group_t) ((unsigned long) v);
+       int i;
+       int err, buddy_loaded = 0;
++      int free = 0;
+       struct ext4_buddy e4b;
+       struct ext4_group_info *grinfo;
+       struct sg {
+@@ -2236,7 +2240,7 @@ static int ext4_mb_seq_groups_show(struc
+       group--;
+       if (group == 0)
+-              seq_puts(seq, "#group: free  frags first ["
++              seq_puts(seq, "#group: bfree gfree frags first pa    ["
+                             " 2^0   2^1   2^2   2^3   2^4   2^5   2^6  "
+                             " 2^7   2^8   2^9   2^10  2^11  2^12  2^13  ]");
+@@ -2256,13 +2260,19 @@ static int ext4_mb_seq_groups_show(struc
+               buddy_loaded = 1;
+       }
++      gdp = ext4_get_group_desc(sb, group, NULL);
++      if (gdp != NULL)
++              free = ext4_free_group_clusters(sb, gdp);
++
+       memcpy(&sg, ext4_get_group_info(sb, group), i);
+       if (buddy_loaded)
+               ext4_mb_unload_buddy(&e4b);
+-      seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free,
+-                      sg.info.bb_fragments, sg.info.bb_first_free);
++      seq_printf(seq, "#%-5lu: %-5u %-5u %-5u %-5u %-5lu [",
++                      (long unsigned int)group, sg.info.bb_free, free,
++                      sg.info.bb_fragments, sg.info.bb_first_free,
++                      sg.info.bb_prealloc_nr);
+       for (i = 0; i <= 13; i++)
+               seq_printf(seq, " %-5u", i <= sb->s_blocksize_bits + 1 ?
+                               sg.info.bb_counters[i] : 0);
+@@ -3507,22 +3517,67 @@ static void ext4_mb_generate_from_freeli
+ }
+ /*
++ * check free blocks in bitmap match free block in group descriptor
++ * do this before taking preallocated blocks into account to be able
++ * to detect on-disk corruptions. The group lock should be hold by the
++ * caller.
++ */
++int ext4_mb_check_ondisk_bitmap(struct super_block *sb, void *bitmap,
++                              struct ext4_group_desc *gdp, int group)
++{
++      unsigned short max = EXT4_CLUSTERS_PER_GROUP(sb);
++      unsigned short i, first, free = 0;
++
++      i = mb_find_next_zero_bit(bitmap, max, 0);
++
++      while (i < max) {
++              first = i;
++              i = mb_find_next_bit(bitmap, max, i);
++              if (i > max)
++                      i = max;
++              free += i - first;
++              if (i < max)
++                      i = mb_find_next_zero_bit(bitmap, max, i);
++      }
++
++      if (free != ext4_free_group_clusters(sb, gdp)) {
++              ext4_error(sb, "on-disk bitmap for group %d"
++                      "corrupted: %u blocks free in bitmap, %u - in gd\n",
++                      group, free, ext4_free_group_clusters(sb, gdp));
++              return -EIO;
++      }
++      return 0;
++}
++
++/*
+  * the function goes through all preallocation in this group and marks them
+  * used in in-core bitmap. buddy must be generated from this bitmap
+  * Need to be called with ext4 group lock held
+  */
+ static noinline_for_stack
+-void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
++int ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
+                                       ext4_group_t group)
+ {
+       struct ext4_group_info *grp = ext4_get_group_info(sb, group);
+       struct ext4_prealloc_space *pa;
++      struct ext4_group_desc *gdp;
+       struct list_head *cur;
+       ext4_group_t groupnr;
+       ext4_grpblk_t start;
+       int preallocated = 0;
++      int skip = 0, count = 0;
++      int err;
+       int len;
++      gdp = ext4_get_group_desc(sb, group, NULL);
++      if (gdp == NULL)
++              return -EIO;
++
++      /* before applying preallocations, check bitmap consistency */
++      err = ext4_mb_check_ondisk_bitmap(sb, bitmap, gdp, group);
++      if (err)
++              return err;
++
+       /* all form of preallocation discards first load group,
+        * so the only competing code is preallocation use.
+        * we don't need any locking here
+@@ -3538,13 +3593,23 @@ void ext4_mb_generate_from_pa(struct sup
+                                            &groupnr, &start);
+               len = pa->pa_len;
+               spin_unlock(&pa->pa_lock);
+-              if (unlikely(len == 0))
++              if (unlikely(len == 0)) {
++                      skip++;
+                       continue;
++              }
+               BUG_ON(groupnr != group);
+               ext4_set_bits(bitmap, start, len);
+               preallocated += len;
++              count++;
++      }
++      if (count + skip != grp->bb_prealloc_nr) {
++              ext4_error(sb, "lost preallocations: "
++                         "count %d, bb_prealloc_nr %lu, skip %d\n",
++                         count, grp->bb_prealloc_nr, skip);
++              return -EIO;
+       }
+       mb_debug(1, "prellocated %u for group %u\n", preallocated, group);
++      return 0;
+ }
+ static void ext4_mb_pa_callback(struct rcu_head *head)
+@@ -3603,6 +3668,7 @@ static void ext4_mb_put_pa(struct ext4_a
+        */
+       ext4_lock_group(sb, grp);
+       list_del(&pa->pa_group_list);
++      ext4_get_group_info(sb, grp)->bb_prealloc_nr--;
+       ext4_unlock_group(sb, grp);
+       spin_lock(pa->pa_obj_lock);
+@@ -3697,6 +3763,7 @@ ext4_mb_new_inode_pa(struct ext4_allocat
+       ext4_lock_group(sb, ac->ac_b_ex.fe_group);
+       list_add(&pa->pa_group_list, &grp->bb_prealloc_list);
++      grp->bb_prealloc_nr++;
+       ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
+       spin_lock(pa->pa_obj_lock);
+@@ -3758,6 +3825,7 @@ ext4_mb_new_group_pa(struct ext4_allocat
+       ext4_lock_group(sb, ac->ac_b_ex.fe_group);
+       list_add(&pa->pa_group_list, &grp->bb_prealloc_list);
++      grp->bb_prealloc_nr++;
+       ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
+       /*
+@@ -3927,6 +3995,8 @@ repeat:
+               spin_unlock(&pa->pa_lock);
++              BUG_ON(grp->bb_prealloc_nr == 0);
++              grp->bb_prealloc_nr--;
+               list_del(&pa->pa_group_list);
+               list_add(&pa->u.pa_tmp_list, &list);
+       }
+@@ -4056,7 +4126,7 @@ repeat:
+               if (err) {
+                       ext4_error(sb, "Error loading buddy information for %u",
+                                       group);
+-                      continue;
++                      return;
+               }
+               bitmap_bh = ext4_read_block_bitmap(sb, group);
+@@ -4068,6 +4138,8 @@ repeat:
+               }
+               ext4_lock_group(sb, group);
++              BUG_ON(e4b.bd_info->bb_prealloc_nr == 0);
++              e4b.bd_info->bb_prealloc_nr--;
+               list_del(&pa->pa_group_list);
+               ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
+               ext4_unlock_group(sb, group);
+@@ -4328,6 +4400,7 @@ ext4_mb_discard_lg_preallocations(struct
+               }
+               ext4_lock_group(sb, group);
+               list_del(&pa->pa_group_list);
++              ext4_get_group_info(sb, group)->bb_prealloc_nr--;
+               ext4_mb_release_group_pa(&e4b, pa);
+               ext4_unlock_group(sb, group);
+Index: linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.h
+===================================================================
+--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/mballoc.h
++++ linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.h
+@@ -82,7 +82,7 @@ extern ushort ext4_mballoc_debug;
+ /*
+  * for which requests use 2^N search using buddies
+  */
+-#define MB_DEFAULT_ORDER2_REQS                2
++#define MB_DEFAULT_ORDER2_REQS                8
+ /*
+  * default group prealloc size 512 blocks
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-misc.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-misc.patch
new file mode 100644 (file)
index 0000000..3943cc6
--- /dev/null
@@ -0,0 +1,153 @@
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
+@@ -1427,6 +1427,8 @@ static inline void ext4_clear_state_flag
+ #define NEXT_ORPHAN(inode) EXT4_I(inode)->i_dtime
++#define JOURNAL_START_HAS_3ARGS       1
++
+ /*
+  * Codes for operating systems
+  */
+@@ -1527,7 +1427,21 @@ static inline void ext4_clear_state_flag
+
+ EXTN_FEATURE_FUNCS(2)
+ EXTN_FEATURE_FUNCS(3)
+-EXTN_FEATURE_FUNCS(4)
++static inline bool ext4_has_unknown_ext4_compat_features(struct super_block *sb)
++{
++      return ((EXT4_SB(sb)->s_es->s_feature_compat &
++              cpu_to_le32(~EXT4_FEATURE_COMPAT_SUPP)) != 0);
++}
++static inline bool ext4_has_unknown_ext4_ro_compat_features(struct super_block *sb)
++{
++      return ((EXT4_SB(sb)->s_es->s_feature_ro_compat &
++              cpu_to_le32(~EXT4_FEATURE_RO_COMPAT_SUPP)) != 0);
++}
++static inline bool ext4_has_unknown_ext4_incompat_features(struct super_block *sb)
++{
++      return ((EXT4_SB(sb)->s_es->s_feature_incompat &
++              cpu_to_le32(~EXT4_FEATURE_INCOMPAT_SUPP)) != 0);
++}
+
+ static inline bool ext4_has_compat_features(struct super_block *sb)
+ {
+@@ -2612,6 +2614,11 @@ struct ext4_extent;
+ extern int ext4_ext_tree_init(handle_t *handle, struct inode *);
+ extern int ext4_ext_writepage_trans_blocks(struct inode *, int);
++extern struct buffer_head *ext4_read_inode_bitmap(struct super_block *sb,
++                                                ext4_group_t block_group);
++extern struct buffer_head *ext4_append(handle_t *handle,
++                                     struct inode *inode,
++                                     ext4_lblk_t *block);
+ extern int ext4_ext_index_trans_blocks(struct inode *inode, int extents);
+                                      int chunk);
+ extern int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/namei.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c
+@@ -48,7 +48,7 @@
+ #define NAMEI_RA_BLOCKS  4
+ #define NAMEI_RA_SIZE      (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
+-static struct buffer_head *ext4_append(handle_t *handle,
++struct buffer_head *ext4_append(handle_t *handle,
+                                       struct inode *inode,
+                                       ext4_lblk_t *block)
+ {
+@@ -155,6 +155,7 @@ static struct buffer_head *__ext4_read_d
+       }
+       return bh;
+ }
++EXPORT_SYMBOL(ext4_append);
+ #ifndef assert
+ #define assert(test) J_ASSERT(test)
+@@ -2210,7 +2211,7 @@ out:
+  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
+  * since this indicates that nlinks count was previously 1.
+  */
+-static void ext4_inc_count(handle_t *handle, struct inode *inode)
++void ext4_inc_count(handle_t *handle, struct inode *inode)
+ {
+       inc_nlink(inode);
+       if (is_dx(inode) && inode->i_nlink > 1) {
+@@ -2222,16 +2223,18 @@ static void ext4_inc_count(handle_t *han
+               }
+       }
+ }
++EXPORT_SYMBOL(ext4_inc_count);
+ /*
+  * If a directory had nlink == 1, then we should let it be 1. This indicates
+  * directory has >EXT4_LINK_MAX subdirs.
+  */
+-static void ext4_dec_count(handle_t *handle, struct inode *inode)
++void ext4_dec_count(handle_t *handle, struct inode *inode)
+ {
+       if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
+               drop_nlink(inode);
+ }
++EXPORT_SYMBOL(ext4_dec_count);
+ static int ext4_add_nondir(handle_t *handle,
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ialloc.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ialloc.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ialloc.c
+@@ -111,7 +111,7 @@ void ext4_end_bitmap_read(struct buffer_
+  *
+  * Return buffer_head of bitmap on success or NULL.
+  */
+-static struct buffer_head *
++struct buffer_head *
+ ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
+ {
+       struct ext4_group_desc *desc;
+@@ -191,6 +191,7 @@ verify:
+       set_buffer_verified(bh);
+       return bh;
+ }
++EXPORT_SYMBOL(ext4_read_inode_bitmap);
+ /*
+  * NOTE! When we get the inode, we're the only people
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inode.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c
+@@ -5281,3 +5281,17 @@ out:
+       sb_end_pagefault(inode->i_sb);
+       return ret;
+ }
++EXPORT_SYMBOL(ext4_map_blocks);
++EXPORT_SYMBOL(ext4_truncate);
++EXPORT_SYMBOL(ext4_iget);
++EXPORT_SYMBOL(ext4_bread);
++EXPORT_SYMBOL(ext4_itable_unused_count);
++EXPORT_SYMBOL(ext4_force_commit);
++EXPORT_SYMBOL(ext4_mark_inode_dirty);
++EXPORT_SYMBOL(ext4_get_group_desc);
++EXPORT_SYMBOL(__ext4_journal_get_write_access);
++EXPORT_SYMBOL(__ext4_journal_start_sb);
++EXPORT_SYMBOL(__ext4_journal_stop);
++EXPORT_SYMBOL(__ext4_handle_dirty_metadata);
++EXPORT_SYMBOL(__ext4_std_error);
++EXPORT_SYMBOL(ext4fs_dirhash);
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/mballoc.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/mballoc.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/mballoc.c
+@@ -5281,7 +5281,6 @@ out:
+                               void *buddy, void *bitmap, ext4_group_t group)
+ {
+       struct ext4_group_info *grp = ext4_get_group_info(sb, group);
+-      struct ext4_sb_info *sbi = EXT4_SB(sb);
+       ext4_grpblk_t max = EXT4_CLUSTERS_PER_GROUP(sb);
+       ext4_grpblk_t i = 0;
+       ext4_grpblk_t first;
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-mmp-brelse.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-mmp-brelse.patch
new file mode 100644 (file)
index 0000000..3d61113
--- /dev/null
@@ -0,0 +1,19 @@
+--- linux-stage.orig/fs/ext4/mmp.c     2015-10-06 08:21:12.013939184 +0530
++++ linux-stage/fs/ext4/mmp.c  2015-10-06 08:20:35.242939292 +0530
+@@ -99,6 +99,8 @@ static int read_mmp_block(struct super_b
+               return 0;
+
+ warn_exit:
++      brelse(*bh);
++      *bh = NULL;
+       ext4_warning(sb, "Error %d while reading MMP block %llu",
+                    ret, mmp_block);
+       return ret;
+@@ -219,6 +221,7 @@ static int kmmpd(void *data)
+                                            "The filesystem seems to have been"
+                                            " multiply mounted.");
+                               ext4_error(sb, "abort");
++                              put_bh(bh_check);
+                               goto failed;
+                       }
+                       put_bh(bh_check);
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-osd-iop-common.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-osd-iop-common.patch
new file mode 100644 (file)
index 0000000..7af24f3
--- /dev/null
@@ -0,0 +1,67 @@
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
+@@ -2145,6 +2145,11 @@ extern int ext4_orphan_add(handle_t *, s
+                                  struct ext4_dir_entry *dirent);
+ extern int ext4_orphan_add(handle_t *, struct inode *);
+ extern int ext4_orphan_del(handle_t *, struct inode *);
++extern struct inode *ext4_create_inode(handle_t *handle,
++                                     struct inode *dir, int mode);
++extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
++                           struct ext4_dir_entry_2 *de_del,
++                           struct buffer_head *bh);
+ extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
+                               __u32 start_minor_hash, __u32 *next_hash);
+ extern int search_dir(struct buffer_head *bh,
+Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c
+===================================================================
+--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/namei.c
++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c
+@@ -2165,7 +2167,7 @@ int ext4_generic_delete_entry(handle_t *
+       return -ENOENT;
+ }
+-static int ext4_delete_entry(handle_t *handle,
++int ext4_delete_entry(handle_t *handle,
+                            struct inode *dir,
+                            struct ext4_dir_entry_2 *de_del,
+                            struct buffer_head *bh)
+@@ -2206,7 +2208,7 @@ out:
+               ext4_std_error(dir->i_sb, err);
+       return err;
+ }
+-
++EXPORT_SYMBOL(ext4_delete_entry);
+ /*
+  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
+  * since this indicates that nlinks count was previously 1.
+@@ -2253,6 +2255,28 @@ static int ext4_add_nondir(handle_t *han
+       return err;
+ }
++ /* Return locked inode, then the caller can modify the inode's states/flags
++  * before others finding it. The caller should unlock the inode by itself. */
++struct inode *ext4_create_inode(handle_t *handle, struct inode *dir, int mode)
++{
++      struct inode *inode;
++
++      inode = ext4_new_inode(handle, dir, mode, NULL, 0, NULL);
++      if (!IS_ERR(inode)) {
++              if (S_ISCHR(mode) || S_ISBLK(mode) || S_ISFIFO(mode)) {
++#ifdef CONFIG_LDISKFS_FS_XATTR
++                      inode->i_op = &ext4_special_inode_operations;
++#endif
++              } else {
++                      inode->i_op = &ext4_file_inode_operations;
++                      inode->i_fop = &ext4_file_operations;
++                      ext4_set_aops(inode);
++              }
++      }
++      return inode;
++}
++EXPORT_SYMBOL(ext4_create_inode);
++
+ /*
+  * By the time this is called, we already have created
+  * the directory cache entry for the new file, but it
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-pdirop.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-pdirop.patch
new file mode 100644 (file)
index 0000000..c977b5e
--- /dev/null
@@ -0,0 +1,1931 @@
+Single directory performance is a critical for HPC workloads. In a
+typical use case an application creates a separate output file for
+each node and task in a job. As nodes and tasks increase, hundreds
+of thousands of files may be created in a single directory within
+a short window of time.
+Today, both filename lookup and file system modifying operations
+(such as create and unlink) are protected with a single lock for
+an entire ldiskfs directory. PDO project will remove this
+bottleneck by introducing a parallel locking mechanism for entire
+ldiskfs directories. This work will enable multiple application
+threads to simultaneously lookup, create and unlink in parallel.
+    
+This patch contains:
+ - pdirops support for ldiskfs
+ - integrate with osd-ldiskfs
+
+Index: linux-3.10.0-229.1.2.fc21.x86_64/include/linux/htree_lock.h
+===================================================================
+--- /dev/null
++++ linux-3.10.0-229.1.2.fc21.x86_64/include/linux/htree_lock.h
+@@ -0,0 +1,187 @@
++/*
++ * include/linux/htree_lock.h
++ *
++ * Copyright (c) 2011, 2012, Intel Corporation.
++ *
++ * Author: Liang Zhen <liang@whamcloud.com>
++ */
++
++/*
++ * htree lock
++ *
++ * htree_lock is an advanced lock, it can support five lock modes (concept is
++ * taken from DLM) and it's a sleeping lock.
++ *
++ * most common use case is:
++ * - create a htree_lock_head for data
++ * - each thread (contender) creates it's own htree_lock
++ * - contender needs to call htree_lock(lock_node, mode) to protect data and
++ *   call htree_unlock to release lock
++ *
++ * Also, there is advanced use-case which is more complex, user can have
++ * PW/PR lock on particular key, it's mostly used while user holding shared
++ * lock on the htree (CW, CR)
++ *
++ * htree_lock(lock_node, HTREE_LOCK_CR); lock the htree with CR
++ * htree_node_lock(lock_node, HTREE_LOCK_PR, key...); lock @key with PR
++ * ...
++ * htree_node_unlock(lock_node);; unlock the key
++ *
++ * Another tip is, we can have N-levels of this kind of keys, all we need to
++ * do is specifying N-levels while creating htree_lock_head, then we can
++ * lock/unlock a specific level by:
++ * htree_node_lock(lock_node, mode1, key1, level1...);
++ * do something;
++ * htree_node_lock(lock_node, mode1, key2, level2...);
++ * do something;
++ * htree_node_unlock(lock_node, level2);
++ * htree_node_unlock(lock_node, level1);
++ *
++ * NB: for multi-level, should be careful about locking order to avoid deadlock
++ */
++
++#ifndef _LINUX_HTREE_LOCK_H
++#define _LINUX_HTREE_LOCK_H
++
++#include <linux/list.h>
++#include <linux/spinlock.h>
++#include <linux/sched.h>
++
++/*
++ * Lock Modes
++ * more details can be found here:
++ * http://en.wikipedia.org/wiki/Distributed_lock_manager
++ */
++typedef enum {
++      HTREE_LOCK_EX   = 0, /* exclusive lock: incompatible with all others */
++      HTREE_LOCK_PW,       /* protected write: allows only CR users */
++      HTREE_LOCK_PR,       /* protected read: allow PR, CR users */
++      HTREE_LOCK_CW,       /* concurrent write: allow CR, CW users */
++      HTREE_LOCK_CR,       /* concurrent read: allow all but EX users */
++      HTREE_LOCK_MAX,      /* number of lock modes */
++} htree_lock_mode_t;
++
++#define HTREE_LOCK_NL         HTREE_LOCK_MAX
++#define HTREE_LOCK_INVAL      0xdead10c
++
++enum {
++      HTREE_HBITS_MIN         = 2,
++      HTREE_HBITS_DEF         = 14,
++      HTREE_HBITS_MAX         = 32,
++};
++
++enum {
++      HTREE_EVENT_DISABLE     = (0),
++      HTREE_EVENT_RD          = (1 << HTREE_LOCK_PR),
++      HTREE_EVENT_WR          = (1 << HTREE_LOCK_PW),
++      HTREE_EVENT_RDWR        = (HTREE_EVENT_RD | HTREE_EVENT_WR),
++};
++
++struct htree_lock;
++
++typedef void (*htree_event_cb_t)(void *target, void *event);
++
++struct htree_lock_child {
++      struct list_head        lc_list;        /* granted list */
++      htree_event_cb_t        lc_callback;    /* event callback */
++      unsigned                lc_events;      /* event types */
++};
++
++struct htree_lock_head {
++      unsigned long           lh_lock;        /* bits lock */
++      /* blocked lock list (htree_lock) */
++      struct list_head        lh_blocked_list;
++      /* # key levels */
++      u16                     lh_depth;
++      /* hash bits for key and limit number of locks */
++      u16                     lh_hbits;
++      /* counters for blocked locks */
++      u16                     lh_nblocked[HTREE_LOCK_MAX];
++      /* counters for granted locks */
++      u16                     lh_ngranted[HTREE_LOCK_MAX];
++      /* private data */
++      void                    *lh_private;
++      /* array of children locks */
++      struct htree_lock_child lh_children[0];
++};
++
++/* htree_lock_node_t is child-lock for a specific key (ln_value) */
++struct htree_lock_node {
++      htree_lock_mode_t       ln_mode;
++      /* major hash key */
++      u16                     ln_major_key;
++      /* minor hash key */
++      u16                     ln_minor_key;
++      struct list_head        ln_major_list;
++      struct list_head        ln_minor_list;
++      /* alive list, all locks (granted, blocked, listening) are on it */
++      struct list_head        ln_alive_list;
++      /* blocked list */
++      struct list_head        ln_blocked_list;
++      /* granted list */
++      struct list_head        ln_granted_list;
++      void                    *ln_ev_target;
++};
++
++struct htree_lock {
++      struct task_struct      *lk_task;
++      struct htree_lock_head  *lk_head;
++      void                    *lk_private;
++      unsigned                lk_depth;
++      htree_lock_mode_t       lk_mode;
++      struct list_head        lk_blocked_list;
++      struct htree_lock_node  lk_nodes[0];
++};
++
++/* create a lock head, which stands for a resource */
++struct htree_lock_head *htree_lock_head_alloc(unsigned depth,
++                                            unsigned hbits, unsigned priv);
++/* free a lock head */
++void htree_lock_head_free(struct htree_lock_head *lhead);
++/* register event callback for child lock at level @depth */
++void htree_lock_event_attach(struct htree_lock_head *lhead, unsigned depth,
++                           unsigned events, htree_event_cb_t callback);
++/* create a lock handle, which stands for a thread */
++struct htree_lock *htree_lock_alloc(unsigned depth, unsigned pbytes);
++/* free a lock handle */
++void htree_lock_free(struct htree_lock *lck);
++/* lock htree, when @wait is true, 0 is returned if the lock can't
++ * be granted immediately */
++int htree_lock_try(struct htree_lock *lck, struct htree_lock_head *lhead,
++                 htree_lock_mode_t mode, int wait);
++/* unlock htree */
++void htree_unlock(struct htree_lock *lck);
++/* unlock and relock htree with @new_mode */
++int htree_change_lock_try(struct htree_lock *lck,
++                        htree_lock_mode_t new_mode, int wait);
++void htree_change_mode(struct htree_lock *lck, htree_lock_mode_t mode);
++/* require child lock (key) of htree at level @dep, @event will be sent to all
++ * listeners on this @key while lock being granted */
++int htree_node_lock_try(struct htree_lock *lck, htree_lock_mode_t mode,
++                      u32 key, unsigned dep, int wait, void *event);
++/* release child lock at level @dep, this lock will listen on it's key
++ * if @event isn't NULL, event_cb will be called against @lck while granting
++ * any other lock at level @dep with the same key */
++void htree_node_unlock(struct htree_lock *lck, unsigned dep, void *event);
++/* stop listening on child lock at level @dep */
++void htree_node_stop_listen(struct htree_lock *lck, unsigned dep);
++/* for debug */
++void htree_lock_stat_print(int depth);
++void htree_lock_stat_reset(void);
++
++#define htree_lock(lck, lh, mode)     htree_lock_try(lck, lh, mode, 1)
++#define htree_change_lock(lck, mode)  htree_change_lock_try(lck, mode, 1)
++
++#define htree_lock_mode(lck)          ((lck)->lk_mode)
++
++#define htree_node_lock(lck, mode, key, dep)  \
++      htree_node_lock_try(lck, mode, key, dep, 1, NULL)
++/* this is only safe in thread context of lock owner */
++#define htree_node_is_granted(lck, dep)               \
++      ((lck)->lk_nodes[dep].ln_mode != HTREE_LOCK_INVAL && \
++       (lck)->lk_nodes[dep].ln_mode != HTREE_LOCK_NL)
++/* this is only safe in thread context of lock owner */
++#define htree_node_is_listening(lck, dep)     \
++      ((lck)->lk_nodes[dep].ln_mode == HTREE_LOCK_NL)
++
++#endif
+Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/htree_lock.c
+===================================================================
+--- /dev/null
++++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/htree_lock.c
+@@ -0,0 +1,880 @@
++/*
++ * fs/ext4/htree_lock.c
++ *
++ * Copyright (c) 2011, 2012, Intel Corporation.
++ *
++ * Author: Liang Zhen <liang@whamcloud.com>
++ */
++#include <linux/jbd2.h>
++#include <linux/hash.h>
++#include <linux/module.h>
++#include <linux/htree_lock.h>
++
++enum {
++      HTREE_LOCK_BIT_EX       = (1 << HTREE_LOCK_EX),
++      HTREE_LOCK_BIT_PW       = (1 << HTREE_LOCK_PW),
++      HTREE_LOCK_BIT_PR       = (1 << HTREE_LOCK_PR),
++      HTREE_LOCK_BIT_CW       = (1 << HTREE_LOCK_CW),
++      HTREE_LOCK_BIT_CR       = (1 << HTREE_LOCK_CR),
++};
++
++enum {
++      HTREE_LOCK_COMPAT_EX    = 0,
++      HTREE_LOCK_COMPAT_PW    = HTREE_LOCK_COMPAT_EX | HTREE_LOCK_BIT_CR,
++      HTREE_LOCK_COMPAT_PR    = HTREE_LOCK_COMPAT_PW | HTREE_LOCK_BIT_PR,
++      HTREE_LOCK_COMPAT_CW    = HTREE_LOCK_COMPAT_PW | HTREE_LOCK_BIT_CW,
++      HTREE_LOCK_COMPAT_CR    = HTREE_LOCK_COMPAT_CW | HTREE_LOCK_BIT_PR |
++                                HTREE_LOCK_BIT_PW,
++};
++
++static int htree_lock_compat[] = {
++      [HTREE_LOCK_EX]         HTREE_LOCK_COMPAT_EX,
++      [HTREE_LOCK_PW]         HTREE_LOCK_COMPAT_PW,
++      [HTREE_LOCK_PR]         HTREE_LOCK_COMPAT_PR,
++      [HTREE_LOCK_CW]         HTREE_LOCK_COMPAT_CW,
++      [HTREE_LOCK_CR]         HTREE_LOCK_COMPAT_CR,
++};
++
++/* max allowed htree-lock depth.
++ * We only need depth=3 for ext4 although user can have higher value. */
++#define HTREE_LOCK_DEP_MAX    16
++
++#ifdef HTREE_LOCK_DEBUG
++
++static char *hl_name[] = {
++      [HTREE_LOCK_EX]         "EX",
++      [HTREE_LOCK_PW]         "PW",
++      [HTREE_LOCK_PR]         "PR",
++      [HTREE_LOCK_CW]         "CW",
++      [HTREE_LOCK_CR]         "CR",
++};
++
++/* lock stats */
++struct htree_lock_node_stats {
++      unsigned long long      blocked[HTREE_LOCK_MAX];
++      unsigned long long      granted[HTREE_LOCK_MAX];
++      unsigned long long      retried[HTREE_LOCK_MAX];
++      unsigned long long      events;
++};
++
++struct htree_lock_stats {
++      struct htree_lock_node_stats    nodes[HTREE_LOCK_DEP_MAX];
++      unsigned long long      granted[HTREE_LOCK_MAX];
++      unsigned long long      blocked[HTREE_LOCK_MAX];
++};
++
++static struct htree_lock_stats hl_stats;
++
++void htree_lock_stat_reset(void)
++{
++      memset(&hl_stats, 0, sizeof(hl_stats));
++}
++
++void htree_lock_stat_print(int depth)
++{
++      int     i;
++      int     j;
++
++      printk(KERN_DEBUG "HTREE LOCK STATS:\n");
++      for (i = 0; i < HTREE_LOCK_MAX; i++) {
++              printk(KERN_DEBUG "[%s]: G [%10llu], B [%10llu]\n",
++                     hl_name[i], hl_stats.granted[i], hl_stats.blocked[i]);
++      }
++      for (i = 0; i < depth; i++) {
++              printk(KERN_DEBUG "HTREE CHILD [%d] STATS:\n", i);
++              for (j = 0; j < HTREE_LOCK_MAX; j++) {
++                      printk(KERN_DEBUG
++                              "[%s]: G [%10llu], B [%10llu], R [%10llu]\n",
++                              hl_name[j], hl_stats.nodes[i].granted[j],
++                              hl_stats.nodes[i].blocked[j],
++                              hl_stats.nodes[i].retried[j]);
++              }
++      }
++}
++
++#define lk_grant_inc(m)       do { hl_stats.granted[m]++; } while (0)
++#define lk_block_inc(m)       do { hl_stats.blocked[m]++; } while (0)
++#define ln_grant_inc(d, m)    do { hl_stats.nodes[d].granted[m]++; } while (0)
++#define ln_block_inc(d, m)    do { hl_stats.nodes[d].blocked[m]++; } while (0)
++#define ln_retry_inc(d, m)    do { hl_stats.nodes[d].retried[m]++; } while (0)
++#define ln_event_inc(d)       do { hl_stats.nodes[d].events++; } while (0)
++
++#else /* !DEBUG */
++
++void htree_lock_stat_reset(void) {}
++void htree_lock_stat_print(int depth) {}
++
++#define lk_grant_inc(m)             do {} while (0)
++#define lk_block_inc(m)             do {} while (0)
++#define ln_grant_inc(d, m)    do {} while (0)
++#define ln_block_inc(d, m)    do {} while (0)
++#define ln_retry_inc(d, m)    do {} while (0)
++#define ln_event_inc(d)             do {} while (0)
++
++#endif /* DEBUG */
++
++EXPORT_SYMBOL(htree_lock_stat_reset);
++EXPORT_SYMBOL(htree_lock_stat_print);
++
++#define HTREE_DEP_ROOT                  (-1)
++
++#define htree_spin_lock(lhead, dep)                           \
++      bit_spin_lock((dep) + 1, &(lhead)->lh_lock)
++#define htree_spin_unlock(lhead, dep)                         \
++      bit_spin_unlock((dep) + 1, &(lhead)->lh_lock)
++
++#define htree_key_event_ignore(child, ln)                     \
++      (!((child)->lc_events & (1 << (ln)->ln_mode)))
++
++static int
++htree_key_list_empty(struct htree_lock_node *ln)
++{
++      return list_empty(&ln->ln_major_list) && list_empty(&ln->ln_minor_list);
++}
++
++static void
++htree_key_list_del_init(struct htree_lock_node *ln)
++{
++      struct htree_lock_node *tmp = NULL;
++
++      if (!list_empty(&ln->ln_minor_list)) {
++              tmp = list_entry(ln->ln_minor_list.next,
++                               struct htree_lock_node, ln_minor_list);
++              list_del_init(&ln->ln_minor_list);
++      }
++
++      if (list_empty(&ln->ln_major_list))
++              return;
++
++      if (tmp == NULL) { /* not on minor key list */
++              list_del_init(&ln->ln_major_list);
++      } else {
++              BUG_ON(!list_empty(&tmp->ln_major_list));
++              list_replace_init(&ln->ln_major_list, &tmp->ln_major_list);
++      }
++}
++
++static void
++htree_key_list_replace_init(struct htree_lock_node *old,
++                          struct htree_lock_node *new)
++{
++      if (!list_empty(&old->ln_major_list))
++              list_replace_init(&old->ln_major_list, &new->ln_major_list);
++
++      if (!list_empty(&old->ln_minor_list))
++              list_replace_init(&old->ln_minor_list, &new->ln_minor_list);
++}
++
++static void
++htree_key_event_enqueue(struct htree_lock_child *child,
++                      struct htree_lock_node *ln, int dep, void *event)
++{
++      struct htree_lock_node *tmp;
++
++      /* NB: ALWAYS called holding lhead::lh_lock(dep) */
++      BUG_ON(ln->ln_mode == HTREE_LOCK_NL);
++      if (event == NULL || htree_key_event_ignore(child, ln))
++              return;
++
++      /* shouldn't be a very long list */
++      list_for_each_entry(tmp, &ln->ln_alive_list, ln_alive_list) {
++              if (tmp->ln_mode == HTREE_LOCK_NL) {
++                      ln_event_inc(dep);
++                      if (child->lc_callback != NULL)
++                              child->lc_callback(tmp->ln_ev_target, event);
++              }
++      }
++}
++
++static int
++htree_node_lock_enqueue(struct htree_lock *newlk, struct htree_lock *curlk,
++                      unsigned dep, int wait, void *event)
++{
++      struct htree_lock_child *child = &newlk->lk_head->lh_children[dep];
++      struct htree_lock_node *newln = &newlk->lk_nodes[dep];
++      struct htree_lock_node *curln = &curlk->lk_nodes[dep];
++
++      /* NB: ALWAYS called holding lhead::lh_lock(dep) */
++      /* NB: we only expect PR/PW lock mode at here, only these two modes are
++       * allowed for htree_node_lock(asserted in htree_node_lock_internal),
++       * NL is only used for listener, user can't directly require NL mode */
++      if ((curln->ln_mode == HTREE_LOCK_NL) ||
++          (curln->ln_mode != HTREE_LOCK_PW &&
++           newln->ln_mode != HTREE_LOCK_PW)) {
++              /* no conflict, attach it on granted list of @curlk */
++              if (curln->ln_mode != HTREE_LOCK_NL) {
++                      list_add(&newln->ln_granted_list,
++                               &curln->ln_granted_list);
++              } else {
++                      /* replace key owner */
++                      htree_key_list_replace_init(curln, newln);
++              }
++
++              list_add(&newln->ln_alive_list, &curln->ln_alive_list);
++              htree_key_event_enqueue(child, newln, dep, event);
++              ln_grant_inc(dep, newln->ln_mode);
++              return 1; /* still hold lh_lock */
++      }
++
++      if (!wait) { /* can't grant and don't want to wait */
++              ln_retry_inc(dep, newln->ln_mode);
++              newln->ln_mode = HTREE_LOCK_INVAL;
++              return -1; /* don't wait and just return -1 */
++      }
++
++      newlk->lk_task = current;
++      set_current_state(TASK_UNINTERRUPTIBLE);
++      /* conflict, attach it on blocked list of curlk */
++      list_add_tail(&newln->ln_blocked_list, &curln->ln_blocked_list);
++      list_add(&newln->ln_alive_list, &curln->ln_alive_list);
++      ln_block_inc(dep, newln->ln_mode);
++
++      htree_spin_unlock(newlk->lk_head, dep);
++      /* wait to be given the lock */
++      if (newlk->lk_task != NULL)
++              schedule();
++      /* granted, no doubt, wake up will set me RUNNING */
++      if (event == NULL || htree_key_event_ignore(child, newln))
++              return 0; /* granted without lh_lock */
++
++      htree_spin_lock(newlk->lk_head, dep);
++      htree_key_event_enqueue(child, newln, dep, event);
++      return 1; /* still hold lh_lock */
++}
++
++/*
++ * get PR/PW access to particular tree-node according to @dep and @key,
++ * it will return -1 if @wait is false and can't immediately grant this lock.
++ * All listeners(HTREE_LOCK_NL) on @dep and with the same @key will get
++ * @event if it's not NULL.
++ * NB: ALWAYS called holding lhead::lh_lock
++ */
++static int
++htree_node_lock_internal(struct htree_lock_head *lhead, struct htree_lock *lck,
++                       htree_lock_mode_t mode, u32 key, unsigned dep,
++                       int wait, void *event)
++{
++      LIST_HEAD(list);
++      struct htree_lock       *tmp;
++      struct htree_lock       *tmp2;
++      u16                     major;
++      u16                     minor;
++      u8                      reverse;
++      u8                      ma_bits;
++      u8                      mi_bits;
++
++      BUG_ON(mode != HTREE_LOCK_PW && mode != HTREE_LOCK_PR);
++      BUG_ON(htree_node_is_granted(lck, dep));
++
++      key = hash_long(key, lhead->lh_hbits);
++
++      mi_bits = lhead->lh_hbits >> 1;
++      ma_bits = lhead->lh_hbits - mi_bits;
++
++      lck->lk_nodes[dep].ln_major_key = major = key & ((1U << ma_bits) - 1);
++      lck->lk_nodes[dep].ln_minor_key = minor = key >> ma_bits;
++      lck->lk_nodes[dep].ln_mode = mode;
++
++      /*
++       * The major key list is an ordered list, so searches are started
++       * at the end of the list that is numerically closer to major_key,
++       * so at most half of the list will be walked (for well-distributed
++       * keys). The list traversal aborts early if the expected key
++       * location is passed.
++       */
++      reverse = (major >= (1 << (ma_bits - 1)));
++
++      if (reverse) {
++              list_for_each_entry_reverse(tmp,
++                                      &lhead->lh_children[dep].lc_list,
++                                      lk_nodes[dep].ln_major_list) {
++                      if (tmp->lk_nodes[dep].ln_major_key == major) {
++                              goto search_minor;
++
++                      } else if (tmp->lk_nodes[dep].ln_major_key < major) {
++                              /* attach _after_ @tmp */
++                              list_add(&lck->lk_nodes[dep].ln_major_list,
++                                       &tmp->lk_nodes[dep].ln_major_list);
++                              goto out_grant_major;
++                      }
++              }
++
++              list_add(&lck->lk_nodes[dep].ln_major_list,
++                       &lhead->lh_children[dep].lc_list);
++              goto out_grant_major;
++
++      } else {
++              list_for_each_entry(tmp, &lhead->lh_children[dep].lc_list,
++                                  lk_nodes[dep].ln_major_list) {
++                      if (tmp->lk_nodes[dep].ln_major_key == major) {
++                              goto search_minor;
++
++                      } else if (tmp->lk_nodes[dep].ln_major_key > major) {
++                              /* insert _before_ @tmp */
++                              list_add_tail(&lck->lk_nodes[dep].ln_major_list,
++                                      &tmp->lk_nodes[dep].ln_major_list);
++                              goto out_grant_major;
++                      }
++              }
++
++              list_add_tail(&lck->lk_nodes[dep].ln_major_list,
++                            &lhead->lh_children[dep].lc_list);
++              goto out_grant_major;
++      }
++
++ search_minor:
++      /*
++       * NB: minor_key list doesn't have a "head", @list is just a
++       * temporary stub for helping list searching, make sure it's removed
++       * after searching.
++       * minor_key list is an ordered list too.
++       */
++      list_add_tail(&list, &tmp->lk_nodes[dep].ln_minor_list);
++
++      reverse = (minor >= (1 << (mi_bits - 1)));
++
++      if (reverse) {
++              list_for_each_entry_reverse(tmp2, &list,
++                                          lk_nodes[dep].ln_minor_list) {
++                      if (tmp2->lk_nodes[dep].ln_minor_key == minor) {
++                              goto out_enqueue;
++
++                      } else if (tmp2->lk_nodes[dep].ln_minor_key < minor) {
++                              /* attach _after_ @tmp2 */
++                              list_add(&lck->lk_nodes[dep].ln_minor_list,
++                                       &tmp2->lk_nodes[dep].ln_minor_list);
++                              goto out_grant_minor;
++                      }
++              }
++
++              list_add(&lck->lk_nodes[dep].ln_minor_list, &list);
++
++      } else {
++              list_for_each_entry(tmp2, &list,
++                                  lk_nodes[dep].ln_minor_list) {
++                      if (tmp2->lk_nodes[dep].ln_minor_key == minor) {
++                              goto out_enqueue;
++
++                      } else if (tmp2->lk_nodes[dep].ln_minor_key > minor) {
++                              /* insert _before_ @tmp2 */
++                              list_add_tail(&lck->lk_nodes[dep].ln_minor_list,
++                                      &tmp2->lk_nodes[dep].ln_minor_list);
++                              goto out_grant_minor;
++                      }
++              }
++
++              list_add_tail(&lck->lk_nodes[dep].ln_minor_list, &list);
++      }
++
++ out_grant_minor:
++      if (list.next == &lck->lk_nodes[dep].ln_minor_list) {
++              /* new lock @lck is the first one on minor_key list, which
++               * means it has the smallest minor_key and it should
++               * replace @tmp as minor_key owner */
++              list_replace_init(&tmp->lk_nodes[dep].ln_major_list,
++                                &lck->lk_nodes[dep].ln_major_list);
++      }
++      /* remove the temporary head */
++      list_del(&list);
++
++ out_grant_major:
++      ln_grant_inc(dep, lck->lk_nodes[dep].ln_mode);
++      return 1; /* granted with holding lh_lock */
++
++ out_enqueue:
++      list_del(&list); /* remove temprary head */
++      return htree_node_lock_enqueue(lck, tmp2, dep, wait, event);
++}
++
++/*
++ * release the key of @lck at level @dep, and grant any blocked locks.
++ * caller will still listen on @key if @event is not NULL, which means
++ * caller can see a event (by event_cb) while granting any lock with
++ * the same key at level @dep.
++ * NB: ALWAYS called holding lhead::lh_lock
++ * NB: listener will not block anyone because listening mode is HTREE_LOCK_NL
++ */
++static void
++htree_node_unlock_internal(struct htree_lock_head *lhead,
++                         struct htree_lock *curlk, unsigned dep, void *event)
++{
++      struct htree_lock_node  *curln = &curlk->lk_nodes[dep];
++      struct htree_lock       *grtlk = NULL;
++      struct htree_lock_node  *grtln;
++      struct htree_lock       *poslk;
++      struct htree_lock       *tmplk;
++
++      if (!htree_node_is_granted(curlk, dep))
++              return;
++
++      if (!list_empty(&curln->ln_granted_list)) {
++              /* there is another granted lock */
++              grtlk = list_entry(curln->ln_granted_list.next,
++                                 struct htree_lock,
++                                 lk_nodes[dep].ln_granted_list);
++              list_del_init(&curln->ln_granted_list);
++      }
++
++      if (grtlk == NULL && !list_empty(&curln->ln_blocked_list)) {
++              /*
++               * @curlk is the only granted lock, so we confirmed:
++               * a) curln is key owner (attached on major/minor_list),
++               *    so if there is any blocked lock, it should be attached
++               *    on curln->ln_blocked_list
++               * b) we always can grant the first blocked lock
++               */
++              grtlk = list_entry(curln->ln_blocked_list.next,
++                                 struct htree_lock,
++                                 lk_nodes[dep].ln_blocked_list);
++              BUG_ON(grtlk->lk_task == NULL);
++              wake_up_process(grtlk->lk_task);
++      }
++
++      if (event != NULL &&
++          lhead->lh_children[dep].lc_events != HTREE_EVENT_DISABLE) {
++              curln->ln_ev_target = event;
++              curln->ln_mode = HTREE_LOCK_NL; /* listen! */
++      } else {
++              curln->ln_mode = HTREE_LOCK_INVAL;
++      }
++
++      if (grtlk == NULL) { /* I must be the only one locking this key */
++              struct htree_lock_node *tmpln;
++
++              BUG_ON(htree_key_list_empty(curln));
++
++              if (curln->ln_mode == HTREE_LOCK_NL) /* listening */
++                      return;
++
++              /* not listening */
++              if (list_empty(&curln->ln_alive_list)) { /* no more listener */
++                      htree_key_list_del_init(curln);
++                      return;
++              }
++
++              tmpln = list_entry(curln->ln_alive_list.next,
++                                 struct htree_lock_node, ln_alive_list);
++
++              BUG_ON(tmpln->ln_mode != HTREE_LOCK_NL);
++
++              htree_key_list_replace_init(curln, tmpln);
++              list_del_init(&curln->ln_alive_list);
++
++              return;
++      }
++
++      /* have a granted lock */
++      grtln = &grtlk->lk_nodes[dep];
++      if (!list_empty(&curln->ln_blocked_list)) {
++              /* only key owner can be on both lists */
++              BUG_ON(htree_key_list_empty(curln));
++
++              if (list_empty(&grtln->ln_blocked_list)) {
++                      list_add(&grtln->ln_blocked_list,
++                               &curln->ln_blocked_list);
++              }
++              list_del_init(&curln->ln_blocked_list);
++      }
++      /*
++       * NB: this is the tricky part:
++       * We have only two modes for child-lock (PR and PW), also,
++       * only owner of the key (attached on major/minor_list) can be on
++       * both blocked_list and granted_list, so @grtlk must be one
++       * of these two cases:
++       *
++       * a) @grtlk is taken from granted_list, which means we've granted
++       *    more than one lock so @grtlk has to be PR, the first blocked
++       *    lock must be PW and we can't grant it at all.
++       *    So even @grtlk is not owner of the key (empty blocked_list),
++       *    we don't care because we can't grant any lock.
++       * b) we just grant a new lock which is taken from head of blocked
++       *    list, and it should be the first granted lock, and it should
++       *    be the first one linked on blocked_list.
++       *
++       * Either way, we can get correct result by iterating blocked_list
++       * of @grtlk, and don't have to bother on how to find out
++       * owner of current key.
++       */
++      list_for_each_entry_safe(poslk, tmplk, &grtln->ln_blocked_list,
++                               lk_nodes[dep].ln_blocked_list) {
++              if (grtlk->lk_nodes[dep].ln_mode == HTREE_LOCK_PW ||
++                  poslk->lk_nodes[dep].ln_mode == HTREE_LOCK_PW)
++                      break;
++              /* grant all readers */
++              list_del_init(&poslk->lk_nodes[dep].ln_blocked_list);
++              list_add(&poslk->lk_nodes[dep].ln_granted_list,
++                       &grtln->ln_granted_list);
++
++              BUG_ON(poslk->lk_task == NULL);
++              wake_up_process(poslk->lk_task);
++      }
++
++      /* if @curln is the owner of this key, replace it with @grtln */
++      if (!htree_key_list_empty(curln))
++              htree_key_list_replace_init(curln, grtln);
++
++      if (curln->ln_mode == HTREE_LOCK_INVAL)
++              list_del_init(&curln->ln_alive_list);
++}
++
++/*
++ * it's just wrapper of htree_node_lock_internal, it returns 1 on granted
++ * and 0 only if @wait is false and can't grant it immediately
++ */
++int
++htree_node_lock_try(struct htree_lock *lck, htree_lock_mode_t mode,
++                  u32 key, unsigned dep, int wait, void *event)
++{
++      struct htree_lock_head *lhead = lck->lk_head;
++      int rc;
++
++      BUG_ON(dep >= lck->lk_depth);
++      BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL);
++
++      htree_spin_lock(lhead, dep);
++      rc = htree_node_lock_internal(lhead, lck, mode, key, dep, wait, event);
++      if (rc != 0)
++              htree_spin_unlock(lhead, dep);
++      return rc >= 0;
++}
++EXPORT_SYMBOL(htree_node_lock_try);
++
++/* it's wrapper of htree_node_unlock_internal */
++void
++htree_node_unlock(struct htree_lock *lck, unsigned dep, void *event)
++{
++      struct htree_lock_head *lhead = lck->lk_head;
++
++      BUG_ON(dep >= lck->lk_depth);
++      BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL);
++
++      htree_spin_lock(lhead, dep);
++      htree_node_unlock_internal(lhead, lck, dep, event);
++      htree_spin_unlock(lhead, dep);
++}
++EXPORT_SYMBOL(htree_node_unlock);
++
++/* stop listening on child-lock level @dep */
++void
++htree_node_stop_listen(struct htree_lock *lck, unsigned dep)
++{
++      struct htree_lock_node *ln = &lck->lk_nodes[dep];
++      struct htree_lock_node *tmp;
++
++      BUG_ON(htree_node_is_granted(lck, dep));
++      BUG_ON(!list_empty(&ln->ln_blocked_list));
++      BUG_ON(!list_empty(&ln->ln_granted_list));
++
++      if (!htree_node_is_listening(lck, dep))
++              return;
++
++      htree_spin_lock(lck->lk_head, dep);
++      ln->ln_mode = HTREE_LOCK_INVAL;
++      ln->ln_ev_target = NULL;
++
++      if (htree_key_list_empty(ln)) { /* not owner */
++              list_del_init(&ln->ln_alive_list);
++              goto out;
++      }
++
++      /* I'm the owner... */
++      if (list_empty(&ln->ln_alive_list)) { /* no more listener */
++              htree_key_list_del_init(ln);
++              goto out;
++      }
++
++      tmp = list_entry(ln->ln_alive_list.next,
++                       struct htree_lock_node, ln_alive_list);
++
++      BUG_ON(tmp->ln_mode != HTREE_LOCK_NL);
++      htree_key_list_replace_init(ln, tmp);
++      list_del_init(&ln->ln_alive_list);
++ out:
++      htree_spin_unlock(lck->lk_head, dep);
++}
++EXPORT_SYMBOL(htree_node_stop_listen);
++
++/* release all child-locks if we have any */
++static void
++htree_node_release_all(struct htree_lock *lck)
++{
++      int     i;
++
++      for (i = 0; i < lck->lk_depth; i++) {
++              if (htree_node_is_granted(lck, i))
++                      htree_node_unlock(lck, i, NULL);
++              else if (htree_node_is_listening(lck, i))
++                      htree_node_stop_listen(lck, i);
++      }
++}
++
++/*
++ * obtain htree lock, it could be blocked inside if there's conflict
++ * with any granted or blocked lock and @wait is true.
++ * NB: ALWAYS called holding lhead::lh_lock
++ */
++static int
++htree_lock_internal(struct htree_lock *lck, int wait)
++{
++      struct htree_lock_head *lhead = lck->lk_head;
++      int     granted = 0;
++      int     blocked = 0;
++      int     i;
++
++      for (i = 0; i < HTREE_LOCK_MAX; i++) {
++              if (lhead->lh_ngranted[i] != 0)
++                      granted |= 1 << i;
++              if (lhead->lh_nblocked[i] != 0)
++                      blocked |= 1 << i;
++      }
++      if ((htree_lock_compat[lck->lk_mode] & granted) != granted ||
++          (htree_lock_compat[lck->lk_mode] & blocked) != blocked) {
++              /* will block current lock even it just conflicts with any
++               * other blocked lock, so lock like EX wouldn't starve */
++              if (!wait)
++                      return -1;
++              lhead->lh_nblocked[lck->lk_mode]++;
++              lk_block_inc(lck->lk_mode);
++
++              lck->lk_task = current;
++              list_add_tail(&lck->lk_blocked_list, &lhead->lh_blocked_list);
++
++              set_current_state(TASK_UNINTERRUPTIBLE);
++              htree_spin_unlock(lhead, HTREE_DEP_ROOT);
++              /* wait to be given the lock */
++              if (lck->lk_task != NULL)
++                      schedule();
++              /* granted, no doubt. wake up will set me RUNNING */
++              return 0; /* without lh_lock */
++      }
++      lhead->lh_ngranted[lck->lk_mode]++;
++      lk_grant_inc(lck->lk_mode);
++      return 1;
++}
++
++/* release htree lock. NB: ALWAYS called holding lhead::lh_lock */
++static void
++htree_unlock_internal(struct htree_lock *lck)
++{
++      struct htree_lock_head *lhead = lck->lk_head;
++      struct htree_lock *tmp;
++      struct htree_lock *tmp2;
++      int granted = 0;
++      int i;
++
++      BUG_ON(lhead->lh_ngranted[lck->lk_mode] == 0);
++
++      lhead->lh_ngranted[lck->lk_mode]--;
++      lck->lk_mode = HTREE_LOCK_INVAL;
++
++      for (i = 0; i < HTREE_LOCK_MAX; i++) {
++              if (lhead->lh_ngranted[i] != 0)
++                      granted |= 1 << i;
++      }
++      list_for_each_entry_safe(tmp, tmp2,
++                               &lhead->lh_blocked_list, lk_blocked_list) {
++              /* conflict with any granted lock? */
++              if ((htree_lock_compat[tmp->lk_mode] & granted) != granted)
++                      break;
++
++              list_del_init(&tmp->lk_blocked_list);
++
++              BUG_ON(lhead->lh_nblocked[tmp->lk_mode] == 0);
++
++              lhead->lh_nblocked[tmp->lk_mode]--;
++              lhead->lh_ngranted[tmp->lk_mode]++;
++              granted |= 1 << tmp->lk_mode;
++
++              BUG_ON(tmp->lk_task == NULL);
++              wake_up_process(tmp->lk_task);
++      }
++}
++
++/* it's wrapper of htree_lock_internal and exported interface.
++ * It always return 1 with granted lock if @wait is true, it can return 0
++ * if @wait is false and locking request can't be granted immediately */
++int
++htree_lock_try(struct htree_lock *lck, struct htree_lock_head *lhead,
++             htree_lock_mode_t mode, int wait)
++{
++      int     rc;
++
++      BUG_ON(lck->lk_depth > lhead->lh_depth);
++      BUG_ON(lck->lk_head != NULL);
++      BUG_ON(lck->lk_task != NULL);
++
++      lck->lk_head = lhead;
++      lck->lk_mode = mode;
++
++      htree_spin_lock(lhead, HTREE_DEP_ROOT);
++      rc = htree_lock_internal(lck, wait);
++      if (rc != 0)
++              htree_spin_unlock(lhead, HTREE_DEP_ROOT);
++      return rc >= 0;
++}
++EXPORT_SYMBOL(htree_lock_try);
++
++/* it's wrapper of htree_unlock_internal and exported interface.
++ * It will release all htree_node_locks and htree_lock */
++void
++htree_unlock(struct htree_lock *lck)
++{
++      BUG_ON(lck->lk_head == NULL);
++      BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL);
++
++      htree_node_release_all(lck);
++
++      htree_spin_lock(lck->lk_head, HTREE_DEP_ROOT);
++      htree_unlock_internal(lck);
++      htree_spin_unlock(lck->lk_head, HTREE_DEP_ROOT);
++      lck->lk_head = NULL;
++      lck->lk_task = NULL;
++}
++EXPORT_SYMBOL(htree_unlock);
++
++/* change lock mode */
++void
++htree_change_mode(struct htree_lock *lck, htree_lock_mode_t mode)
++{
++      BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL);
++      lck->lk_mode = mode;
++}
++EXPORT_SYMBOL(htree_change_mode);
++
++/* release htree lock, and lock it again with new mode.
++ * This function will first release all htree_node_locks and htree_lock,
++ * then try to gain htree_lock with new @mode.
++ * It always return 1 with granted lock if @wait is true, it can return 0
++ * if @wait is false and locking request can't be granted immediately */
++int
++htree_change_lock_try(struct htree_lock *lck, htree_lock_mode_t mode, int wait)
++{
++      struct htree_lock_head *lhead = lck->lk_head;
++      int rc;
++
++      BUG_ON(lhead == NULL);
++      BUG_ON(lck->lk_mode == mode);
++      BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL || mode == HTREE_LOCK_INVAL);
++
++      htree_node_release_all(lck);
++
++      htree_spin_lock(lhead, HTREE_DEP_ROOT);
++      htree_unlock_internal(lck);
++      lck->lk_mode = mode;
++      rc = htree_lock_internal(lck, wait);
++      if (rc != 0)
++              htree_spin_unlock(lhead, HTREE_DEP_ROOT);
++      return rc >= 0;
++}
++EXPORT_SYMBOL(htree_change_lock_try);
++
++/* create a htree_lock head with @depth levels (number of child-locks),
++ * it is a per resoruce structure */
++struct htree_lock_head *
++htree_lock_head_alloc(unsigned depth, unsigned hbits, unsigned priv)
++{
++      struct htree_lock_head *lhead;
++      int  i;
++
++      if (depth > HTREE_LOCK_DEP_MAX) {
++              printk(KERN_ERR "%d is larger than max htree_lock depth %d\n",
++                      depth, HTREE_LOCK_DEP_MAX);
++              return NULL;
++      }
++
++      lhead = kzalloc(offsetof(struct htree_lock_head,
++                               lh_children[depth]) + priv, GFP_NOFS);
++      if (lhead == NULL)
++              return NULL;
++
++      if (hbits < HTREE_HBITS_MIN)
++              lhead->lh_hbits = HTREE_HBITS_MIN;
++      else if (hbits > HTREE_HBITS_MAX)
++              lhead->lh_hbits = HTREE_HBITS_MAX;
++
++      lhead->lh_lock = 0;
++      lhead->lh_depth = depth;
++      INIT_LIST_HEAD(&lhead->lh_blocked_list);
++      if (priv > 0) {
++              lhead->lh_private = (void *)lhead +
++                      offsetof(struct htree_lock_head, lh_children[depth]);
++      }
++
++      for (i = 0; i < depth; i++) {
++              INIT_LIST_HEAD(&lhead->lh_children[i].lc_list);
++              lhead->lh_children[i].lc_events = HTREE_EVENT_DISABLE;
++      }
++      return lhead;
++}
++EXPORT_SYMBOL(htree_lock_head_alloc);
++
++/* free the htree_lock head */
++void
++htree_lock_head_free(struct htree_lock_head *lhead)
++{
++      int     i;
++
++      BUG_ON(!list_empty(&lhead->lh_blocked_list));
++      for (i = 0; i < lhead->lh_depth; i++)
++              BUG_ON(!list_empty(&lhead->lh_children[i].lc_list));
++      kfree(lhead);
++}
++EXPORT_SYMBOL(htree_lock_head_free);
++
++/* register event callback for @events of child-lock at level @dep */
++void
++htree_lock_event_attach(struct htree_lock_head *lhead, unsigned dep,
++                      unsigned events, htree_event_cb_t callback)
++{
++      BUG_ON(lhead->lh_depth <= dep);
++      lhead->lh_children[dep].lc_events = events;
++      lhead->lh_children[dep].lc_callback = callback;
++}
++EXPORT_SYMBOL(htree_lock_event_attach);
++
++/* allocate a htree_lock, which is per-thread structure, @pbytes is some
++ * extra-bytes as private data for caller */
++struct htree_lock *
++htree_lock_alloc(unsigned depth, unsigned pbytes)
++{
++      struct htree_lock *lck;
++      int i = offsetof(struct htree_lock, lk_nodes[depth]);
++
++      if (depth > HTREE_LOCK_DEP_MAX) {
++              printk(KERN_ERR "%d is larger than max htree_lock depth %d\n",
++                      depth, HTREE_LOCK_DEP_MAX);
++              return NULL;
++      }
++      lck = kzalloc(i + pbytes, GFP_NOFS);
++      if (lck == NULL)
++              return NULL;
++
++      if (pbytes != 0)
++              lck->lk_private = (void *)lck + i;
++      lck->lk_mode = HTREE_LOCK_INVAL;
++      lck->lk_depth = depth;
++      INIT_LIST_HEAD(&lck->lk_blocked_list);
++
++      for (i = 0; i < depth; i++) {
++              struct htree_lock_node *node = &lck->lk_nodes[i];
++
++              node->ln_mode = HTREE_LOCK_INVAL;
++              INIT_LIST_HEAD(&node->ln_major_list);
++              INIT_LIST_HEAD(&node->ln_minor_list);
++              INIT_LIST_HEAD(&node->ln_alive_list);
++              INIT_LIST_HEAD(&node->ln_blocked_list);
++              INIT_LIST_HEAD(&node->ln_granted_list);
++      }
++
++      return lck;
++}
++EXPORT_SYMBOL(htree_lock_alloc);
++
++/* free htree_lock node */
++void
++htree_lock_free(struct htree_lock *lck)
++{
++      BUG_ON(lck->lk_mode != HTREE_LOCK_INVAL);
++      kfree(lck);
++}
++EXPORT_SYMBOL(htree_lock_free);
+Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/Makefile
+===================================================================
+--- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/Makefile
++++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/Makefile
+@@ -6,6 +6,7 @@ obj-$(CONFIG_EXT4_FS) += ext4.o
+ ext4-y        := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \
+               ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \
++              htree_lock.o \
+               ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \
+               mmp.o indirect.o extents_status.o xattr.o xattr_user.o \
+               xattr_trusted.o inline.o
+Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/ext4.h
++++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/ext4.h
+@@ -27,6 +27,7 @@
+ #include <linux/mutex.h>
+ #include <linux/timer.h>
+ #include <linux/wait.h>
++#include <linux/htree_lock.h>
+ #include <linux/blockgroup_lock.h>
+ #include <linux/percpu_counter.h>
+ #include <linux/ratelimit.h>
+@@ -821,6 +822,9 @@ struct ext4_inode_info {
+       __u32   i_dtime;
+       ext4_fsblk_t    i_file_acl;
++      /* following fields for parallel directory operations -bzzz */
++      struct semaphore i_append_sem;
++
+       /*
+        * i_block_group is the number of the block group which contains
+        * this file's inode.  Constant across the lifetime of the inode,
+@@ -1846,6 +1850,71 @@ struct dx_hash_info
+  */
+ #define HASH_NB_ALWAYS                1
++/* assume name-hash is protected by upper layer */
++#define EXT4_HTREE_LOCK_HASH  0
++
++enum ext4_pdo_lk_types {
++#if EXT4_HTREE_LOCK_HASH
++      EXT4_LK_HASH,
++#endif
++      EXT4_LK_DX,             /* index block */
++      EXT4_LK_DE,             /* directory entry block */
++      EXT4_LK_SPIN,           /* spinlock */
++      EXT4_LK_MAX,
++};
++
++/* read-only bit */
++#define EXT4_LB_RO(b)         (1 << (b))
++/* read + write, high bits for writer */
++#define EXT4_LB_RW(b)         ((1 << (b)) | (1 << (EXT4_LK_MAX + (b))))
++
++enum ext4_pdo_lock_bits {
++      /* DX lock bits */
++      EXT4_LB_DX_RO           = EXT4_LB_RO(EXT4_LK_DX),
++      EXT4_LB_DX              = EXT4_LB_RW(EXT4_LK_DX),
++      /* DE lock bits */
++      EXT4_LB_DE_RO           = EXT4_LB_RO(EXT4_LK_DE),
++      EXT4_LB_DE              = EXT4_LB_RW(EXT4_LK_DE),
++      /* DX spinlock bits */
++      EXT4_LB_SPIN_RO         = EXT4_LB_RO(EXT4_LK_SPIN),
++      EXT4_LB_SPIN            = EXT4_LB_RW(EXT4_LK_SPIN),
++      /* accurate searching */
++      EXT4_LB_EXACT           = EXT4_LB_RO(EXT4_LK_MAX << 1),
++};
++
++enum ext4_pdo_lock_opc {
++      /* external */
++      EXT4_HLOCK_READDIR      = (EXT4_LB_DE_RO | EXT4_LB_DX_RO),
++      EXT4_HLOCK_LOOKUP       = (EXT4_LB_DE_RO | EXT4_LB_SPIN_RO |
++                                 EXT4_LB_EXACT),
++      EXT4_HLOCK_DEL          = (EXT4_LB_DE | EXT4_LB_SPIN_RO |
++                                 EXT4_LB_EXACT),
++      EXT4_HLOCK_ADD          = (EXT4_LB_DE | EXT4_LB_SPIN_RO),
++
++      /* internal */
++      EXT4_HLOCK_LOOKUP_SAFE  = (EXT4_LB_DE_RO | EXT4_LB_DX_RO |
++                                 EXT4_LB_EXACT),
++      EXT4_HLOCK_DEL_SAFE     = (EXT4_LB_DE | EXT4_LB_DX_RO | EXT4_LB_EXACT),
++      EXT4_HLOCK_SPLIT        = (EXT4_LB_DE | EXT4_LB_DX | EXT4_LB_SPIN),
++};
++
++extern struct htree_lock_head *ext4_htree_lock_head_alloc(unsigned hbits);
++#define ext4_htree_lock_head_free(lhead)      htree_lock_head_free(lhead)
++
++extern struct htree_lock *ext4_htree_lock_alloc(void);
++#define ext4_htree_lock_free(lck)             htree_lock_free(lck)
++
++extern void ext4_htree_lock(struct htree_lock *lck,
++                          struct htree_lock_head *lhead,
++                          struct inode *dir, unsigned flags);
++#define ext4_htree_unlock(lck)                  htree_unlock(lck)
++
++extern struct buffer_head *__ext4_find_entry(struct inode *dir,
++                                      const struct qstr *d_name,
++                                      struct ext4_dir_entry_2 **res_dir,
++                                      int *inlined, struct htree_lock *lck);
++extern int __ext4_add_entry(handle_t *handle, struct dentry *dentry,
++                    struct inode *inode, struct htree_lock *lck);
+ /*
+  * Describe an inode's exact location on disk and in memory
+@@ -2088,8 +2157,16 @@ void ext4_insert_dentry(struct inode *in
+                       const char *name, int namelen, void *data);
+ static inline void ext4_update_dx_flag(struct inode *inode)
+ {
++      /* Disable it for ldiskfs, because going from a DX directory to
++       * a non-DX directory while it is in use will completely break
++       * the htree-locking.
++       * If we really want to support this operation in the future,
++       * we need to exclusively lock the directory at here which will
++       * increase complexity of code */
++#if 0
+       if (!ext4_has_feature_dir_index(inode->i_sb))
+               ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
++#endif
+ }
+ static unsigned char ext4_filetype_table[] = {
+       DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
+Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/namei.c
+===================================================================
+--- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/namei.c
++++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/namei.c
+@@ -53,6 +53,7 @@ struct buffer_head *ext4_append(handle_t
+                                       ext4_lblk_t *block)
+ {
+       struct buffer_head *bh;
++      struct ext4_inode_info *ei = EXT4_I(inode);
+       int err;
+       if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
+@@ -60,15 +61,22 @@ struct buffer_head *ext4_append(handle_t
+                     EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
+               return ERR_PTR(-ENOSPC);
++      /* with parallel dir operations all appends
++      * have to be serialized -bzzz */
++      down(&ei->i_append_sem);
++
+       *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
+       bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
+-      if (IS_ERR(bh))
++      if (IS_ERR(bh)) {
++              up(&ei->i_append_sem);
+               return bh;
++      }
+       inode->i_size += inode->i_sb->s_blocksize;
+       EXT4_I(inode)->i_disksize = inode->i_size;
+       BUFFER_TRACE(bh, "get_write_access");
+       err = ext4_journal_get_write_access(handle, bh);
++      up(&ei->i_append_sem);
+       if (err) {
+               brelse(bh);
+               ext4_std_error(inode->i_sb, err);
+@@ -246,7 +254,8 @@ static struct dx_frame *dx_probe(const s
+                                struct inode *dir,
+                                struct dx_hash_info *hinfo,
+-                               struct dx_frame *frame);
++                               struct dx_frame *frame,
++                               struct htree_lock *lck);
+ static void dx_release(struct dx_frame *frames);
+ static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
+                      struct dx_hash_info *hinfo, struct dx_map_entry map[]);
+@@ -259,12 +267,13 @@ static void dx_insert_block(struct dx_fr
+ static int ext4_htree_next_block(struct inode *dir, __u32 hash,
+                                struct dx_frame *frame,
+                                struct dx_frame *frames,
+-                               __u32 *start_hash);
++                               __u32 *start_hash, struct htree_lock *lck);
+ static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
+               struct ext4_filename *fname,
+-              struct ext4_dir_entry_2 **res_dir);
++              struct ext4_dir_entry_2 **res_dir, struct htree_lock *lck);
+ static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
+-                           struct dentry *dentry, struct inode *inode);
++                           struct dentry *dentry, struct inode *inode,
++                           struct htree_lock *lck);
+ /* checksumming functions */
+ void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
+@@ -668,6 +676,227 @@ struct stats dx_show_entries(struct dx_h
+ }
+ #endif /* DX_DEBUG */
++/* private data for htree_lock */
++struct ext4_dir_lock_data {
++      unsigned                ld_flags;  /* bits-map for lock types */
++      unsigned                ld_count;  /* # entries of the last DX block */
++      struct dx_entry         ld_at_entry; /* copy of leaf dx_entry */
++      struct dx_entry         *ld_at;    /* position of leaf dx_entry */
++};
++
++#define ext4_htree_lock_data(l)       ((struct ext4_dir_lock_data *)(l)->lk_private)
++#define ext4_find_entry(dir, name, dirent, inline) \
++                      __ext4_find_entry(dir, name, dirent, inline, NULL)
++#define ext4_add_entry(handle, dentry, inode) \
++                      __ext4_add_entry(handle, dentry, inode, NULL)
++
++/* NB: ext4_lblk_t is 32 bits so we use high bits to identify invalid blk */
++#define EXT4_HTREE_NODE_CHANGED       (0xcafeULL << 32)
++
++static void ext4_htree_event_cb(void *target, void *event)
++{
++      u64 *block = (u64 *)target;
++
++      if (*block == dx_get_block((struct dx_entry *)event))
++              *block = EXT4_HTREE_NODE_CHANGED;
++}
++
++struct htree_lock_head *ext4_htree_lock_head_alloc(unsigned hbits)
++{
++      struct htree_lock_head *lhead;
++
++      lhead = htree_lock_head_alloc(EXT4_LK_MAX, hbits, 0);
++      if (lhead != NULL) {
++              htree_lock_event_attach(lhead, EXT4_LK_SPIN, HTREE_EVENT_WR,
++                                      ext4_htree_event_cb);
++      }
++      return lhead;
++}
++EXPORT_SYMBOL(ext4_htree_lock_head_alloc);
++
++struct htree_lock *ext4_htree_lock_alloc(void)
++{
++      return htree_lock_alloc(EXT4_LK_MAX,
++                              sizeof(struct ext4_dir_lock_data));
++}
++EXPORT_SYMBOL(ext4_htree_lock_alloc);
++
++static htree_lock_mode_t ext4_htree_mode(unsigned flags)
++{
++      switch (flags) {
++      default: /* 0 or unknown flags require EX lock */
++              return HTREE_LOCK_EX;
++      case EXT4_HLOCK_READDIR:
++              return HTREE_LOCK_PR;
++      case EXT4_HLOCK_LOOKUP:
++              return HTREE_LOCK_CR;
++      case EXT4_HLOCK_DEL:
++      case EXT4_HLOCK_ADD:
++              return HTREE_LOCK_CW;
++      }
++}
++
++/* return PR for read-only operations, otherwise return EX */
++static inline htree_lock_mode_t ext4_htree_safe_mode(unsigned flags)
++{
++      int writer = (flags & EXT4_LB_DE) == EXT4_LB_DE;
++
++      /* 0 requires EX lock */
++      return (flags == 0 || writer) ? HTREE_LOCK_EX : HTREE_LOCK_PR;
++}
++
++static int ext4_htree_safe_locked(struct htree_lock *lck)
++{
++      int writer;
++
++      if (lck == NULL || lck->lk_mode == HTREE_LOCK_EX)
++              return 1;
++
++      writer = (ext4_htree_lock_data(lck)->ld_flags & EXT4_LB_DE) ==
++               EXT4_LB_DE;
++      if (writer) /* all readers & writers are excluded? */
++              return lck->lk_mode == HTREE_LOCK_EX;
++
++      /* all writers are excluded? */
++      return lck->lk_mode == HTREE_LOCK_PR ||
++             lck->lk_mode == HTREE_LOCK_PW ||
++             lck->lk_mode == HTREE_LOCK_EX;
++}
++
++/* relock htree_lock with EX mode if it's change operation, otherwise
++ * relock it with PR mode. It's noop if PDO is disabled. */
++static void ext4_htree_safe_relock(struct htree_lock *lck)
++{
++      if (!ext4_htree_safe_locked(lck)) {
++              unsigned flags = ext4_htree_lock_data(lck)->ld_flags;
++
++              htree_change_lock(lck, ext4_htree_safe_mode(flags));
++      }
++}
++
++void ext4_htree_lock(struct htree_lock *lck, struct htree_lock_head *lhead,
++                   struct inode *dir, unsigned flags)
++{
++      htree_lock_mode_t mode = is_dx(dir) ? ext4_htree_mode(flags) :
++                                            ext4_htree_safe_mode(flags);
++
++      ext4_htree_lock_data(lck)->ld_flags = flags;
++      htree_lock(lck, lhead, mode);
++      if (!is_dx(dir))
++              ext4_htree_safe_relock(lck); /* make sure it's safe locked */
++}
++EXPORT_SYMBOL(ext4_htree_lock);
++
++static int ext4_htree_node_lock(struct htree_lock *lck, struct dx_entry *at,
++                              unsigned lmask, int wait, void *ev)
++{
++      u32     key = (at == NULL) ? 0 : dx_get_block(at);
++      u32     mode;
++
++      /* NOOP if htree is well protected or caller doesn't require the lock */
++      if (ext4_htree_safe_locked(lck) ||
++         !(ext4_htree_lock_data(lck)->ld_flags & lmask))
++              return 1;
++
++      mode = (ext4_htree_lock_data(lck)->ld_flags & lmask) == lmask ?
++              HTREE_LOCK_PW : HTREE_LOCK_PR;
++      while (1) {
++              if (htree_node_lock_try(lck, mode, key, ffz(~lmask), wait, ev))
++                      return 1;
++              if (!(lmask & EXT4_LB_SPIN)) /* not a spinlock */
++                      return 0;
++              cpu_relax(); /* spin until granted */
++      }
++}
++
++static int ext4_htree_node_locked(struct htree_lock *lck, unsigned lmask)
++{
++      return ext4_htree_safe_locked(lck) ||
++             htree_node_is_granted(lck, ffz(~lmask));
++}
++
++static void ext4_htree_node_unlock(struct htree_lock *lck,
++                                 unsigned lmask, void *buf)
++{
++      /* NB: it's safe to call mutiple times or even it's not locked */
++      if (!ext4_htree_safe_locked(lck) &&
++           htree_node_is_granted(lck, ffz(~lmask)))
++              htree_node_unlock(lck, ffz(~lmask), buf);
++}
++
++#define ext4_htree_dx_lock(lck, key)          \
++      ext4_htree_node_lock(lck, key, EXT4_LB_DX, 1, NULL)
++#define ext4_htree_dx_lock_try(lck, key)      \
++      ext4_htree_node_lock(lck, key, EXT4_LB_DX, 0, NULL)
++#define ext4_htree_dx_unlock(lck)             \
++      ext4_htree_node_unlock(lck, EXT4_LB_DX, NULL)
++#define ext4_htree_dx_locked(lck)             \
++      ext4_htree_node_locked(lck, EXT4_LB_DX)
++
++static void ext4_htree_dx_need_lock(struct htree_lock *lck)
++{
++      struct ext4_dir_lock_data *ld;
++
++      if (ext4_htree_safe_locked(lck))
++              return;
++
++      ld = ext4_htree_lock_data(lck);
++      switch (ld->ld_flags) {
++      default:
++              return;
++      case EXT4_HLOCK_LOOKUP:
++              ld->ld_flags = EXT4_HLOCK_LOOKUP_SAFE;
++              return;
++      case EXT4_HLOCK_DEL:
++              ld->ld_flags = EXT4_HLOCK_DEL_SAFE;
++              return;
++      case EXT4_HLOCK_ADD:
++              ld->ld_flags = EXT4_HLOCK_SPLIT;
++              return;
++      }
++}
++
++#define ext4_htree_de_lock(lck, key)          \
++      ext4_htree_node_lock(lck, key, EXT4_LB_DE, 1, NULL)
++#define ext4_htree_de_unlock(lck)             \
++      ext4_htree_node_unlock(lck, EXT4_LB_DE, NULL)
++
++#define ext4_htree_spin_lock(lck, key, event) \
++      ext4_htree_node_lock(lck, key, EXT4_LB_SPIN, 0, event)
++#define ext4_htree_spin_unlock(lck)           \
++      ext4_htree_node_unlock(lck, EXT4_LB_SPIN, NULL)
++#define ext4_htree_spin_unlock_listen(lck, p) \
++      ext4_htree_node_unlock(lck, EXT4_LB_SPIN, p)
++
++static void ext4_htree_spin_stop_listen(struct htree_lock *lck)
++{
++      if (!ext4_htree_safe_locked(lck) &&
++          htree_node_is_listening(lck, ffz(~EXT4_LB_SPIN)))
++              htree_node_stop_listen(lck, ffz(~EXT4_LB_SPIN));
++}
++
++enum {
++      DX_HASH_COL_IGNORE,     /* ignore collision while probing frames */
++      DX_HASH_COL_YES,        /* there is collision and it does matter */
++      DX_HASH_COL_NO,         /* there is no collision */
++};
++
++static int dx_probe_hash_collision(struct htree_lock *lck,
++                                 struct dx_entry *entries,
++                                 struct dx_entry *at, u32 hash)
++{
++      if (!(lck && ext4_htree_lock_data(lck)->ld_flags & EXT4_LB_EXACT)) {
++              return DX_HASH_COL_IGNORE; /* don't care about collision */
++
++      } else if (at == entries + dx_get_count(entries) - 1) {
++              return DX_HASH_COL_IGNORE; /* not in any leaf of this DX */
++
++      } else { /* hash collision? */
++              return ((dx_get_hash(at + 1) & ~1) == hash) ?
++                      DX_HASH_COL_YES : DX_HASH_COL_NO;
++      }
++}
++
+ /*
+  * Probe for a directory leaf block to search.
+  *
+@@ -679,10 +908,11 @@ struct stats dx_show_entries(struct dx_h
+  */
+ static struct dx_frame *
+ dx_probe(struct ext4_filename *fname, struct inode *dir,
+-       struct dx_hash_info *hinfo, struct dx_frame *frame_in)
++       struct dx_hash_info *hinfo, struct dx_frame *frame_in,
++       struct htree_lock *lck)
+ {
+       unsigned count, indirect;
+-      struct dx_entry *at, *entries, *p, *q, *m;
++      struct dx_entry *at, *entries, *p, *q, *m, *dx = NULL;
+       struct dx_root_info *info;
+       struct dx_frame *frame = frame_in;
+       struct buffer_head *bh;
+@@ -750,8 +980,15 @@ dx_probe(const struct qstr *d_name, stru
+       dxtrace(printk("Look up %x", hash));
+       while (1) {
++              if (indirect == 0) { /* the last index level */
++                      /* NB: ext4_htree_dx_lock() could be noop if
++                       * DX-lock flag is not set for current operation */
++                      ext4_htree_dx_lock(lck, dx);
++                      ext4_htree_spin_lock(lck, dx, NULL);
++              }
+               count = dx_get_count(entries);
+-              if (!count || count > dx_get_limit(entries)) {
++              if (count == 0 || count > dx_get_limit(entries)) {
++                      ext4_htree_spin_unlock(lck); /* release spin */
+                       ext4_warning_inode(dir,
+                                    "dx entry: no count or count > limit");
+                       brelse(bh);
+@@ -792,8 +1029,70 @@ dx_probe(const struct qstr *d_name, stru
+               frame->bh = bh;
+               frame->entries = entries;
+               frame->at = at;
+-              if (!indirect--)
+-                      return frame;
++
++              if (indirect == 0) { /* the last index level */
++                      struct ext4_dir_lock_data *ld;
++                      u64 myblock;
++
++                      /* By default we only lock DE-block, however, we will
++                       * also lock the last level DX-block if:
++                       * a) there is hash collision
++                       *    we will set DX-lock flag (a few lines below)
++                       *    and redo to lock DX-block
++                       *    see detail in dx_probe_hash_collision()
++                       * b) it's a retry from splitting
++                       *    we need to lock the last level DX-block so nobody
++                       *    else can split any leaf blocks under the same
++                       *    DX-block, see detail in ext4_dx_add_entry()
++                       */
++                      if (ext4_htree_dx_locked(lck)) {
++                              /* DX-block is locked, just lock DE-block
++                               * and return */
++                              ext4_htree_spin_unlock(lck);
++                              if (!ext4_htree_safe_locked(lck))
++                                      ext4_htree_de_lock(lck, frame->at);
++                              return frame;
++                      }
++                      /* it's pdirop and no DX lock */
++                      if (dx_probe_hash_collision(lck, entries, at, hash) ==
++                          DX_HASH_COL_YES) {
++                              /* found hash collision, set DX-lock flag
++                               * and retry to abtain DX-lock */
++                              ext4_htree_spin_unlock(lck);
++                              ext4_htree_dx_need_lock(lck);
++                              continue;
++                      }
++                      ld = ext4_htree_lock_data(lck);
++                      /* because I don't lock DX, so @at can't be trusted
++                       * after I release spinlock so I have to save it */
++                      ld->ld_at = at;
++                      ld->ld_at_entry = *at;
++                      ld->ld_count = dx_get_count(entries);
++
++                      frame->at = &ld->ld_at_entry;
++                      myblock = dx_get_block(at);
++
++                      /* NB: ordering locking */
++                      ext4_htree_spin_unlock_listen(lck, &myblock);
++                      /* other thread can split this DE-block because:
++                       * a) I don't have lock for the DE-block yet
++                       * b) I released spinlock on DX-block
++                       * if it happened I can detect it by listening
++                       * splitting event on this DE-block */
++                      ext4_htree_de_lock(lck, frame->at);
++                      ext4_htree_spin_stop_listen(lck);
++
++                      if (myblock == EXT4_HTREE_NODE_CHANGED) {
++                              /* someone split this DE-block before
++                               * I locked it, I need to retry and lock
++                               * valid DE-block */
++                              ext4_htree_de_unlock(lck);
++                              continue;
++                      }
++                      return frame;
++              }
++              dx = at;
++              indirect--;
+               frame++;
+               frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
+                       *err = PTR_ERR(bh);
+@@ -860,7 +1160,7 @@ static void dx_release (struct dx_frame
+ static int ext4_htree_next_block(struct inode *dir, __u32 hash,
+                                struct dx_frame *frame,
+                                struct dx_frame *frames,
+-                               __u32 *start_hash)
++                               __u32 *start_hash, struct htree_lock *lck)
+ {
+       struct dx_frame *p;
+       struct buffer_head *bh;
+@@ -875,12 +1175,22 @@ static int ext4_htree_next_block(struct
+        * this loop, num_frames indicates the number of interior
+        * nodes need to be read.
+        */
++      ext4_htree_de_unlock(lck);
+       while (1) {
+-              if (++(p->at) < p->entries + dx_get_count(p->entries))
+-                      break;
++              if (num_frames > 0 || ext4_htree_dx_locked(lck)) {
++                      /* num_frames > 0 :
++                       *   DX block
++                       * ext4_htree_dx_locked:
++                       *   frame->at is reliable pointer returned by dx_probe,
++                       *   otherwise dx_probe already knew no collision */
++                      if (++(p->at) < p->entries + dx_get_count(p->entries))
++                              break;
++              }
+               if (p == frames)
+                       return 0;
+               num_frames++;
++              if (num_frames == 1)
++                      ext4_htree_dx_unlock(lck);
+               p--;
+       }
+@@ -903,6 +1213,13 @@ static int ext4_htree_next_block(struct
+        * block so no check is necessary
+        */
+       while (num_frames--) {
++              if (num_frames == 0) {
++                      /* it's not always necessary, we just don't want to
++                       * detect hash collision again */
++                      ext4_htree_dx_need_lock(lck);
++                      ext4_htree_dx_lock(lck, p->at);
++              }
++
+               bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
+               if (IS_ERR(bh))
+                       return PTR_ERR(bh);
+@@ -911,6 +1228,7 @@ static int ext4_htree_next_block(struct
+               p->bh = bh;
+               p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
+       }
++      ext4_htree_de_lock(lck, p->at);
+       return 1;
+ }
+@@ -1013,10 +1331,10 @@ int ext4_htree_fill_tree(struct file *di
+       }
+       hinfo.hash = start_hash;
+       hinfo.minor_hash = 0;
+-      frame = dx_probe(NULL, dir, &hinfo, frames);
++      /* assume it's PR locked */
++      frame = dx_probe(NULL, dir, &hinfo, frames, NULL);
+       if (IS_ERR(frame))
+               return PTR_ERR(frame);
+-
+       /* Add '.' and '..' from the htree header */
+       if (!start_hash && !start_minor_hash) {
+               de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
+@@ -1043,7 +1361,7 @@ int ext4_htree_fill_tree(struct file *di
+               count += ret;
+               hashval = ~0;
+               ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
+-                                          frame, frames, &hashval);
++                                          frame, frames, &hashval, NULL);
+               *next_hash = hashval;
+               if (ret < 0) {
+                       err = ret;
+@@ -1236,10 +1554,10 @@ static int is_dx_internal_node(struct in
+  * The returned buffer_head has ->b_count elevated.  The caller is expected
+  * to brelse() it when appropriate.
+  */
+-static struct buffer_head * ext4_find_entry (struct inode *dir,
++struct buffer_head *__ext4_find_entry(struct inode *dir,
+                                       const struct qstr *d_name,
+                                       struct ext4_dir_entry_2 **res_dir,
+-                                      int *inlined)
++                                      int *inlined, struct htree_lock *lck)
+ {
+       struct super_block *sb;
+       struct buffer_head *bh_use[NAMEI_RA_SIZE];
+@@ -1283,7 +1601,7 @@ static struct buffer_head * ext4_find_en
+               goto restart;
+       }
+       if (is_dx(dir)) {
+-              ret = ext4_dx_find_entry(dir, &fname, res_dir);
++              ret = ext4_dx_find_entry(dir, &fname, res_dir, lck);
+               /*
+                * On success, or if the error was file not found,
+                * return.  Otherwise, fall back to doing a search the
+@@ -1297,6 +1615,7 @@ static struct buffer_head * ext4_find_en
+                       return bh;
+               dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
+                              "falling back\n"));
++              ext4_htree_safe_relock(lck);
+       }
+       nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
+       start = EXT4_I(dir)->i_dir_start_lookup;
+@@ -1389,10 +1708,12 @@ cleanup_and_exit:
+               brelse(bh_use[ra_ptr]);
+       return ret;
+ }
++EXPORT_SYMBOL(__ext4_find_entry);
+ static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
+                       struct ext4_filename *fname,
+-                      struct ext4_dir_entry_2 **res_dir)
++                      struct ext4_dir_entry_2 **res_dir,
++                      struct htree_lock *lck)
+ {
+       struct super_block * sb = dir->i_sb;
+       struct dx_hash_info     hinfo;
+@@ -1400,7 +1722,7 @@ static struct buffer_head * ext4_dx_find
+ #endif
+-      frame = dx_probe(fname, dir, NULL, frames);
++      frame = dx_probe(fname, dir, NULL, frames, lck);
+       if (IS_ERR(frame))
+               return NULL;
+       do {
+@@ -1424,7 +1746,7 @@ static struct buffer_head * ext4_dx_find
+               /* Check to see if we should continue to search */
+               retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
+-                                             frames, NULL);
++                                             frames, NULL, lck);
+               if (retval < 0) {
+                       ext4_warning(sb,
+                            "error reading index page in directory #%lu",
+@@ -1583,8 +1905,9 @@ static struct ext4_dir_entry_2* dx_pack_
+  * Returns pointer to de in block into which the new entry will be inserted.
+  */
+ static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
+-                      struct buffer_head **bh,struct dx_frame *frame,
+-                      struct dx_hash_info *hinfo)
++                      struct buffer_head **bh, struct dx_frame *frames,
++                      struct dx_frame *frame, struct dx_hash_info *hinfo,
++                      struct htree_lock *lck)
+ {
+       unsigned blocksize = dir->i_sb->s_blocksize;
+       unsigned count, continued;
+@@ -1647,8 +1970,14 @@ static struct ext4_dir_entry_2 *do_split
+                                       hash2, split, count-split));
+       /* Fancy dance to stay within two buffers */
+-      de2 = dx_move_dirents(data1, data2, map + split, count - split,
+-                            blocksize);
++      if (hinfo->hash < hash2) {
++              de2 = dx_move_dirents(data1, data2, map + split,
++                                    count - split, blocksize);
++      } else {
++              /* make sure we will add entry to the same block which
++               * we have already locked */
++              de2 = dx_move_dirents(data1, data2, map, split, blocksize);
++      }
+       de = dx_pack_dirents(data1, blocksize);
+       de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
+                                          (char *) de,
+@@ -1666,12 +1996,21 @@ static struct ext4_dir_entry_2 *do_split
+       dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
+       dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
+-      /* Which block gets the new entry? */
+-      if (hinfo->hash >= hash2) {
+-              swap(*bh, bh2);
+-              de = de2;
++      ext4_htree_spin_lock(lck, frame > frames ? (frame - 1)->at : NULL,
++                           frame->at); /* notify block is being split */
++      if (hinfo->hash < hash2) {
++              dx_insert_block(frame, hash2 + continued, newblock);
++
++      } else {
++              /* switch block number */
++              dx_insert_block(frame, hash2 + continued,
++                              dx_get_block(frame->at));
++              dx_set_block(frame->at, newblock);
++              (frame->at)++;
+       }
+-      dx_insert_block(frame, hash2 + continued, newblock);
++      ext4_htree_spin_unlock(lck);
++      ext4_htree_dx_unlock(lck);
++
+       err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
+       if (err)
+               goto journal_error;
+@@ -1945,7 +2283,7 @@ static int make_indexed_dir(handle_t *ha
+       ext4_handle_dirty_dx_node(handle, dir, frame->bh);
+       ext4_handle_dirty_dirent_node(handle, dir, bh);
+-      de = do_split(handle,dir, &bh, frame, &fname->hinfo);
++      de = do_split(handle, dir, &bh, frames, frame, &fname->hinfo, NULL);
+       if (IS_ERR(de)) {
+               /*
+                * Even if the block split failed, we have to properly write
+@@ -2051,8 +2389,8 @@ out:
+  * may not sleep between calling this and putting something into
+  * the entry, as someone else might have used it while you slept.
+  */
+-static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
+-                        struct inode *inode)
++int __ext4_add_entry(handle_t *handle, struct dentry *dentry,
++                    struct inode *inode, struct htree_lock *lck)
+ {
+       struct inode *dir = dentry->d_parent->d_inode;
+       struct buffer_head *bh;
+@@ -2087,9 +2425,10 @@ static int ext4_add_entry(handle_t *hand
+               if (dentry->d_name.len == 2 &&
+                   memcmp(dentry->d_name.name, "..", 2) == 0)
+                       return ext4_update_dotdot(handle, dentry, inode);
+-              retval = ext4_dx_add_entry(handle, &fname, dentry, inode);
++              retval = ext4_dx_add_entry(handle, &fname, dentry, inode, lck);
+               if (!retval || (retval != ERR_BAD_DX_DIR))
+                       goto out;
++              ext4_htree_safe_relock(lck);
+               ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
+               dx_fallback++;
+               ext4_mark_inode_dirty(handle, dir);
+@@ -2129,12 +2468,14 @@ static int ext4_add_entry(handle_t *hand
+               ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
+       return retval;
+ }
++EXPORT_SYMBOL(__ext4_add_entry);
+ /*
+  * Returns 0 for success, or a negative error value
+  */
+ static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
+-                           struct dentry *dentry, struct inode *inode)
++                           struct dentry *dentry, struct inode *inode,
++                           struct htree_lock *lck)
+ {
+       struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+       struct dx_entry *entries, *at;
+@@ -2148,7 +2488,7 @@ static int ext4_dx_add_entry(handle_t *h
+ again:
+       restart = 0;
+-      frame = dx_probe(fname, dir, NULL, frames);
++      frame = dx_probe(fname, dir, NULL, frames, lck);
+       if (IS_ERR(frame))
+               return PTR_ERR(frame);
+       entries = frame->entries;
+@@ -2178,6 +2518,11 @@ again:
+               struct dx_node *node2;
+               struct buffer_head *bh2;
++              if (!ext4_htree_safe_locked(lck)) { /* retry with EX lock */
++                      ext4_htree_safe_relock(lck);
++                      restart = 1;
++                      goto cleanup;
++              }
+               while (frame > frames) {
+                       if (dx_get_count((frame - 1)->entries) <
+                           dx_get_limit((frame - 1)->entries)) {
+@@ -2277,8 +2622,32 @@ again:
+                       restart = 1;
+                       goto cleanup;
+               }
++      } else if (!ext4_htree_dx_locked(lck)) {
++              struct ext4_dir_lock_data *ld = ext4_htree_lock_data(lck);
++
++              /* not well protected, require DX lock */
++              ext4_htree_dx_need_lock(lck);
++              at = frame > frames ? (frame - 1)->at : NULL;
++
++              /* NB: no risk of deadlock because it's just a try.
++               *
++               * NB: we check ld_count for twice, the first time before
++               * having DX lock, the second time after holding DX lock.
++               *
++               * NB: We never free blocks for directory so far, which
++               * means value returned by dx_get_count() should equal to
++               * ld->ld_count if nobody split any DE-block under @at,
++               * and ld->ld_at still points to valid dx_entry. */
++              if ((ld->ld_count != dx_get_count(entries)) ||
++                  !ext4_htree_dx_lock_try(lck, at) ||
++                  (ld->ld_count != dx_get_count(entries))) {
++                      restart = 1;
++                      goto cleanup;
++              }
++              /* OK, I've got DX lock and nothing changed */
++              frame->at = ld->ld_at;
+       }
+-      de = do_split(handle, dir, &bh, frame, &fname->hinfo);
++      de = do_split(handle, dir, &bh, frames, frame, &fname->hinfo, lck);
+       if (IS_ERR(de)) {
+               err = PTR_ERR(de);
+               goto cleanup;
+@@ -2277,6 +2622,8 @@ again:
+ journal_error:
+       ext4_std_error(dir->i_sb, err);
+ cleanup:
++      ext4_htree_dx_unlock(lck);
++      ext4_htree_de_unlock(lck);
+       brelse(bh);
+       dx_release(frames);
+       /* @restart is true means htree-path has been changed, we need to
+Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/super.c
+===================================================================
+--- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/super.c
++++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/super.c
+@@ -875,6 +875,7 @@ static struct inode *ext4_alloc_inode(st
+       ei->vfs_inode.i_version = 1;
+       spin_lock_init(&ei->i_raw_lock);
++      sema_init(&ei->i_append_sem, 1);
+       INIT_LIST_HEAD(&ei->i_prealloc_list);
+       spin_lock_init(&ei->i_prealloc_lock);
+       ext4_es_init_tree(&ei->i_es_tree);
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-prealloc.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-prealloc.patch
new file mode 100644 (file)
index 0000000..380580d
--- /dev/null
@@ -0,0 +1,375 @@
+Index: linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/ext4.h
++++ linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h
+@@ -1243,11 +1243,14 @@ struct ext4_sb_info {
+       /* tunables */
+       unsigned long s_stripe;
+-      unsigned int s_mb_stream_request;
++      unsigned long s_mb_small_req;
++      unsigned long s_mb_large_req;
+       unsigned int s_mb_max_to_scan;
+       unsigned int s_mb_min_to_scan;
+       unsigned int s_mb_stats;
+       unsigned int s_mb_order2_reqs;
++      unsigned long *s_mb_prealloc_table;
++      unsigned long s_mb_prealloc_table_size;
+       unsigned int s_mb_group_prealloc;
+       unsigned int s_max_writeback_mb_bump;
+       unsigned int s_max_dir_size_kb;
+@@ -2243,6 +1243,7 @@ struct ext4_sb_info {
+ extern void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate);
+ /* mballoc.c */
++extern const struct file_operations ext4_seq_prealloc_table_fops;
+ extern const struct file_operations ext4_seq_mb_groups_fops;
+ extern long ext4_mb_stats;
+ extern long ext4_mb_max_to_scan;
+Index: linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.c
+===================================================================
+--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/mballoc.c
++++ linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.c
+@@ -1828,6 +1828,25 @@ int ext4_mb_find_by_goal(struct ext4_all
+       return 0;
+ }
++static void ext4_mb_prealloc_table_add(struct ext4_sb_info *sbi, int value)
++{
++      int i;
++
++      if (value > (sbi->s_blocks_per_group - 1 - 1 - sbi->s_itb_per_group))
++              return;
++
++      for (i = 0; i < sbi->s_mb_prealloc_table_size; i++) {
++              if (sbi->s_mb_prealloc_table[i] == 0) {
++                      sbi->s_mb_prealloc_table[i] = value;
++                      return;
++              }
++
++              /* they should add values in order */
++              if (value <= sbi->s_mb_prealloc_table[i])
++                      return;
++      }
++}
++
+ /*
+  * The routine scans buddy structures (not bitmap!) from given order
+  * to max order and tries to find big enough chunk to satisfy the req
+@@ -2263,6 +2282,89 @@ static const struct seq_operations ext4_
+       .show   = ext4_mb_seq_groups_show,
+ };
++static ssize_t ext4_mb_prealloc_table_proc_write(struct file *file,
++                                           const char __user *buf,
++                                           size_t cnt, loff_t *pos)
++{
++      struct ext4_sb_info *sbi = EXT4_SB(PDE_DATA(file_inode(file)));
++      unsigned long value;
++      unsigned long prev = 0;
++      char str[128];
++      char *cur;
++      char *end;
++      unsigned long *new_table;
++      int num = 0;
++      int i = 0;
++
++      if (cnt >= sizeof(str))
++              return -EINVAL;
++      if (copy_from_user(str, buf, cnt))
++              return -EFAULT;
++
++      num = 0;
++      cur = str;
++      end = str + cnt;
++      while (cur < end) {
++              int rc;
++              while ((cur < end) && (*cur == ' '))
++                      cur++;
++              rc = kstrtol(cur, 0, &value);
++              if (rc != 0)
++                      return -EINVAL;
++              if (value == 0)
++                      break;
++              if (value <= prev)
++                      return -EINVAL;
++              prev = value;
++              num++;
++      }
++
++      new_table = kmalloc(num * sizeof(*new_table), GFP_KERNEL);
++      if (new_table == NULL)
++              return -ENOMEM;
++      kfree(sbi->s_mb_prealloc_table);
++      memset(new_table, 0, num * sizeof(*new_table));
++      sbi->s_mb_prealloc_table = new_table;
++      sbi->s_mb_prealloc_table_size = num;
++      cur = str;
++      end = str + cnt;
++      while (cur < end && i < num) {
++              while (cur < end && *cur == ' ')
++                      cur++;
++              value = simple_strtol(cur, &cur, 0);
++              ext4_mb_prealloc_table_add(sbi, value);
++              i++;
++      }
++
++      return cnt;
++}
++
++static int mb_prealloc_table_seq_show(struct seq_file *m, void *v)
++{
++      struct ext4_sb_info *sbi = EXT4_SB(m->private);
++      int i;
++
++      for (i = 0; i < sbi->s_mb_prealloc_table_size; i++)
++              seq_printf(m, "%ld ", sbi->s_mb_prealloc_table[i]);
++      seq_printf(m, "\n");
++
++      return 0;
++}
++
++static int mb_prealloc_table_seq_open(struct inode *inode, struct file *file)
++{
++      return single_open(file, mb_prealloc_table_seq_show, PDE_DATA(inode));
++}
++
++const struct file_operations ext4_seq_prealloc_table_fops = {
++      .owner   = THIS_MODULE,
++      .open    = mb_prealloc_table_seq_open,
++      .read    = seq_read,
++      .llseek  = seq_lseek,
++      .release = single_release,
++      .write   = ext4_mb_prealloc_table_proc_write,
++};
++
+ static int ext4_mb_seq_groups_open(struct inode *inode, struct file *file)
+ {
+       struct super_block *sb = PDE_DATA(inode);
+@@ -2557,7 +2656,6 @@ int ext4_mb_init(struct super_block *sb)
+       sbi->s_mb_max_to_scan = MB_DEFAULT_MAX_TO_SCAN;
+       sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
+       sbi->s_mb_stats = MB_DEFAULT_STATS;
+-      sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
+       sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
+       /*
+        * The default group preallocation is 512, which for 4k block
+@@ -2581,9 +2679,48 @@ int ext4_mb_init(struct super_block *sb)
+        * RAID stripe size so that preallocations don't fragment
+        * the stripes.
+        */
+-      if (sbi->s_stripe > 1) {
+-              sbi->s_mb_group_prealloc = roundup(
+-                      sbi->s_mb_group_prealloc, sbi->s_stripe);
++
++      if (sbi->s_stripe == 0) {
++              sbi->s_mb_prealloc_table_size = 10;
++              i = sbi->s_mb_prealloc_table_size * sizeof(unsigned long);
++              sbi->s_mb_prealloc_table = kmalloc(i, GFP_NOFS);
++              if (sbi->s_mb_prealloc_table == NULL) {
++                      ret = -ENOMEM;
++                      goto out;
++              }
++              memset(sbi->s_mb_prealloc_table, 0, i);
++
++              ext4_mb_prealloc_table_add(sbi, 4);
++              ext4_mb_prealloc_table_add(sbi, 8);
++              ext4_mb_prealloc_table_add(sbi, 16);
++              ext4_mb_prealloc_table_add(sbi, 32);
++              ext4_mb_prealloc_table_add(sbi, 64);
++              ext4_mb_prealloc_table_add(sbi, 128);
++              ext4_mb_prealloc_table_add(sbi, 256);
++              ext4_mb_prealloc_table_add(sbi, 512);
++              ext4_mb_prealloc_table_add(sbi, 1024);
++              ext4_mb_prealloc_table_add(sbi, 2048);
++
++              sbi->s_mb_small_req = 256;
++              sbi->s_mb_large_req = 1024;
++              sbi->s_mb_group_prealloc = 512;
++      } else {
++              sbi->s_mb_prealloc_table_size = 3;
++              i = sbi->s_mb_prealloc_table_size * sizeof(unsigned long);
++              sbi->s_mb_prealloc_table = kmalloc(i, GFP_NOFS);
++              if (sbi->s_mb_prealloc_table == NULL) {
++                      ret = -ENOMEM;
++                      goto out;
++              }
++              memset(sbi->s_mb_prealloc_table, 0, i);
++
++              ext4_mb_prealloc_table_add(sbi, sbi->s_stripe);
++              ext4_mb_prealloc_table_add(sbi, sbi->s_stripe * 2);
++              ext4_mb_prealloc_table_add(sbi, sbi->s_stripe * 4);
++
++              sbi->s_mb_small_req = sbi->s_stripe;
++              sbi->s_mb_large_req = sbi->s_stripe * 8;
++              sbi->s_mb_group_prealloc = sbi->s_stripe * 4;
+       }
+       sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
+@@ -2963,9 +3107,9 @@ ext4_mb_normalize_request(struct ext4_al
+                               struct ext4_allocation_request *ar)
+ {
+       struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
+-      int bsbits, max;
++      int bsbits, i, wind;
+       ext4_lblk_t end;
+-      loff_t size, start_off;
++      loff_t size;
+       loff_t orig_size __maybe_unused;
+       ext4_lblk_t start;
+       struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
+@@ -2998,51 +3142,34 @@ ext4_mb_normalize_request(struct ext4_al
+       size = size << bsbits;
+       if (size < i_size_read(ac->ac_inode))
+               size = i_size_read(ac->ac_inode);
+-      orig_size = size;
++      size = (size + ac->ac_sb->s_blocksize - 1) >> bsbits;
+-      /* max size of free chunks */
+-      max = 2 << bsbits;
++      start = wind = 0;
+-#define NRL_CHECK_SIZE(req, size, max, chunk_size)    \
+-              (req <= (size) || max <= (chunk_size))
++      /* let's choose preallocation window depending on file size */
++      for (i = 0; i < sbi->s_mb_prealloc_table_size; i++) {
++              if (size <= sbi->s_mb_prealloc_table[i]) {
++                      wind = sbi->s_mb_prealloc_table[i];
++                      break;
++              }
++      }
++      size = wind;
+-      /* first, try to predict filesize */
+-      /* XXX: should this table be tunable? */
+-      start_off = 0;
+-      if (size <= 16 * 1024) {
+-              size = 16 * 1024;
+-      } else if (size <= 32 * 1024) {
+-              size = 32 * 1024;
+-      } else if (size <= 64 * 1024) {
+-              size = 64 * 1024;
+-      } else if (size <= 128 * 1024) {
+-              size = 128 * 1024;
+-      } else if (size <= 256 * 1024) {
+-              size = 256 * 1024;
+-      } else if (size <= 512 * 1024) {
+-              size = 512 * 1024;
+-      } else if (size <= 1024 * 1024) {
+-              size = 1024 * 1024;
+-      } else if (NRL_CHECK_SIZE(size, 4 * 1024 * 1024, max, 2 * 1024)) {
+-              start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
+-                                              (21 - bsbits)) << 21;
+-              size = 2 * 1024 * 1024;
+-      } else if (NRL_CHECK_SIZE(size, 8 * 1024 * 1024, max, 4 * 1024)) {
+-              start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
+-                                                      (22 - bsbits)) << 22;
+-              size = 4 * 1024 * 1024;
+-      } else if (NRL_CHECK_SIZE(ac->ac_o_ex.fe_len,
+-                                      (8<<20)>>bsbits, max, 8 * 1024)) {
+-              start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
+-                                                      (23 - bsbits)) << 23;
+-              size = 8 * 1024 * 1024;
+-      } else {
+-              start_off = (loff_t) ac->ac_o_ex.fe_logical << bsbits;
+-              size      = (loff_t) EXT4_C2B(EXT4_SB(ac->ac_sb),
+-                                            ac->ac_o_ex.fe_len) << bsbits;
++      if (wind == 0) {
++              __u64 tstart, tend;
++              /* file is quite large, we now preallocate with
++               * the biggest configured window with regart to
++               * logical offset */
++              wind = sbi->s_mb_prealloc_table[i - 1];
++              tstart = ac->ac_o_ex.fe_logical;
++              do_div(tstart, wind);
++              start = tstart * wind;
++              tend = ac->ac_o_ex.fe_logical + ac->ac_o_ex.fe_len - 1;
++              do_div(tend, wind);
++              tend = tend * wind + wind;
++              size = tend - start;
+       }
+-      size = size >> bsbits;
+-      start = start_off >> bsbits;
++      orig_size = size;
+       /* don't cover already allocated blocks in selected range */
+       if (ar->pleft && start <= ar->lleft) {
+@@ -3117,7 +3245,6 @@ ext4_mb_normalize_request(struct ext4_al
+       BUG_ON(start + size <= ac->ac_o_ex.fe_logical &&
+                       start > ac->ac_o_ex.fe_logical);
+       }
+-      BUG_ON(size <= 0 || size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
+       /* now prepare goal request */
+@@ -4056,11 +4183,19 @@ static void ext4_mb_group_or_file(struct
+       /* don't use group allocation for large files */
+       size = max(size, isize);
+-      if (size > sbi->s_mb_stream_request) {
++      if ((ac->ac_o_ex.fe_len >= sbi->s_mb_small_req) ||
++          (size >= sbi->s_mb_large_req)) {
+               ac->ac_flags |= EXT4_MB_STREAM_ALLOC;
+               return;
+       }
++      /*
++       * request is so large that we don't care about
++       * streaming - it overweights any possible seek
++       */
++      if (ac->ac_o_ex.fe_len >= sbi->s_mb_large_req)
++              return;
++
+       BUG_ON(ac->ac_lg != NULL);
+       /*
+        * locality group prealloc space are per cpu. The reason for having
+Index: linux-3.10.0-123.el7.x86_64/fs/ext4/sysfs.c
+===================================================================
+--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/sysfs.c
++++ linux-3.10.0-123.el7.x86_64/fs/ext4/sysfs.c
+@@ -2555,7 +2555,8 @@ EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats
+ EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
+ EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
+ EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
+-EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
++EXT4_RW_ATTR_SBI_UI(mb_small_req, s_mb_small_req);
++EXT4_RW_ATTR_SBI_UI(mb_large_req, s_mb_large_req);
+ EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
+ EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
+ EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
+@@ -2578,7 +2579,8 @@ static struct attribute *ext4_attrs[] =
+       ATTR_LIST(mb_max_to_scan),
+       ATTR_LIST(mb_min_to_scan),
+       ATTR_LIST(mb_order2_req),
+-      ATTR_LIST(mb_stream_req),
++      ATTR_LIST(mb_small_req),
++      ATTR_LIST(mb_large_req),
+       ATTR_LIST(mb_group_prealloc),
+       ATTR_LIST(max_writeback_mb_bump),
+       ATTR_LIST(extent_max_zeroout_kb),
+Index: linux-3.10.0-123.el7.x86_64/fs/ext4/inode.c
+===================================================================
+--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/inode.c
++++ linux-3.10.0-123.el7.x86_64/fs/ext4/inode.c
+@@ -2476,6 +2476,9 @@ static int ext4_da_writepages(struct add
+       if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
+               return -EROFS;
++      if (wbc->nr_to_write < sbi->s_mb_small_req)
++              wbc->nr_to_write = sbi->s_mb_small_req;
++
+       if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
+               range_whole = 1;
+Index: linux-3.10.0-123.el7.x86_64/fs/ext4/sysfs.c
+===================================================================
+--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/sysfs.c
++++ linux-3.10.0-123.el7.x86_64/fs/ext4/sysfs.c
+@@ -1243,6 +1243,7 @@ struct ext4_sb_info {
+       PROC_FILE_LIST(options),
+       PROC_FILE_LIST(es_shrinker_info),
+       PROC_FILE_LIST(mb_groups),
++      PROC_FILE_LIST(prealloc_table),
+       { NULL, NULL },
+ };
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4-print-inum-in-htree-warning.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4-print-inum-in-htree-warning.patch
new file mode 100644 (file)
index 0000000..ba72ec9
--- /dev/null
@@ -0,0 +1,13 @@
+Index: linux-stage/fs/ext4/namei.c
+===================================================================
+--- linux-stage.orig/fs/ext4/namei.c
++++ linux-stage/fs/ext4/namei.c
+@@ -371,6 +371,6 @@ dx_probe(const struct qstr *d_name, stru
+       if (root->info.hash_version != DX_HASH_TEA &&
+           root->info.hash_version != DX_HASH_HALF_MD4 &&
+           root->info.hash_version != DX_HASH_LEGACY) {
+-              ext4_warning_inode(dir, "Unrecognised inode hash code %u",
+-                                 root->info.hash_version);
++              ext4_warning_inode(dir, "Unrecognised inode hash code %u for directory "
++                                 "%lu", root->info.hash_version, dir->i_ino);
+               goto fail;
diff --git a/ldiskfs/kernel_patches/patches/sles12sp2/ext4_s_max_ext_tree_depth.patch b/ldiskfs/kernel_patches/patches/sles12sp2/ext4_s_max_ext_tree_depth.patch
new file mode 100644 (file)
index 0000000..fcd7d26
--- /dev/null
@@ -0,0 +1,121 @@
+Fix ext4_ext_find_extent() to already pre-allocate ext4_ext_path[]
+array of the max depth instead of current depth.
+This will avoid racy cases of concurrent ext_depth() growth in
+current and unsafe implementation with ext4_ext_path[] array
+re-[sizing,allocation], even with more recent and related patches
+that will be integrated in more recent Kernels.
+
+Index: linux-2.6.32-504.el6.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-2.6.32-504.el6.x86_64.orig/fs/ext4/ext4.h
++++ linux-2.6.32-504.el6.x86_64/fs/ext4/ext4.h
+@@ -1147,6 +1147,9 @@
+       unsigned long s_ext_extents;
+ #endif
++      /* maximum possible extents tree depth, to be computed at mount time */
++      unsigned int s_max_ext_tree_depth;
++
+       /* for buddy allocator */
+       struct ext4_group_info ***s_group_info;
+       struct inode *s_buddy_cache;
+Index: linux-2.6.32-504.el6.x86_64/fs/ext4/super.c
+===================================================================
+--- linux-2.6.32-504.el6.x86_64.orig/fs/ext4/super.c
++++ linux-2.6.32-504.el6.x86_64/fs/ext4/super.c
+@@ -4038,6 +4038,8 @@
+               if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
+                       goto failed_mount3;
++      ext4_ext_init(sb); /* needed before using extent-mapped journal */
++
+       /*
+        * The first inode we look at is the journal inode.  Don't try
+        * root first: it may be modified in the journal!
+@@ -4200,7 +4202,6 @@
+               goto failed_mount4a;
+       }
+-      ext4_ext_init(sb);
+       err = ext4_mb_init(sb);
+       if (err) {
+               ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
+Index: linux-2.6.32-504.el6.x86_64/fs/ext4/extents.c
+===================================================================
+--- linux-2.6.32-504.el6.x86_64.orig/fs/ext4/extents.c
++++ linux-2.6.32-504.el6.x86_64/fs/ext4/extents.c
+@@ -699,8 +699,9 @@
+       if (!path) {
+               /* account possible depth increase */
+-              path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
+-                              GFP_NOFS);
++              path = kzalloc(sizeof(struct ext4_ext_path) *
++                             EXT4_SB(inode->i_sb)->s_max_ext_tree_depth,
++                             GFP_NOFS);
+               if (unlikely(!path))
+                       return ERR_PTR(-ENOMEM);
+               alloc = 1;
+@@ -2664,8 +2662,9 @@
+                       path[k].p_block =
+                               le16_to_cpu(path[k].p_hdr->eh_entries)+1;
+       } else {
+-              path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
+-                             GFP_NOFS);
++              path = kzalloc(sizeof(struct ext4_ext_path) *
++                             EXT4_SB(inode->i_sb)->s_max_ext_tree_depth,
++                             GFP_NOFS);
+               if (path == NULL) {
+                       ext4_journal_stop(handle);
+                       return -ENOMEM;
+@@ -3048,13 +3034,14 @@
+  */
+ void ext4_ext_init(struct super_block *sb)
+ {
++      ext4_fsblk_t maxblocks;
++
+       /*
+        * possible initialization would be here
+        */
+       if (ext4_has_feature_extents(sb)) {
+-#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
+-              printk(KERN_INFO "EXT4-fs: file extents enabled"
++              printk(KERN_INFO "EXT4-fs (%s): file extents enabled"
+ #ifdef AGGRESSIVE_TEST
+                      ", aggressive tests"
+ #endif
+@@ -3064,8 +3051,31 @@
+ #ifdef EXTENTS_STATS
+                      ", stats"
+ #endif
+-                     "\n");
+-#endif
++                     , sb->s_id);
++              EXT4_SB(sb)->s_max_ext_tree_depth = 1;
++
++              maxblocks = sb->s_maxbytes / sb->s_blocksize;
++
++              /* 1st/root level/node of extents tree stands in i_data and
++               * entries stored in tree nodes can be of type ext4_extent
++               * (leaf node) or ext4_extent_idx (internal node) */
++              maxblocks /= (sizeof(((struct ext4_inode_info *)0x0)->i_data) -
++                            sizeof(struct ext4_extent_header)) /
++                           max(sizeof(struct ext4_extent),
++                               sizeof(struct ext4_extent_idx));
++
++              /* compute maximum extents tree depth for a fully populated
++               * file of max size made of only minimal/1-block extents */
++              while (maxblocks > 0) {
++                      maxblocks /= (sb->s_blocksize -
++                                    sizeof(struct ext4_extent_header)) /
++                                   max(sizeof(struct ext4_extent),
++                                       sizeof(struct ext4_extent_idx));
++                      EXT4_SB(sb)->s_max_ext_tree_depth++;
++              }
++
++              printk(", maximum tree depth=%u\n",
++                     EXT4_SB(sb)->s_max_ext_tree_depth);
+ #ifdef EXTENTS_STATS
+               spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
+               EXT4_SB(sb)->s_ext_min = 1 << 30;
diff --git a/ldiskfs/kernel_patches/series/ldiskfs-4.4-sles12sp2.series b/ldiskfs/kernel_patches/series/ldiskfs-4.4-sles12sp2.series
new file mode 100644 (file)
index 0000000..0f8ba37
--- /dev/null
@@ -0,0 +1,22 @@
+sles12sp2/ext4-inode-version.patch
+sles12sp2/ext4-lookup-dotdot.patch
+sles12sp2/ext4-print-inum-in-htree-warning.patch
+sles12sp2/ext4-prealloc.patch
+sles12sp2/ext4-osd-iop-common.patch
+sles12sp2/ext4-misc.patch
+sles12sp2/ext4-mballoc-extra-checks.patch
+sles12sp2/ext4-hash-indexed-dir-dotdot-update.patch
+sles12sp2/ext4-kill-dx-root.patch
+rhel7/ext4-mballoc-pa-free-mismatch.patch
+sles12sp2/ext4-data-in-dirent.patch
+sles12sp2/ext4-large-eas.patch
+sles12sp2/ext4-disable-mb-cache.patch
+rhel7/ext4-nocmtime.patch
+sles12sp2/ext4-large-dir.patch
+sles12sp2/ext4-pdirop.patch
+sles12sp2/ext4-max-dir-size.patch
+rhel7/ext4-remove-truncate-warning.patch
+sles12sp2/ext4-corrupted-inode-block-bitmaps-handling-patches.patch
+sles12sp2/ext4-give-warning-with-dir-htree-growing.patch
+sles12sp2/ext4-mmp-brelse.patch
+rhel7/ext4-jcb-optimization.patch
index c493412..4cfc0a1 100644 (file)
@@ -10,7 +10,7 @@ TBD Intel Corporation
          3.10.0-514.2.2.el7  (RHEL7.3)
          3.0.101-0.47.71     (SLES11 SP3)
          3.0.101-91          (SLES11 SP4)
-         3.12.62-60.64.8     (SLES12 SP1)
+         4.4.21-84           (SLES12 SP2)
          vanilla linux 4.5.7 (ZFS only)
        * Client known to build on unpatched kernels:
          2.6.32-431.29.2.el6 (RHEL6.5)
@@ -20,7 +20,7 @@ TBD Intel Corporation
          3.10.0-514.2.2.el7  (RHEL7.3)
          3.0.101-0.47.71     (SLES11 SP3)
          3.0.101-91          (SLES11 SP4)
-         3.12.62-60.64.8     (SLES12 SP1)
+         4.4.21-84           (SLES12 SP2)
          vanilla linux 4.5.7
        * Recommended e2fsprogs version: 1.42.13.wc4 or newer
        * Recommended ZFS / SPL version: 0.6.5.7
index 97a9b87..c8a6efb 100644 (file)
@@ -1713,6 +1713,26 @@ have_bvec_iter, [
 ]) # LC_HAVE_BVEC_ITER
 
 #
+# LC_HAVE_BI_CNT
+#
+# 4.4 redefined bi_cnt as __bi_cnt
+#
+AC_DEFUN([LC_HAVE_BI_CNT], [
+LB_CHECK_COMPILE([if Linux kernel has bi_cnt in struct bio],
+have_bi_cnt, [
+       #include <asm/atomic.h>
+       #include <linux/bio.h>
+],[
+       struct bio bio;
+       int cnt;
+       cnt = atomic_read(&bio.bi_cnt);
+], [
+       AC_DEFINE(HAVE_BI_CNT, 1,
+               [struct bio has bi_cnt])
+])
+]) # LC_HAVE_BI_CNT
+
+#
 # LC_HAVE_TRUNCATE_IPAGE_FINAL
 #
 # 3.14 bring truncate_inode_pages_final for evict_inode
@@ -2429,6 +2449,7 @@ AC_DEFUN([LC_PROG_LINUX], [
        LC_HAVE_LOCKS_LOCK_FILE_WAIT
        LC_HAVE_QC_MAKE_REQUEST_FN
        LC_HAVE_KEY_PAYLOAD_DATA_ARRAY
+       LC_HAVE_BI_CNT
 
        # 4.5
        LC_HAVE_FILE_DENTRY
@@ -2742,6 +2763,7 @@ lustre/kernel_patches/targets/3.0-sles11.target
 lustre/kernel_patches/targets/3.0-sles11sp3.target
 lustre/kernel_patches/targets/3.0-sles11sp4.target
 lustre/kernel_patches/targets/3.12-sles12.target
+lustre/kernel_patches/targets/4.4-sles12.target
 lustre/kernel_patches/targets/2.6-fc11.target
 lustre/kernel_patches/targets/2.6-fc12.target
 lustre/kernel_patches/targets/2.6-fc15.target
index 238e668..18d0105 100644 (file)
@@ -158,7 +158,7 @@ static inline void ll_set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
 #ifdef HAVE_BVEC_ITER
 #define bio_idx(bio)                   (bio->bi_iter.bi_idx)
 #define bio_set_sector(bio, sector)    (bio->bi_iter.bi_sector = sector)
-#define bvl_to_page(bvl)               (bvl.bv_page)
+#define bvl_to_page(bvl)               (bvl->bv_page)
 #else
 #define bio_idx(bio)                   (bio->bi_idx)
 #define bio_set_sector(bio, sector)    (bio->bi_sector = sector)
diff --git a/lustre/kernel_patches/kernel_configs/kernel-4.4-4.4-sles12-x86_64.config b/lustre/kernel_patches/kernel_configs/kernel-4.4-4.4-sles12-x86_64.config
new file mode 100644 (file)
index 0000000..a95ee91
--- /dev/null
@@ -0,0 +1,7000 @@
+#
+# Automatically generated file; DO NOT EDIT.
+# Linux/x86 4.4.21 Kernel Configuration
+#
+CONFIG_64BIT=y
+CONFIG_X86_64=y
+CONFIG_X86=y
+CONFIG_INSTRUCTION_DECODER=y
+CONFIG_PERF_EVENTS_INTEL_UNCORE=y
+CONFIG_OUTPUT_FORMAT="elf64-x86-64"
+CONFIG_ARCH_DEFCONFIG="arch/x86/configs/x86_64_defconfig"
+CONFIG_LOCKDEP_SUPPORT=y
+CONFIG_STACKTRACE_SUPPORT=y
+CONFIG_HAVE_LATENCYTOP_SUPPORT=y
+CONFIG_MMU=y
+CONFIG_NEED_DMA_MAP_STATE=y
+CONFIG_NEED_SG_DMA_LENGTH=y
+CONFIG_GENERIC_ISA_DMA=y
+CONFIG_GENERIC_BUG=y
+CONFIG_GENERIC_BUG_RELATIVE_POINTERS=y
+CONFIG_GENERIC_HWEIGHT=y
+CONFIG_ARCH_MAY_HAVE_PC_FDC=y
+CONFIG_RWSEM_XCHGADD_ALGORITHM=y
+CONFIG_GENERIC_CALIBRATE_DELAY=y
+CONFIG_ARCH_HAS_CPU_RELAX=y
+CONFIG_ARCH_HAS_CACHE_LINE_SIZE=y
+CONFIG_HAVE_SETUP_PER_CPU_AREA=y
+CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK=y
+CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK=y
+CONFIG_ARCH_HIBERNATION_POSSIBLE=y
+CONFIG_ARCH_SUSPEND_POSSIBLE=y
+CONFIG_ARCH_WANT_HUGE_PMD_SHARE=y
+CONFIG_ARCH_WANT_GENERAL_HUGETLB=y
+CONFIG_ZONE_DMA32=y
+CONFIG_AUDIT_ARCH=y
+CONFIG_ARCH_SUPPORTS_OPTIMIZED_INLINING=y
+CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC=y
+CONFIG_HAVE_INTEL_TXT=y
+CONFIG_X86_64_SMP=y
+CONFIG_ARCH_HWEIGHT_CFLAGS="-fcall-saved-rdi -fcall-saved-rsi -fcall-saved-rdx -fcall-saved-rcx -fcall-saved-r8 -fcall-saved-r9 -fcall-saved-r10 -fcall-saved-r11"
+CONFIG_ARCH_SUPPORTS_UPROBES=y
+CONFIG_FIX_EARLYCON_MEM=y
+CONFIG_PGTABLE_LEVELS=4
+CONFIG_SUSE_KERNEL=y
+CONFIG_SUSE_KERNEL_SUPPORTED=y
+CONFIG_SPLIT_PACKAGE=y
+CONFIG_DEFCONFIG_LIST="/lib/modules/$UNAME_RELEASE/.config"
+CONFIG_IRQ_WORK=y
+CONFIG_BUILDTIME_EXTABLE_SORT=y
+
+#
+# General setup
+#
+CONFIG_INIT_ENV_ARG_LIMIT=32
+CONFIG_CROSS_COMPILE=""
+# CONFIG_COMPILE_TEST is not set
+CONFIG_LOCALVERSION="-84-default"
+# CONFIG_LOCALVERSION_AUTO is not set
+CONFIG_HAVE_KERNEL_GZIP=y
+CONFIG_HAVE_KERNEL_BZIP2=y
+CONFIG_HAVE_KERNEL_LZMA=y
+CONFIG_HAVE_KERNEL_XZ=y
+CONFIG_HAVE_KERNEL_LZO=y
+CONFIG_HAVE_KERNEL_LZ4=y
+CONFIG_KERNEL_GZIP=y
+# CONFIG_KERNEL_BZIP2 is not set
+# CONFIG_KERNEL_LZMA is not set
+# CONFIG_KERNEL_XZ is not set
+# CONFIG_KERNEL_LZO is not set
+# CONFIG_KERNEL_LZ4 is not set
+CONFIG_DEFAULT_HOSTNAME="(none)"
+CONFIG_SWAP=y
+CONFIG_SYSVIPC=y
+CONFIG_SYSVIPC_SYSCTL=y
+CONFIG_POSIX_MQUEUE=y
+CONFIG_POSIX_MQUEUE_SYSCTL=y
+CONFIG_CROSS_MEMORY_ATTACH=y
+CONFIG_FHANDLE=y
+CONFIG_USELIB=y
+CONFIG_AUDIT=y
+CONFIG_HAVE_ARCH_AUDITSYSCALL=y
+CONFIG_AUDITSYSCALL=y
+CONFIG_AUDIT_WATCH=y
+CONFIG_AUDIT_TREE=y
+
+#
+# IRQ subsystem
+#
+CONFIG_GENERIC_IRQ_PROBE=y
+CONFIG_GENERIC_IRQ_SHOW=y
+CONFIG_GENERIC_PENDING_IRQ=y
+CONFIG_GENERIC_IRQ_CHIP=y
+CONFIG_IRQ_DOMAIN=y
+CONFIG_IRQ_DOMAIN_HIERARCHY=y
+CONFIG_GENERIC_MSI_IRQ=y
+CONFIG_GENERIC_MSI_IRQ_DOMAIN=y
+# CONFIG_IRQ_DOMAIN_DEBUG is not set
+CONFIG_IRQ_FORCED_THREADING=y
+CONFIG_SPARSE_IRQ=y
+CONFIG_CLOCKSOURCE_WATCHDOG=y
+CONFIG_ARCH_CLOCKSOURCE_DATA=y
+CONFIG_CLOCKSOURCE_VALIDATE_LAST_CYCLE=y
+CONFIG_GENERIC_TIME_VSYSCALL=y
+CONFIG_GENERIC_CLOCKEVENTS=y
+CONFIG_GENERIC_CLOCKEVENTS_BROADCAST=y
+CONFIG_GENERIC_CLOCKEVENTS_MIN_ADJUST=y
+CONFIG_GENERIC_CMOS_UPDATE=y
+
+#
+# Timers subsystem
+#
+CONFIG_TICK_ONESHOT=y
+CONFIG_NO_HZ_COMMON=y
+# CONFIG_HZ_PERIODIC is not set
+# CONFIG_NO_HZ_IDLE is not set
+CONFIG_NO_HZ_FULL=y
+# CONFIG_NO_HZ_FULL_ALL is not set
+# CONFIG_NO_HZ_FULL_SYSIDLE is not set
+CONFIG_NO_HZ=y
+CONFIG_HIGH_RES_TIMERS=y
+
+#
+# CPU/Task time and stats accounting
+#
+CONFIG_VIRT_CPU_ACCOUNTING=y
+CONFIG_VIRT_CPU_ACCOUNTING_GEN=y
+CONFIG_BSD_PROCESS_ACCT=y
+CONFIG_BSD_PROCESS_ACCT_V3=y
+CONFIG_TASKSTATS=y
+CONFIG_TASK_DELAY_ACCT=y
+CONFIG_TASK_XACCT=y
+CONFIG_TASK_IO_ACCOUNTING=y
+
+#
+# RCU Subsystem
+#
+CONFIG_TREE_RCU=y
+# CONFIG_RCU_EXPERT is not set
+CONFIG_SRCU=y
+CONFIG_TASKS_RCU=y
+CONFIG_RCU_STALL_COMMON=y
+CONFIG_CONTEXT_TRACKING=y
+# CONFIG_CONTEXT_TRACKING_FORCE is not set
+# CONFIG_TREE_RCU_TRACE is not set
+CONFIG_RCU_NOCB_CPU=y
+CONFIG_RCU_NOCB_CPU_NONE=y
+# CONFIG_RCU_NOCB_CPU_ZERO is not set
+# CONFIG_RCU_NOCB_CPU_ALL is not set
+# CONFIG_RCU_EXPEDITE_BOOT is not set
+CONFIG_BUILD_BIN2C=y
+CONFIG_IKCONFIG=y
+CONFIG_IKCONFIG_PROC=y
+CONFIG_LOG_BUF_SHIFT=18
+CONFIG_LOG_CPU_MAX_BUF_SHIFT=12
+CONFIG_HAVE_UNSTABLE_SCHED_CLOCK=y
+CONFIG_ARCH_SUPPORTS_NUMA_BALANCING=y
+CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH=y
+CONFIG_ARCH_SUPPORTS_INT128=y
+CONFIG_NUMA_BALANCING=y
+CONFIG_NUMA_BALANCING_DEFAULT_ENABLED=y
+CONFIG_CGROUPS=y
+# CONFIG_CGROUP_DEBUG is not set
+CONFIG_CGROUP_FREEZER=y
+CONFIG_CGROUP_PIDS=y
+CONFIG_CGROUP_DEVICE=y
+CONFIG_CPUSETS=y
+CONFIG_PROC_PID_CPUSET=y
+CONFIG_CGROUP_CPUACCT=y
+CONFIG_PAGE_COUNTER=y
+CONFIG_MEMCG=y
+CONFIG_MEMCG_SWAP=y
+# CONFIG_MEMCG_SWAP_ENABLED is not set
+# CONFIG_MEMCG_KMEM is not set
+CONFIG_CGROUP_HUGETLB=y
+CONFIG_CGROUP_PERF=y
+CONFIG_CGROUP_SCHED=y
+CONFIG_FAIR_GROUP_SCHED=y
+CONFIG_CFS_BANDWIDTH=y
+CONFIG_RT_GROUP_SCHED=y
+CONFIG_BLK_CGROUP=y
+# CONFIG_DEBUG_BLK_CGROUP is not set
+CONFIG_CGROUP_WRITEBACK=y
+CONFIG_CHECKPOINT_RESTORE=y
+CONFIG_NAMESPACES=y
+CONFIG_UTS_NS=y
+CONFIG_IPC_NS=y
+CONFIG_USER_NS=y
+CONFIG_PID_NS=y
+CONFIG_NET_NS=y
+# CONFIG_SCHED_AUTOGROUP is not set
+# CONFIG_SYSFS_DEPRECATED is not set
+CONFIG_RELAY=y
+CONFIG_BLK_DEV_INITRD=y
+CONFIG_INITRAMFS_SOURCE=""
+CONFIG_RD_GZIP=y
+CONFIG_RD_BZIP2=y
+CONFIG_RD_LZMA=y
+CONFIG_RD_XZ=y
+CONFIG_RD_LZO=y
+# CONFIG_RD_LZ4 is not set
+# CONFIG_CC_OPTIMIZE_FOR_SIZE is not set
+CONFIG_SYSCTL=y
+CONFIG_ANON_INODES=y
+CONFIG_HAVE_UID16=y
+CONFIG_SYSCTL_EXCEPTION_TRACE=y
+CONFIG_HAVE_PCSPKR_PLATFORM=y
+CONFIG_BPF=y
+CONFIG_EXPERT=y
+CONFIG_UID16=y
+CONFIG_MULTIUSER=y
+CONFIG_SGETMASK_SYSCALL=y
+CONFIG_SYSFS_SYSCALL=y
+CONFIG_SYSCTL_SYSCALL=y
+CONFIG_KALLSYMS=y
+CONFIG_KALLSYMS_ALL=y
+CONFIG_PRINTK=y
+CONFIG_BUG=y
+CONFIG_ELF_CORE=y
+CONFIG_PCSPKR_PLATFORM=y
+CONFIG_BASE_FULL=y
+CONFIG_FUTEX=y
+CONFIG_EPOLL=y
+CONFIG_SIGNALFD=y
+CONFIG_TIMERFD=y
+CONFIG_EVENTFD=y
+CONFIG_BPF_SYSCALL=y
+CONFIG_SHMEM=y
+CONFIG_AIO=y
+CONFIG_ADVISE_SYSCALLS=y
+CONFIG_USERFAULTFD=y
+CONFIG_PCI_QUIRKS=y
+CONFIG_MEMBARRIER=y
+# CONFIG_EMBEDDED is not set
+CONFIG_HAVE_PERF_EVENTS=y
+
+#
+# Kernel Performance Events And Counters
+#
+CONFIG_PERF_EVENTS=y
+# CONFIG_DEBUG_PERF_USE_VMALLOC is not set
+CONFIG_VM_EVENT_COUNTERS=y
+# CONFIG_COMPAT_BRK is not set
+CONFIG_SLAB=y
+# CONFIG_SLUB is not set
+# CONFIG_SLOB is not set
+CONFIG_SYSTEM_DATA_VERIFICATION=y
+CONFIG_PROFILING=y
+CONFIG_TRACEPOINTS=y
+CONFIG_KEXEC_CORE=y
+CONFIG_OPROFILE=m
+# CONFIG_OPROFILE_EVENT_MULTIPLEX is not set
+CONFIG_HAVE_OPROFILE=y
+CONFIG_OPROFILE_NMI_TIMER=y
+CONFIG_KPROBES=y
+CONFIG_JUMP_LABEL=y
+# CONFIG_STATIC_KEYS_SELFTEST is not set
+CONFIG_OPTPROBES=y
+CONFIG_KPROBES_ON_FTRACE=y
+CONFIG_UPROBES=y
+# CONFIG_HAVE_64BIT_ALIGNED_ACCESS is not set
+CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y
+CONFIG_ARCH_USE_BUILTIN_BSWAP=y
+CONFIG_KRETPROBES=y
+CONFIG_USER_RETURN_NOTIFIER=y
+CONFIG_HAVE_IOREMAP_PROT=y
+CONFIG_HAVE_KPROBES=y
+CONFIG_HAVE_KRETPROBES=y
+CONFIG_HAVE_OPTPROBES=y
+CONFIG_HAVE_KPROBES_ON_FTRACE=y
+CONFIG_HAVE_ARCH_TRACEHOOK=y
+CONFIG_HAVE_DMA_ATTRS=y
+CONFIG_HAVE_DMA_CONTIGUOUS=y
+CONFIG_GENERIC_SMP_IDLE_THREAD=y
+CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT=y
+CONFIG_HAVE_REGS_AND_STACK_ACCESS_API=y
+CONFIG_HAVE_CLK=y
+CONFIG_HAVE_DMA_API_DEBUG=y
+CONFIG_HAVE_HW_BREAKPOINT=y
+CONFIG_HAVE_MIXED_BREAKPOINTS_REGS=y
+CONFIG_HAVE_USER_RETURN_NOTIFIER=y
+CONFIG_HAVE_PERF_EVENTS_NMI=y
+CONFIG_HAVE_PERF_REGS=y
+CONFIG_HAVE_PERF_USER_STACK_DUMP=y
+CONFIG_HAVE_ARCH_JUMP_LABEL=y
+CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG=y
+CONFIG_HAVE_CMPXCHG_LOCAL=y
+CONFIG_HAVE_CMPXCHG_DOUBLE=y
+CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION=y
+CONFIG_ARCH_WANT_OLD_COMPAT_IPC=y
+CONFIG_HAVE_ARCH_SECCOMP_FILTER=y
+CONFIG_SECCOMP_FILTER=y
+CONFIG_HAVE_CC_STACKPROTECTOR=y
+CONFIG_CC_STACKPROTECTOR=y
+# CONFIG_CC_STACKPROTECTOR_NONE is not set
+CONFIG_CC_STACKPROTECTOR_REGULAR=y
+# CONFIG_CC_STACKPROTECTOR_STRONG is not set
+CONFIG_HAVE_CONTEXT_TRACKING=y
+CONFIG_HAVE_VIRT_CPU_ACCOUNTING_GEN=y
+CONFIG_HAVE_IRQ_TIME_ACCOUNTING=y
+CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE=y
+CONFIG_HAVE_ARCH_HUGE_VMAP=y
+CONFIG_HAVE_ARCH_SOFT_DIRTY=y
+CONFIG_MODULES_USE_ELF_RELA=y
+CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK=y
+CONFIG_ARCH_HAS_ELF_RANDOMIZE=y
+CONFIG_HAVE_EXIT_THREAD=y
+CONFIG_HAVE_COPY_THREAD_TLS=y
+CONFIG_OLD_SIGSUSPEND3=y
+CONFIG_COMPAT_OLD_SIGACTION=y
+
+#
+# GCOV-based kernel profiling
+#
+# CONFIG_GCOV_KERNEL is not set
+CONFIG_ARCH_HAS_GCOV_PROFILE_ALL=y
+# CONFIG_HAVE_GENERIC_DMA_COHERENT is not set
+CONFIG_SLABINFO=y
+CONFIG_RT_MUTEXES=y
+CONFIG_BASE_SMALL=0
+CONFIG_MODULES=y
+CONFIG_MODULE_FORCE_LOAD=y
+CONFIG_MODULE_UNLOAD=y
+# CONFIG_MODULE_FORCE_UNLOAD is not set
+CONFIG_MODVERSIONS=y
+CONFIG_MODULE_SRCVERSION_ALL=y
+CONFIG_MODULE_SIG=y
+# CONFIG_MODULE_SIG_FORCE is not set
+# CONFIG_MODULE_SIG_ALL is not set
+CONFIG_MODULE_SIG_UEFI=y
+# CONFIG_MODULE_SIG_SHA1 is not set
+# CONFIG_MODULE_SIG_SHA224 is not set
+CONFIG_MODULE_SIG_SHA256=y
+# CONFIG_MODULE_SIG_SHA384 is not set
+# CONFIG_MODULE_SIG_SHA512 is not set
+CONFIG_MODULE_SIG_HASH="sha256"
+# CONFIG_MODULE_COMPRESS is not set
+CONFIG_MODULES_TREE_LOOKUP=y
+CONFIG_BLOCK=y
+CONFIG_BLK_DEV_BSG=y
+CONFIG_BLK_DEV_BSGLIB=y
+CONFIG_BLK_DEV_INTEGRITY=y
+CONFIG_BLK_DEV_THROTTLING=y
+# CONFIG_BLK_CMDLINE_PARSER is not set
+
+#
+# Partition Types
+#
+CONFIG_PARTITION_ADVANCED=y
+# CONFIG_ACORN_PARTITION is not set
+# CONFIG_AIX_PARTITION is not set
+CONFIG_OSF_PARTITION=y
+# CONFIG_AMIGA_PARTITION is not set
+CONFIG_ATARI_PARTITION=y
+CONFIG_MAC_PARTITION=y
+CONFIG_MSDOS_PARTITION=y
+CONFIG_BSD_DISKLABEL=y
+# CONFIG_MINIX_SUBPARTITION is not set
+CONFIG_SOLARIS_X86_PARTITION=y
+CONFIG_UNIXWARE_DISKLABEL=y
+CONFIG_LDM_PARTITION=y
+# CONFIG_LDM_DEBUG is not set
+CONFIG_SGI_PARTITION=y
+CONFIG_ULTRIX_PARTITION=y
+CONFIG_SUN_PARTITION=y
+CONFIG_KARMA_PARTITION=y
+CONFIG_EFI_PARTITION=y
+CONFIG_SYSV68_PARTITION=y
+# CONFIG_CMDLINE_PARTITION is not set
+CONFIG_BLOCK_COMPAT=y
+
+#
+# IO Schedulers
+#
+CONFIG_IOSCHED_NOOP=y
+CONFIG_IOSCHED_DEADLINE=y
+CONFIG_IOSCHED_CFQ=y
+CONFIG_CFQ_GROUP_IOSCHED=y
+CONFIG_DEFAULT_DEADLINE=y
+# CONFIG_DEFAULT_CFQ is not set
+# CONFIG_DEFAULT_NOOP is not set
+CONFIG_DEFAULT_IOSCHED="deadline"
+CONFIG_PREEMPT_NOTIFIERS=y
+CONFIG_PADATA=y
+CONFIG_ASN1=y
+CONFIG_INLINE_SPIN_UNLOCK_IRQ=y
+CONFIG_INLINE_READ_UNLOCK=y
+CONFIG_INLINE_READ_UNLOCK_IRQ=y
+CONFIG_INLINE_WRITE_UNLOCK=y
+CONFIG_INLINE_WRITE_UNLOCK_IRQ=y
+CONFIG_ARCH_SUPPORTS_ATOMIC_RMW=y
+CONFIG_MUTEX_SPIN_ON_OWNER=y
+CONFIG_RWSEM_SPIN_ON_OWNER=y
+CONFIG_LOCK_SPIN_ON_OWNER=y
+CONFIG_ARCH_USE_QUEUED_SPINLOCKS=y
+CONFIG_QUEUED_SPINLOCKS=y
+CONFIG_ARCH_USE_QUEUED_RWLOCKS=y
+CONFIG_QUEUED_RWLOCKS=y
+CONFIG_FREEZER=y
+CONFIG_HAVE_KGRAFT=y
+CONFIG_KGRAFT=y
+
+#
+# Processor type and features
+#
+CONFIG_ZONE_DMA=y
+CONFIG_SMP=y
+CONFIG_X86_FEATURE_NAMES=y
+CONFIG_X86_X2APIC=y
+CONFIG_X86_MPPARSE=y
+CONFIG_X86_EXTENDED_PLATFORM=y
+# CONFIG_X86_NUMACHIP is not set
+# CONFIG_X86_VSMP is not set
+CONFIG_X86_UV=y
+# CONFIG_X86_GOLDFISH is not set
+CONFIG_X86_INTEL_LPSS=y
+CONFIG_X86_AMD_PLATFORM_DEVICE=y
+CONFIG_IOSF_MBI=y
+# CONFIG_IOSF_MBI_DEBUG is not set
+CONFIG_X86_SUPPORTS_MEMORY_FAILURE=y
+CONFIG_HYPERVISOR_GUEST=y
+CONFIG_PARAVIRT=y
+# CONFIG_PARAVIRT_DEBUG is not set
+CONFIG_PARAVIRT_SPINLOCKS=y
+# CONFIG_QUEUED_LOCK_STAT is not set
+CONFIG_XEN=y
+CONFIG_XEN_DOM0=y
+CONFIG_XEN_PVHVM=y
+# CONFIG_XEN_512GB is not set
+CONFIG_XEN_SAVE_RESTORE=y
+CONFIG_XEN_DEBUG_FS=y
+CONFIG_XEN_PVH=y
+CONFIG_KVM_GUEST=y
+# CONFIG_KVM_DEBUG_FS is not set
+# CONFIG_PARAVIRT_TIME_ACCOUNTING is not set
+CONFIG_PARAVIRT_CLOCK=y
+CONFIG_NO_BOOTMEM=y
+# CONFIG_MK8 is not set
+# CONFIG_MPSC is not set
+# CONFIG_MCORE2 is not set
+# CONFIG_MATOM is not set
+CONFIG_GENERIC_CPU=y
+CONFIG_X86_INTERNODE_CACHE_SHIFT=6
+CONFIG_X86_L1_CACHE_SHIFT=6
+CONFIG_X86_TSC=y
+CONFIG_X86_CMPXCHG64=y
+CONFIG_X86_CMOV=y
+CONFIG_X86_MINIMUM_CPU_FAMILY=64
+CONFIG_X86_DEBUGCTLMSR=y
+# CONFIG_PROCESSOR_SELECT is not set
+CONFIG_CPU_SUP_INTEL=y
+CONFIG_CPU_SUP_AMD=y
+CONFIG_CPU_SUP_CENTAUR=y
+CONFIG_HPET_TIMER=y
+CONFIG_HPET_EMULATE_RTC=y
+CONFIG_DMI=y
+CONFIG_GART_IOMMU=y
+CONFIG_CALGARY_IOMMU=y
+# CONFIG_CALGARY_IOMMU_ENABLED_BY_DEFAULT is not set
+CONFIG_SWIOTLB=y
+CONFIG_IOMMU_HELPER=y
+CONFIG_MAXSMP=y
+CONFIG_NR_CPUS=8192
+CONFIG_SCHED_SMT=y
+CONFIG_SCHED_MC=y
+CONFIG_PREEMPT_NONE=y
+# CONFIG_PREEMPT_VOLUNTARY is not set
+# CONFIG_PREEMPT is not set
+CONFIG_X86_LOCAL_APIC=y
+CONFIG_X86_IO_APIC=y
+CONFIG_X86_REROUTE_FOR_BROKEN_BOOT_IRQS=y
+CONFIG_X86_MCE=y
+CONFIG_X86_MCE_INTEL=y
+CONFIG_X86_MCE_AMD=y
+CONFIG_X86_MCE_THRESHOLD=y
+CONFIG_X86_MCE_INJECT=m
+CONFIG_X86_THERMAL_VECTOR=y
+# CONFIG_VM86 is not set
+# CONFIG_X86_16BIT is not set
+CONFIG_X86_VSYSCALL_EMULATION=y
+CONFIG_I8K=m
+CONFIG_MICROCODE=y
+CONFIG_MICROCODE_INTEL=y
+CONFIG_MICROCODE_AMD=y
+CONFIG_MICROCODE_OLD_INTERFACE=y
+CONFIG_X86_MSR=m
+CONFIG_X86_CPUID=m
+CONFIG_ARCH_PHYS_ADDR_T_64BIT=y
+CONFIG_ARCH_DMA_ADDR_T_64BIT=y
+CONFIG_X86_DIRECT_GBPAGES=y
+CONFIG_NUMA=y
+CONFIG_AMD_NUMA=y
+CONFIG_X86_64_ACPI_NUMA=y
+CONFIG_NODES_SPAN_OTHER_NODES=y
+CONFIG_NUMA_EMU=y
+CONFIG_NODES_SHIFT=10
+CONFIG_ARCH_SPARSEMEM_ENABLE=y
+CONFIG_ARCH_SPARSEMEM_DEFAULT=y
+CONFIG_ARCH_SELECT_MEMORY_MODEL=y
+CONFIG_ARCH_MEMORY_PROBE=y
+CONFIG_ARCH_PROC_KCORE_TEXT=y
+CONFIG_ILLEGAL_POINTER_VALUE=0xdead000000000000
+CONFIG_SELECT_MEMORY_MODEL=y
+CONFIG_SPARSEMEM_MANUAL=y
+CONFIG_SPARSEMEM=y
+CONFIG_NEED_MULTIPLE_NODES=y
+CONFIG_HAVE_MEMORY_PRESENT=y
+CONFIG_SPARSEMEM_EXTREME=y
+CONFIG_SPARSEMEM_VMEMMAP_ENABLE=y
+CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER=y
+CONFIG_SPARSEMEM_VMEMMAP=y
+CONFIG_HAVE_MEMBLOCK=y
+CONFIG_HAVE_MEMBLOCK_NODE_MAP=y
+CONFIG_ARCH_DISCARD_MEMBLOCK=y
+CONFIG_MEMORY_ISOLATION=y
+CONFIG_MOVABLE_NODE=y
+CONFIG_HAVE_BOOTMEM_INFO_NODE=y
+CONFIG_MEMORY_HOTPLUG=y
+CONFIG_MEMORY_HOTPLUG_SPARSE=y
+CONFIG_MEMORY_HOTREMOVE=y
+CONFIG_SPLIT_PTLOCK_CPUS=4
+CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK=y
+CONFIG_MEMORY_BALLOON=y
+CONFIG_BALLOON_COMPACTION=y
+CONFIG_COMPACTION=y
+CONFIG_MIGRATION=y
+CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION=y
+CONFIG_PHYS_ADDR_T_64BIT=y
+CONFIG_ZONE_DMA_FLAG=1
+CONFIG_BOUNCE=y
+CONFIG_VIRT_TO_BUS=y
+CONFIG_MMU_NOTIFIER=y
+CONFIG_KSM=y
+CONFIG_DEFAULT_MMAP_MIN_ADDR=65536
+CONFIG_ARCH_SUPPORTS_MEMORY_FAILURE=y
+CONFIG_MEMORY_FAILURE=y
+CONFIG_HWPOISON_INJECT=m
+CONFIG_TRANSPARENT_HUGEPAGE=y
+CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y
+# CONFIG_TRANSPARENT_HUGEPAGE_MADVISE is not set
+CONFIG_CLEANCACHE=y
+CONFIG_FRONTSWAP=y
+# CONFIG_CMA is not set
+CONFIG_MEM_SOFT_DIRTY=y
+CONFIG_ZSWAP=y
+CONFIG_ZPOOL=y
+CONFIG_ZBUD=y
+CONFIG_ZSMALLOC=y
+# CONFIG_PGTABLE_MAPPING is not set
+# CONFIG_ZSMALLOC_STAT is not set
+CONFIG_GENERIC_EARLY_IOREMAP=y
+CONFIG_ARCH_SUPPORTS_DEFERRED_STRUCT_PAGE_INIT=y
+CONFIG_DEFERRED_STRUCT_PAGE_INIT=y
+# CONFIG_IDLE_PAGE_TRACKING is not set
+CONFIG_ZONE_DEVICE=y
+CONFIG_FRAME_VECTOR=y
+CONFIG_X86_PMEM_LEGACY_DEVICE=y
+CONFIG_X86_PMEM_LEGACY=m
+CONFIG_X86_CHECK_BIOS_CORRUPTION=y
+CONFIG_X86_BOOTPARAM_MEMORY_CORRUPTION_CHECK=y
+CONFIG_X86_RESERVE_LOW=64
+CONFIG_MTRR=y
+CONFIG_MTRR_SANITIZER=y
+CONFIG_MTRR_SANITIZER_ENABLE_DEFAULT=0
+CONFIG_MTRR_SANITIZER_SPARE_REG_NR_DEFAULT=1
+CONFIG_X86_PAT=y
+CONFIG_ARCH_USES_PG_UNCACHED=y
+CONFIG_ARCH_RANDOM=y
+CONFIG_X86_SMAP=y
+CONFIG_X86_INTEL_MPX=y
+CONFIG_EFI=y
+CONFIG_EFI_STUB=y
+CONFIG_EFI_MIXED=y
+CONFIG_EFI_SECURE_BOOT_SECURELEVEL=y
+CONFIG_SECCOMP=y
+# CONFIG_HZ_100 is not set
+CONFIG_HZ_250=y
+# CONFIG_HZ_300 is not set
+# CONFIG_HZ_1000 is not set
+CONFIG_HZ=250
+CONFIG_SCHED_HRTICK=y
+CONFIG_KEXEC=y
+CONFIG_KEXEC_FILE=y
+# CONFIG_KEXEC_VERIFY_SIG is not set
+CONFIG_CRASH_DUMP=y
+# CONFIG_KEXEC_JUMP is not set
+CONFIG_PHYSICAL_START=0x200000
+CONFIG_RELOCATABLE=y
+# CONFIG_RANDOMIZE_BASE is not set
+CONFIG_PHYSICAL_ALIGN=0x1000000
+CONFIG_HOTPLUG_CPU=y
+# CONFIG_BOOTPARAM_HOTPLUG_CPU0 is not set
+# CONFIG_DEBUG_HOTPLUG_CPU0 is not set
+# CONFIG_COMPAT_VDSO is not set
+# CONFIG_LEGACY_VSYSCALL_NATIVE is not set
+CONFIG_LEGACY_VSYSCALL_EMULATE=y
+# CONFIG_LEGACY_VSYSCALL_NONE is not set
+# CONFIG_CMDLINE_BOOL is not set
+CONFIG_MODIFY_LDT_SYSCALL=y
+CONFIG_HAVE_LIVEPATCH=y
+CONFIG_LIVEPATCH=y
+CONFIG_ARCH_ENABLE_MEMORY_HOTPLUG=y
+CONFIG_ARCH_ENABLE_MEMORY_HOTREMOVE=y
+CONFIG_USE_PERCPU_NUMA_NODE_ID=y
+
+#
+# Power management and ACPI options
+#
+CONFIG_ARCH_HIBERNATION_HEADER=y
+CONFIG_SUSPEND=y
+CONFIG_SUSPEND_FREEZER=y
+# CONFIG_SUSPEND_SKIP_SYNC is not set
+CONFIG_HIBERNATE_CALLBACKS=y
+CONFIG_HIBERNATION=y
+CONFIG_HIBERNATE_VERIFICATION=y
+# CONFIG_HIBERNATE_VERIFICATION_FORCE is not set
+CONFIG_PM_STD_PARTITION=""
+CONFIG_PM_SLEEP=y
+CONFIG_PM_SLEEP_SMP=y
+# CONFIG_PM_AUTOSLEEP is not set
+# CONFIG_PM_WAKELOCKS is not set
+CONFIG_PM=y
+CONFIG_PM_DEBUG=y
+CONFIG_PM_ADVANCED_DEBUG=y
+# CONFIG_PM_TEST_SUSPEND is not set
+CONFIG_PM_SLEEP_DEBUG=y
+CONFIG_DPM_WATCHDOG=y
+CONFIG_DPM_WATCHDOG_TIMEOUT=60
+CONFIG_PM_TRACE=y
+CONFIG_PM_TRACE_RTC=y
+CONFIG_PM_CLK=y
+# CONFIG_WQ_POWER_EFFICIENT_DEFAULT is not set
+CONFIG_ACPI=y
+CONFIG_ACPI_LEGACY_TABLES_LOOKUP=y
+CONFIG_ARCH_MIGHT_HAVE_ACPI_PDC=y
+CONFIG_ACPI_SYSTEM_POWER_STATES_SUPPORT=y
+# CONFIG_ACPI_DEBUGGER is not set
+CONFIG_ACPI_SLEEP=y
+CONFIG_ACPI_PROCFS_POWER=y
+CONFIG_ACPI_REV_OVERRIDE_POSSIBLE=y
+CONFIG_ACPI_EC_DEBUGFS=m
+CONFIG_ACPI_AC=m
+CONFIG_ACPI_BATTERY=m
+CONFIG_ACPI_BUTTON=m
+CONFIG_ACPI_VIDEO=m
+CONFIG_ACPI_FAN=m
+CONFIG_ACPI_DOCK=y
+CONFIG_ACPI_CPU_FREQ_PSS=y
+CONFIG_ACPI_PROCESSOR_IDLE=y
+CONFIG_ACPI_PROCESSOR=m
+CONFIG_ACPI_IPMI=m
+CONFIG_ACPI_HOTPLUG_CPU=y
+CONFIG_ACPI_PROCESSOR_AGGREGATOR=m
+CONFIG_ACPI_THERMAL=m
+CONFIG_ACPI_NUMA=y
+CONFIG_ACPI_CUSTOM_DSDT_FILE=""
+# CONFIG_ACPI_CUSTOM_DSDT is not set
+CONFIG_ACPI_INITRD_TABLE_OVERRIDE=y
+CONFIG_ACPI_DEBUG=y
+CONFIG_ACPI_PCI_SLOT=y
+CONFIG_X86_PM_TIMER=y
+CONFIG_ACPI_CONTAINER=y
+CONFIG_ACPI_HOTPLUG_MEMORY=y
+CONFIG_ACPI_HOTPLUG_IOAPIC=y
+CONFIG_ACPI_SBS=m
+CONFIG_ACPI_HED=y
+# CONFIG_ACPI_CUSTOM_METHOD is not set
+CONFIG_ACPI_BGRT=y
+# CONFIG_ACPI_REDUCED_HARDWARE_ONLY is not set
+CONFIG_ACPI_NFIT=m
+# CONFIG_ACPI_NFIT_DEBUG is not set
+CONFIG_HAVE_ACPI_APEI=y
+CONFIG_HAVE_ACPI_APEI_NMI=y
+CONFIG_ACPI_APEI=y
+CONFIG_ACPI_APEI_GHES=y
+CONFIG_ACPI_APEI_PCIEAER=y
+CONFIG_ACPI_APEI_MEMORY_FAILURE=y
+CONFIG_ACPI_APEI_EINJ=m
+# CONFIG_ACPI_APEI_ERST_DEBUG is not set
+CONFIG_ACPI_EXTLOG=m
+# CONFIG_PMIC_OPREGION is not set
+# CONFIG_SFI is not set
+
+#
+# CPU Frequency scaling
+#
+CONFIG_CPU_FREQ=y
+CONFIG_CPU_FREQ_GOV_COMMON=y
+CONFIG_CPU_FREQ_STAT=m
+CONFIG_CPU_FREQ_STAT_DETAILS=y
+# CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not set
+# CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVE is not set
+# CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE is not set
+CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND=y
+# CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE is not set
+CONFIG_CPU_FREQ_GOV_PERFORMANCE=y
+CONFIG_CPU_FREQ_GOV_POWERSAVE=m
+CONFIG_CPU_FREQ_GOV_USERSPACE=m
+CONFIG_CPU_FREQ_GOV_ONDEMAND=y
+CONFIG_CPU_FREQ_GOV_CONSERVATIVE=m
+
+#
+# CPU frequency scaling drivers
+#
+CONFIG_X86_INTEL_PSTATE=y
+CONFIG_X86_PCC_CPUFREQ=m
+CONFIG_X86_ACPI_CPUFREQ=m
+CONFIG_X86_ACPI_CPUFREQ_CPB=y
+CONFIG_X86_POWERNOW_K8=m
+CONFIG_X86_AMD_FREQ_SENSITIVITY=m
+# CONFIG_X86_SPEEDSTEP_CENTRINO is not set
+# CONFIG_X86_P4_CLOCKMOD is not set
+
+#
+# shared options
+#
+# CONFIG_X86_SPEEDSTEP_LIB is not set
+
+#
+# CPU Idle
+#
+CONFIG_CPU_IDLE=y
+CONFIG_CPU_IDLE_GOV_LADDER=y
+CONFIG_CPU_IDLE_GOV_MENU=y
+# CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED is not set
+CONFIG_INTEL_IDLE=y
+
+#
+# Memory power savings
+#
+CONFIG_I7300_IDLE_IOAT_CHANNEL=y
+CONFIG_I7300_IDLE=m
+
+#
+# Bus options (PCI etc.)
+#
+CONFIG_PCI=y
+CONFIG_PCI_DIRECT=y
+CONFIG_PCI_MMCONFIG=y
+CONFIG_PCI_XEN=y
+CONFIG_PCI_DOMAINS=y
+# CONFIG_PCI_CNB20LE_QUIRK is not set
+CONFIG_PCIEPORTBUS=y
+CONFIG_HOTPLUG_PCI_PCIE=y
+CONFIG_PCIEAER=y
+# CONFIG_PCIE_ECRC is not set
+CONFIG_PCIEAER_INJECT=m
+CONFIG_PCIEASPM=y
+# CONFIG_PCIEASPM_DEBUG is not set
+CONFIG_PCIEASPM_DEFAULT=y
+# CONFIG_PCIEASPM_POWERSAVE is not set
+# CONFIG_PCIEASPM_PERFORMANCE is not set
+CONFIG_PCIE_PME=y
+CONFIG_PCI_BUS_ADDR_T_64BIT=y
+CONFIG_PCI_MSI=y
+CONFIG_PCI_MSI_IRQ_DOMAIN=y
+# CONFIG_PCI_DEBUG is not set
+# CONFIG_PCI_REALLOC_ENABLE_AUTO is not set
+CONFIG_PCI_STUB=y
+CONFIG_XEN_PCIDEV_FRONTEND=m
+CONFIG_HT_IRQ=y
+CONFIG_PCI_ATS=y
+CONFIG_PCI_IOV=y
+CONFIG_PCI_PRI=y
+CONFIG_PCI_PASID=y
+CONFIG_PCI_LABEL=y
+CONFIG_PCI_HYPERV=m
+
+#
+# PCI host controller drivers
+#
+CONFIG_ISA_DMA_API=y
+CONFIG_AMD_NB=y
+CONFIG_PCCARD=m
+# CONFIG_PCMCIA is not set
+CONFIG_CARDBUS=y
+
+#
+# PC-card bridges
+#
+CONFIG_YENTA=m
+CONFIG_YENTA_O2=y
+CONFIG_YENTA_RICOH=y
+CONFIG_YENTA_TI=y
+CONFIG_YENTA_ENE_TUNE=y
+CONFIG_YENTA_TOSHIBA=y
+CONFIG_HOTPLUG_PCI=y
+CONFIG_HOTPLUG_PCI_ACPI=y
+CONFIG_HOTPLUG_PCI_ACPI_IBM=m
+CONFIG_HOTPLUG_PCI_CPCI=y
+CONFIG_HOTPLUG_PCI_CPCI_ZT5550=m
+CONFIG_HOTPLUG_PCI_CPCI_GENERIC=m
+CONFIG_HOTPLUG_PCI_SHPC=m
+CONFIG_RAPIDIO=m
+CONFIG_RAPIDIO_TSI721=m
+CONFIG_RAPIDIO_DISC_TIMEOUT=30
+CONFIG_RAPIDIO_ENABLE_RX_TX_PORTS=y
+CONFIG_RAPIDIO_DMA_ENGINE=y
+CONFIG_RAPIDIO_DEBUG=y
+CONFIG_RAPIDIO_ENUM_BASIC=m
+
+#
+# RapidIO Switch drivers
+#
+CONFIG_RAPIDIO_TSI57X=m
+CONFIG_RAPIDIO_CPS_XX=m
+CONFIG_RAPIDIO_TSI568=m
+CONFIG_RAPIDIO_CPS_GEN2=m
+# CONFIG_X86_SYSFB is not set
+
+#
+# Executable file formats / Emulations
+#
+CONFIG_BINFMT_ELF=y
+CONFIG_COMPAT_BINFMT_ELF=y
+CONFIG_CORE_DUMP_DEFAULT_ELF_HEADERS=y
+CONFIG_BINFMT_SCRIPT=y
+# CONFIG_HAVE_AOUT is not set
+CONFIG_BINFMT_MISC=m
+CONFIG_COREDUMP=y
+CONFIG_IA32_EMULATION=y
+CONFIG_IA32_AOUT=m
+# CONFIG_X86_X32 is not set
+CONFIG_COMPAT=y
+CONFIG_COMPAT_FOR_U64_ALIGNMENT=y
+CONFIG_SYSVIPC_COMPAT=y
+CONFIG_KEYS_COMPAT=y
+CONFIG_X86_DEV_DMA_OPS=y
+CONFIG_PMC_ATOM=y
+CONFIG_VMD=y
+CONFIG_NET=y
+CONFIG_COMPAT_NETLINK_MESSAGES=y
+CONFIG_NET_INGRESS=y
+
+#
+# Networking options
+#
+CONFIG_PACKET=m
+CONFIG_PACKET_DIAG=m
+CONFIG_UNIX=y
+CONFIG_UNIX_DIAG=m
+CONFIG_XFRM=y
+CONFIG_XFRM_ALGO=m
+CONFIG_XFRM_USER=m
+CONFIG_XFRM_SUB_POLICY=y
+CONFIG_XFRM_MIGRATE=y
+# CONFIG_XFRM_STATISTICS is not set
+CONFIG_XFRM_IPCOMP=m
+CONFIG_NET_KEY=m
+CONFIG_NET_KEY_MIGRATE=y
+CONFIG_INET=y
+CONFIG_IP_MULTICAST=y
+CONFIG_IP_ADVANCED_ROUTER=y
+# CONFIG_IP_FIB_TRIE_STATS is not set
+CONFIG_IP_MULTIPLE_TABLES=y
+CONFIG_IP_ROUTE_MULTIPATH=y
+CONFIG_IP_ROUTE_VERBOSE=y
+CONFIG_IP_ROUTE_CLASSID=y
+# CONFIG_IP_PNP is not set
+CONFIG_NET_IPIP=m
+CONFIG_NET_IPGRE_DEMUX=m
+CONFIG_NET_IP_TUNNEL=m
+CONFIG_NET_IPGRE=m
+CONFIG_NET_IPGRE_BROADCAST=y
+CONFIG_IP_MROUTE=y
+CONFIG_IP_MROUTE_MULTIPLE_TABLES=y
+CONFIG_IP_PIMSM_V1=y
+CONFIG_IP_PIMSM_V2=y
+CONFIG_SYN_COOKIES=y
+CONFIG_NET_IPVTI=m
+CONFIG_NET_UDP_TUNNEL=m
+CONFIG_NET_FOU=m
+CONFIG_NET_FOU_IP_TUNNELS=y
+CONFIG_INET_AH=m
+CONFIG_INET_ESP=m
+CONFIG_INET_IPCOMP=m
+CONFIG_INET_XFRM_TUNNEL=m
+CONFIG_INET_TUNNEL=m
+CONFIG_INET_XFRM_MODE_TRANSPORT=m
+CONFIG_INET_XFRM_MODE_TUNNEL=m
+CONFIG_INET_XFRM_MODE_BEET=m
+CONFIG_INET_LRO=m
+CONFIG_INET_DIAG=m
+CONFIG_INET_TCP_DIAG=m
+CONFIG_INET_UDP_DIAG=m
+CONFIG_TCP_CONG_ADVANCED=y
+CONFIG_TCP_CONG_BIC=m
+CONFIG_TCP_CONG_CUBIC=y
+CONFIG_TCP_CONG_WESTWOOD=m
+CONFIG_TCP_CONG_HTCP=m
+CONFIG_TCP_CONG_HSTCP=m
+CONFIG_TCP_CONG_HYBLA=m
+CONFIG_TCP_CONG_VEGAS=m
+CONFIG_TCP_CONG_SCALABLE=m
+CONFIG_TCP_CONG_LP=m
+CONFIG_TCP_CONG_VENO=m
+CONFIG_TCP_CONG_YEAH=m
+CONFIG_TCP_CONG_ILLINOIS=m
+CONFIG_TCP_CONG_DCTCP=m
+CONFIG_TCP_CONG_CDG=m
+CONFIG_DEFAULT_CUBIC=y
+# CONFIG_DEFAULT_RENO is not set
+CONFIG_DEFAULT_TCP_CONG="cubic"
+# CONFIG_TCP_MD5SIG is not set
+CONFIG_IPV6=y
+CONFIG_IPV6_ROUTER_PREF=y
+CONFIG_IPV6_ROUTE_INFO=y
+# CONFIG_IPV6_OPTIMISTIC_DAD is not set
+CONFIG_INET6_AH=m
+CONFIG_INET6_ESP=m
+CONFIG_INET6_IPCOMP=m
+CONFIG_IPV6_MIP6=m
+CONFIG_IPV6_ILA=m
+CONFIG_INET6_XFRM_TUNNEL=m
+CONFIG_INET6_TUNNEL=m
+CONFIG_INET6_XFRM_MODE_TRANSPORT=m
+CONFIG_INET6_XFRM_MODE_TUNNEL=m
+CONFIG_INET6_XFRM_MODE_BEET=m
+CONFIG_INET6_XFRM_MODE_ROUTEOPTIMIZATION=m
+CONFIG_IPV6_VTI=m
+CONFIG_IPV6_SIT=m
+# CONFIG_IPV6_SIT_6RD is not set
+CONFIG_IPV6_NDISC_NODETYPE=y
+CONFIG_IPV6_TUNNEL=m
+CONFIG_IPV6_GRE=m
+CONFIG_IPV6_MULTIPLE_TABLES=y
+CONFIG_IPV6_SUBTREES=y
+# CONFIG_IPV6_MROUTE is not set
+CONFIG_NETLABEL=y
+CONFIG_NETWORK_SECMARK=y
+CONFIG_NET_PTP_CLASSIFY=y
+# CONFIG_NETWORK_PHY_TIMESTAMPING is not set
+CONFIG_NETFILTER=y
+# CONFIG_NETFILTER_DEBUG is not set
+CONFIG_NETFILTER_ADVANCED=y
+CONFIG_BRIDGE_NETFILTER=m
+
+#
+# Core Netfilter Configuration
+#
+CONFIG_NETFILTER_INGRESS=y
+CONFIG_NETFILTER_NETLINK=m
+CONFIG_NETFILTER_NETLINK_ACCT=m
+CONFIG_NETFILTER_NETLINK_QUEUE=m
+CONFIG_NETFILTER_NETLINK_LOG=m
+CONFIG_NF_CONNTRACK=m
+CONFIG_NF_LOG_COMMON=m
+CONFIG_NF_CONNTRACK_MARK=y
+CONFIG_NF_CONNTRACK_SECMARK=y
+CONFIG_NF_CONNTRACK_ZONES=y
+CONFIG_NF_CONNTRACK_PROCFS=y
+CONFIG_NF_CONNTRACK_EVENTS=y
+CONFIG_NF_CONNTRACK_TIMEOUT=y
+CONFIG_NF_CONNTRACK_TIMESTAMP=y
+CONFIG_NF_CONNTRACK_LABELS=y
+CONFIG_NF_CT_PROTO_DCCP=m
+CONFIG_NF_CT_PROTO_GRE=m
+CONFIG_NF_CT_PROTO_SCTP=m
+CONFIG_NF_CT_PROTO_UDPLITE=m
+CONFIG_NF_CONNTRACK_AMANDA=m
+CONFIG_NF_CONNTRACK_FTP=m
+CONFIG_NF_CONNTRACK_H323=m
+CONFIG_NF_CONNTRACK_IRC=m
+CONFIG_NF_CONNTRACK_BROADCAST=m
+CONFIG_NF_CONNTRACK_NETBIOS_NS=m
+CONFIG_NF_CONNTRACK_SNMP=m
+CONFIG_NF_CONNTRACK_PPTP=m
+CONFIG_NF_CONNTRACK_SANE=m
+CONFIG_NF_CONNTRACK_SIP=m
+CONFIG_NF_CONNTRACK_TFTP=m
+CONFIG_NF_CONNTRACK_SLP=m
+CONFIG_NF_CT_NETLINK=m
+CONFIG_NF_CT_NETLINK_TIMEOUT=m
+CONFIG_NF_CT_NETLINK_HELPER=m
+CONFIG_NETFILTER_NETLINK_GLUE_CT=y
+CONFIG_NF_NAT=m
+CONFIG_NF_NAT_NEEDED=y
+CONFIG_NF_NAT_PROTO_DCCP=m
+CONFIG_NF_NAT_PROTO_UDPLITE=m
+CONFIG_NF_NAT_PROTO_SCTP=m
+CONFIG_NF_NAT_AMANDA=m
+CONFIG_NF_NAT_FTP=m
+CONFIG_NF_NAT_IRC=m
+CONFIG_NF_NAT_SIP=m
+CONFIG_NF_NAT_TFTP=m
+CONFIG_NF_NAT_REDIRECT=m
+CONFIG_NETFILTER_SYNPROXY=m
+CONFIG_NF_TABLES=m
+CONFIG_NF_TABLES_INET=m
+CONFIG_NF_TABLES_NETDEV=m
+CONFIG_NFT_EXTHDR=m
+CONFIG_NFT_META=m
+CONFIG_NFT_CT=m
+CONFIG_NFT_RBTREE=m
+CONFIG_NFT_HASH=m
+CONFIG_NFT_COUNTER=m
+CONFIG_NFT_LOG=m
+CONFIG_NFT_LIMIT=m
+CONFIG_NFT_MASQ=m
+CONFIG_NFT_REDIR=m
+CONFIG_NFT_NAT=m
+CONFIG_NFT_QUEUE=m
+CONFIG_NFT_REJECT=m
+CONFIG_NFT_REJECT_INET=m
+CONFIG_NFT_COMPAT=m
+CONFIG_NETFILTER_XTABLES=m
+
+#
+# Xtables combined modules
+#
+CONFIG_NETFILTER_XT_MARK=m
+CONFIG_NETFILTER_XT_CONNMARK=m
+CONFIG_NETFILTER_XT_SET=m
+
+#
+# Xtables targets
+#
+CONFIG_NETFILTER_XT_TARGET_AUDIT=m
+CONFIG_NETFILTER_XT_TARGET_CHECKSUM=m
+CONFIG_NETFILTER_XT_TARGET_CLASSIFY=m
+CONFIG_NETFILTER_XT_TARGET_CONNMARK=m
+CONFIG_NETFILTER_XT_TARGET_CONNSECMARK=m
+CONFIG_NETFILTER_XT_TARGET_CT=m
+CONFIG_NETFILTER_XT_TARGET_DSCP=m
+CONFIG_NETFILTER_XT_TARGET_HL=m
+CONFIG_NETFILTER_XT_TARGET_HMARK=m
+CONFIG_NETFILTER_XT_TARGET_IDLETIMER=m
+CONFIG_NETFILTER_XT_TARGET_LED=m
+CONFIG_NETFILTER_XT_TARGET_LOG=m
+CONFIG_NETFILTER_XT_TARGET_MARK=m
+CONFIG_NETFILTER_XT_NAT=m
+CONFIG_NETFILTER_XT_TARGET_NETMAP=m
+CONFIG_NETFILTER_XT_TARGET_NFLOG=m
+CONFIG_NETFILTER_XT_TARGET_NFQUEUE=m
+CONFIG_NETFILTER_XT_TARGET_NOTRACK=m
+CONFIG_NETFILTER_XT_TARGET_RATEEST=m
+CONFIG_NETFILTER_XT_TARGET_REDIRECT=m
+CONFIG_NETFILTER_XT_TARGET_TEE=m
+CONFIG_NETFILTER_XT_TARGET_TPROXY=m
+CONFIG_NETFILTER_XT_TARGET_TRACE=m
+CONFIG_NETFILTER_XT_TARGET_SECMARK=m
+CONFIG_NETFILTER_XT_TARGET_TCPMSS=m
+CONFIG_NETFILTER_XT_TARGET_TCPOPTSTRIP=m
+
+#
+# Xtables matches
+#
+CONFIG_NETFILTER_XT_MATCH_ADDRTYPE=m
+CONFIG_NETFILTER_XT_MATCH_BPF=m
+CONFIG_NETFILTER_XT_MATCH_CGROUP=m
+CONFIG_NETFILTER_XT_MATCH_CLUSTER=m
+CONFIG_NETFILTER_XT_MATCH_COMMENT=m
+CONFIG_NETFILTER_XT_MATCH_CONNBYTES=m
+CONFIG_NETFILTER_XT_MATCH_CONNLABEL=m
+CONFIG_NETFILTER_XT_MATCH_CONNLIMIT=m
+CONFIG_NETFILTER_XT_MATCH_CONNMARK=m
+CONFIG_NETFILTER_XT_MATCH_CONNTRACK=m
+CONFIG_NETFILTER_XT_MATCH_CPU=m
+CONFIG_NETFILTER_XT_MATCH_DCCP=m
+CONFIG_NETFILTER_XT_MATCH_DEVGROUP=m
+CONFIG_NETFILTER_XT_MATCH_DSCP=m
+CONFIG_NETFILTER_XT_MATCH_ECN=m
+CONFIG_NETFILTER_XT_MATCH_ESP=m
+CONFIG_NETFILTER_XT_MATCH_HASHLIMIT=m
+CONFIG_NETFILTER_XT_MATCH_HELPER=m
+CONFIG_NETFILTER_XT_MATCH_HL=m
+CONFIG_NETFILTER_XT_MATCH_IPCOMP=m
+CONFIG_NETFILTER_XT_MATCH_IPRANGE=m
+CONFIG_NETFILTER_XT_MATCH_IPVS=m
+CONFIG_NETFILTER_XT_MATCH_L2TP=m
+CONFIG_NETFILTER_XT_MATCH_LENGTH=m
+CONFIG_NETFILTER_XT_MATCH_LIMIT=m
+CONFIG_NETFILTER_XT_MATCH_MAC=m
+CONFIG_NETFILTER_XT_MATCH_MARK=m
+CONFIG_NETFILTER_XT_MATCH_MULTIPORT=m
+CONFIG_NETFILTER_XT_MATCH_NFACCT=m
+CONFIG_NETFILTER_XT_MATCH_OSF=m
+CONFIG_NETFILTER_XT_MATCH_OWNER=m
+CONFIG_NETFILTER_XT_MATCH_POLICY=m
+CONFIG_NETFILTER_XT_MATCH_PHYSDEV=m
+CONFIG_NETFILTER_XT_MATCH_PKTTYPE=m
+CONFIG_NETFILTER_XT_MATCH_QUOTA=m
+CONFIG_NETFILTER_XT_MATCH_RATEEST=m
+CONFIG_NETFILTER_XT_MATCH_REALM=m
+CONFIG_NETFILTER_XT_MATCH_RECENT=m
+CONFIG_NETFILTER_XT_MATCH_SCTP=m
+CONFIG_NETFILTER_XT_MATCH_SOCKET=m
+CONFIG_NETFILTER_XT_MATCH_STATE=m
+CONFIG_NETFILTER_XT_MATCH_STATISTIC=m
+CONFIG_NETFILTER_XT_MATCH_STRING=m
+CONFIG_NETFILTER_XT_MATCH_TCPMSS=m
+CONFIG_NETFILTER_XT_MATCH_TIME=m
+CONFIG_NETFILTER_XT_MATCH_U32=m
+CONFIG_IP_SET=m
+CONFIG_IP_SET_MAX=256
+CONFIG_IP_SET_BITMAP_IP=m
+CONFIG_IP_SET_BITMAP_IPMAC=m
+CONFIG_IP_SET_BITMAP_PORT=m
+CONFIG_IP_SET_HASH_IP=m
+CONFIG_IP_SET_HASH_IPMARK=m
+CONFIG_IP_SET_HASH_IPPORT=m
+CONFIG_IP_SET_HASH_IPPORTIP=m
+CONFIG_IP_SET_HASH_IPPORTNET=m
+CONFIG_IP_SET_HASH_MAC=m
+CONFIG_IP_SET_HASH_NETPORTNET=m
+CONFIG_IP_SET_HASH_NET=m
+CONFIG_IP_SET_HASH_NETNET=m
+CONFIG_IP_SET_HASH_NETPORT=m
+CONFIG_IP_SET_HASH_NETIFACE=m
+CONFIG_IP_SET_LIST_SET=m
+CONFIG_IP_VS=m
+CONFIG_IP_VS_IPV6=y
+# CONFIG_IP_VS_DEBUG is not set
+CONFIG_IP_VS_TAB_BITS=12
+
+#
+# IPVS transport protocol load balancing support
+#
+CONFIG_IP_VS_PROTO_TCP=y
+CONFIG_IP_VS_PROTO_UDP=y
+CONFIG_IP_VS_PROTO_AH_ESP=y
+CONFIG_IP_VS_PROTO_ESP=y
+CONFIG_IP_VS_PROTO_AH=y
+CONFIG_IP_VS_PROTO_SCTP=y
+
+#
+# IPVS scheduler
+#
+CONFIG_IP_VS_RR=m
+CONFIG_IP_VS_WRR=m
+CONFIG_IP_VS_LC=m
+CONFIG_IP_VS_WLC=m
+CONFIG_IP_VS_FO=m
+CONFIG_IP_VS_OVF=m
+CONFIG_IP_VS_LBLC=m
+CONFIG_IP_VS_LBLCR=m
+CONFIG_IP_VS_DH=m
+CONFIG_IP_VS_SH=m
+CONFIG_IP_VS_SED=m
+CONFIG_IP_VS_NQ=m
+
+#
+# IPVS SH scheduler
+#
+CONFIG_IP_VS_SH_TAB_BITS=8
+
+#
+# IPVS application helper
+#
+CONFIG_IP_VS_FTP=m
+CONFIG_IP_VS_NFCT=y
+CONFIG_IP_VS_PE_SIP=m
+
+#
+# IP: Netfilter Configuration
+#
+CONFIG_NF_DEFRAG_IPV4=m
+CONFIG_NF_CONNTRACK_IPV4=m
+# CONFIG_NF_CONNTRACK_PROC_COMPAT is not set
+CONFIG_NF_TABLES_IPV4=m
+CONFIG_NFT_CHAIN_ROUTE_IPV4=m
+CONFIG_NFT_REJECT_IPV4=m
+CONFIG_NFT_DUP_IPV4=m
+CONFIG_NF_TABLES_ARP=m
+CONFIG_NF_DUP_IPV4=m
+CONFIG_NF_LOG_ARP=m
+CONFIG_NF_LOG_IPV4=m
+CONFIG_NF_REJECT_IPV4=y
+CONFIG_NF_NAT_IPV4=m
+CONFIG_NFT_CHAIN_NAT_IPV4=m
+CONFIG_NF_NAT_MASQUERADE_IPV4=m
+CONFIG_NFT_MASQ_IPV4=m
+CONFIG_NFT_REDIR_IPV4=m
+CONFIG_NF_NAT_SNMP_BASIC=m
+CONFIG_NF_NAT_PROTO_GRE=m
+CONFIG_NF_NAT_PPTP=m
+CONFIG_NF_NAT_H323=m
+CONFIG_IP_NF_IPTABLES=m
+CONFIG_IP_NF_MATCH_AH=m
+CONFIG_IP_NF_MATCH_ECN=m
+CONFIG_IP_NF_MATCH_RPFILTER=m
+CONFIG_IP_NF_MATCH_TTL=m
+CONFIG_IP_NF_FILTER=m
+CONFIG_IP_NF_TARGET_REJECT=m
+CONFIG_IP_NF_TARGET_SYNPROXY=m
+CONFIG_IP_NF_NAT=m
+CONFIG_IP_NF_TARGET_MASQUERADE=m
+CONFIG_IP_NF_TARGET_NETMAP=m
+CONFIG_IP_NF_TARGET_REDIRECT=m
+CONFIG_IP_NF_MANGLE=m
+CONFIG_IP_NF_TARGET_CLUSTERIP=m
+CONFIG_IP_NF_TARGET_ECN=m
+CONFIG_IP_NF_TARGET_TTL=m
+CONFIG_IP_NF_RAW=m
+CONFIG_IP_NF_SECURITY=m
+CONFIG_IP_NF_ARPTABLES=m
+CONFIG_IP_NF_ARPFILTER=m
+CONFIG_IP_NF_ARP_MANGLE=m
+
+#
+# IPv6: Netfilter Configuration
+#
+CONFIG_NF_DEFRAG_IPV6=m
+CONFIG_NF_CONNTRACK_IPV6=m
+CONFIG_NF_TABLES_IPV6=m
+CONFIG_NFT_CHAIN_ROUTE_IPV6=m
+CONFIG_NFT_REJECT_IPV6=m
+CONFIG_NFT_DUP_IPV6=m
+CONFIG_NF_DUP_IPV6=m
+CONFIG_NF_REJECT_IPV6=y
+CONFIG_NF_LOG_IPV6=m
+CONFIG_NF_NAT_IPV6=m
+CONFIG_NFT_CHAIN_NAT_IPV6=m
+CONFIG_NF_NAT_MASQUERADE_IPV6=m
+CONFIG_NFT_MASQ_IPV6=m
+CONFIG_NFT_REDIR_IPV6=m
+CONFIG_IP6_NF_IPTABLES=m
+CONFIG_IP6_NF_MATCH_AH=m
+CONFIG_IP6_NF_MATCH_EUI64=m
+CONFIG_IP6_NF_MATCH_FRAG=m
+CONFIG_IP6_NF_MATCH_OPTS=m
+CONFIG_IP6_NF_MATCH_HL=m
+CONFIG_IP6_NF_MATCH_IPV6HEADER=m
+CONFIG_IP6_NF_MATCH_MH=m
+CONFIG_IP6_NF_MATCH_RPFILTER=m
+CONFIG_IP6_NF_MATCH_RT=m
+CONFIG_IP6_NF_TARGET_HL=m
+CONFIG_IP6_NF_FILTER=m
+CONFIG_IP6_NF_TARGET_REJECT=m
+CONFIG_IP6_NF_TARGET_SYNPROXY=m
+CONFIG_IP6_NF_MANGLE=m
+CONFIG_IP6_NF_RAW=m
+CONFIG_IP6_NF_SECURITY=m
+CONFIG_IP6_NF_NAT=m
+CONFIG_IP6_NF_TARGET_MASQUERADE=m
+CONFIG_IP6_NF_TARGET_NPT=m
+CONFIG_NF_TABLES_BRIDGE=m
+CONFIG_NFT_BRIDGE_META=m
+CONFIG_NFT_BRIDGE_REJECT=m
+CONFIG_NF_LOG_BRIDGE=m
+CONFIG_BRIDGE_NF_EBTABLES=m
+CONFIG_BRIDGE_EBT_BROUTE=m
+CONFIG_BRIDGE_EBT_T_FILTER=m
+CONFIG_BRIDGE_EBT_T_NAT=m
+CONFIG_BRIDGE_EBT_802_3=m
+CONFIG_BRIDGE_EBT_AMONG=m
+CONFIG_BRIDGE_EBT_ARP=m
+CONFIG_BRIDGE_EBT_IP=m
+CONFIG_BRIDGE_EBT_IP6=m
+CONFIG_BRIDGE_EBT_LIMIT=m
+CONFIG_BRIDGE_EBT_MARK=m
+CONFIG_BRIDGE_EBT_PKTTYPE=m
+CONFIG_BRIDGE_EBT_STP=m
+CONFIG_BRIDGE_EBT_VLAN=m
+CONFIG_BRIDGE_EBT_ARPREPLY=m
+CONFIG_BRIDGE_EBT_DNAT=m
+CONFIG_BRIDGE_EBT_MARK_T=m
+CONFIG_BRIDGE_EBT_REDIRECT=m
+CONFIG_BRIDGE_EBT_SNAT=m
+CONFIG_BRIDGE_EBT_LOG=m
+CONFIG_BRIDGE_EBT_NFLOG=m
+# CONFIG_IP_DCCP is not set
+CONFIG_IP_SCTP=m
+CONFIG_NET_SCTPPROBE=m
+# CONFIG_SCTP_DBG_OBJCNT is not set
+CONFIG_SCTP_DEFAULT_COOKIE_HMAC_MD5=y
+# CONFIG_SCTP_DEFAULT_COOKIE_HMAC_SHA1 is not set
+# CONFIG_SCTP_DEFAULT_COOKIE_HMAC_NONE is not set
+CONFIG_SCTP_COOKIE_HMAC_MD5=y
+# CONFIG_SCTP_COOKIE_HMAC_SHA1 is not set
+CONFIG_RDS=m
+CONFIG_RDS_RDMA=m
+CONFIG_RDS_TCP=m
+# CONFIG_RDS_DEBUG is not set
+CONFIG_TIPC=m
+# CONFIG_TIPC_MEDIA_IB is not set
+CONFIG_TIPC_MEDIA_UDP=y
+CONFIG_ATM=m
+CONFIG_ATM_CLIP=m
+# CONFIG_ATM_CLIP_NO_ICMP is not set
+CONFIG_ATM_LANE=m
+CONFIG_ATM_MPOA=m
+CONFIG_ATM_BR2684=m
+# CONFIG_ATM_BR2684_IPFILTER is not set
+CONFIG_L2TP=m
+CONFIG_L2TP_DEBUGFS=m
+CONFIG_L2TP_V3=y
+CONFIG_L2TP_IP=m
+CONFIG_L2TP_ETH=m
+CONFIG_STP=m
+CONFIG_GARP=m
+CONFIG_MRP=m
+CONFIG_BRIDGE=m
+CONFIG_BRIDGE_IGMP_SNOOPING=y
+CONFIG_BRIDGE_VLAN_FILTERING=y
+CONFIG_HAVE_NET_DSA=y
+CONFIG_NET_DSA=m
+CONFIG_NET_DSA_HWMON=y
+CONFIG_NET_DSA_TAG_BRCM=y
+CONFIG_NET_DSA_TAG_DSA=y
+CONFIG_NET_DSA_TAG_EDSA=y
+CONFIG_NET_DSA_TAG_TRAILER=y
+CONFIG_VLAN_8021Q=m
+CONFIG_VLAN_8021Q_GVRP=y
+CONFIG_VLAN_8021Q_MVRP=y
+# CONFIG_DECNET is not set
+CONFIG_LLC=m
+# CONFIG_LLC2 is not set
+CONFIG_IPX=m
+CONFIG_IPX_INTERN=y
+# CONFIG_ATALK is not set
+# CONFIG_X25 is not set
+# CONFIG_LAPB is not set
+# CONFIG_PHONET is not set
+CONFIG_6LOWPAN=m
+CONFIG_6LOWPAN_NHC=m
+CONFIG_6LOWPAN_NHC_DEST=m
+CONFIG_6LOWPAN_NHC_FRAGMENT=m
+CONFIG_6LOWPAN_NHC_HOP=m
+CONFIG_6LOWPAN_NHC_IPV6=m
+CONFIG_6LOWPAN_NHC_MOBILITY=m
+CONFIG_6LOWPAN_NHC_ROUTING=m
+CONFIG_6LOWPAN_NHC_UDP=m
+# CONFIG_IEEE802154 is not set
+CONFIG_NET_SCHED=y
+
+#
+# Queueing/Scheduling
+#
+CONFIG_NET_SCH_CBQ=m
+CONFIG_NET_SCH_HTB=m
+CONFIG_NET_SCH_HFSC=m
+CONFIG_NET_SCH_ATM=m
+CONFIG_NET_SCH_PRIO=m
+CONFIG_NET_SCH_MULTIQ=m
+CONFIG_NET_SCH_RED=m
+CONFIG_NET_SCH_SFB=m
+CONFIG_NET_SCH_SFQ=m
+CONFIG_NET_SCH_TEQL=m
+CONFIG_NET_SCH_TBF=m
+CONFIG_NET_SCH_GRED=m
+CONFIG_NET_SCH_DSMARK=m
+CONFIG_NET_SCH_NETEM=m
+CONFIG_NET_SCH_DRR=m
+CONFIG_NET_SCH_MQPRIO=m
+CONFIG_NET_SCH_CHOKE=m
+CONFIG_NET_SCH_QFQ=m
+CONFIG_NET_SCH_CODEL=m
+CONFIG_NET_SCH_FQ_CODEL=m
+CONFIG_NET_SCH_FQ=m
+CONFIG_NET_SCH_HHF=m
+CONFIG_NET_SCH_PIE=m
+CONFIG_NET_SCH_INGRESS=m
+CONFIG_NET_SCH_PLUG=m
+
+#
+# Classification
+#
+CONFIG_NET_CLS=y
+CONFIG_NET_CLS_BASIC=m
+CONFIG_NET_CLS_TCINDEX=m
+CONFIG_NET_CLS_ROUTE4=m
+CONFIG_NET_CLS_FW=m
+CONFIG_NET_CLS_U32=m
+CONFIG_CLS_U32_PERF=y
+CONFIG_CLS_U32_MARK=y
+CONFIG_NET_CLS_RSVP=m
+CONFIG_NET_CLS_RSVP6=m
+CONFIG_NET_CLS_FLOW=m
+CONFIG_NET_CLS_CGROUP=m
+CONFIG_NET_CLS_BPF=m
+CONFIG_NET_CLS_FLOWER=m
+CONFIG_NET_EMATCH=y
+CONFIG_NET_EMATCH_STACK=32
+CONFIG_NET_EMATCH_CMP=m
+CONFIG_NET_EMATCH_NBYTE=m
+CONFIG_NET_EMATCH_U32=m
+CONFIG_NET_EMATCH_META=m
+CONFIG_NET_EMATCH_TEXT=m
+CONFIG_NET_EMATCH_CANID=m
+CONFIG_NET_EMATCH_IPSET=m
+CONFIG_NET_CLS_ACT=y
+CONFIG_NET_ACT_POLICE=m
+CONFIG_NET_ACT_GACT=m
+CONFIG_GACT_PROB=y
+CONFIG_NET_ACT_MIRRED=m
+CONFIG_NET_ACT_IPT=m
+CONFIG_NET_ACT_NAT=m
+CONFIG_NET_ACT_PEDIT=m
+CONFIG_NET_ACT_SIMP=m
+CONFIG_NET_ACT_SKBEDIT=m
+CONFIG_NET_ACT_CSUM=m
+CONFIG_NET_ACT_VLAN=m
+CONFIG_NET_ACT_BPF=m
+CONFIG_NET_ACT_CONNMARK=m
+CONFIG_NET_CLS_IND=y
+CONFIG_NET_SCH_FIFO=y
+CONFIG_DCB=y
+CONFIG_DNS_RESOLVER=m
+# CONFIG_BATMAN_ADV is not set
+CONFIG_OPENVSWITCH=m
+CONFIG_OPENVSWITCH_GRE=m
+CONFIG_OPENVSWITCH_VXLAN=m
+CONFIG_OPENVSWITCH_GENEVE=m
+CONFIG_VSOCKETS=m
+CONFIG_VMWARE_VMCI_VSOCKETS=m
+CONFIG_NETLINK_MMAP=y
+CONFIG_NETLINK_DIAG=m
+CONFIG_MPLS=y
+CONFIG_NET_MPLS_GSO=m
+CONFIG_MPLS_ROUTING=m
+CONFIG_MPLS_IPTUNNEL=m
+CONFIG_HSR=m
+CONFIG_NET_SWITCHDEV=y
+CONFIG_NET_L3_MASTER_DEV=y
+CONFIG_RPS=y
+CONFIG_RFS_ACCEL=y
+CONFIG_XPS=y
+CONFIG_CGROUP_NET_PRIO=y
+CONFIG_CGROUP_NET_CLASSID=y
+CONFIG_NET_RX_BUSY_POLL=y
+CONFIG_BQL=y
+CONFIG_BPF_JIT=y
+CONFIG_NET_FLOW_LIMIT=y
+
+#
+# Network testing
+#
+CONFIG_NET_PKTGEN=m
+CONFIG_NET_TCPPROBE=m
+CONFIG_NET_DROP_MONITOR=m
+# CONFIG_HAMRADIO is not set
+CONFIG_CAN=m
+CONFIG_CAN_RAW=m
+CONFIG_CAN_BCM=m
+CONFIG_CAN_GW=m
+
+#
+# CAN Device Drivers
+#
+CONFIG_CAN_VCAN=m
+CONFIG_CAN_SLCAN=m
+CONFIG_CAN_DEV=m
+CONFIG_CAN_CALC_BITTIMING=y
+CONFIG_CAN_LEDS=y
+CONFIG_CAN_SJA1000=m
+# CONFIG_CAN_SJA1000_ISA is not set
+# CONFIG_CAN_SJA1000_PLATFORM is not set
+CONFIG_CAN_EMS_PCI=m
+CONFIG_CAN_PEAK_PCI=m
+CONFIG_CAN_PEAK_PCIEC=y
+CONFIG_CAN_KVASER_PCI=m
+CONFIG_CAN_PLX_PCI=m
+CONFIG_CAN_C_CAN=m
+# CONFIG_CAN_C_CAN_PLATFORM is not set
+CONFIG_CAN_C_CAN_PCI=m
+CONFIG_CAN_M_CAN=m
+# CONFIG_CAN_CC770 is not set
+
+#
+# CAN USB interfaces
+#
+CONFIG_CAN_EMS_USB=m
+CONFIG_CAN_ESD_USB2=m
+CONFIG_CAN_GS_USB=m
+CONFIG_CAN_KVASER_USB=m
+CONFIG_CAN_PEAK_USB=m
+CONFIG_CAN_8DEV_USB=m
+# CONFIG_CAN_SOFTING is not set
+# CONFIG_CAN_DEBUG_DEVICES is not set
+CONFIG_IRDA=m
+
+#
+# IrDA protocols
+#
+CONFIG_IRLAN=m
+CONFIG_IRNET=m
+CONFIG_IRCOMM=m
+CONFIG_IRDA_ULTRA=y
+
+#
+# IrDA options
+#
+CONFIG_IRDA_CACHE_LAST_LSAP=y
+CONFIG_IRDA_FAST_RR=y
+# CONFIG_IRDA_DEBUG is not set
+
+#
+# Infrared-port device drivers
+#
+
+#
+# SIR device drivers
+#
+CONFIG_IRTTY_SIR=m
+
+#
+# Dongle support
+#
+CONFIG_DONGLE=y
+CONFIG_ESI_DONGLE=m
+CONFIG_ACTISYS_DONGLE=m
+CONFIG_TEKRAM_DONGLE=m
+CONFIG_TOIM3232_DONGLE=m
+CONFIG_LITELINK_DONGLE=m
+CONFIG_MA600_DONGLE=m
+CONFIG_GIRBIL_DONGLE=m
+CONFIG_MCP2120_DONGLE=m
+CONFIG_OLD_BELKIN_DONGLE=m
+CONFIG_ACT200L_DONGLE=m
+CONFIG_KINGSUN_DONGLE=m
+CONFIG_KSDAZZLE_DONGLE=m
+CONFIG_KS959_DONGLE=m
+
+#
+# FIR device drivers
+#
+CONFIG_USB_IRDA=m
+CONFIG_SIGMATEL_FIR=m
+CONFIG_NSC_FIR=m
+CONFIG_WINBOND_FIR=m
+CONFIG_SMC_IRCC_FIR=m
+CONFIG_ALI_FIR=m
+CONFIG_VLSI_FIR=m
+CONFIG_VIA_FIR=m
+CONFIG_MCS_FIR=m
+CONFIG_BT=m
+CONFIG_BT_BREDR=y
+CONFIG_BT_RFCOMM=m
+CONFIG_BT_RFCOMM_TTY=y
+CONFIG_BT_BNEP=m
+CONFIG_BT_BNEP_MC_FILTER=y
+CONFIG_BT_BNEP_PROTO_FILTER=y
+CONFIG_BT_CMTP=m
+CONFIG_BT_HIDP=m
+CONFIG_BT_HS=y
+CONFIG_BT_LE=y
+CONFIG_BT_6LOWPAN=m
+# CONFIG_BT_SELFTEST is not set
+# CONFIG_BT_DEBUGFS is not set
+
+#
+# Bluetooth device drivers
+#
+CONFIG_BT_INTEL=m
+CONFIG_BT_BCM=m
+CONFIG_BT_RTL=m
+CONFIG_BT_QCA=m
+CONFIG_BT_HCIBTUSB=m
+CONFIG_BT_HCIBTUSB_BCM=y
+CONFIG_BT_HCIBTUSB_RTL=y
+CONFIG_BT_HCIBTSDIO=m
+CONFIG_BT_HCIUART=m
+CONFIG_BT_HCIUART_H4=y
+CONFIG_BT_HCIUART_BCSP=y
+CONFIG_BT_HCIUART_ATH3K=y
+CONFIG_BT_HCIUART_LL=y
+CONFIG_BT_HCIUART_3WIRE=y
+CONFIG_BT_HCIUART_INTEL=y
+CONFIG_BT_HCIUART_BCM=y
+CONFIG_BT_HCIUART_QCA=y
+CONFIG_BT_HCIBCM203X=m
+CONFIG_BT_HCIBPA10X=m
+CONFIG_BT_HCIBFUSB=m
+CONFIG_BT_HCIVHCI=m
+CONFIG_BT_MRVL=m
+CONFIG_BT_MRVL_SDIO=m
+CONFIG_BT_ATH3K=m
+CONFIG_BT_WILINK=m
+# CONFIG_AF_RXRPC is not set
+CONFIG_FIB_RULES=y
+CONFIG_WIRELESS=y
+CONFIG_WIRELESS_EXT=y
+CONFIG_WEXT_CORE=y
+CONFIG_WEXT_PROC=y
+CONFIG_WEXT_SPY=y
+CONFIG_WEXT_PRIV=y
+CONFIG_CFG80211=m
+# CONFIG_NL80211_TESTMODE is not set
+# CONFIG_CFG80211_DEVELOPER_WARNINGS is not set
+# CONFIG_CFG80211_REG_DEBUG is not set
+# CONFIG_CFG80211_CERTIFICATION_ONUS is not set
+CONFIG_CFG80211_DEFAULT_PS=y
+# CONFIG_CFG80211_DEBUGFS is not set
+# CONFIG_CFG80211_INTERNAL_REGDB is not set
+CONFIG_CFG80211_CRDA_SUPPORT=y
+CONFIG_CFG80211_WEXT=y
+CONFIG_CFG80211_WEXT_EXPORT=y
+CONFIG_LIB80211=m
+CONFIG_LIB80211_CRYPT_WEP=m
+CONFIG_LIB80211_CRYPT_CCMP=m
+CONFIG_LIB80211_CRYPT_TKIP=m
+# CONFIG_LIB80211_DEBUG is not set
+CONFIG_MAC80211=m
+CONFIG_MAC80211_HAS_RC=y
+CONFIG_MAC80211_RC_MINSTREL=y
+CONFIG_MAC80211_RC_MINSTREL_HT=y
+CONFIG_MAC80211_RC_MINSTREL_VHT=y
+CONFIG_MAC80211_RC_DEFAULT_MINSTREL=y
+CONFIG_MAC80211_RC_DEFAULT="minstrel_ht"
+CONFIG_MAC80211_MESH=y
+CONFIG_MAC80211_LEDS=y
+CONFIG_MAC80211_DEBUGFS=y
+# CONFIG_MAC80211_MESSAGE_TRACING is not set
+# CONFIG_MAC80211_DEBUG_MENU is not set
+CONFIG_MAC80211_STA_HASH_MAX_SIZE=0
+# CONFIG_WIMAX&nb