if test "x@ENABLEO2IB@" = "xno"; then \
export IB_OPTIONS="--with-o2ib=no"; \
else \
- if test -n "@O2IBPATH@"; then \
+ if test "x@ENABLEO2IB@" != "xyes" && \
+ test -n "@O2IBPATH@"; then \
export IB_OPTIONS="--with-o2ib=@O2IBPATH@"; \
+ else \
+ export IB_OPTIONS="--with-o2ib=yes"; \
fi; \
fi; \
export KSRC_TREE=$(LINUX) && \
)], [LDISKFS_SERIES="4.4-sles12sp3.series"],
[LDISKFS_SERIES="4.4-sles12sp3.series"])
], [test x$UBUNTU_KERNEL = xyes], [
- AS_VERSION_COMPARE([$LINUXRELEASE],[4.4.0],
- [],
- [
- KPLEV=$(echo $LINUXRELEASE | sed -n 's/.*-\([0-9]\+\).*/\1/p')
- AS_IF(
- [test -z "$KPLEV"], [
- AC_MSG_WARN([Failed to determine Kernel patch level. Assume latest.])
- LDISKFS_SERIES="4.4.0-73-ubuntu14+16.series"
- ],
- [test $KPLEV -ge 73], [LDISKFS_SERIES="4.4.0-73-ubuntu14+16.series"],
- [test $KPLEV -ge 62], [LDISKFS_SERIES="4.4.0-62-ubuntu14+16.series"],
- [test $KPLEV -ge 49], [LDISKFS_SERIES="4.4.0-49-ubuntu14+16.series"],
- [LDISKFS_SERIES="4.4.0-45-ubuntu14+16.series"]
- )
- ],
- [LDISKFS_SERIES="4.4.0-73-ubuntu14+16.series"]
- )
+ AS_VERSION_COMPARE([$LINUXRELEASE],[4.15.0],[
+ AS_VERSION_COMPARE([$LINUXRELEASE],[4.4.0], [],
+ [
+ KPLEV=$(echo $LINUXRELEASE | sed -n 's/.*-\([0-9]\+\).*/\1/p')
+ AS_IF(
+ [test -z "$KPLEV"], [
+ AC_MSG_WARN([Failed to determine Kernel patch level. Assume latest.])
+ LDISKFS_SERIES="4.4.0-73-ubuntu14+16.series"
+ ],
+ [test $KPLEV -ge 73], [LDISKFS_SERIES="4.4.0-73-ubuntu14+16.series"],
+ [test $KPLEV -ge 62], [LDISKFS_SERIES="4.4.0-62-ubuntu14+16.series"],
+ [test $KPLEV -ge 49], [LDISKFS_SERIES="4.4.0-49-ubuntu14+16.series"],
+ [LDISKFS_SERIES="4.4.0-45-ubuntu14+16.series"]
+ )
+ ],
+ [LDISKFS_SERIES="4.4.0-73-ubuntu14+16.series"])],
+ [
+ KPLEV=$(echo $LINUXRELEASE | sed -n 's/.*-\([0-9]\+\).*/\1/p')
+ AS_IF(
+ [test -z "$KPLEV"], [
+ AC_MSG_WARN([Failed to determine Kernel patch level. Assume latest.])
+ LDISKFS_SERIES="4.15.0-20-ubuntu18.series"
+ ],
+ [test $KPLEV -ge 20], [LDISKFS_SERIES="4.15.0-20-ubuntu18.series"]
+ )
+ ],
+ [LDISKFS_SERIES="4.15.0-20-ubuntu18.series"])
])
])
AS_IF([test -z "$LDISKFS_SERIES"],
# Find paths for linux, handling kernel-source rpms
#
AC_DEFUN([LB_LINUX_PATH], [
-for DEFAULT_LINUX in /lib/modules/$(uname -r)/{source,build} /usr/src/linux $(find /usr/src/kernels/ -maxdepth 1 -name @<:@0-9@:>@\* | xargs -r ls -d | tail -n 1); do
+for DEFAULT_LINUX in /usr/src/linux-source-* /lib/modules/$(uname -r)/{source,build} /usr/src/linux $(find /usr/src/kernels/ -maxdepth 1 -name @<:@0-9@:>@\* | xargs -r ls -d | tail -n 1); do
AS_IF([readlink -q -e $DEFAULT_LINUX >/dev/null], [break])
done
if test "$DEFAULT_LINUX" = "/lib/modules/$(uname -r)/source"; then
PATHS="/lib/modules/$(uname -r)/build"
+else
+ PATHS="/usr/src/linux-headers-$(uname -r)"
fi
PATHS+=" $DEFAULT_LINUX"
for DEFAULT_LINUX_OBJ in $PATHS; do
dnl # * /var/lib/dkms/zfs/${VERSION}/source
dnl # * /usr/src/zfs-${VERSION}/${LINUXRELEASE}
dnl # * /usr/src/zfs-${VERSION}
-dnl # * ../spl/
-dnl # * $LINUX
+dnl # * ../zfs/
+dnl # * $LINUX/zfs
dnl #
dnl # --with-zfs-devel=path
dnl # - User provided directory where zfs development headers
dnl # * /usr/src/spl-${VERSION}/${LINUXRELEASE}
dnl # * /usr/src/spl-${VERSION}
dnl # * ../spl/
-dnl # * $LINUX
+dnl # * $LINUX/spl
dnl #
dnl # --with-spl=path - Enable spl support and use the spl headers in the
dnl # provided path. No autodetection is performed.
splsrc1="/usr/src/spl-${splver}/${LINUXRELEASE}"
splsrc2="/usr/src/spl-${splver}"
splsrc3="../spl/"
- splsrc4="$LINUX"
+ splsrc4="$LINUX/spl"
AC_MSG_CHECKING([spl source directory])
AS_IF([test -z "${splsrc}"], [
zfssrc1="/usr/src/zfs-${zfsver}/${LINUXRELEASE}"
zfssrc2="/usr/src/zfs-${zfsver}"
zfssrc3="../zfs/"
- zfssrc4="$LINUX"
+ zfssrc4="$LINUX/zfs"
AC_MSG_CHECKING([zfs source directory])
AS_IF([test -z "${zfssrc}"], [
+lustre (2.11.52-108-g258f0cf-dirty-1) unstable; urgency=low
+
+ * Automated changelog entry update
+
+ -- Brian J. Murrell <brian@interlinx.bc.ca> Wed, 18 Jul 2018 08:36:49 -0600
+
lustre (2.7.50-1) unstable; urgency=low
* Automated changelog entry update
Maintainer: Brian J. Murrell <brian.murrell@intel.com>
Uploaders: Brian J. Murrell <brian.murrell@intel.com>
Standards-Version: 3.8.3
-Build-Depends: module-assistant, libreadline-dev, debhelper (>=7.0.0), dpatch, automake (>=1.7) | automake1.7 | automake1.8 | automake1.9, libtool, libyaml-dev, libselinux-dev, libsnmp-dev, mpi-default-dev, bzip2, quilt, linux-headers-generic | linux-headers | linux-headers-amd64, rsync
-Homepage: https://wiki.hpdd.intel.com/
-Vcs-Git: git://git.hpdd.intel.com/fs/lustre-release.git
+Build-Depends: module-assistant, libreadline-dev, debhelper (>=9.0.0), dpatch, automake (>=1.7) | automake1.7 | automake1.8 | automake1.9, pkg-config, libtool, libyaml-dev, libselinux-dev, libsnmp-dev, mpi-default-dev, bzip2, quilt, linux-headers-generic | linux-headers | linux-headers-amd64, rsync
+Homepage: https://wiki.whamcloud.com/
+Vcs-Git: git://git.whamcloud.com/fs/lustre-release.git
Package: lustre-source
Section: admin
Architecture: all
Priority: optional
-Depends: module-assistant, bzip2, debhelper (>= 7.0.0), libtool, libyaml-dev, libselinux-dev, libsnmp-dev, mpi-default-dev, dpatch
+Depends: module-assistant, bzip2, debhelper (>= 9.0.0), libtool, libyaml-dev, libselinux-dev, libsnmp-dev, mpi-default-dev, dpatch, pkg-config
Description: source for Lustre filesystem client kernel modules
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package contains the module source. The client kernel modules
can be built for kernels 2.6.32+ with the use of module-assistant
Depends: ${shlibs:Depends}, ${misc:Depends}, libyaml-0-2, libselinux1, libsnmp30, zlib1g, perl
Description: Userspace utilities for the Lustre filesystem (client)
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides a number of userspace utilities for
accessing Lustre filesystems from a client. If you need server utils,
Architecture: i386 powerpc ppc64el amd64 ia64 arm64
Priority: optional
Depends: ${shlibs:Depends}, ${misc:Depends}, libyaml-0-2, libselinux1, libsnmp30, zlib1g, perl
-Provides: lustre-server-utils, lustre-client-utils
+Provides: lustre-server-utils, lustre-client-utils (= ${binary:Version})
Conflicts: lustre-client-utils
Replaces: lustre-client-utils
Description: Userspace utilities for the Lustre filesystem (server)
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides a number of userspace utilities for
accessing and maintaining Lustre filesystems from a server.
Depends: lustre-client-utils (= ${binary:Version}), python, perl, sg3-utils
Description: Collection of benchmark tools for the Lustre filesystem
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides a collection of benchmark tools
Depends: lustre-iokit (= ${binary:Version}), attr, rsync, perl, lsof, mpi-default-bin
Description: Test suite for the Lustre filesystem
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides a number of test utilities for the Lustre filesystem.
Depends: lustre-client-utils (= ${binary:Version})
Description: Development files for the Lustre filesystem
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides development libraries for the Lustre filesystem.
Maintainer: Brian J. Murrell <brian.murrell@intel.com>
Uploaders: Brian J. Murrell <brian.murrell@intel.com>
Standards-Version: 3.8.3
-Build-Depends: module-assistant, libreadline-dev, debhelper (>=7.0.0), dpatch, automake (>=1.7) | automake1.7 | automake1.8 | automake1.9, libtool, libyaml-dev, libselinux-dev, libsnmp-dev, mpi-default-dev, bzip2, quilt, linux-headers-generic | linux-headers | linux-headers-amd64, rsync
-Homepage: https://wiki.hpdd.intel.com/
-Vcs-Git: git://git.hpdd.intel.com/fs/lustre-release.git
+Build-Depends: module-assistant, libreadline-dev, debhelper (>=9.0.0), dpatch, automake (>=1.7) | automake1.7 | automake1.8 | automake1.9, pkg-config, libtool, libyaml-dev, libselinux-dev, libsnmp-dev, mpi-default-dev, bzip2, quilt, linux-headers-generic | linux-headers | linux-headers-amd64, rsync
+Homepage: https://wiki.whamcloud.com/
+Vcs-Git: git://git.whamcloud.com/fs/lustre-release.git
Package: lustre-source
Section: admin
Architecture: all
Priority: optional
-Depends: module-assistant, bzip2, debhelper (>= 7.0.0), libtool, libyaml-dev, libselinux-dev, libsnmp-dev, mpi-default-dev, dpatch
+Depends: module-assistant, bzip2, debhelper (>= 9.0.0), libtool, libyaml-dev, libselinux-dev, libsnmp-dev, mpi-default-dev, dpatch, pkg-config
Description: source for Lustre filesystem client kernel modules
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package contains the module source. The client kernel modules
can be built for kernels 2.6.32+ with the use of module-assistant
Depends: ${shlibs:Depends}, ${misc:Depends}, libyaml-0-2, libselinux1, libsnmp30, zlib1g, perl
Description: Userspace utilities for the Lustre filesystem (client)
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides a number of userspace utilities for
accessing Lustre filesystems from a client. If you need server utils,
Architecture: i386 powerpc ppc64el amd64 ia64 arm64
Priority: optional
Depends: ${shlibs:Depends}, ${misc:Depends}, libyaml-0-2, libselinux1, libsnmp30, zlib1g, perl
-Provides: lustre-server-utils, lustre-client-utils
+Provides: lustre-server-utils, lustre-client-utils (= ${binary:Version})
Conflicts: lustre-client-utils
Replaces: lustre-client-utils
Description: Userspace utilities for the Lustre filesystem (server)
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides a number of userspace utilities for
accessing and maintaining Lustre filesystems from a server.
Depends: lustre-client-utils (= ${binary:Version}), python, perl, sg3-utils
Description: Collection of benchmark tools for the Lustre filesystem
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides a collection of benchmark tools
Depends: lustre-iokit (= ${binary:Version}), attr, rsync, perl, lsof, mpi-default-bin
Description: Test suite for the Lustre filesystem
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides a number of test utilities for the Lustre filesystem.
Depends: lustre-client-utils (= ${binary:Version})
Description: Development files for the Lustre filesystem
Lustre is a scalable, secure, robust, highly-available cluster file system.
- This release is maintained by Intel Corporation and available from
- https://wiki.hpdd.intel.com/
+ This release is maintained by Whamcloud and available from
+ https://wiki.whamcloud.com/
.
This package provides development libraries for the Lustre filesystem.
Maintainer: Brian J. Murrell <brian.murrell@intel.com>
Uploaders: Brian J. Murrell <brian.murrell@intel.com>
Standards-Version: 3.7.2
-Build-Depends: debhelper (>= 7.0.0), bzip2
+Build-Depends: debhelper (>= 9.0.0), bzip2
Package: lustre-client-modules-_KVERS_
Architecture: any
+++ /dev/null
-/etc/lnet.conf
-/etc/lnet_routes.conf
-/etc/modprobe.d/ko2iblnd.conf
-/etc/udev/rules.d/99-lustre.rules
+++ /dev/null
-/etc/ldev.conf
-/etc/lnet.conf
-/etc/lnet_routes.conf
-/etc/modprobe.d/ko2iblnd.conf
-/etc/udev/rules.d/99-lustre.rules
EXT4_FEATURE_INCOMPAT_MMP| \
EXT4_FEATURE_INCOMPAT_DIRDATA)
-@@ -1695,6 +1697,12 @@ 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
- */
-
@@ -1706,6 +1714,10 @@ struct mmpd_data {
# define ATTRIB_NORET __attribute__((noreturn))
# define NORET_AND noreturn,
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,
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,
EXT4_FEATURE_INCOMPAT_MMP| \
EXT4_FEATURE_INCOMPAT_DIRDATA)
-@@ -1764,6 +1765,12 @@ 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
- */
-
@@ -1775,6 +1782,10 @@ struct mmpd_data {
# define ATTRIB_NORET __attribute__((noreturn))
# define NORET_AND noreturn,
EXT4_FEATURE_INCOMPAT_MMP | \
EXT4_FEATURE_INCOMPAT_DIRDATA| \
EXT4_FEATURE_INCOMPAT_INLINE_DATA)
-@@ -1945,6 +1946,12 @@ 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
- */
-
@@ -1956,6 +1963,10 @@ struct mmpd_data {
# define ATTRIB_NORET __attribute__((noreturn))
# define NORET_AND noreturn,
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,
EXT4_FEATURE_INCOMPAT_MMP | \
EXT4_FEATURE_INCOMPAT_DIRDATA| \
EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
-@@ -2241,6 +2242,12 @@ 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
- */
-
@@ -2252,6 +2259,10 @@ struct mmpd_data {
# define ATTRIB_NORET __attribute__((noreturn))
# define NORET_AND noreturn,
--- /dev/null
+Index: linux-4.15.0/fs/ext4/inode.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/inode.c
++++ linux-4.15.0/fs/ext4/inode.c
+@@ -728,6 +728,9 @@ out_sem:
+ !(flags & EXT4_GET_BLOCKS_ZERO) &&
+ !ext4_is_quota_file(inode) &&
+ ext4_should_order_data(inode)) {
++ ret = ext4_inode_attach_jinode(inode);
++ if (ret)
++ return ret;
+ if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
+ ret = ext4_jbd2_inode_add_wait(handle, inode);
+ else
+@@ -2806,6 +2809,9 @@ static int ext4_writepages(struct addres
+ mpd.last_page = wbc->range_end >> PAGE_SHIFT;
+ }
+
++ ret = ext4_inode_attach_jinode(inode);
++ if (ret)
++ goto out_writepages;
+ mpd.inode = inode;
+ mpd.wbc = wbc;
+ ext4_io_submit_init(&mpd.io_submit, wbc);
+@@ -4356,6 +4362,7 @@ int ext4_inode_attach_jinode(struct inod
+ jbd2_free_inode(jinode);
+ return 0;
+ }
++EXPORT_SYMBOL(ext4_inode_attach_jinode);
+
+ /*
+ * ext4_truncate()
+Index: linux-4.15.0/fs/ext4/ext4.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4.h
++++ linux-4.15.0/fs/ext4/ext4.h
+@@ -2595,6 +2595,7 @@ extern int ext4_trim_fs(struct super_blo
+ extern void ext4_process_freed_data(struct super_block *sb, tid_t commit_tid);
+
+ /* inode.c */
++#define HAVE_LDISKFS_INFO_JINODE
+ int ext4_inode_is_fast_symlink(struct inode *inode);
+ struct buffer_head *ext4_getblk(handle_t *, struct inode *, ext4_lblk_t, int);
+ struct buffer_head *ext4_bread(handle_t *, struct inode *, ext4_lblk_t, int);
--- /dev/null
+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
+
+Index: linux-4.15.0/fs/ext4/balloc.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/balloc.c
++++ linux-4.15.0/fs/ext4/balloc.c
+@@ -185,25 +185,17 @@ static int ext4_init_block_bitmap(struct
+ 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);
+@@ -371,7 +363,6 @@ static int ext4_validate_block_bitmap(st
+ {
+ 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;
+@@ -382,22 +373,19 @@ static int ext4_validate_block_bitmap(st
+ 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);
+@@ -451,8 +439,6 @@ ext4_read_block_bitmap_nowait(struct sup
+ 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;
+Index: linux-4.15.0/fs/ext4/ext4.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4.h
++++ linux-4.15.0/fs/ext4/ext4.h
+@@ -101,6 +101,18 @@ 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)
++
+ enum SHIFT_DIRECTION {
+ SHIFT_LEFT = 0,
+ SHIFT_RIGHT,
+@@ -3032,7 +3044,11 @@ 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)))
+Index: linux-4.15.0/fs/ext4/ialloc.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ialloc.c
++++ linux-4.15.0/fs/ext4/ialloc.c
+@@ -72,25 +72,15 @@ static int ext4_init_inode_bitmap(struct
+ 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;
+ }
+
+@@ -194,8 +184,6 @@ ext4_read_inode_bitmap(struct super_bloc
+ 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;
+@@ -371,14 +359,9 @@ 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:
+Index: linux-4.15.0/fs/ext4/mballoc.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/mballoc.c
++++ linux-4.15.0/fs/ext4/mballoc.c
+@@ -751,10 +751,18 @@ int ext4_mb_generate_buddy(struct super_
+ 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);
+@@ -1110,7 +1118,7 @@ ext4_mb_load_buddy_gfp(struct super_bloc
+ 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);
+@@ -1136,7 +1144,7 @@ ext4_mb_load_buddy_gfp(struct super_bloc
+ */
+ ret = ext4_mb_init_group(sb, group, gfp);
+ if (ret)
+- return ret;
++ goto err;
+ }
+
+ /*
+@@ -1239,6 +1247,7 @@ err:
+ put_page(e4b->bd_buddy_page);
+ e4b->bd_buddy = NULL;
+ e4b->bd_bitmap = NULL;
++ ext4_warning(sb, "Error loading buddy information for %u", group);
+ return ret;
+ }
+
+@@ -3689,9 +3698,11 @@ int ext4_mb_check_ondisk_bitmap(struct s
+ }
+
+ if (free != free_in_gdp) {
+- ext4_error(sb, "on-disk bitmap for group %d"
+- "corrupted: %u blocks free in bitmap, %u - in gd\n",
+- group, free, free_in_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,
++ free_in_gdp);
+ return -EIO;
+ }
+ return 0;
+@@ -4052,16 +4063,8 @@ ext4_mb_release_inode_pa(struct ext4_bud
+ /* "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.
+@@ -4123,15 +4126,11 @@ ext4_mb_discard_group_preallocations(str
+ 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_warning(sb, "Error %d loading buddy information for %u",
+- err, group);
+ put_bh(bitmap_bh);
+ return 0;
+ }
+@@ -4292,17 +4291,12 @@ repeat:
+
+ err = ext4_mb_load_buddy_gfp(sb, group, &e4b,
+ GFP_NOFS|__GFP_NOFAIL);
+- if (err) {
+- ext4_error(sb, "Error %d loading buddy information for %u",
+- err, 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;
+ }
+@@ -4565,11 +4559,8 @@ ext4_mb_discard_lg_preallocations(struct
+ group = ext4_get_group_number(sb, pa->pa_pstart);
+ err = ext4_mb_load_buddy_gfp(sb, group, &e4b,
+ GFP_NOFS|__GFP_NOFAIL);
+- if (err) {
+- ext4_error(sb, "Error %d loading buddy information for %u",
+- err, group);
++ if (err)
+ continue;
+- }
+ ext4_lock_group(sb, group);
+ list_del(&pa->pa_group_list);
+ ext4_get_group_info(sb, group)->bb_prealloc_nr--;
+@@ -4822,17 +4813,18 @@ 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);
+@@ -5118,7 +5110,7 @@ do_more:
+ err = ext4_mb_load_buddy_gfp(sb, block_group, &e4b,
+ GFP_NOFS|__GFP_NOFAIL);
+ if (err)
+- goto error_return;
++ goto error_brelse;
+
+ /*
+ * We need to make sure we don't reuse the freed block until after the
+@@ -5200,8 +5192,9 @@ do_more:
+ goto do_more;
+ }
+ error_return:
+- brelse(bitmap_bh);
+ ext4_std_error(sb, err);
++error_brelse:
++ brelse(bitmap_bh);
+ return;
+ }
+
+@@ -5301,7 +5294,7 @@ int ext4_group_add_blocks(handle_t *hand
+
+ 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
+@@ -5339,8 +5332,9 @@ int ext4_group_add_blocks(handle_t *hand
+ err = ret;
+
+ error_return:
+- brelse(bitmap_bh);
+ ext4_std_error(sb, err);
++error_brelse:
++ brelse(bitmap_bh);
+ return err;
+ }
+
+@@ -5415,11 +5409,9 @@ ext4_trim_all_free(struct super_block *s
+ trace_ext4_trim_all_free(sb, group, start, max);
+
+ ret = ext4_mb_load_buddy(sb, group, &e4b);
+- if (ret) {
+- ext4_warning(sb, "Error %d loading buddy information for %u",
+- ret, group);
++ if (ret)
+ return ret;
+- }
++
+ bitmap = e4b.bd_bitmap;
+
+ ext4_lock_group(sb, group);
+Index: linux-4.15.0/fs/ext4/super.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/super.c
++++ linux-4.15.0/fs/ext4/super.c
+@@ -708,6 +708,37 @@ void __ext4_warning_inode(const struct i
+ 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,
--- /dev/null
+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-4.15.0/fs/ext4/dir.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/dir.c
++++ linux-4.15.0/fs/ext4/dir.c
+@@ -68,11 +68,11 @@ 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";
+@@ -219,7 +219,7 @@ static int ext4_readdir(struct file *fil
+ * 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);
+@@ -442,12 +442,17 @@ 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;
+@@ -456,7 +461,7 @@ int ext4_htree_store_dirent(struct file
+ new_fn->inode = le32_to_cpu(dirent->inode);
+ new_fn->name_len = ent_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) {
+Index: linux-4.15.0/fs/ext4/ext4.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4.h
++++ linux-4.15.0/fs/ext4/ext4.h
+@@ -1055,6 +1055,7 @@ struct ext4_inode_info {
+ __u32 i_csum_seed;
+
+ kprojid_t i_projid;
++ void *i_dirdata;
+ };
+
+ /*
+@@ -1098,6 +1099,7 @@ 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 0x40000 /* Some quota option set */
+ #define EXT4_MOUNT_USRQUOTA 0x80000 /* "old" user quota,
+ * enable enforcement for hidden
+@@ -1768,6 +1770,7 @@ EXT4_FEATURE_INCOMPAT_FUNCS(encrypt, EN
+ EXT4_FEATURE_INCOMPAT_FLEX_BG| \
+ EXT4_FEATURE_INCOMPAT_EA_INODE| \
+ EXT4_FEATURE_INCOMPAT_MMP | \
++ EXT4_FEATURE_INCOMPAT_DIRDATA| \
+ EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
+ EXT4_FEATURE_INCOMPAT_ENCRYPT | \
+ EXT4_FEATURE_INCOMPAT_CSUM_SEED | \
+@@ -1943,6 +1946,43 @@ 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_feature_dirdata(sb))
++ return NULL;
++ if (p && p->edp_magic == EXT4_LUFID_MAGIC)
++ return &p->edp_len;
++ else
++ return NULL;
++}
+
+ #define EXT4_FT_DIR_CSUM 0xDE
+
+@@ -1953,8 +1993,11 @@ 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)
+
+ /*
+@@ -2357,11 +2400,11 @@ 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);
+ 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);
+ static inline void ext4_update_dx_flag(struct inode *inode)
+ {
+ if (!ext4_has_feature_dir_index(inode->i_sb))
+@@ -2373,10 +2416,17 @@ static const unsigned char ext4_filetype
+
+ static inline unsigned char get_dtype(struct super_block *sb, int filetype)
+ {
+- if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
++ int fl_index = filetype & EXT4_FT_MASK;
++
++ 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);
++
+ }
+ extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
+ void *buf, int buf_size);
+@@ -2525,6 +2575,8 @@ extern struct inode *ext4_create_inode(h
+ 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,
+@@ -3265,6 +3317,28 @@ static inline void ext4_clear_io_unwritt
+
+ extern const struct iomap_ops ext4_iomap_ops;
+
++/*
++ * 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__ */
+
+ #define EFSBADCRC EBADMSG /* Bad CRC detected */
+Index: linux-4.15.0/fs/ext4/namei.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/namei.c
++++ linux-4.15.0/fs/ext4/namei.c
+@@ -242,7 +242,8 @@ 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(struct ext4_filename *fname,
+ struct inode *dir,
+@@ -505,11 +506,12 @@ ext4_next_entry(struct ext4_dir_entry_2
+ */
+ struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
+ {
++ BUG_ON(de->name_len != 1);
+ /* get dotdot first */
+- de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
++ 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));
++ de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
+
+ return (struct dx_root_info *)de;
+ }
+@@ -554,10 +556,16 @@ 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_metadata_csum(dir->i_sb))
+ entry_space -= sizeof(struct dx_tail);
+@@ -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_metadata_csum(dir->i_sb))
+ entry_space -= sizeof(struct dx_tail);
+@@ -678,7 +686,7 @@ static struct stats dx_show_leaf(struct
+ (unsigned) ((char *) de - base));
+ #endif
+ }
+- space += EXT4_DIR_REC_LEN(de->name_len);
++ space += EXT4_DIR_REC_LEN(de);
+ names++;
+ }
+ de = ext4_next_entry(de, size);
+@@ -785,11 +793,14 @@ dx_probe(struct ext4_filename *fname, st
+
+ 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;
+ }
+
+@@ -981,7 +992,7 @@ 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)) {
+@@ -1653,7 +1664,7 @@ 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);
+@@ -1677,7 +1688,7 @@ 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);
+@@ -1808,14 +1819,16 @@ 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;
+
++ dlen ? *dlen = 0 : 0; /* default set to 0 */
+ de = (struct ext4_dir_entry_2 *)buf;
+ top = buf + buf_size - reclen;
+ while ((char *) de <= top) {
+@@ -1824,10 +1837,26 @@ int ext4_find_dest_de(struct inode *dir,
+ return -EFSCORRUPTED;
+ if (ext4_match(fname, de))
+ return -EEXIST;
+- 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;
+ }
+@@ -1841,12 +1870,12 @@ 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 =
+@@ -1860,6 +1889,11 @@ 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;
++ }
+ }
+
+ /*
+@@ -1877,14 +1911,19 @@ static int add_dirent_to_buf(handle_t *h
+ {
+ 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 *)
++ EXT4_I(inode)->i_dirdata);
+ 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;
+ }
+@@ -1896,7 +1935,10 @@ static int add_dirent_to_buf(handle_t *h
+ }
+
+ /* By now the buffer is marked for journaling */
+- ext4_insert_dentry(inode, de, blocksize, fname);
++ /* If writing the short form of "dotdot", don't add the data section */
++ if (dlen == 1)
++ data = NULL;
++ ext4_insert_dentry(inode, de, blocksize, fname, data);
+
+ /*
+ * XXX shouldn't update any times until successful
+@@ -2005,7 +2047,8 @@ 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 */
+ fname->hinfo.hash_version = dx_info->hash_version;
+@@ -2055,6 +2098,8 @@ 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);
+@@ -2072,19 +2117,24 @@ 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) {
+@@ -2098,10 +2148,15 @@ 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) {
+@@ -2140,6 +2195,7 @@ static int ext4_add_entry(handle_t *hand
+ ext4_lblk_t block, blocks;
+ int csum_size = 0;
+
++ EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
+ if (ext4_has_metadata_csum(inode->i_sb))
+ csum_size = sizeof(struct ext4_dir_entry_tail);
+
+@@ -2661,37 +2717,70 @@ err_unlock_inode:
+ 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);
+- strcpy(de->name, "..");
+- ext4_set_de_type(inode->i_sb, de, S_IFDIR);
++ EXT4_DIR_REC_LEN(de), blocksize);
+
+ return ext4_next_entry(de, blocksize);
+ }
+
+ 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;
+@@ -2716,7 +2805,11 @@ static int ext4_init_new_dir(handle_t *h
+ if (IS_ERR(dir_block))
+ return PTR_ERR(dir_block);
+ 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 *)(¶m), de, blocksize,
++ csum_size, dir->i_ino, 10);
+ set_nlink(inode, 2);
+ if (csum_size) {
+ t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
+@@ -2733,6 +2826,29 @@ 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;
+@@ -2759,7 +2875,7 @@ 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);
+@@ -2811,7 +2927,7 @@ bool ext4_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 true;
+ }
+Index: linux-4.15.0/fs/ext4/inline.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/inline.c
++++ linux-4.15.0/fs/ext4/inline.c
+@@ -1017,7 +1017,7 @@ static int ext4_add_dirent_to_inline(han
+ struct ext4_dir_entry_2 *de;
+
+ err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
+- inline_size, fname, &de);
++ inline_size, fname, &de, NULL);
+ if (err)
+ return err;
+
+@@ -1025,7 +1025,7 @@ static int ext4_add_dirent_to_inline(han
+ err = ext4_journal_get_write_access(handle, iloc->bh);
+ if (err)
+ return err;
+- ext4_insert_dentry(inode, de, inline_size, fname);
++ ext4_insert_dentry(inode, de, inline_size, fname, NULL);
+
+ ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
+
+@@ -1094,7 +1094,7 @@ 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,
+@@ -1375,7 +1375,7 @@ 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;
+@@ -1385,7 +1385,7 @@ 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;
+@@ -1483,8 +1483,8 @@ 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;
+
+@@ -1519,7 +1519,7 @@ int ext4_read_inline_dir(struct file *fi
+ * 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-4.15.0/fs/ext4/super.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/super.c
++++ linux-4.15.0/fs/ext4/super.c
+@@ -1347,7 +1347,7 @@ enum {
+ Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
+ 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_prjquota, Opt_i_version, Opt_dax,
+ Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
+ Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
+@@ -1420,6 +1420,7 @@ static const match_table_t tokens = {
+ {Opt_nolazytime, "nolazytime"},
+ {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
+ {Opt_nodelalloc, "nodelalloc"},
++ {Opt_dirdata, "dirdata"},
+ {Opt_removed, "mblk_io_submit"},
+ {Opt_removed, "nomblk_io_submit"},
+ {Opt_block_validity, "block_validity"},
+@@ -1641,6 +1642,7 @@ 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},
--- /dev/null
+When ldiskfs run in failover mode whith read-only disk.
+Part of allocation updates are lost and ldiskfs may fail
+while mounting this is due to inconsistent state of
+group-descriptor. Group-descriptor check is added after
+journal replay.
+
+Index: linux-4.15.0/fs/ext4/super.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/super.c
++++ linux-4.15.0/fs/ext4/super.c
+@@ -4051,11 +4051,6 @@ static int ext4_fill_super(struct super_
+ goto failed_mount2;
+ }
+ }
+- if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
+- ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
+- ret = -EFSCORRUPTED;
+- goto failed_mount2;
+- }
+
+ sbi->s_gdb_count = db_count;
+
+@@ -4196,6 +4191,13 @@ static int ext4_fill_super(struct super_
+ sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
+
+ no_journal:
++
++ if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
++ ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
++ ret = -EFSCORRUPTED;
++ goto failed_mount_wq;
++ }
++
+ if (!test_opt(sb, NO_MBCACHE)) {
+ sbi->s_ea_block_cache = ext4_xattr_create_cache();
+ if (!sbi->s_ea_block_cache) {
--- /dev/null
+Index: linux-4.15.0/fs/ext4/ext4.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4.h
++++ linux-4.15.0/fs/ext4/ext4.h
+@@ -1450,6 +1450,7 @@ 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;
+Index: linux-4.15.0/fs/ext4/namei.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/namei.c
++++ linux-4.15.0/fs/ext4/namei.c
+@@ -751,12 +751,20 @@ 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;
+@@ -2508,6 +2516,54 @@ out:
+ return err;
+ }
+
++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;
++}
++
+ /*
+ * ext4_add_entry()
+ *
+@@ -2629,6 +2685,7 @@ static int ext4_dx_add_entry(handle_t *h
+ struct ext4_dir_entry_2 *de;
+ int restart;
+ int err;
++ bool ret_warn = false;
+
+ again:
+ restart = 0;
+@@ -2657,6 +2714,11 @@ 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;
+@@ -2819,6 +2881,8 @@ cleanup:
+ */
+ if (restart && err == 0)
+ goto again;
++ if (err == 0 && ret_warn)
++ err = -ENOBUFS;
+ return err;
+ }
+
+Index: linux-4.15.0/fs/ext4/super.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/super.c
++++ linux-4.15.0/fs/ext4/super.c
+@@ -1804,6 +1804,8 @@ static int handle_mount_opt(struct super
+ 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) {
+Index: linux-4.15.0/fs/ext4/sysfs.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/sysfs.c
++++ linux-4.15.0/fs/ext4/sysfs.c
+@@ -173,6 +173,7 @@ EXT4_ATTR_OFFSET(inode_readahead_blks, 0
+ 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);
+@@ -204,6 +205,7 @@ 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),
--- /dev/null
+Index: linux-4.15.0/fs/ext4/namei.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/namei.c
++++ linux-4.15.0/fs/ext4/namei.c
+@@ -2043,6 +2043,74 @@ out_frames:
+ 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()
+ *
+@@ -2091,6 +2159,9 @@ static int ext4_add_entry(handle_t *hand
+ }
+
+ 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, dir, inode);
+ if (!retval || (retval != ERR_BAD_DX_DIR))
+ goto out;
--- /dev/null
+Change list_add_tail to list_add. It gives advantages to ldiskfs
+in tgt_cb_last_committed. In the beginning of list will be placed
+thandles with the highest transaction numbers. So at the first
+iterations we will have the highest transno. It will save from
+extra call of ptlrpc_commit_replies.
+
+Index: linux-4.15.0/fs/ext4/ext4_jbd2.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4_jbd2.h
++++ linux-4.15.0/fs/ext4/ext4_jbd2.h
+@@ -172,7 +172,7 @@ static inline void _ext4_journal_callbac
+ struct ext4_journal_cb_entry *jce)
+ {
+ /* Add the jce to transaction's private list */
+- list_add_tail(&jce->jce_list, &handle->h_transaction->t_private_list);
++ list_add(&jce->jce_list, &handle->h_transaction->t_private_list);
+ }
+
+ static inline void ext4_journal_callback_add(handle_t *handle,
--- /dev/null
+Index: linux-4.15.0/fs/ext4/namei.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/namei.c
++++ linux-4.15.0/fs/ext4/namei.c
+@@ -196,22 +196,13 @@ 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
+@@ -512,6 +503,16 @@ 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)
+ {
+@@ -736,7 +737,7 @@ dx_probe(struct ext4_filename *fname, st
+ {
+ 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;
+ struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
+ u32 hash;
+@@ -745,18 +746,17 @@ dx_probe(struct ext4_filename *fname, st
+ frame->bh = ext4_read_dirblock(dir, 0, INDEX);
+ 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) {
+- ext4_warning_inode(dir, "Unrecognised inode hash code %u for directory "
+- "%lu", root->info.hash_version, dir->i_ino);
++ 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) {
++ ldiskfs_warning(dir->i_sb, "Unrecognised inode hash code %d for directory "
++ "#%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;
+@@ -764,13 +764,13 @@ dx_probe(struct ext4_filename *fname, st
+ ext4fs_dirhash(fname_name(fname), fname_len(fname), 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 >= ext4_dir_htree_level(dir->i_sb)) {
+ ext4_warning(dir->i_sb,
+ "Directory (ino: %lu) htree depth %#06x exceed"
+@@ -783,14 +783,13 @@ dx_probe(struct ext4_filename *fname, st
+ 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;
+ }
+
+@@ -874,7 +873,7 @@ static void dx_release(struct dx_frame *
+ if (frames[0].bh == NULL)
+ return;
+
+- info = &((struct dx_root *)frames[0].bh->b_data)->info;
++ info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data);
+ for (i = 0; i <= info->indirect_levels; i++) {
+ if (frames[i].bh == NULL)
+ break;
+@@ -1930,17 +1929,16 @@ static int make_indexed_dir(handle_t *ha
+ struct inode *inode, struct buffer_head *bh)
+ {
+ struct buffer_head *bh2;
+- struct dx_root *root;
+ struct dx_frame frames[EXT4_HTREE_LEVEL], *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;
+ int retval;
+ unsigned blocksize;
+ ext4_lblk_t block;
+- struct fake_dirent *fde;
++ struct dx_root_info *dx_info;
+ int csum_size = 0;
+
+ if (ext4_has_metadata_csum(inode->i_sb))
+@@ -1955,18 +1953,19 @@ 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);
+@@ -1992,19 +1991,24 @@ 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)
+ fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
+ fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
+@@ -2346,7 +2350,7 @@ again:
+ goto journal_error;
+ }
+ } else {
+- struct dx_root *dxroot;
++ struct dx_root_info *info;
+ memcpy((char *) entries2, (char *) entries,
+ icount * sizeof(struct dx_entry));
+ dx_set_limit(entries2, dx_node_limit(dir));
+@@ -2354,8 +2358,9 @@ again:
+ /* Set up root */
+ dx_set_count(entries, 1);
+ dx_set_block(entries + 0, newblock);
+- dxroot = (struct dx_root *)frames[0].bh->b_data;
+- dxroot->info.indirect_levels += 1;
++ info = dx_get_dx_info((struct ext4_dir_entry_2 *)
++ frames[0].bh->b_data);
++ info->indirect_levels = 1;
+ dxtrace(printk(KERN_DEBUG
+ "Creating %d level index...\n",
+ info->indirect_levels));
--- /dev/null
+Index: linux-4.15.0/fs/ext4/ext4.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4.h
++++ linux-4.15.0/fs/ext4/ext4.h
+@@ -2874,6 +2874,7 @@ 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-4.15.0/fs/ext4/mballoc.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/mballoc.c
++++ linux-4.15.0/fs/ext4/mballoc.c
+@@ -363,7 +363,7 @@ static const char * const ext4_groupinfo
+ "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);
+@@ -717,7 +717,7 @@ 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);
+@@ -749,19 +749,13 @@ 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);
+
+@@ -772,6 +766,8 @@ 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)
+@@ -892,7 +888,7 @@ 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;
+@@ -936,7 +932,7 @@ 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 {
+@@ -951,7 +947,7 @@ 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);
+
+@@ -961,7 +957,8 @@ static int ext4_mb_init_cache(struct pag
+ incore = data;
+ }
+ }
+- SetPageUptodate(page);
++ if (likely(err == 0))
++ SetPageUptodate(page);
+
+ out:
+ if (bh) {
+@@ -2309,9 +2306,11 @@ 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;
+ unsigned char blocksize_bits = min_t(unsigned char,
+@@ -2324,7 +2323,7 @@ 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 ]\n");
+
+@@ -2342,13 +2341,19 @@ 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 <= blocksize_bits + 1 ?
+ sg.info.bb_counters[i] : 0);
+@@ -3656,22 +3661,71 @@ 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;
++ unsigned short free_in_gdp = ext4_free_group_clusters(sb, gdp);
++
++ if (free_in_gdp == 0 && gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))
++ return 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 != free_in_gdp) {
++ ext4_error(sb, "on-disk bitmap for group %d"
++ "corrupted: %u blocks free in bitmap, %u - in gd\n",
++ group, free, free_in_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
+@@ -3687,13 +3741,23 @@ 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, "preallocated %u for group %u\n", preallocated, group);
++ return 0;
+ }
+
+ static void ext4_mb_pa_callback(struct rcu_head *head)
+@@ -3757,6 +3821,7 @@ 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);
+@@ -3851,6 +3916,7 @@ 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);
+@@ -3912,6 +3978,7 @@ 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);
+
+ /*
+@@ -4084,6 +4151,8 @@ 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);
+ }
+@@ -4214,7 +4283,7 @@ repeat:
+ if (err) {
+ ext4_error(sb, "Error %d loading buddy information for %u",
+ err, group);
+- continue;
++ return;
+ }
+
+ bitmap_bh = ext4_read_block_bitmap(sb, group);
+@@ -4227,6 +4296,8 @@ 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);
+@@ -4489,6 +4560,7 @@ 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-4.15.0/fs/ext4/mballoc.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/mballoc.h
++++ linux-4.15.0/fs/ext4/mballoc.h
+@@ -70,7 +70,7 @@ do { \
+ /*
+ * 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
--- /dev/null
+Index: linux-4.15.0/fs/ext4/ext4.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4.h
++++ linux-4.15.0/fs/ext4/ext4.h
+@@ -1587,6 +1587,8 @@ 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
+ */
+@@ -1801,7 +1803,21 @@ static inline bool ext4_has_unknown_ext#
+
+ 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)
+ {
+@@ -3103,6 +3119,11 @@ 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);
+ extern int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
+ struct ext4_map_blocks *map, int flags);
+Index: linux-4.15.0/fs/ext4/ialloc.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ialloc.c
++++ linux-4.15.0/fs/ext4/ialloc.c
+@@ -156,7 +156,7 @@ static int ext4_validate_inode_bitmap(st
+ *
+ * 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;
+@@ -236,6 +236,7 @@ out:
+ put_bh(bh);
+ return ERR_PTR(err);
+ }
++EXPORT_SYMBOL(ext4_read_inode_bitmap);
+
+ /*
+ * NOTE! When we get the inode, we're the only people
+Index: linux-4.15.0/fs/ext4/inode.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/inode.c
++++ linux-4.15.0/fs/ext4/inode.c
+@@ -6179,3 +6179,18 @@ int ext4_filemap_fault(struct vm_fault *
+
+ return err;
+ }
++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);
++EXPORT_SYMBOL(ext4_get_inode_loc);
+Index: linux-4.15.0/fs/ext4/mballoc.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/mballoc.c
++++ linux-4.15.0/fs/ext4/mballoc.c
+@@ -721,7 +721,6 @@ void ext4_mb_generate_buddy(struct super
+ 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;
+Index: linux-4.15.0/fs/ext4/namei.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/namei.c
++++ linux-4.15.0/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)
+ {
+@@ -158,6 +158,7 @@ static struct buffer_head *__ext4_read_d
+ }
+ return bh;
+ }
++EXPORT_SYMBOL(ext4_append);
+
+ #ifndef assert
+ #define assert(test) J_ASSERT(test)
+@@ -2412,23 +2413,25 @@ EXPORT_SYMBOL(ext4_delete_entry);
+ * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
+ * on regular files) and to avoid creating huge/slow non-HTREE directories.
+ */
+-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 > EXT4_LINK_MAX || inode->i_nlink == 2))
+ set_nlink(inode, 1);
+ }
++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-4.15.0/fs/ext4/super.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/super.c
++++ linux-4.15.0/fs/ext4/super.c
+@@ -5860,16 +5860,12 @@ static int __init ext4_init_fs(void)
+ err = init_inodecache();
+ if (err)
+ goto out1;
+- register_as_ext3();
+- register_as_ext2();
+ err = register_filesystem(&ext4_fs_type);
+ if (err)
+ goto out;
+
+ return 0;
+ out:
+- unregister_as_ext2();
+- unregister_as_ext3();
+ destroy_inodecache();
+ out1:
+ ext4_exit_mballoc();
+@@ -5888,8 +5884,6 @@ out5:
+ static void __exit ext4_exit_fs(void)
+ {
+ ext4_destroy_lazyinit_thread();
+- unregister_as_ext2();
+- unregister_as_ext3();
+ unregister_filesystem(&ext4_fs_type);
+ destroy_inodecache();
+ ext4_exit_mballoc();
--- /dev/null
+We won't change i_xtime in ldiskfs code path. But also
+need keep normal function out of Lustre. So we using
+S_NOCMTIME to indicate invoked from Lustre.
+
+Index: linux-4.15.0/fs/ext4/ext4.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4.h
++++ linux-4.15.0/fs/ext4/ext4.h
+@@ -655,6 +655,20 @@ enum {
+ #define EXT4_GOING_FLAGS_LOGFLUSH 0x1 /* flush log but not data */
+ #define EXT4_GOING_FLAGS_NOLOGFLUSH 0x2 /* don't flush log nor data */
+
++#define current_time(a) ext4_current_time(a)
++static inline struct timespec ext4_current_time(struct inode *inode)
++{
++ struct timespec now = current_kernel_time();
++
++ if (IS_NOCMTIME(inode))
++ return inode->i_ctime;
++
++ /* Copy from fs/inode.c */
++ if (unlikely(!inode->i_sb))
++ return now;
++
++ return timespec_trunc(now, inode->i_sb->s_time_gran);
++}
+
+ #if defined(__KERNEL__) && defined(CONFIG_COMPAT)
+ /*
--- /dev/null
+Index: linux-4.15.0/fs/ext4/ext4.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4.h
++++ linux-4.15.0/fs/ext4/ext4.h
+@@ -2504,6 +2504,11 @@ extern int ext4_dirent_csum_verify(struc
+ 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 ext4_search_dir(struct buffer_head *bh,
+Index: linux-4.15.0/fs/ext4/namei.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/namei.c
++++ linux-4.15.0/fs/ext4/namei.c
+@@ -2360,7 +2360,7 @@ 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)
+@@ -2400,7 +2400,7 @@ out:
+ ext4_std_error(dir->i_sb, err);
+ return err;
+ }
+-
++EXPORT_SYMBOL(ext4_delete_entry);
+ /*
+ * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
+ * since this indicates that nlinks count was previously 1 to avoid overflowing
+@@ -2447,6 +2447,28 @@ 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, 0);
++ 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
--- /dev/null
+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-4.15.0/fs/ext4/Makefile
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/Makefile
++++ linux-4.15.0/fs/ext4/Makefile
+@@ -7,6 +7,7 @@ obj-$(CONFIG_EXT4_FS) += ext4.o
+
+ ext4-y := balloc.o bitmap.o block_validity.o dir.o ext4_jbd2.o extents.o \
+ extents_status.o file.o fsmap.o fsync.o hash.o ialloc.o \
++ htree_lock.o \
+ indirect.o inline.o inode.o ioctl.o mballoc.o migrate.o \
+ mmp.o move_extent.o namei.o page-io.o readpage.o resize.o \
+ super.o symlink.o sysfs.o xattr.o xattr_trusted.o xattr_user.o
+Index: linux-4.15.0/fs/ext4/ext4.h
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/ext4.h
++++ linux-4.15.0/fs/ext4/ext4.h
+@@ -29,6 +29,7 @@
+ #include <linux/timer.h>
+ #include <linux/version.h>
+ #include <linux/wait.h>
++#include <linux/htree_lock.h>
+ #include <linux/sched/signal.h>
+ #include <linux/blockgroup_lock.h>
+ #include <linux/percpu_counter.h>
+@@ -932,6 +933,9 @@ 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,
+@@ -2114,6 +2118,72 @@ 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);
++
+ struct ext4_filename {
+ const struct qstr *usr_fname;
+ struct fscrypt_str disk_name;
+@@ -2421,8 +2491,16 @@ void ext4_insert_dentry(struct inode *in
+ struct ext4_filename *fname, 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 const unsigned char ext4_filetype_table[] = {
+ DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
+Index: linux-4.15.0/fs/ext4/htree_lock.c
+===================================================================
+--- /dev/null
++++ linux-4.15.0/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-4.15.0/fs/ext4/namei.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/namei.c
++++ linux-4.15.0/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);
+@@ -248,7 +256,8 @@ static unsigned dx_node_limit(struct ino
+ static struct dx_frame *dx_probe(struct ext4_filename *fname,
+ 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 inode *dir, struct ext4_dir_entry_2 *de,
+ unsigned blocksize, struct dx_hash_info *hinfo,
+@@ -262,12 +271,13 @@ 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 inode *dir, struct inode *inode);
++ struct inode *dir, struct inode *inode,
++ struct htree_lock *lck);
+
+ /* checksumming functions */
+ void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
+@@ -730,6 +740,227 @@ 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.
+ *
+@@ -741,10 +972,11 @@ 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 dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
+@@ -806,8 +1038,15 @@ dx_probe(struct ext4_filename *fname, st
+
+ 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: count %u beyond limit %u",
+ count, dx_get_limit(entries));
+@@ -846,8 +1085,70 @@ dx_probe(struct ext4_filename *fname, st
+ dx_get_block(at)));
+ frame->entries = entries;
+ frame->at = at;
+- if (!indirect--)
++
++ 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);
+ if (IS_ERR(frame->bh)) {
+@@ -913,7 +1214,7 @@ 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;
+@@ -928,12 +1229,22 @@ 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--;
+ }
+
+@@ -956,6 +1267,13 @@ 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);
+@@ -964,6 +1282,7 @@ 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;
+ }
+
+@@ -1111,10 +1430,10 @@ 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;
+@@ -1154,7 +1473,7 @@ 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;
+@@ -1346,10 +1665,10 @@ 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];
+@@ -1398,7 +1717,7 @@ 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
+@@ -1408,6 +1727,7 @@ static struct buffer_head * ext4_find_en
+ goto cleanup_and_exit;
+ 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);
+ if (!nblocks) {
+@@ -1495,10 +1815,12 @@ cleanup_and_exit:
+ ext4_fname_free_filename(&fname);
+ 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_frame frames[EXT4_HTREE_LEVEL], *frame;
+@@ -1509,7 +1831,7 @@ static struct buffer_head * ext4_dx_find
+ #ifdef CONFIG_EXT4_FS_ENCRYPTION
+ *res_dir = NULL;
+ #endif
+- frame = dx_probe(fname, dir, NULL, frames);
++ frame = dx_probe(fname, dir, NULL, frames, lck);
+ if (IS_ERR(frame))
+ return (struct buffer_head *) frame;
+ do {
+@@ -1531,7 +1853,7 @@ 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_inode(dir,
+ "error %d reading directory index block",
+@@ -1706,8 +2028,9 @@ 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;
+@@ -1769,8 +2092,14 @@ 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,
+@@ -1791,12 +2120,21 @@ static struct ext4_dir_entry_2 *do_split
+ dxtrace(dx_show_leaf(dir, 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;
+@@ -2070,7 +2408,7 @@ static int make_indexed_dir(handle_t *ha
+ if (retval)
+ goto out_frames;
+
+- de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
++ de = do_split(handle, dir, &bh2, frames, frame, &fname->hinfo, NULL);
+ if (IS_ERR(de)) {
+ retval = PTR_ERR(de);
+ goto out_frames;
+@@ -2180,8 +2518,8 @@ 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 = d_inode(dentry->d_parent);
+ struct buffer_head *bh = NULL;
+@@ -2222,9 +2560,10 @@ 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, dir, inode);
++ retval = ext4_dx_add_entry(handle, &fname, dir, 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);
+@@ -2274,12 +2613,14 @@ out:
+ 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 inode *dir, struct inode *inode)
++ struct inode *dir, struct inode *inode,
++ struct htree_lock *lck)
+ {
+ struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+ struct dx_entry *entries, *at;
+@@ -2291,7 +2632,7 @@ 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;
+@@ -2326,6 +2667,11 @@ 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)) {
+@@ -2428,8 +2774,32 @@ again:
+ restart = 1;
+ goto journal_error;
+ }
++ } 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;
+@@ -2440,6 +2810,8 @@ again:
+ journal_error:
+ ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
+ 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-4.15.0/fs/ext4/super.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/super.c
++++ linux-4.15.0/fs/ext4/super.c
+@@ -975,6 +975,7 @@ 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);
+Index: linux-4.15.0/include/linux/htree_lock.h
+===================================================================
+--- /dev/null
++++ linux-4.15.0/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
--- /dev/null
+There will cause a deadlock if invoke ext4_truncate with i_mutex locked
+in lustre. Since lustre has own lock to provide protect so we don't
+need this check at all.
+
+Index: linux-4.15.0/fs/ext4/inode.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/inode.c
++++ linux-4.15.0/fs/ext4/inode.c
+@@ -4398,8 +4398,6 @@ int ext4_truncate(struct inode *inode)
+ * or it's a completely new inode. In those cases we might not
+ * have i_mutex locked because it's not necessary.
+ */
+- if (!(inode->i_state & (I_NEW|I_FREEING)))
+- WARN_ON(!inode_is_locked(inode));
+ trace_ext4_truncate_enter(inode);
+
+ if (!ext4_can_truncate(inode))
+Index: linux-4.15.0/fs/ext4/namei.c
+===================================================================
+--- linux-4.15.0.orig/fs/ext4/namei.c
++++ linux-4.15.0/fs/ext4/namei.c
+@@ -3370,8 +3370,6 @@ int ext4_orphan_add(handle_t *handle, st
+ if (!sbi->s_journal || is_bad_inode(inode))
+ return 0;
+
+- WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
+- !inode_is_locked(inode));
+ /*
+ * Exit early if inode already is on orphan list. This is a big speedup
+ * since we don't have to contend on the global s_orphan_lock.
+@@ -3452,8 +3450,6 @@ int ext4_orphan_del(handle_t *handle, st
+ if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
+ return 0;
+
+- WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
+- !inode_is_locked(inode));
+ /* Do this quick check before taking global s_orphan_lock. */
+ if (list_empty(&ei->i_orphan))
+ return 0;
--- /dev/null
+sles12sp2/ext4-inode-version.patch
+sles12sp2/ext4-lookup-dotdot.patch
+sles12sp2/ext4-print-inum-in-htree-warning.patch
+sles12sp2/ext4-prealloc.patch
+ubuntu18/ext4-osd-iop-common.patch
+ubuntu18/ext4-misc.patch
+ubuntu18/ext4-mballoc-extra-checks.patch
+ubuntu18/ext4-hash-indexed-dir-dotdot-update.patch
+ubuntu18/ext4-kill-dx-root.patch
+rhel7/ext4-mballoc-pa-free-mismatch.patch
+ubuntu18/ext4-data-in-dirent.patch
+ubuntu18/ext4-nocmtime.patch
+ubuntu18/ext4-pdirop.patch
+sles12sp2/ext4-max-dir-size.patch
+ubuntu18/ext4-remove-truncate-warning.patch
+ubuntu18/ext4-corrupted-inode-block-bitmaps-handling-patches.patch
+ubuntu18/ext4-give-warning-with-dir-htree-growing.patch
+ubuntu18/ext4-jcb-optimization.patch
+ubuntu18/ext4-attach-jinode-in-writepages.patch
+ubuntu18/ext4-dont-check-before-replay.patch
+rhel7/ext4-use-GFP_NOFS-in-ext4_inode_attach_jinode.patch
+rhel7/ext4-export-orphan-add.patch
# we know at this point that the found OFED source is good
O2IB_SYMVER=""
if test $ENABLEO2IB = "withpath" -o "x$OFED" = "xyes" ; then
- # OFED default rpm not handle sles10 Modules.symvers name
- for name in Module.symvers Modules.symvers; do
- if test -f $O2IBPATH/$name; then
- O2IB_SYMVER=$name;
- break;
- fi
- done
+ if test -f $O2IBPATH/Module.symvers; then
+ O2IB_SYMVER=$O2IBPATH/Module.symvers;
+ break;
+ fi
if test -n "$O2IB_SYMVER"; then
- AC_MSG_NOTICE([adding $O2IBPATH/$O2IB_SYMVER to Symbol Path])
- EXTRA_SYMBOLS="$EXTRA_SYMBOLS $O2IBPATH/$O2IB_SYMVER"
+ AC_MSG_NOTICE([adding $O2IB_SYMVER to Symbol Path])
+ EXTRA_SYMBOLS="$EXTRA_SYMBOLS $O2IB_SYMVER"
AC_SUBST(EXTRA_SYMBOLS)
else
AC_MSG_ERROR([an external source tree was, either specified or detected, for o2iblnd however I could not find a $O2IBPATH/Module.symvers there])
]) # LC_GENERIC_WRITE_SYNC_2ARGS
#
+# LC_FOP_ITERATE_SHARED
+#
+# Kernel version 4.7 adds iterate_shared method to file_operations
+#
+AC_DEFUN([LC_FOP_ITERATE_SHARED], [
+LB_CHECK_COMPILE([if 'file_operations' has 'iterate_shared'],
+fop_iterate_shared, [
+ #include <linux/fs.h>
+],[
+ struct file_operations fop;
+ fop.iterate_shared = NULL;
+],[
+ AC_DEFINE(HAVE_FOP_ITERATE_SHARED, 1,
+ [file_operations has iterate_shared])
+])
+]) # LC_FOP_ITERATE_SHARED
+
+#
# LC_FOPS_ITERATE_SHARED
#
# 4.7 commit ae05327a00fd47c34dfe25294b359a3f3fef96e8
void osd_process_truncates(struct list_head *list);
void osd_execute_truncate(struct osd_object *obj);
+/*
+ * Maximum size of xattr attributes for FEATURE_INCOMPAT_EA_INODE 1Mb
+ * This limit is arbitrary, but is reasonable for the xattr API.
+ */
+#define LDISKFS_XATTR_MAX_LARGE_EA_SIZE (1024 * 1024)
+
#endif /* _OSD_INTERNAL_H */
git://git.whamcloud.com / fs / lustre-release.git / commitdiff