]> git.proxmox.com Git - mirror_zfs.git/commitdiff
Linux 3.1 compat, super_block->s_shrink
authorBrian Behlendorf <behlendorf1@llnl.gov>
Thu, 22 Dec 2011 20:20:43 +0000 (12:20 -0800)
committerBrian Behlendorf <behlendorf1@llnl.gov>
Wed, 11 Jan 2012 19:46:02 +0000 (11:46 -0800)
The Linux 3.1 kernel has introduced the concept of per-filesystem
shrinkers which are directly assoicated with a super block.  Prior
to this change there was one shared global shrinker.

The zfs code relied on being able to call the global shrinker when
the arc_meta_limit was exceeded.  This would cause the VFS to drop
references on a fraction of the dentries in the dcache.  The ARC
could then safely reclaim the memory used by these entries and
honor the arc_meta_limit.  Unfortunately, when per-filesystem
shrinkers were added the old interfaces were made unavailable.

This change adds support to use the new per-filesystem shrinker
interface so we can continue to honor the arc_meta_limit.  The
major benefit of the new interface is that we can now target
only the zfs filesystem for dentry and inode pruning.  Thus we
can minimize any impact on the caching of other filesystems.

In the context of making this change several other important
issues related to managing the ARC were addressed, they include:

* The dnlc_reduce_cache() function which was called by the ARC
to drop dentries for the Posix layer was replaced with a generic
zfs_prune_t callback.  The ZPL layer now registers a callback to
drop these dentries removing a layering violation which dates
back to the Solaris code.  This callback can also be used by
other ARC consumers such as Lustre.

  arc_add_prune_callback()
  arc_remove_prune_callback()

* The arc_reduce_dnlc_percent module option has been changed to
arc_meta_prune for clarity.  The dnlc functions are specific to
Solaris's VFS and have already been largely eliminated already.
The replacement tunable now represents the number of bytes the
prune callback will request when invoked.

* Less aggressively invoke the prune callback.  We used to call
this whenever we exceeded the arc_meta_limit however that's not
strictly correct since it results in over zeleous reclaim of
dentries and inodes.  It is now only called once the arc_meta_limit
is exceeded and every effort has been made to evict other data from
the ARC cache.

* More promptly manage exceeding the arc_meta_limit.  When reading
meta data in to the cache if a buffer was unable to be recycled
notify the arc_reclaim thread to invoke the required prune.

* Added arcstat_prune kstat which is incremented when the ARC
is forced to request that a consumer prune its cache.  Remember
this will only occur when the ARC has no other choice.  If it
can evict buffers safely without invoking the prune callback
it will.

* This change is also expected to resolve the unexpect collapses
of the ARC cache.  This would occur because when exceeded just the
arc_meta_limit reclaim presure would be excerted on the arc_c
value via arc_shrink().  This effectively shrunk the entire cache
when really we just needed to reclaim meta data.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #466
Closes #292

65 files changed:
Makefile.in
cmd/Makefile.in
cmd/mount_zfs/Makefile.in
cmd/sas_switch_id/Makefile.in
cmd/zdb/Makefile.in
cmd/zfs/Makefile.in
cmd/zinject/Makefile.in
cmd/zpios/Makefile.in
cmd/zpool/Makefile.in
cmd/zpool_id/Makefile.in
cmd/zpool_layout/Makefile.in
cmd/ztest/Makefile.in
cmd/zvol_id/Makefile.in
config/kernel-shrink.m4 [new file with mode: 0644]
config/kernel.m4
configure
dracut/90zfs/Makefile.in
dracut/Makefile.in
etc/Makefile.in
etc/init.d/Makefile.in
etc/zfs/Makefile.in
include/Makefile.in
include/linux/Makefile.in
include/sys/Makefile.in
include/sys/arc.h
include/sys/fm/Makefile.in
include/sys/fm/fs/Makefile.in
include/sys/fs/Makefile.in
include/sys/zfs_vfsops.h
include/sys/zpl.h
lib/Makefile.in
lib/libavl/Makefile.in
lib/libefi/Makefile.in
lib/libnvpair/Makefile.in
lib/libshare/Makefile.in
lib/libspl/Makefile.in
lib/libspl/asm-generic/Makefile.in
lib/libspl/asm-i386/Makefile.in
lib/libspl/asm-x86_64/Makefile.in
lib/libspl/include/Makefile.in
lib/libspl/include/ia32/Makefile.in
lib/libspl/include/ia32/sys/Makefile.in
lib/libspl/include/rpc/Makefile.in
lib/libspl/include/sys/Makefile.in
lib/libspl/include/sys/dktp/Makefile.in
lib/libspl/include/sys/sysevent/Makefile.in
lib/libspl/include/util/Makefile.in
lib/libunicode/Makefile.in
lib/libuutil/Makefile.in
lib/libzfs/Makefile.in
lib/libzpool/Makefile.in
man/Makefile.in
man/man8/Makefile.in
module/zfs/arc.c
module/zfs/zfs_vfsops.c
module/zfs/zfs_znode.c
module/zfs/zpl_super.c
scripts/Makefile.in
scripts/zpios-profile/Makefile.in
scripts/zpios-test/Makefile.in
scripts/zpool-config/Makefile.in
scripts/zpool-layout/Makefile.in
udev/Makefile.in
udev/rules.d/Makefile.in
zfs_config.h.in

index 332efc173292fbf266b0250a426a789eac005d1e..37c813cd3dbfd3612f83d1dab00cc7578049a69e 100644 (file)
@@ -92,6 +92,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index ee3f4bef26d86181be1bbb50ac4ebe50604f843b..eb86e279604a7ea2d37abdcab2fbb52e2a516a68 100644 (file)
@@ -69,6 +69,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 73660d21badb582a05d0e1b74a3896132c60c36e..44d07667e00a84bb24bbf5a9eb622a44af80f858 100644 (file)
@@ -72,6 +72,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index fb6ea65ec0bba9e68a5b462e8167b04a7abf71c9..889da17859c36570293dc80a5658fa173512cdf9 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index da3119592f0691af62e3ba1b78e049de72ffc00e..4fd55e5fa904fdd104b0423b8992ccca7b5aab0b 100644 (file)
@@ -72,6 +72,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 55ca80892fcdd59d03b68794132a72c5a3af84c4..0b9b4dcc6808c8202fe27987f09284c758e579b0 100644 (file)
@@ -72,6 +72,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 5e6f90ecbcc473bfcd38fbc19b60846bc4c9f89a..8adfaab39086649d08cb613b9e2a25598bf8ecab 100644 (file)
@@ -72,6 +72,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index cf672f1e12d7f89b93a939db3f490488885e38fd..c0a84df6bdaa865904a88ca2a842b245cc4d12cc 100644 (file)
@@ -72,6 +72,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index b915006f4114360247654d9ebbd9f63637734a7d..127b93690ceedc05da0542249b89b77c4d314804 100644 (file)
@@ -72,6 +72,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index aadd124133c67f0abfc6e4ee1d5f8db8d5545837..29f83a3ffaa5a16264061f0cb02383cceb1b1a1c 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index e1b16431313e9eb4787cbf4e665a6f08fa71ee9f..514e88f5ffd2d47a2bb4866cfe678eb9c6278c0b 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index e99cc9e85271accf199444fd0c3f5851ecadbd01..dc6549b6c9c7e64142ed25510c4c20a445b32bd8 100644 (file)
@@ -72,6 +72,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index d6719fdb6a318ce27c57f4708ae7e0d09ac15f68..d657c8007472fbdd1444cd7224e05fb115b6cd44 100644 (file)
@@ -72,6 +72,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
diff --git a/config/kernel-shrink.m4 b/config/kernel-shrink.m4
new file mode 100644 (file)
index 0000000..ea748c9
--- /dev/null
@@ -0,0 +1,64 @@
+dnl #
+dnl # 3.1 API change
+dnl # The super_block structure now stores a per-filesystem shrinker.
+dnl # This interface is preferable because it can be used to specifically
+dnl # target only the zfs filesystem for pruning.
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SHRINK], [
+       AC_MSG_CHECKING([whether super_block has s_shrink])
+       ZFS_LINUX_TRY_COMPILE([
+               #include <linux/fs.h>
+       ],[
+               int (*shrink)(struct shrinker *, struct shrink_control *sc)
+                       __attribute__ ((unused)) = NULL;
+               struct super_block sb __attribute__ ((unused)) = {
+                       .s_shrink.shrink = shrink,
+                       .s_shrink.seeks = DEFAULT_SEEKS,
+                       .s_shrink.batch = 0,
+               };
+       ],[
+               AC_MSG_RESULT(yes)
+               AC_DEFINE(HAVE_SHRINK, 1, [struct super_block has s_shrink])
+
+       ],[
+               AC_MSG_RESULT(no)
+       ])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_NR_CACHED_OBJECTS], [
+       AC_MSG_CHECKING([whether sops->nr_cached_objects() exists])
+       ZFS_LINUX_TRY_COMPILE([
+               #include <linux/fs.h>
+       ],[
+               int (*nr_cached_objects)(struct super_block *)
+                       __attribute__ ((unused)) = NULL;
+               struct super_operations sops __attribute__ ((unused)) = {
+                       .nr_cached_objects = nr_cached_objects,
+               };
+       ],[
+               AC_MSG_RESULT(yes)
+               AC_DEFINE(HAVE_NR_CACHED_OBJECTS, 1,
+                       [sops->nr_cached_objects() exists])
+       ],[
+               AC_MSG_RESULT(no)
+       ])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_FREE_CACHED_OBJECTS], [
+       AC_MSG_CHECKING([whether sops->free_cached_objects() exists])
+       ZFS_LINUX_TRY_COMPILE([
+               #include <linux/fs.h>
+       ],[
+               void (*free_cached_objects)(struct super_block *, int)
+                       __attribute__ ((unused)) = NULL;
+               struct super_operations sops __attribute__ ((unused)) = {
+                       .free_cached_objects = free_cached_objects,
+               };
+       ],[
+               AC_MSG_RESULT(yes)
+               AC_DEFINE(HAVE_FREE_CACHED_OBJECTS, 1,
+                       [sops->free_cached_objects() exists])
+       ],[
+               AC_MSG_RESULT(no)
+       ])
+])
index 63a3237ec8b76c4ce500c00bc7066a6a2d815dfe..2afe77e63b45c5fb6d2ab954608e3749d87d9b3d 100644 (file)
@@ -34,12 +34,15 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [
        ZFS_AC_KERNEL_XATTR_HANDLER_SET
        ZFS_AC_KERNEL_FSYNC
        ZFS_AC_KERNEL_EVICT_INODE
+       ZFS_AC_KERNEL_NR_CACHED_OBJECTS
+       ZFS_AC_KERNEL_FREE_CACHED_OBJECTS
        ZFS_AC_KERNEL_INSERT_INODE_LOCKED
        ZFS_AC_KERNEL_D_OBTAIN_ALIAS
        ZFS_AC_KERNEL_CHECK_DISK_SIZE_CHANGE
        ZFS_AC_KERNEL_TRUNCATE_SETSIZE
        ZFS_AC_KERNEL_6ARGS_SECURITY_INODE_INIT_SECURITY
        ZFS_AC_KERNEL_MOUNT_NODEV
+       ZFS_AC_KERNEL_SHRINK
        ZFS_AC_KERNEL_BDI
        ZFS_AC_KERNEL_BDI_SETUP_AND_REGISTER
        ZFS_AC_KERNEL_SET_NLINK
index d6f405bbbd09ce7e8b5ae0413084dad2377eca80..f30b1ffcbbf76a54e47b71783d495419e9a41308 100755 (executable)
--- a/configure
+++ b/configure
 
 
 
+       { $as_echo "$as_me:$LINENO: checking whether sops->nr_cached_objects() exists" >&5
+$as_echo_n "checking whether sops->nr_cached_objects() exists... " >&6; }
+
+
+cat confdefs.h - <<_ACEOF >conftest.c
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+
+
+               #include <linux/fs.h>
+
+int
+main (void)
+{
+
+               int (*nr_cached_objects)(struct super_block *)
+                       __attribute__ ((unused)) = NULL;
+               struct super_operations sops __attribute__ ((unused)) = {
+                       .nr_cached_objects = nr_cached_objects,
+               };
+
+  ;
+  return 0;
+}
+
+_ACEOF
+
+
+       rm -Rf build && mkdir -p build
+       echo "obj-m := conftest.o" >build/Makefile
+       if { ac_try='cp conftest.c build && make modules -C $LINUX_OBJ EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM M=$PWD/build'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; } >/dev/null && { ac_try='test -s build/conftest.o'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; }; then
+
+               { $as_echo "$as_me:$LINENO: result: yes" >&5
+$as_echo "yes" >&6; }
+
+cat >>confdefs.h <<\_ACEOF
+#define HAVE_NR_CACHED_OBJECTS 1
+_ACEOF
+
+
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+               { $as_echo "$as_me:$LINENO: result: no" >&5
+$as_echo "no" >&6; }
+
+
+
+fi
+
+       rm -Rf build
+
+
+
+
+       { $as_echo "$as_me:$LINENO: checking whether sops->free_cached_objects() exists" >&5
+$as_echo_n "checking whether sops->free_cached_objects() exists... " >&6; }
+
+
+cat confdefs.h - <<_ACEOF >conftest.c
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+
+
+               #include <linux/fs.h>
+
+int
+main (void)
+{
+
+               void (*free_cached_objects)(struct super_block *, int)
+                       __attribute__ ((unused)) = NULL;
+               struct super_operations sops __attribute__ ((unused)) = {
+                       .free_cached_objects = free_cached_objects,
+               };
+
+  ;
+  return 0;
+}
+
+_ACEOF
+
+
+       rm -Rf build && mkdir -p build
+       echo "obj-m := conftest.o" >build/Makefile
+       if { ac_try='cp conftest.c build && make modules -C $LINUX_OBJ EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM M=$PWD/build'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; } >/dev/null && { ac_try='test -s build/conftest.o'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; }; then
+
+               { $as_echo "$as_me:$LINENO: result: yes" >&5
+$as_echo "yes" >&6; }
+
+cat >>confdefs.h <<\_ACEOF
+#define HAVE_FREE_CACHED_OBJECTS 1
+_ACEOF
+
+
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+               { $as_echo "$as_me:$LINENO: result: no" >&5
+$as_echo "no" >&6; }
+
+
+
+fi
+
+       rm -Rf build
+
+
+
+
        { $as_echo "$as_me:$LINENO: checking whether symbol insert_inode_locked is exported" >&5
 $as_echo_n "checking whether symbol insert_inode_locked is exported... " >&6; }
        grep -q -E '[[:space:]]insert_inode_locked[[:space:]]' \
 
 
 
+       { $as_echo "$as_me:$LINENO: checking whether super_block has s_shrink" >&5
+$as_echo_n "checking whether super_block has s_shrink... " >&6; }
+
+
+cat confdefs.h - <<_ACEOF >conftest.c
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+
+
+               #include <linux/fs.h>
+
+int
+main (void)
+{
+
+               int (*shrink)(struct shrinker *, struct shrink_control *sc)
+                       __attribute__ ((unused)) = NULL;
+               struct super_block sb __attribute__ ((unused)) = {
+                       .s_shrink.shrink = shrink,
+                       .s_shrink.seeks = DEFAULT_SEEKS,
+                       .s_shrink.batch = 0,
+               };
+
+  ;
+  return 0;
+}
+
+_ACEOF
+
+
+       rm -Rf build && mkdir -p build
+       echo "obj-m := conftest.o" >build/Makefile
+       if { ac_try='cp conftest.c build && make modules -C $LINUX_OBJ EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM M=$PWD/build'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; } >/dev/null && { ac_try='test -s build/conftest.o'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; }; then
+
+               { $as_echo "$as_me:$LINENO: result: yes" >&5
+$as_echo "yes" >&6; }
+
+cat >>confdefs.h <<\_ACEOF
+#define HAVE_SHRINK 1
+_ACEOF
+
+
+
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+               { $as_echo "$as_me:$LINENO: result: no" >&5
+$as_echo "no" >&6; }
+
+
+
+fi
+
+       rm -Rf build
+
+
+
+
        { $as_echo "$as_me:$LINENO: checking whether super_block has s_bdi" >&5
 $as_echo_n "checking whether super_block has s_bdi... " >&6; }
 
 
 
 
+       { $as_echo "$as_me:$LINENO: checking whether sops->nr_cached_objects() exists" >&5
+$as_echo_n "checking whether sops->nr_cached_objects() exists... " >&6; }
+
+
+cat confdefs.h - <<_ACEOF >conftest.c
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+
+
+               #include <linux/fs.h>
+
+int
+main (void)
+{
+
+               int (*nr_cached_objects)(struct super_block *)
+                       __attribute__ ((unused)) = NULL;
+               struct super_operations sops __attribute__ ((unused)) = {
+                       .nr_cached_objects = nr_cached_objects,
+               };
+
+  ;
+  return 0;
+}
+
+_ACEOF
+
+
+       rm -Rf build && mkdir -p build
+       echo "obj-m := conftest.o" >build/Makefile
+       if { ac_try='cp conftest.c build && make modules -C $LINUX_OBJ EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM M=$PWD/build'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; } >/dev/null && { ac_try='test -s build/conftest.o'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; }; then
+
+               { $as_echo "$as_me:$LINENO: result: yes" >&5
+$as_echo "yes" >&6; }
+
+cat >>confdefs.h <<\_ACEOF
+#define HAVE_NR_CACHED_OBJECTS 1
+_ACEOF
+
+
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+               { $as_echo "$as_me:$LINENO: result: no" >&5
+$as_echo "no" >&6; }
+
+
+
+fi
+
+       rm -Rf build
+
+
+
+
+       { $as_echo "$as_me:$LINENO: checking whether sops->free_cached_objects() exists" >&5
+$as_echo_n "checking whether sops->free_cached_objects() exists... " >&6; }
+
+
+cat confdefs.h - <<_ACEOF >conftest.c
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+
+
+               #include <linux/fs.h>
+
+int
+main (void)
+{
+
+               void (*free_cached_objects)(struct super_block *, int)
+                       __attribute__ ((unused)) = NULL;
+               struct super_operations sops __attribute__ ((unused)) = {
+                       .free_cached_objects = free_cached_objects,
+               };
+
+  ;
+  return 0;
+}
+
+_ACEOF
+
+
+       rm -Rf build && mkdir -p build
+       echo "obj-m := conftest.o" >build/Makefile
+       if { ac_try='cp conftest.c build && make modules -C $LINUX_OBJ EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM M=$PWD/build'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; } >/dev/null && { ac_try='test -s build/conftest.o'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; }; then
+
+               { $as_echo "$as_me:$LINENO: result: yes" >&5
+$as_echo "yes" >&6; }
+
+cat >>confdefs.h <<\_ACEOF
+#define HAVE_FREE_CACHED_OBJECTS 1
+_ACEOF
+
+
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+               { $as_echo "$as_me:$LINENO: result: no" >&5
+$as_echo "no" >&6; }
+
+
+
+fi
+
+       rm -Rf build
+
+
+
+
        { $as_echo "$as_me:$LINENO: checking whether symbol insert_inode_locked is exported" >&5
 $as_echo_n "checking whether symbol insert_inode_locked is exported... " >&6; }
        grep -q -E '[[:space:]]insert_inode_locked[[:space:]]' \
 
 
 
+       { $as_echo "$as_me:$LINENO: checking whether super_block has s_shrink" >&5
+$as_echo_n "checking whether super_block has s_shrink... " >&6; }
+
+
+cat confdefs.h - <<_ACEOF >conftest.c
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+
+
+               #include <linux/fs.h>
+
+int
+main (void)
+{
+
+               int (*shrink)(struct shrinker *, struct shrink_control *sc)
+                       __attribute__ ((unused)) = NULL;
+               struct super_block sb __attribute__ ((unused)) = {
+                       .s_shrink.shrink = shrink,
+                       .s_shrink.seeks = DEFAULT_SEEKS,
+                       .s_shrink.batch = 0,
+               };
+
+  ;
+  return 0;
+}
+
+_ACEOF
+
+
+       rm -Rf build && mkdir -p build
+       echo "obj-m := conftest.o" >build/Makefile
+       if { ac_try='cp conftest.c build && make modules -C $LINUX_OBJ EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM M=$PWD/build'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; } >/dev/null && { ac_try='test -s build/conftest.o'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; }; then
+
+               { $as_echo "$as_me:$LINENO: result: yes" >&5
+$as_echo "yes" >&6; }
+
+cat >>confdefs.h <<\_ACEOF
+#define HAVE_SHRINK 1
+_ACEOF
+
+
+
+else
+  $as_echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+               { $as_echo "$as_me:$LINENO: result: no" >&5
+$as_echo "no" >&6; }
+
+
+
+fi
+
+       rm -Rf build
+
+
+
+
        { $as_echo "$as_me:$LINENO: checking whether super_block has s_bdi" >&5
 $as_echo_n "checking whether super_block has s_bdi... " >&6; }
 
index cc79618127ca39ad90f9d8983e0bf04acb5ae034..ca1d86063154557044e02ebc3cea400cb4dae047 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 1c6b3e609ef9cf7c0f75a74cacddc56b87d29005..f385985d227f0d673b39c132abfa862d69ed19b2 100644 (file)
@@ -69,6 +69,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index a2e398604dc65c1365d1df20f27c4bfa7a45ffd9..481d0cdd6d8eb575766c475f2051350bebf9ece2 100644 (file)
@@ -69,6 +69,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 9eb8ae7b4be25c8c47291266f7af72fc537dde2d..f4d0854850f2520e1a96e01e84e1557b5910702d 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 6c81f5d8c96cb01753693b0cffbc22ec90d49775..b421d59bc92571175c9ab69de1e5682a03191487 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index f82fdb267b26110f046976f3c07942ea4c7c3c29..18e235722d46988c0b3675183f311761d5c7402f 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index e50b89b9b84755ba76ffdcdb495005ec0fbbbe15..030437b442bcdcd92fa6876ca3aa647196fe986c 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index b17649b028ca104aa75f051491ce5bf88fc4d981..012ffe9f7ecd20f9fe20281b3ed381f3b4a07b55 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 8f189c62d31dd417314ad4057939c0d2189059bb..dd9b128bc1ea157101f6efe9c6dd4f370cf29f66 100644 (file)
@@ -34,16 +34,27 @@ extern "C" {
 #include <sys/zio.h>
 #include <sys/dmu.h>
 #include <sys/spa.h>
+#include <sys/refcount.h>
 
 typedef struct arc_buf_hdr arc_buf_hdr_t;
 typedef struct arc_buf arc_buf_t;
+typedef struct arc_prune arc_prune_t;
 typedef void arc_done_func_t(zio_t *zio, arc_buf_t *buf, void *private);
+typedef void arc_prune_func_t(int64_t bytes, void *private);
 typedef int arc_evict_func_t(void *private);
 
 /* generic arc_done_func_t's which you can use */
 arc_done_func_t arc_bcopy_func;
 arc_done_func_t arc_getbuf_func;
 
+/* generic arc_prune_func_t wrapper for callbacks */
+struct arc_prune {
+       arc_prune_func_t        *p_pfunc;
+       void                    *p_private;
+       list_node_t             p_node;
+       refcount_t              p_refcnt;
+};
+
 struct arc_buf {
        arc_buf_hdr_t           *b_hdr;
        arc_buf_t               *b_next;
@@ -113,9 +124,13 @@ zio_t *arc_write(zio_t *pio, spa_t *spa, uint64_t txg,
     arc_done_func_t *ready, arc_done_func_t *done, void *private,
     int priority, int zio_flags, const zbookmark_t *zb);
 
+arc_prune_t *arc_add_prune_callback(arc_prune_func_t *func, void *private);
+void arc_remove_prune_callback(arc_prune_t *p);
+
 void arc_set_callback(arc_buf_t *buf, arc_evict_func_t *func, void *private);
 int arc_buf_evict(arc_buf_t *buf);
 
+void arc_adjust_meta(int64_t adjustment, boolean_t may_prune);
 void arc_flush(spa_t *spa);
 void arc_tempreserve_clear(uint64_t reserve);
 int arc_tempreserve_space(uint64_t reserve, uint64_t txg);
index e3fd84cfdda725d574bb9fdee0b89743eb4704a3..932e5e7329f610be84e234b6d0bc4354b6770553 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 40952616b81d4b02a3703de0e3433bf722d55ba2..bbcb1c258e50924849ff89e07f1e6785f4e5dd50 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index db73ab8622cf87b420d19c16a4e2facb96ce3459..da86a6fbc2812f96605537f7d2ceb0757f76f878 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 6d4d713cebb0392d853cb54448202ef5400f7386..b7badeed8e58816844e63a423b1e8884600e4112 100644 (file)
@@ -67,7 +67,8 @@ typedef struct zfs_sb {
        boolean_t       z_unmounted;    /* unmounted */
        rrwlock_t       z_teardown_lock;
        krwlock_t       z_teardown_inactive_lock;
-       list_t          z_all_znodes;   /* all vnodes in the fs */
+       list_t          z_all_znodes;   /* all znodes in the fs */
+       uint64_t        z_nr_znodes;    /* number of znodes in the fs */
        kmutex_t        z_znodes_lock;  /* lock for z_all_znodes */
        struct inode    *z_ctldir;      /* .zfs directory inode */
        boolean_t       z_show_ctldir;  /* expose .zfs in the root dir */
@@ -185,6 +186,8 @@ extern int zfs_get_zplprop(objset_t *os, zfs_prop_t prop,
 extern int zfs_sb_create(const char *name, zfs_sb_t **zsbp);
 extern int zfs_sb_setup(zfs_sb_t *zsb, boolean_t mounting);
 extern void zfs_sb_free(zfs_sb_t *zsb);
+extern int zfs_sb_prune(struct super_block *sb, unsigned long nr_to_scan,
+    int *objects);
 extern int zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting);
 extern int zfs_check_global_label(const char *dsname, const char *hexsl);
 extern boolean_t zfs_is_readonly(zfs_sb_t *zsb);
index 0aacce8bdf43e7bec2f46918b55a8fac9311227d..2195ec9a3d98f4d9addd48f160b078341b7fa7c5 100644 (file)
@@ -48,6 +48,8 @@ extern const struct file_operations zpl_file_operations;
 extern const struct file_operations zpl_dir_file_operations;
 
 /* zpl_super.c */
+extern void zpl_prune_sbs(int64_t bytes_to_scan, void *private);
+
 typedef struct zpl_mount_data {
        const char *z_osname;   /* Dataset name */
        void *z_data;           /* Mount options string */
index 17f84a5482ddecae9440fe99affa7f77ce8ccd42..5929cc468246b252e64fb5829a75ec21c84d36ad 100644 (file)
@@ -69,6 +69,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 55bb04d9f5b7a6a56bf773d2c204fcd23e6d8309..9a6ea3a7537aa97ac6d21d85022ddb1d4312c724 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 6e4a83b3c09f09a6b23b9987fa75dbec9b9b038c..7abee8cab28f35bf1cf709f7d1a4acc1bf2187cc 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 60439aece27e995c5dda1b1bcf2a93c9da882525..33d3f3b1d9fc3a4394b28323a47fbf6309c873a5 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 94ea82e90ac1f138c89f92bd6bba09537417031d..67d23d27967d3e4847918bbd48c414b0a04222be 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 194679a93839eb2c03189de23f7c3d0b8534f0b5..db3499a6d1bb3c12eb8c5f21045a35381568d47b 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index a9ffb578c8e5ddd4753e0f358123221df3bef18a..cae5d5678da827d72072c1d20b2e5958e389c9bf 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index ca6573040465c3257e3567cab9ba5a9568425f10..652782e66764e70eaf2480749cdeceb72107ff29 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 2db49b35a68e5d09d2019d9e7991bb94c38e29b9..ad6c870dc06dbe12da6906aca3f6ebba1fbc8fcc 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index d7a8f52fcefe9e2f816cba32d641ca60a53869e9..419041afa43d5d5585e9814befc2bc5f6a1a1085 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index b60892865a9ebe2f3d0e533a6d32672c45a29424..c0418ee60ded8210558569c8f36c05e5b424b132 100644 (file)
@@ -69,6 +69,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 36629967ef2472dca878be53b1bb487e571a65df..86b08fd796c77c5ff699db498ec72b0c4b138e58 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 68e3c2f12948ac27a4412b4143abe00301f4c3a7..92529aecf7740403861b8afc370ab50e63e61a79 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 6851505dd96b5354b13ea4625f177c16f9d73ed5..09e1ec95797c15a91ff95e3be23329f3e86e692e 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 89c3285e1d4832014b5d84a82ba1be1ad2dcd999..84961afae1a7a9ec286d34600596fb11940b9112 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index de9db421b254724a606840f1b8c6f84685ef2fbd..5d6a93dbe5cf6237b5c56e72b60d6730c99e24eb 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index d7cd17c1508654e4668ac150fa309e78c8dfe831..068a2f560037f62467a0462daa059be4ed17e709 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index f6c8690cdecca6cb5f29138800293d7ab9b19f4f..fbcb32b9998def1c026fe4127c29361ec13865e9 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index d44bba2d6581f74798b4e979bf820d2ed7c4c278..6507a7eadb9a1fc86353b46532e75e3b08afe3ca 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 29642b1ac189e8a7b1fdf37c02da57ba1e46c673..a975a193f29a13d6829add359b3ecbef7545e4b0 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 0e46c1516d7f14ab9beec415eccab74135262f4e..5e1fded91ca91b41d4a21ba64252fc8b5a494bcd 100644 (file)
@@ -71,6 +71,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 84c6c1706b69cf76f5d96f638ab04181e4a77acb..1ad3c35fdd060b2cd229d96fb6f9e545734ca89a 100644 (file)
@@ -69,6 +69,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index be187c9c03e2091de6a6c82c801eca2bc7a783ad..86e691d2ef95af361eb6d072fab069b5470c18ff 100644 (file)
@@ -69,6 +69,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index f71c24844e8b3821d220de97ab90c0932476f7e5..95d14a9e75110192d80367d53019122b09ad763a 100644 (file)
  * protected from simultaneous callbacks from arc_buf_evict()
  * and arc_do_user_evicts().
  *
+ * It as also possible to register a callback which is run when the
+ * arc_meta_limit is reached and no buffers can be safely evicted.  In
+ * this case the arc user should drop a reference on some arc buffers so
+ * they can be reclaimed and the arc_meta_limit honored.  For example,
+ * when using the ZPL each dentry holds a references on a znode.  These
+ * dentries must be pruned before the arc buffer holding the znode can
+ * be safely evicted.
+ *
  * Note that the majority of the performance stats are manipulated
  * with atomic operations.
  *
 #include <sys/zio.h>
 #include <sys/zfs_context.h>
 #include <sys/arc.h>
-#include <sys/refcount.h>
 #include <sys/vdev.h>
 #include <sys/vdev_impl.h>
 #ifdef _KERNEL
 #include <sys/vmsystm.h>
 #include <vm/anon.h>
 #include <sys/fs/swapnode.h>
-#include <sys/dnlc.h>
+#include <sys/zpl.h>
 #endif
 #include <sys/callb.h>
 #include <sys/kstat.h>
@@ -141,8 +148,8 @@ extern int zfs_write_limit_shift;
 extern uint64_t zfs_write_limit_max;
 extern kmutex_t zfs_write_limit_lock;
 
-#define        ARC_REDUCE_DNLC_PERCENT 3
-uint_t arc_reduce_dnlc_percent = ARC_REDUCE_DNLC_PERCENT;
+/* number of bytes to prune from caches when at arc_meta_limit is reached */
+uint_t arc_meta_prune = 1048576;
 
 typedef enum arc_reclaim_strategy {
        ARC_RECLAIM_AGGR,               /* Aggressive reclaim strategy */
@@ -180,7 +187,7 @@ unsigned long zfs_arc_meta_limit = 0;
 int zfs_arc_grow_retry = 0;
 int zfs_arc_shrink_shift = 0;
 int zfs_arc_p_min_shift = 0;
-int zfs_arc_reduce_dnlc_percent = 0;
+int zfs_arc_meta_prune = 0;
 
 /*
  * Note that buffers can be in one of 6 states:
@@ -288,6 +295,7 @@ typedef struct arc_stats {
        kstat_named_t arcstat_no_grow;
        kstat_named_t arcstat_tempreserve;
        kstat_named_t arcstat_loaned_bytes;
+       kstat_named_t arcstat_prune;
        kstat_named_t arcstat_meta_used;
        kstat_named_t arcstat_meta_limit;
        kstat_named_t arcstat_meta_max;
@@ -352,6 +360,7 @@ static arc_stats_t arc_stats = {
        { "arc_no_grow",                KSTAT_DATA_UINT64 },
        { "arc_tempreserve",            KSTAT_DATA_UINT64 },
        { "arc_loaned_bytes",           KSTAT_DATA_UINT64 },
+       { "arc_prune",                  KSTAT_DATA_UINT64 },
        { "arc_meta_used",              KSTAT_DATA_UINT64 },
        { "arc_meta_limit",             KSTAT_DATA_UINT64 },
        { "arc_meta_max",               KSTAT_DATA_UINT64 },
@@ -481,6 +490,8 @@ struct arc_buf_hdr {
        list_node_t             b_l2node;
 };
 
+static list_t arc_prune_list;
+static kmutex_t arc_prune_mtx;
 static arc_buf_t *arc_eviction_list;
 static kmutex_t arc_eviction_mtx;
 static arc_buf_hdr_t arc_eviction_hdr;
@@ -1925,6 +1936,48 @@ arc_adjust(void)
        }
 }
 
+/*
+ * Request that arc user drop references so that N bytes can be released
+ * from the cache.  This provides a mechanism to ensure the arc can honor
+ * the arc_meta_limit and reclaim buffers which are pinned in the cache
+ * by higher layers.  (i.e. the zpl)
+ */
+static void
+arc_do_user_prune(int64_t adjustment)
+{
+       arc_prune_func_t *func;
+       void *private;
+       arc_prune_t *cp, *np;
+
+       mutex_enter(&arc_prune_mtx);
+
+       cp = list_head(&arc_prune_list);
+       while (cp != NULL) {
+               func = cp->p_pfunc;
+               private = cp->p_private;
+               np = list_next(&arc_prune_list, cp);
+               refcount_add(&cp->p_refcnt, func);
+               mutex_exit(&arc_prune_mtx);
+
+               if (func != NULL)
+                       func(adjustment, private);
+
+               mutex_enter(&arc_prune_mtx);
+
+               /* User removed prune callback concurrently with execution */
+               if (refcount_remove(&cp->p_refcnt, func) == 0) {
+                       ASSERT(!list_link_active(&cp->p_node));
+                       refcount_destroy(&cp->p_refcnt);
+                       kmem_free(cp, sizeof (*cp));
+               }
+
+               cp = np;
+       }
+
+       ARCSTAT_BUMP(arcstat_prune);
+       mutex_exit(&arc_prune_mtx);
+}
+
 static void
 arc_do_user_evicts(void)
 {
@@ -1948,6 +2001,32 @@ arc_do_user_evicts(void)
        mutex_exit(&arc_eviction_mtx);
 }
 
+/*
+ * Evict only meta data objects from the cache leaving the data objects.
+ * This is only used to enforce the tunable arc_meta_limit, if we are
+ * unable to evict enough buffers notify the user via the prune callback.
+ */
+void
+arc_adjust_meta(int64_t adjustment, boolean_t may_prune)
+{
+       int64_t delta;
+
+       if (adjustment > 0 && arc_mru->arcs_lsize[ARC_BUFC_METADATA] > 0) {
+               delta = MIN(arc_mru->arcs_lsize[ARC_BUFC_METADATA], adjustment);
+               arc_evict(arc_mru, 0, delta, FALSE, ARC_BUFC_METADATA);
+               adjustment -= delta;
+       }
+
+       if (adjustment > 0 && arc_mfu->arcs_lsize[ARC_BUFC_METADATA] > 0) {
+               delta = MIN(arc_mfu->arcs_lsize[ARC_BUFC_METADATA], adjustment);
+               arc_evict(arc_mfu, 0, delta, FALSE, ARC_BUFC_METADATA);
+               adjustment -= delta;
+       }
+
+       if (may_prune && (adjustment > 0) && (arc_meta_used > arc_meta_limit))
+               arc_do_user_prune(arc_meta_prune);
+}
+
 /*
  * Flush all *evictable* data from the cache for the given spa.
  * NOTE: this will not touch "active" (i.e. referenced) data.
@@ -2085,24 +2164,6 @@ arc_kmem_reap_now(arc_reclaim_strategy_t strat)
        kmem_cache_t            *prev_data_cache = NULL;
        extern kmem_cache_t     *zio_buf_cache[];
        extern kmem_cache_t     *zio_data_buf_cache[];
-#ifdef _KERNEL
-       int                     retry = 0;
-
-       while ((arc_meta_used >= arc_meta_limit) && (retry < 10)) {
-               /*
-                * We are exceeding our meta-data cache limit.
-                * Purge some DNLC entries to release holds on meta-data.
-                */
-               dnlc_reduce_cache((void *)(uintptr_t)arc_reduce_dnlc_percent);
-               retry++;
-       }
-#if defined(__i386)
-       /*
-        * Reclaim unused memory from all kmem caches.
-        */
-       kmem_reap();
-#endif
-#endif
 
        /*
         * An aggressive reclamation will shrink the cache size as well as
@@ -2121,6 +2182,7 @@ arc_kmem_reap_now(arc_reclaim_strategy_t strat)
                        kmem_cache_reap_now(zio_data_buf_cache[i]);
                }
        }
+
        kmem_cache_reap_now(buf_cache);
        kmem_cache_reap_now(hdr_cache);
 }
@@ -2131,6 +2193,7 @@ arc_reclaim_thread(void)
        clock_t                 growtime = 0;
        arc_reclaim_strategy_t  last_reclaim = ARC_RECLAIM_CONS;
        callb_cpr_t             cpr;
+       int64_t                 prune;
 
        CALLB_CPR_INIT(&cpr, &arc_reclaim_thr_lock, callb_generic_cpr, FTAG);
 
@@ -2160,9 +2223,14 @@ arc_reclaim_thread(void)
                        arc_no_grow = FALSE;
                }
 
-               /* Keep meta data usage within limits */
-               if (arc_meta_used >= arc_meta_limit)
-                       arc_kmem_reap_now(ARC_RECLAIM_CONS);
+               /*
+                * Keep meta data usage within limits, arc_shrink() is not
+                * used to avoid collapsing the arc_c value when only the
+                * arc_meta_limit is being exceeded.
+                */
+               prune = (int64_t)arc_meta_used - (int64_t)arc_meta_limit;
+               if (prune > 0)
+                       arc_adjust_meta(prune, B_TRUE);
 
                arc_adjust();
 
@@ -2399,16 +2467,27 @@ arc_get_data_buf(arc_buf_t *buf)
                state =  (arc_mru->arcs_lsize[type] >= size &&
                    mfu_space > arc_mfu->arcs_size) ? arc_mru : arc_mfu;
        }
+
        if ((buf->b_data = arc_evict(state, 0, size, TRUE, type)) == NULL) {
                if (type == ARC_BUFC_METADATA) {
                        buf->b_data = zio_buf_alloc(size);
                        arc_space_consume(size, ARC_SPACE_DATA);
+
+                       /*
+                        * If we are unable to recycle an existing meta buffer
+                        * signal the reclaim thread.  It will notify users
+                        * via the prune callback to drop references.  The
+                        * prune callback in run in the context of the reclaim
+                        * thread to avoid deadlocking on the hash_lock.
+                        */
+                       cv_signal(&arc_reclaim_thr_cv);
                } else {
                        ASSERT(type == ARC_BUFC_DATA);
                        buf->b_data = zio_data_buf_alloc(size);
                        ARCSTAT_INCR(arcstat_data_size, size);
                        atomic_add_64(&arc_size, size);
                }
+
                ARCSTAT_BUMP(arcstat_recycle_miss);
        }
        ASSERT(buf->b_data != NULL);
@@ -3021,6 +3100,37 @@ top:
        return (0);
 }
 
+arc_prune_t *
+arc_add_prune_callback(arc_prune_func_t *func, void *private)
+{
+       arc_prune_t *p;
+
+       p = kmem_alloc(sizeof(*p), KM_SLEEP);
+       p->p_pfunc = func;
+       p->p_private = private;
+       list_link_init(&p->p_node);
+       refcount_create(&p->p_refcnt);
+
+       mutex_enter(&arc_prune_mtx);
+       refcount_add(&p->p_refcnt, &arc_prune_list);
+       list_insert_head(&arc_prune_list, p);
+       mutex_exit(&arc_prune_mtx);
+
+       return (p);
+}
+
+void
+arc_remove_prune_callback(arc_prune_t *p)
+{
+       mutex_enter(&arc_prune_mtx);
+       list_remove(&arc_prune_list, p);
+       if (refcount_remove(&p->p_refcnt, &arc_prune_list) == 0) {
+               refcount_destroy(&p->p_refcnt);
+               kmem_free(p, sizeof (*p));
+       }
+       mutex_exit(&arc_prune_mtx);
+}
+
 void
 arc_set_callback(arc_buf_t *buf, arc_evict_func_t *func, void *private)
 {
@@ -3598,8 +3708,8 @@ arc_init(void)
        if (zfs_arc_p_min_shift > 0)
                arc_p_min_shift = zfs_arc_p_min_shift;
 
-       if (zfs_arc_reduce_dnlc_percent > 0)
-               arc_reduce_dnlc_percent = zfs_arc_reduce_dnlc_percent;
+       if (zfs_arc_meta_prune > 0)
+               arc_meta_prune = zfs_arc_meta_prune;
 
        /* if kmem_flags are set, lets try to use less memory */
        if (kmem_debugging())
@@ -3646,7 +3756,10 @@ arc_init(void)
        buf_init();
 
        arc_thread_exit = 0;
+       list_create(&arc_prune_list, sizeof (arc_prune_t),
+           offsetof(arc_prune_t, p_node));
        arc_eviction_list = NULL;
+       mutex_init(&arc_prune_mtx, NULL, MUTEX_DEFAULT, NULL);
        mutex_init(&arc_eviction_mtx, NULL, MUTEX_DEFAULT, NULL);
        bzero(&arc_eviction_hdr, sizeof (arc_buf_hdr_t));
 
@@ -3674,6 +3787,8 @@ arc_init(void)
 void
 arc_fini(void)
 {
+       arc_prune_t *p;
+
        mutex_enter(&arc_reclaim_thr_lock);
 #ifdef _KERNEL
        spl_unregister_shrinker(&arc_shrinker);
@@ -3693,6 +3808,17 @@ arc_fini(void)
                arc_ksp = NULL;
        }
 
+       mutex_enter(&arc_prune_mtx);
+       while ((p = list_head(&arc_prune_list)) != NULL) {
+               list_remove(&arc_prune_list, p);
+               refcount_remove(&p->p_refcnt, &arc_prune_list);
+               refcount_destroy(&p->p_refcnt);
+               kmem_free(p, sizeof (*p));
+       }
+       mutex_exit(&arc_prune_mtx);
+
+       list_destroy(&arc_prune_list);
+       mutex_destroy(&arc_prune_mtx);
        mutex_destroy(&arc_eviction_mtx);
        mutex_destroy(&arc_reclaim_thr_lock);
        cv_destroy(&arc_reclaim_thr_cv);
@@ -4774,6 +4900,8 @@ l2arc_stop(void)
 EXPORT_SYMBOL(arc_read);
 EXPORT_SYMBOL(arc_buf_remove_ref);
 EXPORT_SYMBOL(arc_getbuf_func);
+EXPORT_SYMBOL(arc_add_prune_callback);
+EXPORT_SYMBOL(arc_remove_prune_callback);
 
 module_param(zfs_arc_min, ulong, 0444);
 MODULE_PARM_DESC(zfs_arc_min, "Min arc size");
@@ -4784,8 +4912,8 @@ MODULE_PARM_DESC(zfs_arc_max, "Max arc size");
 module_param(zfs_arc_meta_limit, ulong, 0444);
 MODULE_PARM_DESC(zfs_arc_meta_limit, "Meta limit for arc size");
 
-module_param(zfs_arc_reduce_dnlc_percent, int, 0444);
-MODULE_PARM_DESC(zfs_arc_reduce_dnlc_percent, "Meta reclaim percentage");
+module_param(zfs_arc_meta_prune, int, 0444);
+MODULE_PARM_DESC(zfs_arc_meta_prune, "Bytes of meta data to prune");
 
 module_param(zfs_arc_grow_retry, int, 0444);
 MODULE_PARM_DESC(zfs_arc_grow_retry, "Seconds before growing arc size");
index a0726e117eaea75c0ed3cfa2d0f1788df82d626d..fb319a547a93d9b305bb39df77979d5c5ed66e69 100644 (file)
@@ -986,6 +986,26 @@ zfs_root(zfs_sb_t *zsb, struct inode **ipp)
 }
 EXPORT_SYMBOL(zfs_root);
 
+#ifdef HAVE_SHRINK
+int
+zfs_sb_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
+{
+       zfs_sb_t *zsb = sb->s_fs_info;
+       struct shrinker *shrinker = &sb->s_shrink;
+       struct shrink_control sc = {
+               .nr_to_scan = nr_to_scan,
+               .gfp_mask = GFP_KERNEL,
+       };
+
+       ZFS_ENTER(zsb);
+       *objects = (*shrinker->shrink)(shrinker, &sc);
+       ZFS_EXIT(zsb);
+
+       return (0);
+}
+EXPORT_SYMBOL(zfs_sb_prune);
+#endif /* HAVE_SHRINK */
+
 /*
  * Teardown the zfs_sb_t::z_os.
  *
@@ -1533,6 +1553,7 @@ zfs_init(void)
        zfs_znode_init();
        dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
        register_filesystem(&zpl_fs_type);
+       (void) arc_add_prune_callback(zpl_prune_sbs, NULL);
 }
 
 void
index a35e3b5f2c7849d379d8c46e5de559c1aed10e88..709ae74f8f5cfd54a50322349c5b263bbcf7056d 100644 (file)
@@ -269,6 +269,7 @@ zfs_inode_destroy(struct inode *ip)
 
        mutex_enter(&zsb->z_znodes_lock);
        list_remove(&zsb->z_all_znodes, zp);
+       zsb->z_nr_znodes--;
        mutex_exit(&zsb->z_znodes_lock);
 
        if (zp->z_acl_cached) {
@@ -401,6 +402,7 @@ zfs_znode_alloc(zfs_sb_t *zsb, dmu_buf_t *db, int blksz,
 
        mutex_enter(&zsb->z_znodes_lock);
        list_insert_tail(&zsb->z_all_znodes, zp);
+       zsb->z_nr_znodes++;
        membar_producer();
        mutex_exit(&zsb->z_znodes_lock);
 
index 650e9c0d3791639851c5a29e47f87af43f226227..3abb26a9e418f3e425b96034f6ca3941755bf900 100644 (file)
@@ -199,34 +199,120 @@ zpl_kill_sb(struct super_block *sb)
        kill_anon_super(sb);
 }
 
+#ifdef HAVE_SHRINK
+/*
+ * Linux 3.1 - 3.x API
+ *
+ * The Linux 3.1 API introduced per-sb cache shrinkers to replace the
+ * global ones.  This allows us a mechanism to cleanly target a specific
+ * zfs file system when the dnode and inode caches grow too large.
+ *
+ * In addition, the 3.0 kernel added the iterate_supers_type() helper
+ * function which is used to safely walk all of the zfs file systems.
+ */
+static void
+zpl_prune_sb(struct super_block *sb, void *arg)
+{
+       int objects = 0;
+       int error;
+
+       error = -zfs_sb_prune(sb, *(unsigned long *)arg, &objects);
+       ASSERT3S(error, <=, 0);
+
+       return;
+}
+
+void
+zpl_prune_sbs(int64_t bytes_to_scan, void *private)
+{
+       unsigned long nr_to_scan = (bytes_to_scan / sizeof(znode_t));
+
+       iterate_supers_type(&zpl_fs_type, zpl_prune_sb, &nr_to_scan);
+       kmem_reap();
+}
+#else
+/*
+ * Linux 2.6.x - 3.0 API
+ *
+ * These are best effort interfaces are provided by the SPL to induce
+ * the Linux VM subsystem to reclaim a fraction of the both dnode and
+ * inode caches.  Ideally, we want to just target the zfs file systems
+ * however our only option is to reclaim from them all.
+ */
+void
+zpl_prune_sbs(int64_t bytes_to_scan, void *private)
+{
+       unsigned long nr_to_scan = (bytes_to_scan / sizeof(znode_t));
+
+        shrink_dcache_memory(nr_to_scan, GFP_KERNEL);
+        shrink_icache_memory(nr_to_scan, GFP_KERNEL);
+        kmem_reap();
+}
+#endif /* HAVE_SHRINK */
+
+#ifdef HAVE_NR_CACHED_OBJECTS
+static int
+zpl_nr_cached_objects(struct super_block *sb)
+{
+       zfs_sb_t *zsb = sb->s_fs_info;
+       int nr;
+
+       mutex_enter(&zsb->z_znodes_lock);
+       nr = zsb->z_nr_znodes;
+       mutex_exit(&zsb->z_znodes_lock);
+
+       return (nr);
+}
+#endif /* HAVE_NR_CACHED_OBJECTS */
+
+#ifdef HAVE_FREE_CACHED_OBJECTS
+/*
+ * Attempt to evict some meta data from the cache.  The ARC operates in
+ * terms of bytes while the Linux VFS uses objects.  Now because this is
+ * just a best effort eviction and the exact values aren't critical so we
+ * extrapolate from an object count to a byte size using the znode_t size.
+ */
+static void
+zpl_free_cached_objects(struct super_block *sb, int nr_to_scan)
+{
+       arc_adjust_meta(nr_to_scan * sizeof(znode_t), B_FALSE);
+}
+#endif /* HAVE_FREE_CACHED_OBJECTS */
+
 const struct super_operations zpl_super_operations = {
-       .alloc_inode    = zpl_inode_alloc,
-       .destroy_inode  = zpl_inode_destroy,
-       .dirty_inode    = NULL,
-       .write_inode    = NULL,
-       .drop_inode     = NULL,
+       .alloc_inode            = zpl_inode_alloc,
+       .destroy_inode          = zpl_inode_destroy,
+       .dirty_inode            = NULL,
+       .write_inode            = NULL,
+       .drop_inode             = NULL,
 #ifdef HAVE_EVICT_INODE
-       .evict_inode    = zpl_evict_inode,
+       .evict_inode            = zpl_evict_inode,
 #else
-       .clear_inode    = zpl_clear_inode,
-       .delete_inode   = zpl_inode_delete,
+       .clear_inode            = zpl_clear_inode,
+       .delete_inode           = zpl_inode_delete,
 #endif /* HAVE_EVICT_INODE */
-       .put_super      = zpl_put_super,
-       .write_super    = NULL,
-       .sync_fs        = zpl_sync_fs,
-       .statfs         = zpl_statfs,
-       .remount_fs     = zpl_remount_fs,
-       .show_options   = zpl_show_options,
-       .show_stats     = NULL,
+       .put_super              = zpl_put_super,
+       .write_super            = NULL,
+       .sync_fs                = zpl_sync_fs,
+       .statfs                 = zpl_statfs,
+       .remount_fs             = zpl_remount_fs,
+       .show_options           = zpl_show_options,
+       .show_stats             = NULL,
+#ifdef HAVE_NR_CACHED_OBJECTS
+       .nr_cached_objects      = zpl_nr_cached_objects,
+#endif /* HAVE_NR_CACHED_OBJECTS */
+#ifdef HAVE_FREE_CACHED_OBJECTS
+       .free_cached_objects    = zpl_free_cached_objects,
+#endif /* HAVE_FREE_CACHED_OBJECTS */
 };
 
 struct file_system_type zpl_fs_type = {
-       .owner          = THIS_MODULE,
-       .name           = ZFS_DRIVER,
+       .owner                  = THIS_MODULE,
+       .name                   = ZFS_DRIVER,
 #ifdef HAVE_MOUNT_NODEV
-       .mount          = zpl_mount,
+       .mount                  = zpl_mount,
 #else
-       .get_sb         = zpl_get_sb,
+       .get_sb                 = zpl_get_sb,
 #endif /* HAVE_MOUNT_NODEV */
-       .kill_sb        = zpl_kill_sb,
+       .kill_sb                = zpl_kill_sb,
 };
index 5115a9e0b344d8dd5ddca70706ef2d1051bda310..4fdc85b676a97fa809b27c567b94aad1aa2796b8 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 1c59d7475d7e420f980f58a1674274a1d2b37330..3390621115779be97b602b9b720377b7c49701ce 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index fb1114530a799adcbbed18f901bcd501f93916d1..1adbef8cf89f661d482c9ef1ef7a53e1c3093a1b 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 2c4fcb976855ae4781d1eb1fba7618abbb9807f3..c78e79667b4ff1626801e76811b4f5d3b39ef656 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index bebd47b8feaedbcf77aa055192cf97414951a62e..59034e069af934d0fc6ac75b0076f5f2f318aa57 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 856d11aaf4f13e1415396f8b4c215c403041e999..6232f65dbbd06d4165f46e36c83efc0b8de0c111 100644 (file)
@@ -69,6 +69,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index 37eb4c732eacefaa0350dc760b1d71640f74da88..f9a72cfb63cbd4bdbbf8c43616f528223ecbed45 100644 (file)
@@ -70,6 +70,7 @@ am__aclocal_m4_deps =  \
        $(top_srcdir)/config/kernel-rq-is_sync.m4 \
        $(top_srcdir)/config/kernel-security-inode-init.m4 \
        $(top_srcdir)/config/kernel-set-nlink.m4 \
+       $(top_srcdir)/config/kernel-shrink.m4 \
        $(top_srcdir)/config/kernel-truncate-setsize.m4 \
        $(top_srcdir)/config/kernel-xattr-handler.m4 \
        $(top_srcdir)/config/kernel.m4 \
index e0f746704b990b289c916d8cf666a17bf5f1e7aa..b53ca52d75a68337b033168b23052c0939f93442 100644 (file)
@@ -90,6 +90,9 @@
 /* kernel defines fmode_t */
 #undef HAVE_FMODE_T
 
+/* sops->free_cached_objects() exists */
+#undef HAVE_FREE_CACHED_OBJECTS
+
 /* fops->fsync() with range */
 #undef HAVE_FSYNC_RANGE
 
 /* mount_nodev() is available */
 #undef HAVE_MOUNT_NODEV
 
+/* sops->nr_cached_objects() exists */
+#undef HAVE_NR_CACHED_OBJECTS
+
 /* open_bdev_exclusive() is available */
 #undef HAVE_OPEN_BDEV_EXCLUSIVE
 
 /* set_nlink() is available */
 #undef HAVE_SET_NLINK
 
+/* struct super_block has s_shrink */
+#undef HAVE_SHRINK
+
 /* Define to 1 if you have the <stdint.h> header file. */
 #undef HAVE_STDINT_H