staging: ramster: cluster/messaging foundation

RAMster implements peer-to-peer transcendent memory, allowing a "cluster"
of kernels to dynamically pool their RAM.

This patch provides the cluster and messaging foundation for RAMster,
implementing the basic cluster discovery, mapping, heartbeat / keepalive,
and messaging ("r2net") that RAMster requires for internode communication.
This code heavily leverages code from the ocfs2 cluster layer but
has been extended, interfaces to userland changed, and external functions
renamed so that RAMster and ocfs2 can co-exist in the kernel and userland.
Signed-off-by: Dan Magenheimer <dan.magenheimer@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/staging/ramster/cluster/Makefile

+obj-$(CONFIG_RAMSTER) += ramster_nodemanager.o
+
+ramster_nodemanager-objs := heartbeat.o masklog.o nodemanager.o tcp.o

drivers/staging/ramster/cluster/heartbeat.h

+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * heartbeat.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef R2CLUSTER_HEARTBEAT_H
+#define R2CLUSTER_HEARTBEAT_H
+
+#define R2HB_REGION_TIMEOUT_MS		2000
+
+#define R2HB_MAX_REGION_NAME_LEN	32
+
+/* number of changes to be seen as live */
+#define R2HB_LIVE_THRESHOLD	   2
+/* number of equal samples to be seen as dead */
+extern unsigned int r2hb_dead_threshold;
+#define R2HB_DEFAULT_DEAD_THRESHOLD	   31
+/* Otherwise MAX_WRITE_TIMEOUT will be zero... */
+#define R2HB_MIN_DEAD_THRESHOLD	  2
+#define R2HB_MAX_WRITE_TIMEOUT_MS \
+	(R2HB_REGION_TIMEOUT_MS * (r2hb_dead_threshold - 1))
+
+#define R2HB_CB_MAGIC		0x51d1e4ec
+
+/* callback stuff */
+enum r2hb_callback_type {
+	R2HB_NODE_DOWN_CB = 0,
+	R2HB_NODE_UP_CB,
+	R2HB_NUM_CB
+};
+
+struct r2nm_node;
+typedef void (r2hb_cb_func)(struct r2nm_node *, int, void *);
+
+struct r2hb_callback_func {
+	u32			hc_magic;
+	struct list_head	hc_item;
+	r2hb_cb_func		*hc_func;
+	void			*hc_data;
+	int			hc_priority;
+	enum r2hb_callback_type hc_type;
+};
+
+struct config_group *r2hb_alloc_hb_set(void);
+void r2hb_free_hb_set(struct config_group *group);
+
+void r2hb_setup_callback(struct r2hb_callback_func *hc,
+			 enum r2hb_callback_type type,
+			 r2hb_cb_func *func,
+			 void *data,
+			 int priority);
+int r2hb_register_callback(const char *region_uuid,
+			   struct r2hb_callback_func *hc);
+void r2hb_unregister_callback(const char *region_uuid,
+			      struct r2hb_callback_func *hc);
+void r2hb_fill_node_map(unsigned long *map,
+			unsigned bytes);
+void r2hb_exit(void);
+int r2hb_init(void);
+int r2hb_check_node_heartbeating_from_callback(u8 node_num);
+void r2hb_stop_all_regions(void);
+int r2hb_get_all_regions(char *region_uuids, u8 numregions);
+int r2hb_global_heartbeat_active(void);
+void r2hb_manual_set_node_heartbeating(int);
+
+#endif /* R2CLUSTER_HEARTBEAT_H */

drivers/staging/ramster/cluster/masklog.c

+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2004, 2005, 2012 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include <linux/string.h>
+#include <linux/uaccess.h>
+
+#include "masklog.h"
+
+struct mlog_bits mlog_and_bits = MLOG_BITS_RHS(MLOG_INITIAL_AND_MASK);
+EXPORT_SYMBOL_GPL(mlog_and_bits);
+struct mlog_bits mlog_not_bits = MLOG_BITS_RHS(0);
+EXPORT_SYMBOL_GPL(mlog_not_bits);
+
+static ssize_t mlog_mask_show(u64 mask, char *buf)
+{
+	char *state;
+
+	if (__mlog_test_u64(mask, mlog_and_bits))
+		state = "allow";
+	else if (__mlog_test_u64(mask, mlog_not_bits))
+		state = "deny";
+	else
+		state = "off";
+
+	return snprintf(buf, PAGE_SIZE, "%s\n", state);
+}
+
+static ssize_t mlog_mask_store(u64 mask, const char *buf, size_t count)
+{
+	if (!strnicmp(buf, "allow", 5)) {
+		__mlog_set_u64(mask, mlog_and_bits);
+		__mlog_clear_u64(mask, mlog_not_bits);
+	} else if (!strnicmp(buf, "deny", 4)) {
+		__mlog_set_u64(mask, mlog_not_bits);
+		__mlog_clear_u64(mask, mlog_and_bits);
+	} else if (!strnicmp(buf, "off", 3)) {
+		__mlog_clear_u64(mask, mlog_not_bits);
+		__mlog_clear_u64(mask, mlog_and_bits);
+	} else
+		return -EINVAL;
+
+	return count;
+}
+
+struct mlog_attribute {
+	struct attribute attr;
+	u64 mask;
+};
+
+#define to_mlog_attr(_attr) container_of(_attr, struct mlog_attribute, attr)
+
+#define define_mask(_name) {			\
+	.attr = {				\
+		.name = #_name,			\
+		.mode = S_IRUGO | S_IWUSR,	\
+	},					\
+	.mask = ML_##_name,			\
+}
+
+static struct mlog_attribute mlog_attrs[MLOG_MAX_BITS] = {
+	define_mask(TCP),
+	define_mask(MSG),
+	define_mask(SOCKET),
+	define_mask(HEARTBEAT),
+	define_mask(HB_BIO),
+	define_mask(DLMFS),
+	define_mask(DLM),
+	define_mask(DLM_DOMAIN),
+	define_mask(DLM_THREAD),
+	define_mask(DLM_MASTER),
+	define_mask(DLM_RECOVERY),
+	define_mask(DLM_GLUE),
+	define_mask(VOTE),
+	define_mask(CONN),
+	define_mask(QUORUM),
+	define_mask(BASTS),
+	define_mask(CLUSTER),
+	define_mask(ERROR),
+	define_mask(NOTICE),
+	define_mask(KTHREAD),
+};
+
+static struct attribute *mlog_attr_ptrs[MLOG_MAX_BITS] = {NULL, };
+
+static ssize_t mlog_show(struct kobject *obj, struct attribute *attr,
+			 char *buf)
+{
+	struct mlog_attribute *mlog_attr = to_mlog_attr(attr);
+
+	return mlog_mask_show(mlog_attr->mask, buf);
+}
+
+static ssize_t mlog_store(struct kobject *obj, struct attribute *attr,
+			  const char *buf, size_t count)
+{
+	struct mlog_attribute *mlog_attr = to_mlog_attr(attr);
+
+	return mlog_mask_store(mlog_attr->mask, buf, count);
+}
+
+static const struct sysfs_ops mlog_attr_ops = {
+	.show  = mlog_show,
+	.store = mlog_store,
+};
+
+static struct kobj_type mlog_ktype = {
+	.default_attrs = mlog_attr_ptrs,
+	.sysfs_ops     = &mlog_attr_ops,
+};
+
+static struct kset mlog_kset = {
+	.kobj   = {.ktype = &mlog_ktype},
+};
+
+int mlog_sys_init(struct kset *r2cb_kset)
+{
+	int i = 0;
+
+	while (mlog_attrs[i].attr.mode) {
+		mlog_attr_ptrs[i] = &mlog_attrs[i].attr;
+		i++;
+	}
+	mlog_attr_ptrs[i] = NULL;
+
+	kobject_set_name(&mlog_kset.kobj, "logmask");
+	mlog_kset.kobj.kset = r2cb_kset;
+	return kset_register(&mlog_kset);
+}
+
+void mlog_sys_shutdown(void)
+{
+	kset_unregister(&mlog_kset);
+}

drivers/staging/ramster/cluster/masklog.h

+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2005, 2012 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef R2CLUSTER_MASKLOG_H
+#define R2CLUSTER_MASKLOG_H
+
+/*
+ * For now this is a trivial wrapper around printk() that gives the critical
+ * ability to enable sets of debugging output at run-time.  In the future this
+ * will almost certainly be redirected to relayfs so that it can pay a
+ * substantially lower heisenberg tax.
+ *
+ * Callers associate the message with a bitmask and a global bitmask is
+ * maintained with help from /proc.  If any of the bits match the message is
+ * output.
+ *
+ * We must have efficient bit tests on i386 and it seems gcc still emits crazy
+ * code for the 64bit compare.  It emits very good code for the dual unsigned
+ * long tests, though, completely avoiding tests that can never pass if the
+ * caller gives a constant bitmask that fills one of the longs with all 0s.  So
+ * the desire is to have almost all of the calls decided on by comparing just
+ * one of the longs.  This leads to having infrequently given bits that are
+ * frequently matched in the high bits.
+ *
+ * _ERROR and _NOTICE are used for messages that always go to the console and
+ * have appropriate KERN_ prefixes.  We wrap these in our function instead of
+ * just calling printk() so that this can eventually make its way through
+ * relayfs along with the debugging messages.  Everything else gets KERN_DEBUG.
+ * The inline tests and macro dance give GCC the opportunity to quite cleverly
+ * only emit the appropriage printk() when the caller passes in a constant
+ * mask, as is almost always the case.
+ *
+ * All this bitmask nonsense is managed from the files under
+ * /sys/fs/r2cb/logmask/.  Reading the files gives a straightforward
+ * indication of which bits are allowed (allow) or denied (off/deny).
+ *	ENTRY deny
+ *	EXIT deny
+ *	TCP off
+ *	MSG off
+ *	SOCKET off
+ *	ERROR allow
+ *	NOTICE allow
+ *
+ * Writing changes the state of a given bit and requires a strictly formatted
+ * single write() call:
+ *
+ *	write(fd, "allow", 5);
+ *
+ * Echoing allow/deny/off string into the logmask files can flip the bits
+ * on or off as expected; here is the bash script for example:
+ *
+ * log_mask="/sys/fs/r2cb/log_mask"
+ * for node in ENTRY EXIT TCP MSG SOCKET ERROR NOTICE; do
+ *	echo allow >"$log_mask"/"$node"
+ * done
+ *
+ * The debugfs.ramster tool can also flip the bits with the -l option:
+ *
+ * debugfs.ramster -l TCP allow
+ */
+
+/* for task_struct */
+#include <linux/sched.h>
+
+/* bits that are frequently given and infrequently matched in the low word */
+/* NOTE: If you add a flag, you need to also update masklog.c! */
+#define ML_TCP		0x0000000000000001ULL /* net cluster/tcp.c */
+#define ML_MSG		0x0000000000000002ULL /* net network messages */
+#define ML_SOCKET	0x0000000000000004ULL /* net socket lifetime */
+#define ML_HEARTBEAT	0x0000000000000008ULL /* hb all heartbeat tracking */
+#define ML_HB_BIO	0x0000000000000010ULL /* hb io tracing */
+#define ML_DLMFS	0x0000000000000020ULL /* dlm user dlmfs */
+#define ML_DLM		0x0000000000000040ULL /* dlm general debugging */
+#define ML_DLM_DOMAIN	0x0000000000000080ULL /* dlm domain debugging */
+#define ML_DLM_THREAD	0x0000000000000100ULL /* dlm domain thread */
+#define ML_DLM_MASTER	0x0000000000000200ULL /* dlm master functions */
+#define ML_DLM_RECOVERY	0x0000000000000400ULL /* dlm master functions */
+#define ML_DLM_GLUE	0x0000000000000800ULL /* ramster dlm glue layer */
+#define ML_VOTE		0x0000000000001000ULL /* ramster node messaging  */
+#define ML_CONN		0x0000000000002000ULL /* net connection management */
+#define ML_QUORUM	0x0000000000004000ULL /* net connection quorum */
+#define ML_BASTS	0x0000000000008000ULL /* dlmglue asts and basts */
+#define ML_CLUSTER	0x0000000000010000ULL /* cluster stack */
+
+/* bits that are infrequently given and frequently matched in the high word */
+#define ML_ERROR	0x1000000000000000ULL /* sent to KERN_ERR */
+#define ML_NOTICE	0x2000000000000000ULL /* setn to KERN_NOTICE */
+#define ML_KTHREAD	0x4000000000000000ULL /* kernel thread activity */
+
+#define MLOG_INITIAL_AND_MASK (ML_ERROR|ML_NOTICE)
+#ifndef MLOG_MASK_PREFIX
+#define MLOG_MASK_PREFIX 0
+#endif
+
+/*
+ * When logging is disabled, force the bit test to 0 for anything other
+ * than errors and notices, allowing gcc to remove the code completely.
+ * When enabled, allow all masks.
+ */
+#if defined(CONFIG_RAMSTER_DEBUG_MASKLOG)
+#define ML_ALLOWED_BITS (~0)
+#else
+#define ML_ALLOWED_BITS (ML_ERROR|ML_NOTICE)
+#endif
+
+#define MLOG_MAX_BITS 64
+
+struct mlog_bits {
+	unsigned long words[MLOG_MAX_BITS / BITS_PER_LONG];
+};
+
+extern struct mlog_bits mlog_and_bits, mlog_not_bits;
+
+#if BITS_PER_LONG == 32
+
+#define __mlog_test_u64(mask, bits)			\
+	((u32)(mask & 0xffffffff) & bits.words[0] ||	\
+	  ((u64)(mask) >> 32) & bits.words[1])
+#define __mlog_set_u64(mask, bits) do {			\
+	bits.words[0] |= (u32)(mask & 0xffffffff);	\
+	bits.words[1] |= (u64)(mask) >> 32;		\
+} while (0)
+#define __mlog_clear_u64(mask, bits) do {		\
+	bits.words[0] &= ~((u32)(mask & 0xffffffff));	\
+	bits.words[1] &= ~((u64)(mask) >> 32);		\
+} while (0)
+#define MLOG_BITS_RHS(mask) {				\
+	{						\
+		[0] = (u32)(mask & 0xffffffff),		\
+		[1] = (u64)(mask) >> 32,		\
+	}						\
+}
+
+#else /* 32bit long above, 64bit long below */
+
+#define __mlog_test_u64(mask, bits)	((mask) & bits.words[0])
+#define __mlog_set_u64(mask, bits) do {		\
+	bits.words[0] |= (mask);		\
+} while (0)
+#define __mlog_clear_u64(mask, bits) do {	\
+	bits.words[0] &= ~(mask);		\
+} while (0)
+#define MLOG_BITS_RHS(mask) { { (mask) } }
+
+#endif
+
+/*
+ * smp_processor_id() "helpfully" screams when called outside preemptible
+ * regions in current kernels.  sles doesn't have the variants that don't
+ * scream.  just do this instead of trying to guess which we're building
+ * against.. *sigh*.
+ */
+#define __mlog_cpu_guess ({		\
+	unsigned long _cpu = get_cpu();	\
+	put_cpu();			\
+	_cpu;				\
+})
+
+/* In the following two macros, the whitespace after the ',' just
+ * before ##args is intentional. Otherwise, gcc 2.95 will eat the
+ * previous token if args expands to nothing.
+ */
+#define __mlog_printk(level, fmt, args...)				\
+	printk(level "(%s,%u,%lu):%s:%d " fmt, current->comm,		\
+	       task_pid_nr(current), __mlog_cpu_guess,			\
+	       __PRETTY_FUNCTION__, __LINE__ , ##args)
+
+#define mlog(mask, fmt, args...) do {					\
+	u64 __m = MLOG_MASK_PREFIX | (mask);				\
+	if ((__m & ML_ALLOWED_BITS) &&					\
+	    __mlog_test_u64(__m, mlog_and_bits) &&			\
+	    !__mlog_test_u64(__m, mlog_not_bits)) {			\
+		if (__m & ML_ERROR)					\
+			__mlog_printk(KERN_ERR, "ERROR: "fmt , ##args);	\
+		else if (__m & ML_NOTICE)				\
+			__mlog_printk(KERN_NOTICE, fmt , ##args);	\
+		else							\
+			__mlog_printk(KERN_INFO, fmt , ##args);		\
+	}								\
+} while (0)
+
+#define mlog_errno(st) do {						\
+	int _st = (st);							\
+	if (_st != -ERESTARTSYS && _st != -EINTR &&			\
+	    _st != AOP_TRUNCATED_PAGE && _st != -ENOSPC)		\
+		mlog(ML_ERROR, "status = %lld\n", (long long)_st);	\
+} while (0)
+
+#define mlog_bug_on_msg(cond, fmt, args...) do {			\
+	if (cond) {							\
+		mlog(ML_ERROR, "bug expression: " #cond "\n");		\
+		mlog(ML_ERROR, fmt, ##args);				\
+		BUG();							\
+	}								\
+} while (0)
+
+#include <linux/kobject.h>
+#include <linux/sysfs.h>
+int mlog_sys_init(struct kset *r2cb_subsys);
+void mlog_sys_shutdown(void);
+
+#endif /* R2CLUSTER_MASKLOG_H */

drivers/staging/ramster/cluster/nodemanager.h

+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * nodemanager.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef R2CLUSTER_NODEMANAGER_H
+#define R2CLUSTER_NODEMANAGER_H
+
+#include "ramster_nodemanager.h"
+
+/* This totally doesn't belong here. */
+#include <linux/configfs.h>
+#include <linux/rbtree.h>
+
+enum r2nm_fence_method {
+	R2NM_FENCE_RESET	= 0,
+	R2NM_FENCE_PANIC,
+	R2NM_FENCE_METHODS,	/* Number of fence methods */
+};
+
+struct r2nm_node {
+	spinlock_t		nd_lock;
+	struct config_item	nd_item;
+	char			nd_name[R2NM_MAX_NAME_LEN+1]; /* replace? */
+	__u8			nd_num;
+	/* only one address per node, as attributes, for now. */
+	__be32			nd_ipv4_address;
+	__be16			nd_ipv4_port;
+	struct rb_node		nd_ip_node;
+	/* there can be only one local node for now */
+	int			nd_local;
+
+	unsigned long		nd_set_attributes;
+};
+
+struct r2nm_cluster {
+	struct config_group	cl_group;
+	unsigned		cl_has_local:1;
+	u8			cl_local_node;
+	rwlock_t		cl_nodes_lock;
+	struct r2nm_node	*cl_nodes[R2NM_MAX_NODES];
+	struct rb_root		cl_node_ip_tree;
+	unsigned int		cl_idle_timeout_ms;
+	unsigned int		cl_keepalive_delay_ms;
+	unsigned int		cl_reconnect_delay_ms;
+	enum r2nm_fence_method	cl_fence_method;
+
+	/* part of a hack for disk bitmap.. will go eventually. - zab */
+	unsigned long	cl_nodes_bitmap[BITS_TO_LONGS(R2NM_MAX_NODES)];
+};
+
+extern struct r2nm_cluster *r2nm_single_cluster;
+
+u8 r2nm_this_node(void);
+
+int r2nm_configured_node_map(unsigned long *map, unsigned bytes);
+struct r2nm_node *r2nm_get_node_by_num(u8 node_num);
+struct r2nm_node *r2nm_get_node_by_ip(__be32 addr);
+void r2nm_node_get(struct r2nm_node *node);
+void r2nm_node_put(struct r2nm_node *node);
+
+int r2nm_depend_item(struct config_item *item);
+void r2nm_undepend_item(struct config_item *item);
+int r2nm_depend_this_node(void);
+void r2nm_undepend_this_node(void);
+
+#endif /* R2CLUSTER_NODEMANAGER_H */

drivers/staging/ramster/cluster/ramster_nodemanager.h

+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ramster_nodemanager.h
+ *
+ * Header describing the interface between userspace and the kernel
+ * for the ramster_nodemanager module.
+ *
+ * Copyright (C) 2002, 2004, 2012 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef _RAMSTER_NODEMANAGER_H
+#define _RAMSTER_NODEMANAGER_H
+
+#define R2NM_API_VERSION	5
+
+#define R2NM_MAX_NODES		255
+#define R2NM_INVALID_NODE_NUM	255
+
+/* host name, group name, cluster name all 64 bytes */
+#define R2NM_MAX_NAME_LEN        64    /* __NEW_UTS_LEN */
+
+#endif /* _RAMSTER_NODEMANAGER_H */

drivers/staging/ramster/cluster/tcp.h

+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * tcp.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef R2CLUSTER_TCP_H
+#define R2CLUSTER_TCP_H
+
+#include <linux/socket.h>
+#ifdef __KERNEL__
+#include <net/sock.h>
+#include <linux/tcp.h>
+#else
+#include <sys/socket.h>
+#endif
+#include <linux/inet.h>
+#include <linux/in.h>
+
+struct r2net_msg {
+	__be16 magic;
+	__be16 data_len;
+	__be16 msg_type;
+	__be16 pad1;
+	__be32 sys_status;
+	__be32 status;
+	__be32 key;
+	__be32 msg_num;
+	__u8  buf[0];
+};
+
+typedef int (r2net_msg_handler_func)(struct r2net_msg *msg, u32 len, void *data,
+				     void **ret_data);
+typedef void (r2net_post_msg_handler_func)(int status, void *data,
+					   void *ret_data);
+
+#define R2NET_MAX_PAYLOAD_BYTES  (4096 - sizeof(struct r2net_msg))
+
+/* same as hb delay, we're waiting for another node to recognize our hb */
+#define R2NET_RECONNECT_DELAY_MS_DEFAULT	2000
+
+#define R2NET_KEEPALIVE_DELAY_MS_DEFAULT	2000
+#define R2NET_IDLE_TIMEOUT_MS_DEFAULT		30000
+
+
+/* TODO: figure this out.... */
+static inline int r2net_link_down(int err, struct socket *sock)
+{
+	if (sock) {
+		if (sock->sk->sk_state != TCP_ESTABLISHED &&
+			sock->sk->sk_state != TCP_CLOSE_WAIT)
+			return 1;
+	}
+
+	if (err >= 0)
+		return 0;
+	switch (err) {
+
+	/* ????????????????????????? */
+	case -ERESTARTSYS:
+	case -EBADF:
+	/* When the server has died, an ICMP port unreachable
+	 * message prompts ECONNREFUSED. */
+	case -ECONNREFUSED:
+	case -ENOTCONN:
+	case -ECONNRESET:
+	case -EPIPE:
+		return 1;
+
+	}
+	return 0;
+}
+
+enum {
+	R2NET_DRIVER_UNINITED,
+	R2NET_DRIVER_READY,
+};
+
+int r2net_send_message(u32 msg_type, u32 key, void *data, u32 len,
+		       u8 target_node, int *status);
+int r2net_send_message_vec(u32 msg_type, u32 key, struct kvec *vec,
+			   size_t veclen, u8 target_node, int *status);
+
+int r2net_register_handler(u32 msg_type, u32 key, u32 max_len,
+			   r2net_msg_handler_func *func, void *data,
+			   r2net_post_msg_handler_func *post_func,
+			   struct list_head *unreg_list);
+void r2net_unregister_handler_list(struct list_head *list);
+
+void r2net_fill_node_map(unsigned long *map, unsigned bytes);
+
+void r2net_force_data_magic(struct r2net_msg *, u16, u32);
+void r2net_hb_node_up_manual(int);
+struct r2net_node *r2net_nn_from_num(u8);
+
+struct r2nm_node;
+int r2net_register_hb_callbacks(void);
+void r2net_unregister_hb_callbacks(void);
+int r2net_start_listening(struct r2nm_node *node);
+void r2net_stop_listening(struct r2nm_node *node);
+void r2net_disconnect_node(struct r2nm_node *node);
+int r2net_num_connected_peers(void);
+
+int r2net_init(void);
+void r2net_exit(void);
+
+struct r2net_send_tracking;
+struct r2net_sock_container;
+
+#if 0
+int r2net_debugfs_init(void);
+void r2net_debugfs_exit(void);
+void r2net_debug_add_nst(struct r2net_send_tracking *nst);
+void r2net_debug_del_nst(struct r2net_send_tracking *nst);
+void r2net_debug_add_sc(struct r2net_sock_container *sc);
+void r2net_debug_del_sc(struct r2net_sock_container *sc);
+#else
+static inline int r2net_debugfs_init(void)
+{
+	return 0;
+}
+static inline void r2net_debugfs_exit(void)
+{
+}
+static inline void r2net_debug_add_nst(struct r2net_send_tracking *nst)
+{
+}
+static inline void r2net_debug_del_nst(struct r2net_send_tracking *nst)
+{
+}
+static inline void r2net_debug_add_sc(struct r2net_sock_container *sc)
+{
+}
+static inline void r2net_debug_del_sc(struct r2net_sock_container *sc)
+{
+}
+#endif	/* CONFIG_DEBUG_FS */
+
+#endif /* R2CLUSTER_TCP_H */

drivers/staging/ramster/cluster/tcp_internal.h

+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef R2CLUSTER_TCP_INTERNAL_H
+#define R2CLUSTER_TCP_INTERNAL_H
+
+#define R2NET_MSG_MAGIC           ((u16)0xfa55)
+#define R2NET_MSG_STATUS_MAGIC    ((u16)0xfa56)
+#define R2NET_MSG_KEEP_REQ_MAGIC  ((u16)0xfa57)
+#define R2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58)
+/*
+ * "data magic" is a long version of "status magic" where the message
+ * payload actually contains data to be passed in reply to certain messages
+ */
+#define R2NET_MSG_DATA_MAGIC      ((u16)0xfa59)
+
+/* we're delaying our quorum decision so that heartbeat will have timed
+ * out truly dead nodes by the time we come around to making decisions
+ * on their number */
+#define R2NET_QUORUM_DELAY_MS	\
+		((r2hb_dead_threshold + 2) * R2HB_REGION_TIMEOUT_MS)
+
+/*
+ * This version number represents quite a lot, unfortunately.  It not
+ * only represents the raw network message protocol on the wire but also
+ * locking semantics of the file system using the protocol.  It should
+ * be somewhere else, I'm sure, but right now it isn't.
+ *
+ * With version 11, we separate out the filesystem locking portion.  The
+ * filesystem now has a major.minor version it negotiates.  Version 11
+ * introduces this negotiation to the r2dlm protocol, and as such the
+ * version here in tcp_internal.h should not need to be bumped for
+ * filesystem locking changes.
+ *
+ * New in version 11
+ *	- Negotiation of filesystem locking in the dlm join.
+ *
+ * New in version 10:
+ *	- Meta/data locks combined
+ *
+ * New in version 9:
+ *	- All votes removed
+ *
+ * New in version 8:
+ *	- Replace delete inode votes with a cluster lock
+ *
+ * New in version 7:
+ *	- DLM join domain includes the live nodemap
+ *
+ * New in version 6:
+ *	- DLM lockres remote refcount fixes.
+ *
+ * New in version 5:
+ *	- Network timeout checking protocol
+ *
+ * New in version 4:
+ *	- Remove i_generation from lock names for better stat performance.
+ *
+ * New in version 3:
+ *	- Replace dentry votes with a cluster lock
+ *
+ * New in version 2:
+ *	- full 64 bit i_size in the metadata lock lvbs
+ *	- introduction of "rw" lock and pushing meta/data locking down
+ */
+#define R2NET_PROTOCOL_VERSION 11ULL
+struct r2net_handshake {
+	__be64	protocol_version;
+	__be64	connector_id;
+	__be32  r2hb_heartbeat_timeout_ms;
+	__be32  r2net_idle_timeout_ms;
+	__be32  r2net_keepalive_delay_ms;
+	__be32  r2net_reconnect_delay_ms;
+};
+
+struct r2net_node {
+	/* this is never called from int/bh */
+	spinlock_t			nn_lock;
+
+	/* set the moment an sc is allocated and a connect is started */
+	struct r2net_sock_container	*nn_sc;
+	/* _valid is only set after the handshake passes and tx can happen */
+	unsigned			nn_sc_valid:1;
+	/* if this is set tx just returns it */
+	int				nn_persistent_error;
+	/* It is only set to 1 after the idle time out. */
+	atomic_t			nn_timeout;
+
+	/* threads waiting for an sc to arrive wait on the wq for generation
+	 * to increase.  it is increased when a connecting socket succeeds
+	 * or fails or when an accepted socket is attached. */
+	wait_queue_head_t		nn_sc_wq;
+
+	struct idr			nn_status_idr;
+	struct list_head		nn_status_list;
+
+	/* connects are attempted from when heartbeat comes up until either hb
+	 * goes down, the node is unconfigured, no connect attempts succeed
+	 * before R2NET_CONN_IDLE_DELAY, or a connect succeeds.  connect_work
+	 * is queued from set_nn_state both from hb up and from itself if a
+	 * connect attempt fails and so can be self-arming.  shutdown is
+	 * careful to first mark the nn such that no connects will be attempted
+	 * before canceling delayed connect work and flushing the queue. */
+	struct delayed_work		nn_connect_work;
+	unsigned long			nn_last_connect_attempt;
+
+	/* this is queued as nodes come up and is canceled when a connection is
+	 * established.  this expiring gives up on the node and errors out
+	 * transmits */
+	struct delayed_work		nn_connect_expired;
+
+	/* after we give up on a socket we wait a while before deciding
+	 * that it is still heartbeating and that we should do some
+	 * quorum work */
+	struct delayed_work		nn_still_up;
+};
+
+struct r2net_sock_container {
+	struct kref		sc_kref;
+	/* the next two are valid for the life time of the sc */
+	struct socket		*sc_sock;
+	struct r2nm_node	*sc_node;
+
+	/* all of these sc work structs hold refs on the sc while they are
+	 * queued.  they should not be able to ref a freed sc.  the teardown
+	 * race is with r2net_wq destruction in r2net_stop_listening() */
+
+	/* rx and connect work are generated from socket callbacks.  sc
+	 * shutdown removes the callbacks and then flushes the work queue */
+	struct work_struct	sc_rx_work;
+	struct work_struct	sc_connect_work;
+	/* shutdown work is triggered in two ways.  the simple way is
+	 * for a code path calls ensure_shutdown which gets a lock, removes
+	 * the sc from the nn, and queues the work.  in this case the
+	 * work is single-shot.  the work is also queued from a sock
+	 * callback, though, and in this case the work will find the sc
+	 * still on the nn and will call ensure_shutdown itself.. this
+	 * ends up triggering the shutdown work again, though nothing
+	 * will be done in that second iteration.  so work queue teardown
+	 * has to be careful to remove the sc from the nn before waiting
+	 * on the work queue so that the shutdown work doesn't remove the
+	 * sc and rearm itself.
+	 */
+	struct work_struct	sc_shutdown_work;
+
+	struct timer_list	sc_idle_timeout;
+	struct delayed_work	sc_keepalive_work;
+
+	unsigned		sc_handshake_ok:1;
+
+	struct page		*sc_page;
+	size_t			sc_page_off;
+
+	/* original handlers for the sockets */
+	void			(*sc_state_change)(struct sock *sk);
+	void			(*sc_data_ready)(struct sock *sk, int bytes);
+
+	u32			sc_msg_key;
+	u16			sc_msg_type;
+
+#ifdef CONFIG_DEBUG_FS
+	struct list_head        sc_net_debug_item;
+	ktime_t			sc_tv_timer;
+	ktime_t			sc_tv_data_ready;
+	ktime_t			sc_tv_advance_start;
+	ktime_t			sc_tv_advance_stop;
+	ktime_t			sc_tv_func_start;
+	ktime_t			sc_tv_func_stop;
+#endif
+#ifdef CONFIG_RAMSTER_FS_STATS
+	ktime_t			sc_tv_acquiry_total;
+	ktime_t			sc_tv_send_total;
+	ktime_t			sc_tv_status_total;
+	u32			sc_send_count;
+	u32			sc_recv_count;
+	ktime_t			sc_tv_process_total;
+#endif
+	struct mutex		sc_send_lock;
+};
+
+struct r2net_msg_handler {
+	struct rb_node		nh_node;
+	u32			nh_max_len;
+	u32			nh_msg_type;
+	u32			nh_key;
+	r2net_msg_handler_func	*nh_func;
+	r2net_msg_handler_func	*nh_func_data;
+	r2net_post_msg_handler_func
+				*nh_post_func;
+	struct kref		nh_kref;
+	struct list_head	nh_unregister_item;
+};
+
+enum r2net_system_error {
+	R2NET_ERR_NONE = 0,
+	R2NET_ERR_NO_HNDLR,
+	R2NET_ERR_OVERFLOW,
+	R2NET_ERR_DIED,
+	R2NET_ERR_MAX
+};
+
+struct r2net_status_wait {
+	enum r2net_system_error	ns_sys_status;
+	s32			ns_status;
+	int			ns_id;
+	wait_queue_head_t	ns_wq;
+	struct list_head	ns_node_item;
+};
+
+#ifdef CONFIG_DEBUG_FS
+/* just for state dumps */
+struct r2net_send_tracking {
+	struct list_head		st_net_debug_item;
+	struct task_struct		*st_task;
+	struct r2net_sock_container	*st_sc;
+	u32				st_id;
+	u32				st_msg_type;
+	u32				st_msg_key;
+	u8				st_node;
+	ktime_t				st_sock_time;
+	ktime_t				st_send_time;
+	ktime_t				st_status_time;
+};
+#else
+struct r2net_send_tracking {
+	u32	dummy;
+};
+#endif	/* CONFIG_DEBUG_FS */
+
+#endif /* R2CLUSTER_TCP_INTERNAL_H */