[mirror_ubuntu-artful-kernel.git] / fs / ocfs2 / dlmfs / userdlm.c

/* -*- mode: c; c-basic-offset: 8; -*-
 * vim: noexpandtab sw=8 ts=8 sts=0:
 *
 * userdlm.c
 *
 * Code which implements the kernel side of a minimal userspace
 * interface to our DLM.
 *
 * Many of the functions here are pared down versions of dlmglue.c
 * functions.
 *
 * Copyright (C) 2003, 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.
 */

#include <linux/signal.h>

#include <linux/module.h>
#include <linux/fs.h>
#include <linux/types.h>
#include <linux/crc32.h>


#include "cluster/nodemanager.h"
#include "cluster/heartbeat.h"
#include "cluster/tcp.h"

#include "dlm/dlmapi.h"

#include "userdlm.h"

#define MLOG_MASK_PREFIX ML_DLMFS
#include "cluster/masklog.h"

static inline int user_check_wait_flag(struct user_lock_res *lockres,
				       int flag)
{
	int ret;

	spin_lock(&lockres->l_lock);
	ret = lockres->l_flags & flag;
	spin_unlock(&lockres->l_lock);

	return ret;
}

static inline void user_wait_on_busy_lock(struct user_lock_res *lockres)

{
	wait_event(lockres->l_event,
		   !user_check_wait_flag(lockres, USER_LOCK_BUSY));
}

static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres)

{
	wait_event(lockres->l_event,
		   !user_check_wait_flag(lockres, USER_LOCK_BLOCKED));
}

/* I heart container_of... */
static inline struct dlm_ctxt *
dlm_ctxt_from_user_lockres(struct user_lock_res *lockres)
{
	struct dlmfs_inode_private *ip;

	ip = container_of(lockres,
			  struct dlmfs_inode_private,
			  ip_lockres);
	return ip->ip_dlm;
}

static struct inode *
user_dlm_inode_from_user_lockres(struct user_lock_res *lockres)
{
	struct dlmfs_inode_private *ip;

	ip = container_of(lockres,
			  struct dlmfs_inode_private,
			  ip_lockres);
	return &ip->ip_vfs_inode;
}

static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
{
	spin_lock(&lockres->l_lock);
	lockres->l_flags &= ~USER_LOCK_BUSY;
	spin_unlock(&lockres->l_lock);
}

#define user_log_dlm_error(_func, _stat, _lockres) do {			\
	mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on "		\
		"resource %.*s: %s\n", dlm_errname(_stat), _func,	\
		_lockres->l_namelen, _lockres->l_name, dlm_errmsg(_stat)); \
} while (0)

/* WARNING: This function lives in a world where the only three lock
 * levels are EX, PR, and NL. It *will* have to be adjusted when more
 * lock types are added. */
static inline int user_highest_compat_lock_level(int level)
{
	int new_level = LKM_EXMODE;

	if (level == LKM_EXMODE)
		new_level = LKM_NLMODE;
	else if (level == LKM_PRMODE)
		new_level = LKM_PRMODE;
	return new_level;
}

static void user_ast(void *opaque)
{
	struct user_lock_res *lockres = opaque;
	struct dlm_lockstatus *lksb;

	mlog(0, "AST fired for lockres %.*s\n", lockres->l_namelen,
	     lockres->l_name);

	spin_lock(&lockres->l_lock);

	lksb = &(lockres->l_lksb);
	if (lksb->status != DLM_NORMAL) {
		mlog(ML_ERROR, "lksb status value of %u on lockres %.*s\n",
		     lksb->status, lockres->l_namelen, lockres->l_name);
		spin_unlock(&lockres->l_lock);
		return;
	}

	mlog_bug_on_msg(lockres->l_requested == LKM_IVMODE,
			"Lockres %.*s, requested ivmode. flags 0x%x\n",
			lockres->l_namelen, lockres->l_name, lockres->l_flags);

	/* we're downconverting. */
	if (lockres->l_requested < lockres->l_level) {
		if (lockres->l_requested <=
		    user_highest_compat_lock_level(lockres->l_blocking)) {
			lockres->l_blocking = LKM_NLMODE;
			lockres->l_flags &= ~USER_LOCK_BLOCKED;
		}
	}

	lockres->l_level = lockres->l_requested;
	lockres->l_requested = LKM_IVMODE;
	lockres->l_flags |= USER_LOCK_ATTACHED;
	lockres->l_flags &= ~USER_LOCK_BUSY;

	spin_unlock(&lockres->l_lock);

	wake_up(&lockres->l_event);
}

static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres)
{
	struct inode *inode;
	inode = user_dlm_inode_from_user_lockres(lockres);
	if (!igrab(inode))
		BUG();
}

static void user_dlm_unblock_lock(struct work_struct *work);

static void __user_dlm_queue_lockres(struct user_lock_res *lockres)
{
	if (!(lockres->l_flags & USER_LOCK_QUEUED)) {
		user_dlm_grab_inode_ref(lockres);

		INIT_WORK(&lockres->l_work, user_dlm_unblock_lock);

		queue_work(user_dlm_worker, &lockres->l_work);
		lockres->l_flags |= USER_LOCK_QUEUED;
	}
}

static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres)
{
	int queue = 0;

	if (!(lockres->l_flags & USER_LOCK_BLOCKED))
		return;

	switch (lockres->l_blocking) {
	case LKM_EXMODE:
		if (!lockres->l_ex_holders && !lockres->l_ro_holders)
			queue = 1;
		break;
	case LKM_PRMODE:
		if (!lockres->l_ex_holders)
			queue = 1;
		break;
	default:
		BUG();
	}

	if (queue)
		__user_dlm_queue_lockres(lockres);
}

static void user_bast(void *opaque, int level)
{
	struct user_lock_res *lockres = opaque;

	mlog(0, "Blocking AST fired for lockres %.*s. Blocking level %d\n",
	     lockres->l_namelen, lockres->l_name, level);

	spin_lock(&lockres->l_lock);
	lockres->l_flags |= USER_LOCK_BLOCKED;
	if (level > lockres->l_blocking)
		lockres->l_blocking = level;

	__user_dlm_queue_lockres(lockres);
	spin_unlock(&lockres->l_lock);

	wake_up(&lockres->l_event);
}

static void user_unlock_ast(void *opaque, enum dlm_status status)
{
	struct user_lock_res *lockres = opaque;

	mlog(0, "UNLOCK AST called on lock %.*s\n", lockres->l_namelen,
	     lockres->l_name);

	if (status != DLM_NORMAL && status != DLM_CANCELGRANT)
		mlog(ML_ERROR, "Dlm returns status %d\n", status);

	spin_lock(&lockres->l_lock);
	/* The teardown flag gets set early during the unlock process,
	 * so test the cancel flag to make sure that this ast isn't
	 * for a concurrent cancel. */
	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN
	    && !(lockres->l_flags & USER_LOCK_IN_CANCEL)) {
		lockres->l_level = LKM_IVMODE;
	} else if (status == DLM_CANCELGRANT) {
		/* We tried to cancel a convert request, but it was
		 * already granted. Don't clear the busy flag - the
		 * ast should've done this already. */
		BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
		lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
		goto out_noclear;
	} else {
		BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
		/* Cancel succeeded, we want to re-queue */
		lockres->l_requested = LKM_IVMODE; /* cancel an
						    * upconvert
						    * request. */
		lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
		/* we want the unblock thread to look at it again
		 * now. */
		if (lockres->l_flags & USER_LOCK_BLOCKED)
			__user_dlm_queue_lockres(lockres);
	}

	lockres->l_flags &= ~USER_LOCK_BUSY;
out_noclear:
	spin_unlock(&lockres->l_lock);

	wake_up(&lockres->l_event);
}

static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres)
{
	struct inode *inode;
	inode = user_dlm_inode_from_user_lockres(lockres);
	iput(inode);
}

static void user_dlm_unblock_lock(struct work_struct *work)
{
	int new_level, status;
	struct user_lock_res *lockres =
		container_of(work, struct user_lock_res, l_work);
	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);

	mlog(0, "processing lockres %.*s\n", lockres->l_namelen,
	     lockres->l_name);

	spin_lock(&lockres->l_lock);

	mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
			"Lockres %.*s, flags 0x%x\n",
			lockres->l_namelen, lockres->l_name, lockres->l_flags);

	/* notice that we don't clear USER_LOCK_BLOCKED here. If it's
	 * set, we want user_ast clear it. */
	lockres->l_flags &= ~USER_LOCK_QUEUED;

	/* It's valid to get here and no longer be blocked - if we get
	 * several basts in a row, we might be queued by the first
	 * one, the unblock thread might run and clear the queued
	 * flag, and finally we might get another bast which re-queues
	 * us before our ast for the downconvert is called. */
	if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
		spin_unlock(&lockres->l_lock);
		goto drop_ref;
	}

	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
		spin_unlock(&lockres->l_lock);
		goto drop_ref;
	}

	if (lockres->l_flags & USER_LOCK_BUSY) {
		if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
			spin_unlock(&lockres->l_lock);
			goto drop_ref;
		}

		lockres->l_flags |= USER_LOCK_IN_CANCEL;
		spin_unlock(&lockres->l_lock);

		status = dlmunlock(dlm,
				   &lockres->l_lksb,
				   LKM_CANCEL,
				   user_unlock_ast,
				   lockres);
		if (status != DLM_NORMAL)
			user_log_dlm_error("dlmunlock", status, lockres);
		goto drop_ref;
	}

	/* If there are still incompat holders, we can exit safely
	 * without worrying about re-queueing this lock as that will
	 * happen on the last call to user_cluster_unlock. */
	if ((lockres->l_blocking == LKM_EXMODE)
	    && (lockres->l_ex_holders || lockres->l_ro_holders)) {
		spin_unlock(&lockres->l_lock);
		mlog(0, "can't downconvert for ex: ro = %u, ex = %u\n",
			lockres->l_ro_holders, lockres->l_ex_holders);
		goto drop_ref;
	}

	if ((lockres->l_blocking == LKM_PRMODE)
	    && lockres->l_ex_holders) {
		spin_unlock(&lockres->l_lock);
		mlog(0, "can't downconvert for pr: ex = %u\n",
			lockres->l_ex_holders);
		goto drop_ref;
	}

	/* yay, we can downconvert now. */
	new_level = user_highest_compat_lock_level(lockres->l_blocking);
	lockres->l_requested = new_level;
	lockres->l_flags |= USER_LOCK_BUSY;
	mlog(0, "Downconvert lock from %d to %d\n",
		lockres->l_level, new_level);
	spin_unlock(&lockres->l_lock);

	/* need lock downconvert request now... */
	status = dlmlock(dlm,
			 new_level,
			 &lockres->l_lksb,
			 LKM_CONVERT|LKM_VALBLK,
			 lockres->l_name,
			 lockres->l_namelen,
			 user_ast,
			 lockres,
			 user_bast);
	if (status != DLM_NORMAL) {
		user_log_dlm_error("dlmlock", status, lockres);
		user_recover_from_dlm_error(lockres);
	}

drop_ref:
	user_dlm_drop_inode_ref(lockres);
}

static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
					int level)
{
	switch(level) {
	case LKM_EXMODE:
		lockres->l_ex_holders++;
		break;
	case LKM_PRMODE:
		lockres->l_ro_holders++;
		break;
	default:
		BUG();
	}
}

/* predict what lock level we'll be dropping down to on behalf
 * of another node, and return true if the currently wanted
 * level will be compatible with it. */
static inline int
user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
				  int wanted)
{
	BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));

	return wanted <= user_highest_compat_lock_level(lockres->l_blocking);
}

int user_dlm_cluster_lock(struct user_lock_res *lockres,
			  int level,
			  int lkm_flags)
{
	int status, local_flags;
	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);

	if (level != LKM_EXMODE &&
	    level != LKM_PRMODE) {
		mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
		     lockres->l_namelen, lockres->l_name);
		status = -EINVAL;
		goto bail;
	}

	mlog(0, "lockres %.*s: asking for %s lock, passed flags = 0x%x\n",
	     lockres->l_namelen, lockres->l_name,
	     (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE",
	     lkm_flags);

again:
	if (signal_pending(current)) {
		status = -ERESTARTSYS;
		goto bail;
	}

	spin_lock(&lockres->l_lock);

	/* We only compare against the currently granted level
	 * here. If the lock is blocked waiting on a downconvert,
	 * we'll get caught below. */
	if ((lockres->l_flags & USER_LOCK_BUSY) &&
	    (level > lockres->l_level)) {
		/* is someone sitting in dlm_lock? If so, wait on
		 * them. */
		spin_unlock(&lockres->l_lock);

		user_wait_on_busy_lock(lockres);
		goto again;
	}

	if ((lockres->l_flags & USER_LOCK_BLOCKED) &&
	    (!user_may_continue_on_blocked_lock(lockres, level))) {
		/* is the lock is currently blocked on behalf of
		 * another node */
		spin_unlock(&lockres->l_lock);

		user_wait_on_blocked_lock(lockres);
		goto again;
	}

	if (level > lockres->l_level) {
		local_flags = lkm_flags | LKM_VALBLK;
		if (lockres->l_level != LKM_IVMODE)
			local_flags |= LKM_CONVERT;

		lockres->l_requested = level;
		lockres->l_flags |= USER_LOCK_BUSY;
		spin_unlock(&lockres->l_lock);

		BUG_ON(level == LKM_IVMODE);
		BUG_ON(level == LKM_NLMODE);

		/* call dlm_lock to upgrade lock now */
		status = dlmlock(dlm,
				 level,
				 &lockres->l_lksb,
				 local_flags,
				 lockres->l_name,
				 lockres->l_namelen,
				 user_ast,
				 lockres,
				 user_bast);
		if (status != DLM_NORMAL) {
			if ((lkm_flags & LKM_NOQUEUE) &&
			    (status == DLM_NOTQUEUED))
				status = -EAGAIN;
			else {
				user_log_dlm_error("dlmlock", status, lockres);
				status = -EINVAL;
			}
			user_recover_from_dlm_error(lockres);
			goto bail;
		}

		user_wait_on_busy_lock(lockres);
		goto again;
	}

	user_dlm_inc_holders(lockres, level);
	spin_unlock(&lockres->l_lock);

	status = 0;
bail:
	return status;
}

static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
					int level)
{
	switch(level) {
	case LKM_EXMODE:
		BUG_ON(!lockres->l_ex_holders);
		lockres->l_ex_holders--;
		break;
	case LKM_PRMODE:
		BUG_ON(!lockres->l_ro_holders);
		lockres->l_ro_holders--;
		break;
	default:
		BUG();
	}
}

void user_dlm_cluster_unlock(struct user_lock_res *lockres,
			     int level)
{
	if (level != LKM_EXMODE &&
	    level != LKM_PRMODE) {
		mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
		     lockres->l_namelen, lockres->l_name);
		return;
	}

	spin_lock(&lockres->l_lock);
	user_dlm_dec_holders(lockres, level);
	__user_dlm_cond_queue_lockres(lockres);
	spin_unlock(&lockres->l_lock);
}

void user_dlm_write_lvb(struct inode *inode,
			const char *val,
			unsigned int len)
{
	struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
	char *lvb = lockres->l_lksb.lvb;

	BUG_ON(len > DLM_LVB_LEN);

	spin_lock(&lockres->l_lock);

	BUG_ON(lockres->l_level < LKM_EXMODE);
	memcpy(lvb, val, len);

	spin_unlock(&lockres->l_lock);
}

void user_dlm_read_lvb(struct inode *inode,
		       char *val,
		       unsigned int len)
{
	struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
	char *lvb = lockres->l_lksb.lvb;

	BUG_ON(len > DLM_LVB_LEN);

	spin_lock(&lockres->l_lock);

	BUG_ON(lockres->l_level < LKM_PRMODE);
	memcpy(val, lvb, len);

	spin_unlock(&lockres->l_lock);
}

void user_dlm_lock_res_init(struct user_lock_res *lockres,
			    struct dentry *dentry)
{
	memset(lockres, 0, sizeof(*lockres));

	spin_lock_init(&lockres->l_lock);
	init_waitqueue_head(&lockres->l_event);
	lockres->l_level = LKM_IVMODE;
	lockres->l_requested = LKM_IVMODE;
	lockres->l_blocking = LKM_IVMODE;

	/* should have been checked before getting here. */
	BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);

	memcpy(lockres->l_name,
	       dentry->d_name.name,
	       dentry->d_name.len);
	lockres->l_namelen = dentry->d_name.len;
}

int user_dlm_destroy_lock(struct user_lock_res *lockres)
{
	int status = -EBUSY;
	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);

	mlog(0, "asked to destroy %.*s\n", lockres->l_namelen, lockres->l_name);

	spin_lock(&lockres->l_lock);
	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
		spin_unlock(&lockres->l_lock);
		return 0;
	}

	lockres->l_flags |= USER_LOCK_IN_TEARDOWN;

	while (lockres->l_flags & USER_LOCK_BUSY) {
		spin_unlock(&lockres->l_lock);

		user_wait_on_busy_lock(lockres);

		spin_lock(&lockres->l_lock);
	}

	if (lockres->l_ro_holders || lockres->l_ex_holders) {
		spin_unlock(&lockres->l_lock);
		goto bail;
	}

	status = 0;
	if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
		spin_unlock(&lockres->l_lock);
		goto bail;
	}

	lockres->l_flags &= ~USER_LOCK_ATTACHED;
	lockres->l_flags |= USER_LOCK_BUSY;
	spin_unlock(&lockres->l_lock);

	status = dlmunlock(dlm,
			   &lockres->l_lksb,
			   LKM_VALBLK,
			   user_unlock_ast,
			   lockres);
	if (status != DLM_NORMAL) {
		user_log_dlm_error("dlmunlock", status, lockres);
		status = -EINVAL;
		goto bail;
	}

	user_wait_on_busy_lock(lockres);

	status = 0;
bail:
	return status;
}

struct dlm_ctxt *user_dlm_register_context(struct qstr *name,
					   struct dlm_protocol_version *proto)
{
	struct dlm_ctxt *dlm;
	u32 dlm_key;
	char *domain;

	domain = kmalloc(name->len + 1, GFP_NOFS);
	if (!domain) {
		mlog_errno(-ENOMEM);
		return ERR_PTR(-ENOMEM);
	}

	dlm_key = crc32_le(0, name->name, name->len);

	snprintf(domain, name->len + 1, "%.*s", name->len, name->name);

	dlm = dlm_register_domain(domain, dlm_key, proto);
	if (IS_ERR(dlm))
		mlog_errno(PTR_ERR(dlm));

	kfree(domain);
	return dlm;
}

void user_dlm_unregister_context(struct dlm_ctxt *dlm)
{
	dlm_unregister_domain(dlm);
}
Commit	Line	Data
8df08c89 MF	1	/* -- mode: c; c-basic-offset: 8; --
	2	* vim: noexpandtab sw=8 ts=8 sts=0:
	3	*
	4	* userdlm.c
	5	*
	6	* Code which implements the kernel side of a minimal userspace
	7	* interface to our DLM.
	8	*
	9	* Many of the functions here are pared down versions of dlmglue.c
	10	* functions.
	11	*
	12	* Copyright (C) 2003, 2004 Oracle. All rights reserved.
	13	*
	14	* This program is free software; you can redistribute it and/or
	15	* modify it under the terms of the GNU General Public
	16	* License as published by the Free Software Foundation; either
	17	* version 2 of the License, or (at your option) any later version.
	18	*
	19	* This program is distributed in the hope that it will be useful,
	20	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	21	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	22	* General Public License for more details.
	23	*
	24	* You should have received a copy of the GNU General Public
	25	* License along with this program; if not, write to the
	26	* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
	27	* Boston, MA 021110-1307, USA.
	28	*/
	29
aee93ac4	30	#include <linux/signal.h>
8df08c89 MF	31
	32	#include <linux/module.h>
	33	#include <linux/fs.h>
	34	#include <linux/types.h>
	35	#include <linux/crc32.h>
	36
	37
	38	#include "cluster/nodemanager.h"
	39	#include "cluster/heartbeat.h"
	40	#include "cluster/tcp.h"
	41
34a9dd7e	42	#include "dlm/dlmapi.h"
8df08c89 MF	43
	44	#include "userdlm.h"
	45
	46	#define MLOG_MASK_PREFIX ML_DLMFS
	47	#include "cluster/masklog.h"
	48
	49	static inline int user_check_wait_flag(struct user_lock_res *lockres,
	50	int flag)
	51	{
	52	int ret;
	53
	54	spin_lock(&lockres->l_lock);
	55	ret = lockres->l_flags & flag;
	56	spin_unlock(&lockres->l_lock);
	57
	58	return ret;
	59	}
	60
	61	static inline void user_wait_on_busy_lock(struct user_lock_res *lockres)
	62
	63	{
	64	wait_event(lockres->l_event,
	65	!user_check_wait_flag(lockres, USER_LOCK_BUSY));
	66	}
	67
	68	static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres)
	69
	70	{
	71	wait_event(lockres->l_event,
	72	!user_check_wait_flag(lockres, USER_LOCK_BLOCKED));
	73	}
	74
	75	/* I heart container_of... */
	76	static inline struct dlm_ctxt *
	77	dlm_ctxt_from_user_lockres(struct user_lock_res *lockres)
	78	{
	79	struct dlmfs_inode_private *ip;
	80
	81	ip = container_of(lockres,
	82	struct dlmfs_inode_private,
	83	ip_lockres);
	84	return ip->ip_dlm;
	85	}
	86
	87	static struct inode *
	88	user_dlm_inode_from_user_lockres(struct user_lock_res *lockres)
	89	{
	90	struct dlmfs_inode_private *ip;
	91
	92	ip = container_of(lockres,
	93	struct dlmfs_inode_private,
	94	ip_lockres);
	95	return &ip->ip_vfs_inode;
	96	}
	97
	98	static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
	99	{
	100	spin_lock(&lockres->l_lock);
	101	lockres->l_flags &= ~USER_LOCK_BUSY;
	102	spin_unlock(&lockres->l_lock);
	103	}
	104
ea5b3a18 MF	105	#define user_log_dlm_error(_func, _stat, _lockres) do { \
	106	mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \
	107	"resource %.*s: %s\n", dlm_errname(_stat), _func, \
	108	_lockres->l_namelen, _lockres->l_name, dlm_errmsg(_stat)); \
8df08c89 MF	109	} while (0)
	110
	111	/* WARNING: This function lives in a world where the only three lock
	112	* levels are EX, PR, and NL. It will have to be adjusted when more
	113	* lock types are added. */
	114	static inline int user_highest_compat_lock_level(int level)
	115	{
	116	int new_level = LKM_EXMODE;
	117
	118	if (level == LKM_EXMODE)
	119	new_level = LKM_NLMODE;
	120	else if (level == LKM_PRMODE)
	121	new_level = LKM_PRMODE;
	122	return new_level;
	123	}
	124
	125	static void user_ast(void *opaque)
	126	{
	127	struct user_lock_res *lockres = opaque;
	128	struct dlm_lockstatus *lksb;
	129
ea5b3a18 MF	130	mlog(0, "AST fired for lockres %.*s\n", lockres->l_namelen,
ea5b3a18 MF	131	lockres->l_name);
8df08c89 MF	132
	133	spin_lock(&lockres->l_lock);
	134
	135	lksb = &(lockres->l_lksb);
	136	if (lksb->status != DLM_NORMAL) {
ea5b3a18 MF	137	mlog(ML_ERROR, "lksb status value of %u on lockres %.*s\n",
ea5b3a18 MF	138	lksb->status, lockres->l_namelen, lockres->l_name);
8df08c89 MF	139	spin_unlock(&lockres->l_lock);
	140	return;
	141	}
	142
cc6eb725	143	mlog_bug_on_msg(lockres->l_requested == LKM_IVMODE,
ea5b3a18 MF	144	"Lockres %.*s, requested ivmode. flags 0x%x\n",
ea5b3a18 MF	145	lockres->l_namelen, lockres->l_name, lockres->l_flags);
cc6eb725	146
8df08c89 MF	147	/* we're downconverting. */
	148	if (lockres->l_requested < lockres->l_level) {
	149	if (lockres->l_requested <=
	150	user_highest_compat_lock_level(lockres->l_blocking)) {
	151	lockres->l_blocking = LKM_NLMODE;
	152	lockres->l_flags &= ~USER_LOCK_BLOCKED;
	153	}
	154	}
	155
	156	lockres->l_level = lockres->l_requested;
	157	lockres->l_requested = LKM_IVMODE;
	158	lockres->l_flags \|= USER_LOCK_ATTACHED;
	159	lockres->l_flags &= ~USER_LOCK_BUSY;
	160
	161	spin_unlock(&lockres->l_lock);
	162
	163	wake_up(&lockres->l_event);
	164	}
	165
	166	static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres)
	167	{
	168	struct inode *inode;
	169	inode = user_dlm_inode_from_user_lockres(lockres);
	170	if (!igrab(inode))
	171	BUG();
	172	}
	173
c4028958	174	static void user_dlm_unblock_lock(struct work_struct *work);
8df08c89 MF	175
	176	static void __user_dlm_queue_lockres(struct user_lock_res *lockres)
	177	{
	178	if (!(lockres->l_flags & USER_LOCK_QUEUED)) {
	179	user_dlm_grab_inode_ref(lockres);
	180
c4028958	181	INIT_WORK(&lockres->l_work, user_dlm_unblock_lock);
8df08c89 MF	182
	183	queue_work(user_dlm_worker, &lockres->l_work);
	184	lockres->l_flags \|= USER_LOCK_QUEUED;
	185	}
	186	}
	187
	188	static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres)
	189	{
	190	int queue = 0;
	191
	192	if (!(lockres->l_flags & USER_LOCK_BLOCKED))
	193	return;
	194
	195	switch (lockres->l_blocking) {
	196	case LKM_EXMODE:
	197	if (!lockres->l_ex_holders && !lockres->l_ro_holders)
	198	queue = 1;
	199	break;
	200	case LKM_PRMODE:
	201	if (!lockres->l_ex_holders)
	202	queue = 1;
	203	break;
	204	default:
	205	BUG();
	206	}
	207
	208	if (queue)
	209	__user_dlm_queue_lockres(lockres);
	210	}
	211
	212	static void user_bast(void *opaque, int level)
	213	{
	214	struct user_lock_res *lockres = opaque;
	215
ea5b3a18 MF	216	mlog(0, "Blocking AST fired for lockres %.*s. Blocking level %d\n",
ea5b3a18 MF	217	lockres->l_namelen, lockres->l_name, level);
8df08c89 MF	218
	219	spin_lock(&lockres->l_lock);
	220	lockres->l_flags \|= USER_LOCK_BLOCKED;
	221	if (level > lockres->l_blocking)
	222	lockres->l_blocking = level;
	223
	224	__user_dlm_queue_lockres(lockres);
	225	spin_unlock(&lockres->l_lock);
	226
	227	wake_up(&lockres->l_event);
	228	}
	229
	230	static void user_unlock_ast(void *opaque, enum dlm_status status)
	231	{
	232	struct user_lock_res *lockres = opaque;
	233
ea5b3a18 MF	234	mlog(0, "UNLOCK AST called on lock %.*s\n", lockres->l_namelen,
ea5b3a18 MF	235	lockres->l_name);
8df08c89	236
f43e6918	237	if (status != DLM_NORMAL && status != DLM_CANCELGRANT)
8df08c89 MF	238	mlog(ML_ERROR, "Dlm returns status %d\n", status);
	239
	240	spin_lock(&lockres->l_lock);
2cd98885 MF	241	/* The teardown flag gets set early during the unlock process,
	242	* so test the cancel flag to make sure that this ast isn't
	243	* for a concurrent cancel. */
	244	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN
	245	&& !(lockres->l_flags & USER_LOCK_IN_CANCEL)) {
8df08c89	246	lockres->l_level = LKM_IVMODE;
2cd98885	247	} else if (status == DLM_CANCELGRANT) {
f43e6918 MF	248	/* We tried to cancel a convert request, but it was
	249	* already granted. Don't clear the busy flag - the
	250	* ast should've done this already. */
	251	BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
	252	lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
	253	goto out_noclear;
	254	} else {
	255	BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
	256	/* Cancel succeeded, we want to re-queue */
8df08c89 MF	257	lockres->l_requested = LKM_IVMODE; /* cancel an
	258	* upconvert
	259	* request. */
	260	lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
	261	/* we want the unblock thread to look at it again
	262	* now. */
f43e6918 MF	263	if (lockres->l_flags & USER_LOCK_BLOCKED)
f43e6918 MF	264	__user_dlm_queue_lockres(lockres);
8df08c89 MF	265	}
	266
	267	lockres->l_flags &= ~USER_LOCK_BUSY;
f43e6918	268	out_noclear:
8df08c89 MF	269	spin_unlock(&lockres->l_lock);
	270
	271	wake_up(&lockres->l_event);
	272	}
	273
	274	static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres)
	275	{
	276	struct inode *inode;
	277	inode = user_dlm_inode_from_user_lockres(lockres);
	278	iput(inode);
	279	}
	280
c4028958	281	static void user_dlm_unblock_lock(struct work_struct *work)
8df08c89 MF	282	{
8df08c89 MF	283	int new_level, status;
c4028958 DH	284	struct user_lock_res *lockres =
c4028958 DH	285	container_of(work, struct user_lock_res, l_work);
8df08c89 MF	286	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
8df08c89 MF	287
ea5b3a18 MF	288	mlog(0, "processing lockres %.*s\n", lockres->l_namelen,
ea5b3a18 MF	289	lockres->l_name);
8df08c89 MF	290
	291	spin_lock(&lockres->l_lock);
	292
1f7bc828	293	mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
ea5b3a18 MF	294	"Lockres %.*s, flags 0x%x\n",
ea5b3a18 MF	295	lockres->l_namelen, lockres->l_name, lockres->l_flags);
8df08c89	296
1f7bc828 MF	297	/* notice that we don't clear USER_LOCK_BLOCKED here. If it's
1f7bc828 MF	298	* set, we want user_ast clear it. */
8df08c89 MF	299	lockres->l_flags &= ~USER_LOCK_QUEUED;
8df08c89 MF	300
1f7bc828 MF	301	/* It's valid to get here and no longer be blocked - if we get
	302	* several basts in a row, we might be queued by the first
	303	* one, the unblock thread might run and clear the queued
	304	* flag, and finally we might get another bast which re-queues
	305	* us before our ast for the downconvert is called. */
	306	if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
1f7bc828 MF	307	spin_unlock(&lockres->l_lock);
	308	goto drop_ref;
	309	}
	310
8df08c89	311	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
8df08c89 MF	312	spin_unlock(&lockres->l_lock);
	313	goto drop_ref;
	314	}
	315
	316	if (lockres->l_flags & USER_LOCK_BUSY) {
8df08c89 MF	317	if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
	318	spin_unlock(&lockres->l_lock);
	319	goto drop_ref;
	320	}
	321
	322	lockres->l_flags \|= USER_LOCK_IN_CANCEL;
	323	spin_unlock(&lockres->l_lock);
	324
	325	status = dlmunlock(dlm,
	326	&lockres->l_lksb,
	327	LKM_CANCEL,
	328	user_unlock_ast,
	329	lockres);
f43e6918	330	if (status != DLM_NORMAL)
8df08c89 MF	331	user_log_dlm_error("dlmunlock", status, lockres);
	332	goto drop_ref;
	333	}
	334
	335	/* If there are still incompat holders, we can exit safely
	336	* without worrying about re-queueing this lock as that will
	337	* happen on the last call to user_cluster_unlock. */
	338	if ((lockres->l_blocking == LKM_EXMODE)
	339	&& (lockres->l_ex_holders \|\| lockres->l_ro_holders)) {
	340	spin_unlock(&lockres->l_lock);
	341	mlog(0, "can't downconvert for ex: ro = %u, ex = %u\n",
	342	lockres->l_ro_holders, lockres->l_ex_holders);
	343	goto drop_ref;
	344	}
	345
	346	if ((lockres->l_blocking == LKM_PRMODE)
	347	&& lockres->l_ex_holders) {
	348	spin_unlock(&lockres->l_lock);
	349	mlog(0, "can't downconvert for pr: ex = %u\n",
	350	lockres->l_ex_holders);
	351	goto drop_ref;
	352	}
	353
	354	/* yay, we can downconvert now. */
	355	new_level = user_highest_compat_lock_level(lockres->l_blocking);
	356	lockres->l_requested = new_level;
	357	lockres->l_flags \|= USER_LOCK_BUSY;
	358	mlog(0, "Downconvert lock from %d to %d\n",
	359	lockres->l_level, new_level);
	360	spin_unlock(&lockres->l_lock);
	361
	362	/* need lock downconvert request now... */
	363	status = dlmlock(dlm,
	364	new_level,
	365	&lockres->l_lksb,
	366	LKM_CONVERT\|LKM_VALBLK,
	367	lockres->l_name,
ea5b3a18	368	lockres->l_namelen,
8df08c89 MF	369	user_ast,
	370	lockres,
	371	user_bast);
	372	if (status != DLM_NORMAL) {
	373	user_log_dlm_error("dlmlock", status, lockres);
	374	user_recover_from_dlm_error(lockres);
	375	}
	376
	377	drop_ref:
	378	user_dlm_drop_inode_ref(lockres);
	379	}
	380
	381	static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
	382	int level)
	383	{
	384	switch(level) {
	385	case LKM_EXMODE:
	386	lockres->l_ex_holders++;
	387	break;
	388	case LKM_PRMODE:
	389	lockres->l_ro_holders++;
	390	break;
	391	default:
	392	BUG();
	393	}
	394	}
	395
	396	/* predict what lock level we'll be dropping down to on behalf
	397	* of another node, and return true if the currently wanted
	398	* level will be compatible with it. */
	399	static inline int
	400	user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
	401	int wanted)
	402	{
	403	BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
	404
	405	return wanted <= user_highest_compat_lock_level(lockres->l_blocking);
	406	}
	407
	408	int user_dlm_cluster_lock(struct user_lock_res *lockres,
	409	int level,
	410	int lkm_flags)
	411	{
	412	int status, local_flags;
	413	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
	414
	415	if (level != LKM_EXMODE &&
	416	level != LKM_PRMODE) {
ea5b3a18 MF	417	mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
ea5b3a18 MF	418	lockres->l_namelen, lockres->l_name);
8df08c89 MF	419	status = -EINVAL;
	420	goto bail;
	421	}
	422
ea5b3a18 MF	423	mlog(0, "lockres %.*s: asking for %s lock, passed flags = 0x%x\n",
	424	lockres->l_namelen, lockres->l_name,
	425	(level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE",
	426	lkm_flags);
8df08c89 MF	427
	428	again:
	429	if (signal_pending(current)) {
	430	status = -ERESTARTSYS;
	431	goto bail;
	432	}
	433
	434	spin_lock(&lockres->l_lock);
	435
	436	/* We only compare against the currently granted level
	437	* here. If the lock is blocked waiting on a downconvert,
	438	* we'll get caught below. */
	439	if ((lockres->l_flags & USER_LOCK_BUSY) &&
	440	(level > lockres->l_level)) {
	441	/* is someone sitting in dlm_lock? If so, wait on
	442	* them. */
	443	spin_unlock(&lockres->l_lock);
	444
	445	user_wait_on_busy_lock(lockres);
	446	goto again;
	447	}
	448
	449	if ((lockres->l_flags & USER_LOCK_BLOCKED) &&
	450	(!user_may_continue_on_blocked_lock(lockres, level))) {
	451	/* is the lock is currently blocked on behalf of
	452	* another node */
	453	spin_unlock(&lockres->l_lock);
	454
	455	user_wait_on_blocked_lock(lockres);
	456	goto again;
	457	}
	458
	459	if (level > lockres->l_level) {
	460	local_flags = lkm_flags \| LKM_VALBLK;
	461	if (lockres->l_level != LKM_IVMODE)
	462	local_flags \|= LKM_CONVERT;
	463
	464	lockres->l_requested = level;
	465	lockres->l_flags \|= USER_LOCK_BUSY;
	466	spin_unlock(&lockres->l_lock);
	467
	468	BUG_ON(level == LKM_IVMODE);
	469	BUG_ON(level == LKM_NLMODE);
	470
8df08c89 MF	471	/* call dlm_lock to upgrade lock now */
	472	status = dlmlock(dlm,
	473	level,
	474	&lockres->l_lksb,
	475	local_flags,
	476	lockres->l_name,
ea5b3a18	477	lockres->l_namelen,
8df08c89 MF	478	user_ast,
	479	lockres,
	480	user_bast);
	481	if (status != DLM_NORMAL) {
	482	if ((lkm_flags & LKM_NOQUEUE) &&
	483	(status == DLM_NOTQUEUED))
	484	status = -EAGAIN;
	485	else {
	486	user_log_dlm_error("dlmlock", status, lockres);
	487	status = -EINVAL;
	488	}
	489	user_recover_from_dlm_error(lockres);
	490	goto bail;
	491	}
	492
8df08c89 MF	493	user_wait_on_busy_lock(lockres);
	494	goto again;
	495	}
	496
	497	user_dlm_inc_holders(lockres, level);
	498	spin_unlock(&lockres->l_lock);
	499
8df08c89 MF	500	status = 0;
	501	bail:
	502	return status;
	503	}
	504
	505	static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
	506	int level)
	507	{
	508	switch(level) {
	509	case LKM_EXMODE:
	510	BUG_ON(!lockres->l_ex_holders);
	511	lockres->l_ex_holders--;
	512	break;
	513	case LKM_PRMODE:
	514	BUG_ON(!lockres->l_ro_holders);
	515	lockres->l_ro_holders--;
	516	break;
	517	default:
	518	BUG();
	519	}
	520	}
	521
	522	void user_dlm_cluster_unlock(struct user_lock_res *lockres,
	523	int level)
	524	{
	525	if (level != LKM_EXMODE &&
	526	level != LKM_PRMODE) {
ea5b3a18 MF	527	mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
ea5b3a18 MF	528	lockres->l_namelen, lockres->l_name);
8df08c89 MF	529	return;
	530	}
	531
8df08c89 MF	532	spin_lock(&lockres->l_lock);
	533	user_dlm_dec_holders(lockres, level);
	534	__user_dlm_cond_queue_lockres(lockres);
	535	spin_unlock(&lockres->l_lock);
	536	}
	537
	538	void user_dlm_write_lvb(struct inode *inode,
	539	const char *val,
	540	unsigned int len)
	541	{
	542	struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
	543	char *lvb = lockres->l_lksb.lvb;
	544
	545	BUG_ON(len > DLM_LVB_LEN);
	546
	547	spin_lock(&lockres->l_lock);
	548
	549	BUG_ON(lockres->l_level < LKM_EXMODE);
	550	memcpy(lvb, val, len);
	551
	552	spin_unlock(&lockres->l_lock);
	553	}
	554
	555	void user_dlm_read_lvb(struct inode *inode,
	556	char *val,
	557	unsigned int len)
	558	{
	559	struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
	560	char *lvb = lockres->l_lksb.lvb;
	561
	562	BUG_ON(len > DLM_LVB_LEN);
	563
	564	spin_lock(&lockres->l_lock);
	565
	566	BUG_ON(lockres->l_level < LKM_PRMODE);
	567	memcpy(val, lvb, len);
	568
	569	spin_unlock(&lockres->l_lock);
	570	}
	571
	572	void user_dlm_lock_res_init(struct user_lock_res *lockres,
	573	struct dentry *dentry)
	574	{
	575	memset(lockres, 0, sizeof(*lockres));
	576
	577	spin_lock_init(&lockres->l_lock);
	578	init_waitqueue_head(&lockres->l_event);
	579	lockres->l_level = LKM_IVMODE;
	580	lockres->l_requested = LKM_IVMODE;
	581	lockres->l_blocking = LKM_IVMODE;
	582
	583	/* should have been checked before getting here. */
	584	BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);
	585
	586	memcpy(lockres->l_name,
	587	dentry->d_name.name,
	588	dentry->d_name.len);
ea5b3a18	589	lockres->l_namelen = dentry->d_name.len;
8df08c89 MF	590	}
	591
	592	int user_dlm_destroy_lock(struct user_lock_res *lockres)
	593	{
	594	int status = -EBUSY;
	595	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
	596
ea5b3a18	597	mlog(0, "asked to destroy %.*s\n", lockres->l_namelen, lockres->l_name);
8df08c89 MF	598
8df08c89 MF	599	spin_lock(&lockres->l_lock);
2cd98885	600	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
2cd98885 MF	601	spin_unlock(&lockres->l_lock);
	602	return 0;
	603	}
	604
	605	lockres->l_flags \|= USER_LOCK_IN_TEARDOWN;
	606
8df08c89 MF	607	while (lockres->l_flags & USER_LOCK_BUSY) {
	608	spin_unlock(&lockres->l_lock);
	609
8df08c89 MF	610	user_wait_on_busy_lock(lockres);
	611
	612	spin_lock(&lockres->l_lock);
	613	}
	614
	615	if (lockres->l_ro_holders \|\| lockres->l_ex_holders) {
	616	spin_unlock(&lockres->l_lock);
8df08c89 MF	617	goto bail;
	618	}
	619
	620	status = 0;
	621	if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
	622	spin_unlock(&lockres->l_lock);
8df08c89 MF	623	goto bail;
	624	}
	625
	626	lockres->l_flags &= ~USER_LOCK_ATTACHED;
	627	lockres->l_flags \|= USER_LOCK_BUSY;
8df08c89 MF	628	spin_unlock(&lockres->l_lock);
8df08c89 MF	629
8df08c89 MF	630	status = dlmunlock(dlm,
	631	&lockres->l_lksb,
	632	LKM_VALBLK,
	633	user_unlock_ast,
	634	lockres);
	635	if (status != DLM_NORMAL) {
	636	user_log_dlm_error("dlmunlock", status, lockres);
	637	status = -EINVAL;
	638	goto bail;
	639	}
	640
	641	user_wait_on_busy_lock(lockres);
	642
	643	status = 0;
	644	bail:
	645	return status;
	646	}
	647
d24fbcda JB	648	struct dlm_ctxt user_dlm_register_context(struct qstr name,
d24fbcda JB	649	struct dlm_protocol_version *proto)
8df08c89 MF	650	{
	651	struct dlm_ctxt *dlm;
	652	u32 dlm_key;
	653	char *domain;
	654
ad8100e0	655	domain = kmalloc(name->len + 1, GFP_NOFS);
8df08c89 MF	656	if (!domain) {
	657	mlog_errno(-ENOMEM);
	658	return ERR_PTR(-ENOMEM);
	659	}
	660
	661	dlm_key = crc32_le(0, name->name, name->len);
	662
	663	snprintf(domain, name->len + 1, "%.*s", name->len, name->name);
	664
d24fbcda	665	dlm = dlm_register_domain(domain, dlm_key, proto);
8df08c89 MF	666	if (IS_ERR(dlm))
	667	mlog_errno(PTR_ERR(dlm));
	668
	669	kfree(domain);
	670	return dlm;
	671	}
	672
	673	void user_dlm_unregister_context(struct dlm_ctxt *dlm)
	674	{
	675	dlm_unregister_domain(dlm);
	676	}