linux/fs/gfs2/locking/dlm/lock.c
Steven Whitehouse 41d7db0ab4 [GFS2] Reduce size of struct gdlm_lock
This patch removes the completion (which is rather large) from struct
gdlm_lock in favour of using the wait_on_bit() functions. We don't need
to add any extra fields to the structure to do this, so we save 32 bytes
(on x86_64) per structure. This adds up to quite a lot when we may
potentially have millions of these lock structures,

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Acked-by: David Teigland <teigland@redhat.com>
2007-07-09 08:22:21 +01:00

528 lines
12 KiB
C

/*
* Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
* Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
*
* This copyrighted material is made available to anyone wishing to use,
* modify, copy, or redistribute it subject to the terms and conditions
* of the GNU General Public License version 2.
*/
#include "lock_dlm.h"
static char junk_lvb[GDLM_LVB_SIZE];
static void queue_complete(struct gdlm_lock *lp)
{
struct gdlm_ls *ls = lp->ls;
clear_bit(LFL_ACTIVE, &lp->flags);
spin_lock(&ls->async_lock);
list_add_tail(&lp->clist, &ls->complete);
spin_unlock(&ls->async_lock);
wake_up(&ls->thread_wait);
}
static inline void gdlm_ast(void *astarg)
{
queue_complete(astarg);
}
static inline void gdlm_bast(void *astarg, int mode)
{
struct gdlm_lock *lp = astarg;
struct gdlm_ls *ls = lp->ls;
if (!mode) {
printk(KERN_INFO "lock_dlm: bast mode zero %x,%llx\n",
lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number);
return;
}
spin_lock(&ls->async_lock);
if (!lp->bast_mode) {
list_add_tail(&lp->blist, &ls->blocking);
lp->bast_mode = mode;
} else if (lp->bast_mode < mode)
lp->bast_mode = mode;
spin_unlock(&ls->async_lock);
wake_up(&ls->thread_wait);
}
void gdlm_queue_delayed(struct gdlm_lock *lp)
{
struct gdlm_ls *ls = lp->ls;
spin_lock(&ls->async_lock);
list_add_tail(&lp->delay_list, &ls->delayed);
spin_unlock(&ls->async_lock);
}
/* convert gfs lock-state to dlm lock-mode */
static s16 make_mode(s16 lmstate)
{
switch (lmstate) {
case LM_ST_UNLOCKED:
return DLM_LOCK_NL;
case LM_ST_EXCLUSIVE:
return DLM_LOCK_EX;
case LM_ST_DEFERRED:
return DLM_LOCK_CW;
case LM_ST_SHARED:
return DLM_LOCK_PR;
}
gdlm_assert(0, "unknown LM state %d", lmstate);
return -1;
}
/* convert dlm lock-mode to gfs lock-state */
s16 gdlm_make_lmstate(s16 dlmmode)
{
switch (dlmmode) {
case DLM_LOCK_IV:
case DLM_LOCK_NL:
return LM_ST_UNLOCKED;
case DLM_LOCK_EX:
return LM_ST_EXCLUSIVE;
case DLM_LOCK_CW:
return LM_ST_DEFERRED;
case DLM_LOCK_PR:
return LM_ST_SHARED;
}
gdlm_assert(0, "unknown DLM mode %d", dlmmode);
return -1;
}
/* verify agreement with GFS on the current lock state, NB: DLM_LOCK_NL and
DLM_LOCK_IV are both considered LM_ST_UNLOCKED by GFS. */
static void check_cur_state(struct gdlm_lock *lp, unsigned int cur_state)
{
s16 cur = make_mode(cur_state);
if (lp->cur != DLM_LOCK_IV)
gdlm_assert(lp->cur == cur, "%d, %d", lp->cur, cur);
}
static inline unsigned int make_flags(struct gdlm_lock *lp,
unsigned int gfs_flags,
s16 cur, s16 req)
{
unsigned int lkf = 0;
if (gfs_flags & LM_FLAG_TRY)
lkf |= DLM_LKF_NOQUEUE;
if (gfs_flags & LM_FLAG_TRY_1CB) {
lkf |= DLM_LKF_NOQUEUE;
lkf |= DLM_LKF_NOQUEUEBAST;
}
if (gfs_flags & LM_FLAG_PRIORITY) {
lkf |= DLM_LKF_NOORDER;
lkf |= DLM_LKF_HEADQUE;
}
if (gfs_flags & LM_FLAG_ANY) {
if (req == DLM_LOCK_PR)
lkf |= DLM_LKF_ALTCW;
else if (req == DLM_LOCK_CW)
lkf |= DLM_LKF_ALTPR;
}
if (lp->lksb.sb_lkid != 0) {
lkf |= DLM_LKF_CONVERT;
/* Conversion deadlock avoidance by DLM */
if (!test_bit(LFL_FORCE_PROMOTE, &lp->flags) &&
!(lkf & DLM_LKF_NOQUEUE) &&
cur > DLM_LOCK_NL && req > DLM_LOCK_NL && cur != req)
lkf |= DLM_LKF_CONVDEADLK;
}
if (lp->lvb)
lkf |= DLM_LKF_VALBLK;
return lkf;
}
/* make_strname - convert GFS lock numbers to a string */
static inline void make_strname(const struct lm_lockname *lockname,
struct gdlm_strname *str)
{
sprintf(str->name, "%8x%16llx", lockname->ln_type,
(unsigned long long)lockname->ln_number);
str->namelen = GDLM_STRNAME_BYTES;
}
static int gdlm_create_lp(struct gdlm_ls *ls, struct lm_lockname *name,
struct gdlm_lock **lpp)
{
struct gdlm_lock *lp;
lp = kzalloc(sizeof(struct gdlm_lock), GFP_KERNEL);
if (!lp)
return -ENOMEM;
lp->lockname = *name;
make_strname(name, &lp->strname);
lp->ls = ls;
lp->cur = DLM_LOCK_IV;
lp->lvb = NULL;
lp->hold_null = NULL;
INIT_LIST_HEAD(&lp->clist);
INIT_LIST_HEAD(&lp->blist);
INIT_LIST_HEAD(&lp->delay_list);
spin_lock(&ls->async_lock);
list_add(&lp->all_list, &ls->all_locks);
ls->all_locks_count++;
spin_unlock(&ls->async_lock);
*lpp = lp;
return 0;
}
void gdlm_delete_lp(struct gdlm_lock *lp)
{
struct gdlm_ls *ls = lp->ls;
spin_lock(&ls->async_lock);
if (!list_empty(&lp->clist))
list_del_init(&lp->clist);
if (!list_empty(&lp->blist))
list_del_init(&lp->blist);
if (!list_empty(&lp->delay_list))
list_del_init(&lp->delay_list);
gdlm_assert(!list_empty(&lp->all_list), "%x,%llx", lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number);
list_del_init(&lp->all_list);
ls->all_locks_count--;
spin_unlock(&ls->async_lock);
kfree(lp);
}
int gdlm_get_lock(void *lockspace, struct lm_lockname *name,
void **lockp)
{
struct gdlm_lock *lp;
int error;
error = gdlm_create_lp(lockspace, name, &lp);
*lockp = lp;
return error;
}
void gdlm_put_lock(void *lock)
{
gdlm_delete_lp(lock);
}
unsigned int gdlm_do_lock(struct gdlm_lock *lp)
{
struct gdlm_ls *ls = lp->ls;
int error, bast = 1;
/*
* When recovery is in progress, delay lock requests for submission
* once recovery is done. Requests for recovery (NOEXP) and unlocks
* can pass.
*/
if (test_bit(DFL_BLOCK_LOCKS, &ls->flags) &&
!test_bit(LFL_NOBLOCK, &lp->flags) && lp->req != DLM_LOCK_NL) {
gdlm_queue_delayed(lp);
return LM_OUT_ASYNC;
}
/*
* Submit the actual lock request.
*/
if (test_bit(LFL_NOBAST, &lp->flags))
bast = 0;
set_bit(LFL_ACTIVE, &lp->flags);
log_debug("lk %x,%llx id %x %d,%d %x", lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number, lp->lksb.sb_lkid,
lp->cur, lp->req, lp->lkf);
error = dlm_lock(ls->dlm_lockspace, lp->req, &lp->lksb, lp->lkf,
lp->strname.name, lp->strname.namelen, 0, gdlm_ast,
lp, bast ? gdlm_bast : NULL);
if ((error == -EAGAIN) && (lp->lkf & DLM_LKF_NOQUEUE)) {
lp->lksb.sb_status = -EAGAIN;
queue_complete(lp);
error = 0;
}
if (error) {
log_error("%s: gdlm_lock %x,%llx err=%d cur=%d req=%d lkf=%x "
"flags=%lx", ls->fsname, lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number, error,
lp->cur, lp->req, lp->lkf, lp->flags);
return LM_OUT_ERROR;
}
return LM_OUT_ASYNC;
}
static unsigned int gdlm_do_unlock(struct gdlm_lock *lp)
{
struct gdlm_ls *ls = lp->ls;
unsigned int lkf = 0;
int error;
set_bit(LFL_DLM_UNLOCK, &lp->flags);
set_bit(LFL_ACTIVE, &lp->flags);
if (lp->lvb)
lkf = DLM_LKF_VALBLK;
log_debug("un %x,%llx %x %d %x", lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number,
lp->lksb.sb_lkid, lp->cur, lkf);
error = dlm_unlock(ls->dlm_lockspace, lp->lksb.sb_lkid, lkf, NULL, lp);
if (error) {
log_error("%s: gdlm_unlock %x,%llx err=%d cur=%d req=%d lkf=%x "
"flags=%lx", ls->fsname, lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number, error,
lp->cur, lp->req, lp->lkf, lp->flags);
return LM_OUT_ERROR;
}
return LM_OUT_ASYNC;
}
unsigned int gdlm_lock(void *lock, unsigned int cur_state,
unsigned int req_state, unsigned int flags)
{
struct gdlm_lock *lp = lock;
clear_bit(LFL_DLM_CANCEL, &lp->flags);
if (flags & LM_FLAG_NOEXP)
set_bit(LFL_NOBLOCK, &lp->flags);
check_cur_state(lp, cur_state);
lp->req = make_mode(req_state);
lp->lkf = make_flags(lp, flags, lp->cur, lp->req);
return gdlm_do_lock(lp);
}
unsigned int gdlm_unlock(void *lock, unsigned int cur_state)
{
struct gdlm_lock *lp = lock;
clear_bit(LFL_DLM_CANCEL, &lp->flags);
if (lp->cur == DLM_LOCK_IV)
return 0;
return gdlm_do_unlock(lp);
}
void gdlm_cancel(void *lock)
{
struct gdlm_lock *lp = lock;
struct gdlm_ls *ls = lp->ls;
int error, delay_list = 0;
if (test_bit(LFL_DLM_CANCEL, &lp->flags))
return;
log_info("gdlm_cancel %x,%llx flags %lx", lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number, lp->flags);
spin_lock(&ls->async_lock);
if (!list_empty(&lp->delay_list)) {
list_del_init(&lp->delay_list);
delay_list = 1;
}
spin_unlock(&ls->async_lock);
if (delay_list) {
set_bit(LFL_CANCEL, &lp->flags);
set_bit(LFL_ACTIVE, &lp->flags);
queue_complete(lp);
return;
}
if (!test_bit(LFL_ACTIVE, &lp->flags) ||
test_bit(LFL_DLM_UNLOCK, &lp->flags)) {
log_info("gdlm_cancel skip %x,%llx flags %lx",
lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number, lp->flags);
return;
}
/* the lock is blocked in the dlm */
set_bit(LFL_DLM_CANCEL, &lp->flags);
set_bit(LFL_ACTIVE, &lp->flags);
error = dlm_unlock(ls->dlm_lockspace, lp->lksb.sb_lkid, DLM_LKF_CANCEL,
NULL, lp);
log_info("gdlm_cancel rv %d %x,%llx flags %lx", error,
lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number, lp->flags);
if (error == -EBUSY)
clear_bit(LFL_DLM_CANCEL, &lp->flags);
}
static int gdlm_add_lvb(struct gdlm_lock *lp)
{
char *lvb;
lvb = kzalloc(GDLM_LVB_SIZE, GFP_KERNEL);
if (!lvb)
return -ENOMEM;
lp->lksb.sb_lvbptr = lvb;
lp->lvb = lvb;
return 0;
}
static void gdlm_del_lvb(struct gdlm_lock *lp)
{
kfree(lp->lvb);
lp->lvb = NULL;
lp->lksb.sb_lvbptr = NULL;
}
static int gdlm_ast_wait(void *word)
{
schedule();
return 0;
}
/* This can do a synchronous dlm request (requiring a lock_dlm thread to get
the completion) because gfs won't call hold_lvb() during a callback (from
the context of a lock_dlm thread). */
static int hold_null_lock(struct gdlm_lock *lp)
{
struct gdlm_lock *lpn = NULL;
int error;
if (lp->hold_null) {
printk(KERN_INFO "lock_dlm: lvb already held\n");
return 0;
}
error = gdlm_create_lp(lp->ls, &lp->lockname, &lpn);
if (error)
goto out;
lpn->lksb.sb_lvbptr = junk_lvb;
lpn->lvb = junk_lvb;
lpn->req = DLM_LOCK_NL;
lpn->lkf = DLM_LKF_VALBLK | DLM_LKF_EXPEDITE;
set_bit(LFL_NOBAST, &lpn->flags);
set_bit(LFL_INLOCK, &lpn->flags);
set_bit(LFL_AST_WAIT, &lpn->flags);
gdlm_do_lock(lpn);
wait_on_bit(&lpn->flags, LFL_AST_WAIT, gdlm_ast_wait, TASK_UNINTERRUPTIBLE);
error = lpn->lksb.sb_status;
if (error) {
printk(KERN_INFO "lock_dlm: hold_null_lock dlm error %d\n",
error);
gdlm_delete_lp(lpn);
lpn = NULL;
}
out:
lp->hold_null = lpn;
return error;
}
/* This cannot do a synchronous dlm request (requiring a lock_dlm thread to get
the completion) because gfs may call unhold_lvb() during a callback (from
the context of a lock_dlm thread) which could cause a deadlock since the
other lock_dlm thread could be engaged in recovery. */
static void unhold_null_lock(struct gdlm_lock *lp)
{
struct gdlm_lock *lpn = lp->hold_null;
gdlm_assert(lpn, "%x,%llx", lp->lockname.ln_type,
(unsigned long long)lp->lockname.ln_number);
lpn->lksb.sb_lvbptr = NULL;
lpn->lvb = NULL;
set_bit(LFL_UNLOCK_DELETE, &lpn->flags);
gdlm_do_unlock(lpn);
lp->hold_null = NULL;
}
/* Acquire a NL lock because gfs requires the value block to remain
intact on the resource while the lvb is "held" even if it's holding no locks
on the resource. */
int gdlm_hold_lvb(void *lock, char **lvbp)
{
struct gdlm_lock *lp = lock;
int error;
error = gdlm_add_lvb(lp);
if (error)
return error;
*lvbp = lp->lvb;
error = hold_null_lock(lp);
if (error)
gdlm_del_lvb(lp);
return error;
}
void gdlm_unhold_lvb(void *lock, char *lvb)
{
struct gdlm_lock *lp = lock;
unhold_null_lock(lp);
gdlm_del_lvb(lp);
}
void gdlm_submit_delayed(struct gdlm_ls *ls)
{
struct gdlm_lock *lp, *safe;
spin_lock(&ls->async_lock);
list_for_each_entry_safe(lp, safe, &ls->delayed, delay_list) {
list_del_init(&lp->delay_list);
list_add_tail(&lp->delay_list, &ls->submit);
}
spin_unlock(&ls->async_lock);
wake_up(&ls->thread_wait);
}
int gdlm_release_all_locks(struct gdlm_ls *ls)
{
struct gdlm_lock *lp, *safe;
int count = 0;
spin_lock(&ls->async_lock);
list_for_each_entry_safe(lp, safe, &ls->all_locks, all_list) {
list_del_init(&lp->all_list);
if (lp->lvb && lp->lvb != junk_lvb)
kfree(lp->lvb);
kfree(lp);
count++;
}
spin_unlock(&ls->async_lock);
return count;
}