mariadb/storage/myisam/mi_locking.c
Davi Arnaut a2e4449bfc Bug#41098: Query Cache returns wrong result with concurrent insert
The problem is that select queries executed concurrently with
a concurrent insert on a MyISAM table could be cached if the
select started after the query cache invalidation but before
the unlock of tables performed by the concurrent insert. This
race could happen because the concurrent insert was failing
to prevent cache of select queries happening at the same time.

The solution is to add a 'uncacheable' status flag to signal
that a concurrent insert is being performed on the table and
that queries executing at the same time shouldn't cache the
results.
2009-02-19 18:09:35 -03:00

579 lines
17 KiB
C

/* Copyright (C) 2000-2006 MySQL AB
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; version 2 of the License.
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 02111-1307 USA */
/*
locking of isam-tables.
reads info from a isam-table. Must be first request before doing any furter
calls to any isamfunktion. Is used to allow many process use the same
isamdatabase.
*/
#include "ftdefs.h"
/* lock table by F_UNLCK, F_RDLCK or F_WRLCK */
int mi_lock_database(MI_INFO *info, int lock_type)
{
int error;
uint count;
MYISAM_SHARE *share=info->s;
uint flag;
DBUG_ENTER("mi_lock_database");
DBUG_PRINT("enter",("lock_type: %d old lock %d r_locks: %u w_locks: %u "
"global_changed: %d open_count: %u name: '%s'",
lock_type, info->lock_type, share->r_locks,
share->w_locks,
share->global_changed, share->state.open_count,
share->index_file_name));
if (share->options & HA_OPTION_READ_ONLY_DATA ||
info->lock_type == lock_type)
DBUG_RETURN(0);
if (lock_type == F_EXTRA_LCK) /* Used by TMP tables */
{
++share->w_locks;
++share->tot_locks;
info->lock_type= lock_type;
DBUG_RETURN(0);
}
flag=error=0;
pthread_mutex_lock(&share->intern_lock);
if (share->kfile >= 0) /* May only be false on windows */
{
switch (lock_type) {
case F_UNLCK:
ftparser_call_deinitializer(info);
if (info->lock_type == F_RDLCK)
count= --share->r_locks;
else
count= --share->w_locks;
--share->tot_locks;
if (info->lock_type == F_WRLCK && !share->w_locks &&
!share->delay_key_write && flush_key_blocks(share->key_cache,
share->kfile,FLUSH_KEEP))
{
error=my_errno;
mi_print_error(info->s, HA_ERR_CRASHED);
mi_mark_crashed(info); /* Mark that table must be checked */
}
if (info->opt_flag & (READ_CACHE_USED | WRITE_CACHE_USED))
{
if (end_io_cache(&info->rec_cache))
{
error=my_errno;
mi_print_error(info->s, HA_ERR_CRASHED);
mi_mark_crashed(info);
}
}
if (!count)
{
DBUG_PRINT("info",("changed: %u w_locks: %u",
(uint) share->changed, share->w_locks));
if (share->changed && !share->w_locks)
{
#ifdef HAVE_MMAP
if ((info->s->mmaped_length != info->s->state.state.data_file_length) &&
(info->s->nonmmaped_inserts > MAX_NONMAPPED_INSERTS))
{
if (info->s->concurrent_insert)
rw_wrlock(&info->s->mmap_lock);
mi_remap_file(info, info->s->state.state.data_file_length);
info->s->nonmmaped_inserts= 0;
if (info->s->concurrent_insert)
rw_unlock(&info->s->mmap_lock);
}
#endif
share->state.process= share->last_process=share->this_process;
share->state.unique= info->last_unique= info->this_unique;
share->state.update_count= info->last_loop= ++info->this_loop;
if (mi_state_info_write(share->kfile, &share->state, 1))
error=my_errno;
share->changed=0;
if (myisam_flush)
{
if (my_sync(share->kfile, MYF(0)))
error= my_errno;
if (my_sync(info->dfile, MYF(0)))
error= my_errno;
}
else
share->not_flushed=1;
if (error)
{
mi_print_error(info->s, HA_ERR_CRASHED);
mi_mark_crashed(info);
}
}
if (info->lock_type != F_EXTRA_LCK)
{
if (share->r_locks)
{ /* Only read locks left */
flag=1;
if (my_lock(share->kfile,F_RDLCK,0L,F_TO_EOF,
MYF(MY_WME | MY_SEEK_NOT_DONE)) && !error)
error=my_errno;
}
else if (!share->w_locks)
{ /* No more locks */
flag=1;
if (my_lock(share->kfile,F_UNLCK,0L,F_TO_EOF,
MYF(MY_WME | MY_SEEK_NOT_DONE)) && !error)
error=my_errno;
}
}
}
info->opt_flag&= ~(READ_CACHE_USED | WRITE_CACHE_USED);
info->lock_type= F_UNLCK;
break;
case F_RDLCK:
if (info->lock_type == F_WRLCK)
{
/*
Change RW to READONLY
mysqld does not turn write locks to read locks,
so we're never here in mysqld.
*/
if (share->w_locks == 1)
{
flag=1;
if (my_lock(share->kfile,lock_type,0L,F_TO_EOF,
MYF(MY_SEEK_NOT_DONE)))
{
error=my_errno;
break;
}
}
share->w_locks--;
share->r_locks++;
info->lock_type=lock_type;
break;
}
if (!share->r_locks && !share->w_locks)
{
flag=1;
if (my_lock(share->kfile,lock_type,0L,F_TO_EOF,
info->lock_wait | MY_SEEK_NOT_DONE))
{
error=my_errno;
break;
}
if (mi_state_info_read_dsk(share->kfile, &share->state, 1))
{
error=my_errno;
VOID(my_lock(share->kfile,F_UNLCK,0L,F_TO_EOF,MYF(MY_SEEK_NOT_DONE)));
my_errno=error;
break;
}
}
VOID(_mi_test_if_changed(info));
share->r_locks++;
share->tot_locks++;
info->lock_type=lock_type;
break;
case F_WRLCK:
if (info->lock_type == F_RDLCK)
{ /* Change READONLY to RW */
if (share->r_locks == 1)
{
flag=1;
if (my_lock(share->kfile,lock_type,0L,F_TO_EOF,
MYF(info->lock_wait | MY_SEEK_NOT_DONE)))
{
error=my_errno;
break;
}
share->r_locks--;
share->w_locks++;
info->lock_type=lock_type;
break;
}
}
if (!(share->options & HA_OPTION_READ_ONLY_DATA))
{
if (!share->w_locks)
{
flag=1;
if (my_lock(share->kfile,lock_type,0L,F_TO_EOF,
info->lock_wait | MY_SEEK_NOT_DONE))
{
error=my_errno;
break;
}
if (!share->r_locks)
{
if (mi_state_info_read_dsk(share->kfile, &share->state, 1))
{
error=my_errno;
VOID(my_lock(share->kfile,F_UNLCK,0L,F_TO_EOF,
info->lock_wait | MY_SEEK_NOT_DONE));
my_errno=error;
break;
}
}
}
}
VOID(_mi_test_if_changed(info));
info->lock_type=lock_type;
info->invalidator=info->s->invalidator;
share->w_locks++;
share->tot_locks++;
break;
default:
break; /* Impossible */
}
}
#ifdef __WIN__
else
{
/*
Check for bad file descriptors if this table is part
of a merge union. Failing to capture this may cause
a crash on windows if the table is renamed and
later on referenced by the merge table.
*/
if( info->owned_by_merge && (info->s)->kfile < 0 )
{
error = HA_ERR_NO_SUCH_TABLE;
}
}
#endif
pthread_mutex_unlock(&share->intern_lock);
#if defined(FULL_LOG) || defined(_lint)
lock_type|=(int) (flag << 8); /* Set bit to set if real lock */
myisam_log_command(MI_LOG_LOCK,info,(uchar*) &lock_type,sizeof(lock_type),
error);
#endif
DBUG_RETURN(error);
} /* mi_lock_database */
/****************************************************************************
The following functions are called by thr_lock() in threaded applications
****************************************************************************/
/*
Create a copy of the current status for the table
SYNOPSIS
mi_get_status()
param Pointer to Myisam handler
concurrent_insert Set to 1 if we are going to do concurrent inserts
(THR_WRITE_CONCURRENT_INSERT was used)
*/
void mi_get_status(void* param, int concurrent_insert)
{
MI_INFO *info=(MI_INFO*) param;
DBUG_ENTER("mi_get_status");
DBUG_PRINT("info",("key_file: %ld data_file: %ld concurrent_insert: %d",
(long) info->s->state.state.key_file_length,
(long) info->s->state.state.data_file_length,
concurrent_insert));
#ifndef DBUG_OFF
if (info->state->key_file_length > info->s->state.state.key_file_length ||
info->state->data_file_length > info->s->state.state.data_file_length)
DBUG_PRINT("warning",("old info: key_file: %ld data_file: %ld",
(long) info->state->key_file_length,
(long) info->state->data_file_length));
#endif
info->save_state=info->s->state.state;
info->state= &info->save_state;
info->append_insert_at_end= concurrent_insert;
if (concurrent_insert)
info->s->state.state.uncacheable= TRUE;
DBUG_VOID_RETURN;
}
void mi_update_status(void* param)
{
MI_INFO *info=(MI_INFO*) param;
/*
Because someone may have closed the table we point at, we only
update the state if its our own state. This isn't a problem as
we are always pointing at our own lock or at a read lock.
(This is enforced by thr_multi_lock.c)
*/
if (info->state == &info->save_state)
{
#ifndef DBUG_OFF
DBUG_PRINT("info",("updating status: key_file: %ld data_file: %ld",
(long) info->state->key_file_length,
(long) info->state->data_file_length));
if (info->state->key_file_length < info->s->state.state.key_file_length ||
info->state->data_file_length < info->s->state.state.data_file_length)
DBUG_PRINT("warning",("old info: key_file: %ld data_file: %ld",
(long) info->s->state.state.key_file_length,
(long) info->s->state.state.data_file_length));
#endif
info->s->state.state= *info->state;
info->state= &info->s->state.state;
}
info->append_insert_at_end= 0;
/*
We have to flush the write cache here as other threads may start
reading the table before mi_lock_database() is called
*/
if (info->opt_flag & WRITE_CACHE_USED)
{
if (end_io_cache(&info->rec_cache))
{
mi_print_error(info->s, HA_ERR_CRASHED);
mi_mark_crashed(info);
}
info->opt_flag&= ~WRITE_CACHE_USED;
}
}
void mi_restore_status(void *param)
{
MI_INFO *info= (MI_INFO*) param;
info->state= &info->s->state.state;
info->append_insert_at_end= 0;
}
void mi_copy_status(void* to,void *from)
{
((MI_INFO*) to)->state= &((MI_INFO*) from)->save_state;
}
/*
Check if should allow concurrent inserts
IMPLEMENTATION
Allow concurrent inserts if we don't have a hole in the table or
if there is no active write lock and there is active read locks and
myisam_concurrent_insert == 2. In this last case the new
row('s) are inserted at end of file instead of filling up the hole.
The last case is to allow one to inserts into a heavily read-used table
even if there is holes.
NOTES
If there is a an rtree indexes in the table, concurrent inserts are
disabled in mi_open()
RETURN
0 ok to use concurrent inserts
1 not ok
*/
my_bool mi_check_status(void *param)
{
MI_INFO *info=(MI_INFO*) param;
/*
The test for w_locks == 1 is here because this thread has already done an
external lock (in other words: w_locks == 1 means no other threads has
a write lock)
*/
DBUG_PRINT("info",("dellink: %ld r_locks: %u w_locks: %u",
(long) info->s->state.dellink, (uint) info->s->r_locks,
(uint) info->s->w_locks));
return (my_bool) !(info->s->state.dellink == HA_OFFSET_ERROR ||
(myisam_concurrent_insert == 2 && info->s->r_locks &&
info->s->w_locks == 1));
}
/****************************************************************************
** functions to read / write the state
****************************************************************************/
int _mi_readinfo(register MI_INFO *info, int lock_type, int check_keybuffer)
{
DBUG_ENTER("_mi_readinfo");
if (info->lock_type == F_UNLCK)
{
MYISAM_SHARE *share=info->s;
if (!share->tot_locks)
{
if (my_lock(share->kfile,lock_type,0L,F_TO_EOF,
info->lock_wait | MY_SEEK_NOT_DONE))
DBUG_RETURN(1);
if (mi_state_info_read_dsk(share->kfile, &share->state, 1))
{
int error=my_errno ? my_errno : -1;
VOID(my_lock(share->kfile,F_UNLCK,0L,F_TO_EOF,
MYF(MY_SEEK_NOT_DONE)));
my_errno=error;
DBUG_RETURN(1);
}
}
if (check_keybuffer)
VOID(_mi_test_if_changed(info));
info->invalidator=info->s->invalidator;
}
else if (lock_type == F_WRLCK && info->lock_type == F_RDLCK)
{
my_errno=EACCES; /* Not allowed to change */
DBUG_RETURN(-1); /* when have read_lock() */
}
DBUG_RETURN(0);
} /* _mi_readinfo */
/*
Every isam-function that uppdates the isam-database MUST end with this
request
*/
int _mi_writeinfo(register MI_INFO *info, uint operation)
{
int error,olderror;
MYISAM_SHARE *share=info->s;
DBUG_ENTER("_mi_writeinfo");
DBUG_PRINT("info",("operation: %u tot_locks: %u", operation,
share->tot_locks));
error=0;
if (share->tot_locks == 0)
{
olderror=my_errno; /* Remember last error */
if (operation)
{ /* Two threads can't be here */
share->state.process= share->last_process= share->this_process;
share->state.unique= info->last_unique= info->this_unique;
share->state.update_count= info->last_loop= ++info->this_loop;
if ((error=mi_state_info_write(share->kfile, &share->state, 1)))
olderror=my_errno;
#ifdef __WIN__
if (myisam_flush)
{
_commit(share->kfile);
_commit(info->dfile);
}
#endif
}
if (!(operation & WRITEINFO_NO_UNLOCK) &&
my_lock(share->kfile,F_UNLCK,0L,F_TO_EOF,
MYF(MY_WME | MY_SEEK_NOT_DONE)) && !error)
DBUG_RETURN(1);
my_errno=olderror;
}
else if (operation)
share->changed= 1; /* Mark keyfile changed */
DBUG_RETURN(error);
} /* _mi_writeinfo */
/* Test if someone has changed the database */
/* (Should be called after readinfo) */
int _mi_test_if_changed(register MI_INFO *info)
{
MYISAM_SHARE *share=info->s;
if (share->state.process != share->last_process ||
share->state.unique != info->last_unique ||
share->state.update_count != info->last_loop)
{ /* Keyfile has changed */
DBUG_PRINT("info",("index file changed"));
if (share->state.process != share->this_process)
VOID(flush_key_blocks(share->key_cache, share->kfile, FLUSH_RELEASE));
share->last_process=share->state.process;
info->last_unique= share->state.unique;
info->last_loop= share->state.update_count;
info->update|= HA_STATE_WRITTEN; /* Must use file on next */
info->data_changed= 1; /* For mi_is_changed */
return 1;
}
return (!(info->update & HA_STATE_AKTIV) ||
(info->update & (HA_STATE_WRITTEN | HA_STATE_DELETED |
HA_STATE_KEY_CHANGED)));
} /* _mi_test_if_changed */
/*
Put a mark in the .MYI file that someone is updating the table
DOCUMENTATION
state.open_count in the .MYI file is used the following way:
- For the first change of the .MYI file in this process open_count is
incremented by mi_mark_file_change(). (We have a write lock on the file
when this happens)
- In mi_close() it's decremented by _mi_decrement_open_count() if it
was incremented in the same process.
This mean that if we are the only process using the file, the open_count
tells us if the MYISAM file wasn't properly closed. (This is true if
my_disable_locking is set).
*/
int _mi_mark_file_changed(MI_INFO *info)
{
uchar buff[3];
register MYISAM_SHARE *share=info->s;
DBUG_ENTER("_mi_mark_file_changed");
if (!(share->state.changed & STATE_CHANGED) || ! share->global_changed)
{
share->state.changed|=(STATE_CHANGED | STATE_NOT_ANALYZED |
STATE_NOT_OPTIMIZED_KEYS);
if (!share->global_changed)
{
share->global_changed=1;
share->state.open_count++;
}
if (!share->temporary)
{
mi_int2store(buff,share->state.open_count);
buff[2]=1; /* Mark that it's changed */
DBUG_RETURN(my_pwrite(share->kfile,buff,sizeof(buff),
sizeof(share->state.header),
MYF(MY_NABP)));
}
}
DBUG_RETURN(0);
}
/*
This is only called by close or by extra(HA_FLUSH) if the OS has the pwrite()
call. In these context the following code should be safe!
*/
int _mi_decrement_open_count(MI_INFO *info)
{
uchar buff[2];
register MYISAM_SHARE *share=info->s;
int lock_error=0,write_error=0;
if (share->global_changed)
{
uint old_lock=info->lock_type;
share->global_changed=0;
lock_error=mi_lock_database(info,F_WRLCK);
/* Its not fatal even if we couldn't get the lock ! */
if (share->state.open_count > 0)
{
share->state.open_count--;
mi_int2store(buff,share->state.open_count);
write_error=my_pwrite(share->kfile,buff,sizeof(buff),
sizeof(share->state.header),
MYF(MY_NABP));
}
if (!lock_error)
lock_error=mi_lock_database(info,old_lock);
}
return test(lock_error || write_error);
}