mariadb/sql/threadpool_common.cc
Monty bddbef3573 MDEV-34533 asan error about stack overflow when writing record in Aria
The problem was that when using clang + asan, we do not get a correct value
for the thread stack as some local variables are not allocated at the
normal stack.

It looks like that for example clang 18.1.3, when compiling with
-O2 -fsanitize=addressan it puts local variables and things allocated by
alloca() in other areas than on the stack.

The following code shows the issue

Thread 6 "mariadbd" hit Breakpoint 3, do_handle_one_connection
    (connect=0x5080000027b8,
    put_in_cache=<optimized out>) at sql/sql_connect.cc:1399

THD *thd;
1399      thd->thread_stack= (char*) &thd;
(gdb) p &thd
(THD **) 0x7fffedee7060
(gdb) p $sp
(void *) 0x7fffef4e7bc0

The address of thd is 24M away from the stack pointer

(gdb) info reg
...
rsp            0x7fffef4e7bc0      0x7fffef4e7bc0
...
r13            0x7fffedee7060      140737185214560

r13 is pointing to the address of the thd. Probably some kind of
"local stack" used by the sanitizer

I have verified this with gdb on a recursive call that calls alloca()
in a loop. In this case all objects was stored in a local heap,
not on the stack.

To solve this issue in a portable way, I have added two functions:

my_get_stack_pointer() returns the address of the current stack pointer.
The code is using asm instructions for intel 32/64 bit, powerpc,
arm 32/64 bit and sparc 32/64 bit.
Supported compilers are gcc, clang and MSVC.
For MSVC 64 bit we are using _AddressOfReturnAddress()

As a fallback for other compilers/arch we use the address of a local
variable.

my_get_stack_bounds() that will return the address of the base stack
and stack size using pthread_attr_getstack() or NtCurrentTed() with
fallback to using the address of a local variable and user provided
stack size.

Server changes are:

- Moving setting of thread_stack to THD::store_globals() using
  my_get_stack_bounds().
- Removing setting of thd->thread_stack, except in functions that
  allocates a lot on the stack before calling store_globals().  When
  using estimates for stack start, we reduce stack_size with
  MY_STACK_SAFE_MARGIN (8192) to take into account the stack used
  before calling store_globals().

I also added a unittest, stack_allocation-t, to verify the new code.

Reviewed-by: Sergei Golubchik <serg@mariadb.org>
2024-10-16 17:24:46 +03:00

542 lines
12 KiB
C++

/* Copyright (C) 2012, 2020, MariaDB
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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1335 USA */
#include "mariadb.h"
#include <violite.h>
#include <sql_priv.h>
#include <sql_class.h>
#include <my_pthread.h>
#include <scheduler.h>
#include <sql_connect.h>
#include <sql_audit.h>
#include <debug_sync.h>
#include <threadpool.h>
#ifdef WITH_WSREP
#include "wsrep_trans_observer.h"
#endif /* WITH_WSREP */
/* Threadpool parameters */
uint threadpool_min_threads;
uint threadpool_idle_timeout;
uint threadpool_size;
uint threadpool_max_size;
uint threadpool_stall_limit;
uint threadpool_max_threads;
uint threadpool_oversubscribe;
uint threadpool_mode;
uint threadpool_prio_kickup_timer;
my_bool threadpool_exact_stats;
my_bool threadpool_dedicated_listener;
/* Stats */
TP_STATISTICS tp_stats;
static void threadpool_remove_connection(THD *thd);
static int threadpool_process_request(THD *thd);
static THD* threadpool_add_connection(CONNECT *connect, void *scheduler_data);
extern bool do_command(THD*);
static inline TP_connection *get_TP_connection(THD *thd)
{
return (TP_connection *)thd->event_scheduler.data;
}
/*
Worker threads contexts, and THD contexts.
=========================================
Both worker threads and connections have their sets of thread local variables
At the moment it is mysys_var (this has specific data for dbug, my_error and
similar goodies), and PSI per-client structure.
Whenever query is executed following needs to be done:
1. Save worker thread context.
2. Change TLS variables to connection specific ones using thread_attach(THD*).
This function does some additional work , e.g setting up
thread_stack/thread_ends_here pointers.
3. Process query
4. Restore worker thread context.
Connection login and termination follows similar schema w.r.t saving and
restoring contexts.
For both worker thread, and for the connection, mysys variables are created
using my_thread_init() and freed with my_thread_end().
*/
struct Worker_thread_context
{
PSI_thread *psi_thread;
st_my_thread_var* mysys_var;
void save()
{
psi_thread= PSI_CALL_get_thread();
mysys_var= my_thread_var;
}
void restore()
{
PSI_CALL_set_thread(psi_thread);
set_mysys_var(mysys_var);
set_current_thd(nullptr);
}
};
#ifdef HAVE_PSI_INTERFACE
/*
The following fixes PSI "idle" psi instrumentation.
The server assumes that connection becomes idle
just before net_read_packet() and switches to active after it.
In out setup, server becomes idle when async socket io is made.
*/
extern void net_before_header_psi(struct st_net *net, void *user_data, size_t);
static void dummy_before_header(struct st_net *, void *, size_t)
{
}
static void re_init_net_server_extension(THD *thd)
{
thd->m_net_server_extension.m_before_header = dummy_before_header;
}
#else
#define re_init_net_server_extension(thd)
#endif /* HAVE_PSI_INTERFACE */
static inline bool has_unread_compressed_data(const NET *net)
{
return net->compress && net->remain_in_buf;
}
static inline void set_thd_idle(THD *thd)
{
thd->net.reading_or_writing= 1;
#ifdef HAVE_PSI_INTERFACE
if (!has_unread_compressed_data(&thd->net))
net_before_header_psi(&thd->net, thd, 0);
#endif
}
/*
Attach/associate the connection with the OS thread,
*/
static void thread_attach(THD* thd)
{
#ifdef WITH_WSREP
/* Wait until possible background rollback has finished before
attaching the thd. */
wsrep_wait_rollback_complete_and_acquire_ownership(thd);
#endif /* WITH_WSREP */
set_mysys_var(thd->mysys_var);
thd->store_globals();
PSI_CALL_set_thread(thd->get_psi());
mysql_socket_set_thread_owner(thd->net.vio->mysql_socket);
}
/*
Determine connection priority , using current
transaction state and 'threadpool_priority' variable value.
*/
static TP_PRIORITY get_priority(TP_connection *c)
{
DBUG_ASSERT(c->thd == current_thd);
TP_PRIORITY prio= (TP_PRIORITY)c->thd->variables.threadpool_priority;
if (prio == TP_PRIORITY_AUTO)
prio= c->thd->transaction->is_active() ? TP_PRIORITY_HIGH : TP_PRIORITY_LOW;
return prio;
}
void tp_callback(TP_connection *c)
{
DBUG_ASSERT(c);
Worker_thread_context worker_context;
worker_context.save();
THD *thd= c->thd;
c->state = TP_STATE_RUNNING;
if (unlikely(!thd))
{
/* No THD, need to login first. */
DBUG_ASSERT(c->connect);
thd= c->thd= threadpool_add_connection(c->connect, c);
if (!thd)
{
/* Bail out on connect error.*/
goto error;
}
c->connect= 0;
}
else if (threadpool_process_request(thd))
{
/* QUIT or an error occurred. */
goto error;
}
/* Set priority */
c->priority= get_priority(c);
/* Read next command from client. */
c->set_io_timeout(thd->get_net_wait_timeout());
c->state= TP_STATE_IDLE;
if (c->start_io())
goto error;
worker_context.restore();
return;
error:
c->thd= 0;
if (thd)
{
threadpool_remove_connection(thd);
}
delete c;
worker_context.restore();
}
static THD* threadpool_add_connection(CONNECT *connect, void *scheduler_data)
{
THD *thd= NULL;
/*
Create a new connection context: mysys_thread_var and PSI thread
Store them in THD.
*/
set_mysys_var(NULL);
my_thread_init();
st_my_thread_var* mysys_var= my_thread_var;
PSI_CALL_set_thread(PSI_CALL_new_thread(key_thread_one_connection, connect, 0));
if (!mysys_var ||!(thd= connect->create_thd(NULL)))
{
/* Out of memory? */
connect->close_and_delete(0);
if (mysys_var)
my_thread_end();
return NULL;
}
thd->event_scheduler.data = scheduler_data;
server_threads.insert(thd); // Make THD visible in show processlist
delete connect; // must be after server_threads.insert, see close_connections()
thd->set_mysys_var(mysys_var);
/* Login. */
thread_attach(thd);
re_init_net_server_extension(thd);
ulonglong now= microsecond_interval_timer();
thd->prior_thr_create_utime= now;
thd->start_utime= now;
thd->thr_create_utime= now;
setup_connection_thread_globals(thd);
if (thd_prepare_connection(thd))
goto end;
/*
Check if THD is ok, as prepare_new_connection_state()
can fail, for example if init command failed.
*/
if (!thd_is_connection_alive(thd))
goto end;
thd->skip_wait_timeout= true;
set_thd_idle(thd);
return thd;
end:
threadpool_remove_connection(thd);
return NULL;
}
static void threadpool_remove_connection(THD *thd)
{
thread_attach(thd);
thd->net.reading_or_writing = 0;
end_connection(thd);
close_connection(thd, 0);
unlink_thd(thd);
PSI_CALL_delete_current_thread(); // before THD is destroyed
delete thd;
/*
Free resources associated with this connection:
mysys thread_var and PSI thread.
*/
my_thread_end();
}
/*
Ensure that proper error message is sent to client,
and "aborted" message appears in the log in case of
wait timeout.
See also timeout handling in net_serv.cc
*/
static void handle_wait_timeout(THD *thd)
{
thd->get_stmt_da()->reset_diagnostics_area();
thd->reset_killed();
my_error(ER_NET_READ_INTERRUPTED, MYF(0));
thd->net.last_errno= ER_NET_READ_INTERRUPTED;
thd->net.error= 2;
}
/** Check if some client data is cached in thd->net or thd->net.vio */
static bool has_unread_data(THD* thd)
{
NET *net= &thd->net;
Vio *vio= net->vio;
return vio->has_data(vio) || has_unread_compressed_data(net);
}
/**
Process a single client request or a single batch.
*/
static int threadpool_process_request(THD *thd)
{
int retval= 0;
thread_attach(thd);
if (thd->killed >= KILL_CONNECTION)
{
/*
killed flag was set by timeout handler
or KILL command. Return error.
*/
retval= 1;
if(thd->killed == KILL_WAIT_TIMEOUT)
handle_wait_timeout(thd);
goto end;
}
/*
In the loop below, the flow is essentially the copy of
thead-per-connections
logic, see do_handle_one_connection() in sql_connect.c
The goal is to execute a single query, thus the loop is normally executed
only once. However for SSL connections, it can be executed multiple times
(SSL can preread and cache incoming data, and vio->has_data() checks if it
was the case).
*/
for(;;)
{
thd->net.reading_or_writing= 0;
if (mysql_audit_release_required(thd))
mysql_audit_release(thd);
if ((retval= do_command(thd)) != 0)
goto end;
if (!thd_is_connection_alive(thd))
{
retval= 1;
goto end;
}
set_thd_idle(thd);
if (!has_unread_data(thd))
{
/* More info on this debug sync is in sql_parse.cc*/
DEBUG_SYNC(thd, "before_do_command_net_read");
goto end;
}
}
end:
return retval;
}
static TP_pool *pool;
static bool tp_init()
{
#ifdef _WIN32
if (threadpool_mode == TP_MODE_WINDOWS)
pool= new (std::nothrow) TP_pool_win;
else
pool= new (std::nothrow) TP_pool_generic;
#else
pool= new (std::nothrow) TP_pool_generic;
#endif
if (!pool)
return true;
if (pool->init())
{
delete pool;
pool= 0;
return true;
}
return false;
}
static void tp_add_connection(CONNECT *connect)
{
TP_connection *c= pool->new_connection(connect);
DBUG_EXECUTE_IF("simulate_failed_connection_1", delete c ; c= 0;);
if (c)
pool->add(c);
else
connect->close_and_delete(0);
}
int tp_get_idle_thread_count()
{
return pool? pool->get_idle_thread_count(): 0;
}
int tp_get_thread_count()
{
return pool ? pool->get_thread_count() : 0;
}
void tp_set_min_threads(uint val)
{
if (pool)
pool->set_min_threads(val);
}
void tp_set_max_threads(uint val)
{
if (pool)
pool->set_max_threads(val);
}
void tp_set_threadpool_size(uint val)
{
if (pool)
pool->set_pool_size(val);
}
void tp_set_threadpool_stall_limit(uint val)
{
if (pool)
pool->set_stall_limit(val);
}
void tp_timeout_handler(TP_connection *c)
{
if (c->state != TP_STATE_IDLE)
return;
THD *thd= c->thd;
mysql_mutex_lock(&thd->LOCK_thd_kill);
Vio *vio= thd->net.vio;
if (vio && (vio_pending(vio) > 0 || vio->has_data(vio)) &&
c->state == TP_STATE_IDLE)
{
/*
There is some data on that connection, i.e
i.e there was no inactivity timeout.
Don't kill.
*/
c->state= TP_STATE_PENDING;
}
else if (c->state == TP_STATE_IDLE)
{
thd->set_killed_no_mutex(KILL_WAIT_TIMEOUT);
c->priority= TP_PRIORITY_HIGH;
post_kill_notification(thd);
}
mysql_mutex_unlock(&thd->LOCK_thd_kill);
}
MY_ALIGNED(CPU_LEVEL1_DCACHE_LINESIZE) Atomic_counter<unsigned long long> tp_waits[THD_WAIT_LAST];
static void tp_wait_begin(THD *thd, int type)
{
TP_connection *c = get_TP_connection(thd);
if (c)
{
DBUG_ASSERT(type > 0 && type < THD_WAIT_LAST);
tp_waits[type]++;
c->wait_begin(type);
}
}
static void tp_wait_end(THD *thd)
{
TP_connection *c = get_TP_connection(thd);
if (c)
c->wait_end();
}
static void tp_end()
{
delete pool;
}
static void tp_post_kill_notification(THD *thd)
{
TP_connection *c= get_TP_connection(thd);
if (c)
c->priority= TP_PRIORITY_HIGH;
post_kill_notification(thd);
}
static scheduler_functions tp_scheduler_functions=
{
0, // max_threads
NULL,
NULL,
tp_init, // init
tp_add_connection, // add_connection
tp_wait_begin, // thd_wait_begin
tp_wait_end, // thd_wait_end
tp_post_kill_notification, // post kill notification
tp_end // end
};
void pool_of_threads_scheduler(struct scheduler_functions *func,
ulong *arg_max_connections,
Atomic_counter<uint> *arg_connection_count)
{
*func = tp_scheduler_functions;
func->max_threads= threadpool_max_threads;
func->max_connections= arg_max_connections;
func->connection_count= arg_connection_count;
scheduler_init();
}