alex/esp-idf
1
0
Fork 0
mirror of https://github.com/espressif/esp-idf.git synced 2025-09-24 17:27:21 +00:00
Code Issues Packages Projects Releases Wiki Activity

lwip: fix mbox thread-safe issue

Browse Source 
Fix a mbox free thread-safe issue that can lead to crash in sys_arch_mbox_fetch.
This commit is contained in:
Liu Zhi Fu
2018-12-20 14:03:11 +08:00
parent 0700cde229
commit 6933c103e5
3 changed files with 22 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
components/lwip/port/esp32/include/arch/sys_arch.h
View File

@@ -62,7 +62,25 @@ typedef struct sys_mbox_s {
#endif
#define sys_mbox_valid( x ) ( ( ( *x ) == NULL) ? pdFALSE : pdTRUE )
#define sys_mbox_set_invalid( x ) ( ( *x ) = NULL )
/* Define the sys_mbox_set_invalid() to empty to support lock-free mbox in ESP LWIP.
*
* The basic idea about the lock-free mbox is that the mbox should always be valid unless
* no socket APIs are using the socket and the socket is closed. ESP LWIP achieves this by
* following two changes to official LWIP:
* 1. Postpone the deallocation of mbox to netconn_free(), in other words, free the mbox when
* no one is using the socket.
* 2. Define the sys_mbox_set_invalid() to empty if the mbox is not actually freed.
* The second change is necessary. Consider a common scenario: the application task calls
* recv() to receive packets from the socket, the sys_mbox_valid() returns true. Because there
* is no lock for the mbox, the LWIP CORE can call sys_mbox_set_invalid() to set the mbox at
* anytime and the thread-safe issue may happen.
*
* However, if the sys_mbox_set_invalid() is not called after sys_mbox_free(), e.g. in netconn_alloc(),
* we need to initialize the mbox to invalid explicitly since sys_mbox_set_invalid() now is empty.
*/
#define sys_mbox_set_invalid( x )
#define sys_sem_valid( x ) ( ( ( *x ) == NULL) ? pdFALSE : pdTRUE )
#define sys_sem_set_invalid( x ) ( ( *x ) = NULL )