在MySQL中使用getopt lib时可疑的内存泄漏？

Question

I am posting this post on behalf of my college.

He found a suspicious memory leak when using the handle_option (MySQL getopt lib) to read config file (/etc/my.cnf)

He execute the Source code below after malloc host_name, user name:

char* host_name;
char* user_name;

struct my_option mysql_confgs[] = 
{
  {"host", "h", "MySQL Server", (uchar**) & host_name, NULL, NULL, GET_STR, 
REQUIRED_ARG, 0,0,0,0,0,0},
    {"user", "u", "userID", "h",(uchar**) & user_name, NULL, NULL, GET_STR, 
REQUIRED_ARG, 0,0,0,0,0,0}
}

handle_options(&argc, &argv, mysql_configs, aux_config_reader);

He mention about the method above is using Error(Segment) instead of using free(host_name) and free(user_name)? So this is the possible reason of causing memory leakage?

Well.. I have zero basic on MySQL, so I might not able to delivery 100% of the problem description. So, feel free to query on this and I will update the details of problem descriptions as per the query.

My college is having language barrier so I am posting on behalf of him.

Valgrind Report:

480 bytes in 1 blocks are possibly lost in loss record 26 of 43
at 0x4A068FE: malloc (vg_replace_malloc.c:270)
by 0x33E4E293C1: my_malloc (in /usr/lib64/mysql/libmysqlclient.so.16.0.0)
by 0x33E4E2C974: alloc_root (in /usr/lib64/mysql/libmysqlclient.so.16.0.0)
by 0x33E4E2E620: ??? (in /usr/lib64/mysql/libmysqlclient.so.16.0.0)
by 0x33E4E2F838: my_load_defaults (in /usr/lib64/mysql/libmysqlclient.so.16.0.0)
by 0x408BF1: MS_MYSQL_init (MS_MYSQL_O.h:109)
by 0x438A39: main_proc (AccLab.c:221)
by 0x437F8A: main (AccLab.c:67)

75,840 bytes in 158 blocks are definitely lost in loss record 41 of 43
at 0x4A068FE: malloc (vg_replace_malloc.c:270)
by 0x33E4E293C1: my_malloc (in /usr/lib64/mysql/libmysqlclient.so.16.0.0)
by 0x33E4E2C974: alloc_root (in /usr/lib64/mysql/libmysqlclient.so.16.0.0)
by 0x33E4E2E620: ??? (in /usr/lib64/mysql/libmysqlclient.so.16.0.0)
by 0x33E4E2F838: my_load_defaults (in /usr/lib64/mysql/libmysqlclient.so.16.0.0)
by 0x408BF1: MS_MYSQL_init (MS_MYSQL_O.h:109)
by 0x438A39: main_proc (AccLab.c:221)
by 0x437F8A: main (AccLab.c:67)

LEAK SUMMARY:

definitely lost: 75,840 bytes in 158 blocks
indirectly lost: 0 bytes in 0 blocks
possibly lost: 2,304 bytes in 7 blocks
still reachable: 9,675,408 bytes in 2,387 blocks
suppressed: 0 bytes in 0 blocks
Reachable blocks (those to which a pointer was found) are not shown.
To see them, rerun with: --leak-check=full --show-reachable=yes

For counts of detected and suppressed errors, rerun with: -v
ERROR SUMMARY: 8 errors from 8 contexts (suppressed: 4 from 4)

Answer 1

Put printf("user_name: %p; host_name: %p\\n", (void *) user_name, (void *) host_name); before and after the call to handle_options and run your code. Do the additional two lines printed as a result differ from each other? If so, your diagnosis is correct and user_name and host_name are changed by handle_options, and perhaps using a malloc'd pointer is unsuitable for this function.

If not, your diagnosis is incorrect and the memory leak lies elsewhere. You'd want to look at the source code for MS_MYSQL_init, main_proc and main, in that order, which are the three functions listed by valgrind from your project. Let me know if you need my help...

Answer 2

I'd say the combination of allocating memory to the pointer my_option.value is pointing to along with the use of GET_STR is leading to leaking what had been allocated to my_option.value , as GET_PTR initalises what my_option.value is pointing to, to point right to somewhere into what had been passed as argv to handle_options , without freeing what the value my_option.value is pointing, previously pointed to.

To get around this, either do not allocate any memory to what my_option.value is pointing to prior to passing it to handle_options or allocate it using my_alloc() and use GET_PTR_ALLOC as value type, as GET_PTR_ALLOC implies a call to my_free() on what my_option.value is pointing to prior to re-initialising what it is pointing to.

---

Just out of curiosity: What is uchar , and why do you cast to uchar ** and not to void * , the type of my_option.value ?

---

Also this

"user", "u", "userID", "h",(uchar**) & user_name ...

should read

"user", "u", "userID", (uchar**) & user_name ...

shouldn't it?