Wireshark-dev: [Wireshark-dev] Column data memory allocation - excessive?

From: Stephen Fisher <stephentfisher@xxxxxxxxx>
Date: Sun, 2 Mar 2008 15:31:36 -0700

One place where there are a lot of calls to the unsafe strcpy() is 
epan/column-utils.c.  However, before replacing them with g_strpcpy(), I 
wanted to ask if there is any reason we can't start using dynamic memory 
allocation in those places (as Guy recently suggested) instead of 
g_strlcpy()?

We are allocating full field widths worth of memory for every column 
(256 bytes each) plus the info column (4096 bytes each) for every row in 
the build_column_format_array() function (in epan/column.c) ahead of 
time and then strcpy() data into them.  With our default number of 
columns, that's 5,376 bytes per row!  A sample capture file I just 
opened would use 49 bytes for the first row (~0.9% of the allocated 
space).  Worse yet, I haven't found a place where these allocations 
(done with g_malloc()) are freed.  If they really aren't freed, this 
would lead to increasing memory usage every time you close a capture and 
open a new one until the program itself was closed.

We would probably need to change all of the column data handling 
functions to allocate memory as needed and realloc it if data is 
appended.  Perhaps we could use g_string_* functions to make that 
easier.

Any thoughts?


Steve