There was a memory leak due to the vector storing the split data not being freed after being copied into the http struct.
Thanks for the suggestion.
Can you please explain why you think there's a memory leak, how you compiled your patch, and how you tested it?
v is a stack-local array of pointers into buf, so trying to free it is not a good idea.
I can't even get your patch through gcc45, and even if I could, I wouldn't expect it to work:
gcc45 -c -pipe -fstack-protector-all -ggdb -Wshadow -Wconversion -I/usr/local/include/ -pthread -Wall urlmatch.c -o urlmatch.o
urlmatch.c: In function 'parse_http_request':
urlmatch.c:534:4: warning: the address of 'v' will always evaluate as 'true'
urlmatch.c:534:4: error: incompatible types when assigning to type 'char *' from type 'void *'
gmake: *** [urlmatch.o] Error 1
Sorry about that, while reading the code for parse_http_request I noticed that the vector was being passed to ssplit but was being statically defined and set to zero within parse_http_request which I thought was odd because ssplit wouldn't be able grow and shrink it. I looked into the ssplit function and saw that the vector indeed does not grow and simply returns an error when/if it is filled. When I came back to my computer later I finished looking through the parse_http_request, not remembering that it was a statically defined array, and thinking it was a dynamically allocated vector didn't see it being freed at the end of the function. I should have spent more time looking over the code and making sure that it was actually an issue.
While testing the patch I must have build the original source tree because I made sure to build it and check to be sure that it at least compiled before submitting the patch. I probably should have run it as well and checked that it worked functionally as well.
Log in to post a comment.
Sign up for the SourceForge newsletter:
You seem to have CSS turned off.
Please don't fill out this field.