Обсуждение: a potential size overflow issue
Hi hackers, "InitBufTable" is the function used to initialize the buffer lookup table for buffer manager. With the memory size increasing nowadays, there is a potential overflow issue for the parameter "int size" used by "InitBufTable". This function is invoked in freelist.c as below: InitBufTable(NBuffers + NUM_BUFFER_PARTITIONS); The number of buffer block “NBuffers” is also defined as "int", and "NUM_BUFFER_PARTITIONS" has a default value 128. In theory, it may get the chance to overflow the "size" parameter in "InitBufTable". The "size" parameter is later used by "ShmemInitHash" as "init_size" and "max_size", which are all defined as "long". SharedBufHash = ShmemInitHash("Shared Buffer Lookup Table", size, size, &info, HASH_ELEM | HASH_BLOBS | HASH_PARTITION); Therefore, it would be better to change "InitBufTable(int size)" to "InitBufTable(long size)". A simple patch is attached and it passed the “make installcheck-world” test. -- David Software Engineer Highgo Software Inc. (Canada) www.highgo.ca
Вложения
David Zhang <david.zhang@highgo.ca> writes: > "InitBufTable" is the function used to initialize the buffer lookup > table for buffer manager. With the memory size increasing nowadays, > there is a potential overflow issue for the parameter "int size" used by > "InitBufTable". This function is invoked in freelist.c as below: > InitBufTable(NBuffers + NUM_BUFFER_PARTITIONS); > The number of buffer block “NBuffers” is also defined as "int", and > "NUM_BUFFER_PARTITIONS" has a default value 128. In theory, it may get > the chance to overflow the "size" parameter in "InitBufTable". No, because guc.c limits NBuffers to INT_MAX/2. > Therefore, it would be better to change "InitBufTable(int size)" to > "InitBufTable(long size)". If I was worried about this, that wouldn't be much of a fix, since on many platforms "long" is not any wider than "int". (We really oughta try to move away from relying on "long", because its size is so poorly standardized.) regards, tom lane