Out-of-bounds Write

msgpack_unpacker_expand_buffer in src/unpack.c, reached through the public msgpack_unpacker_reserve_buffer API, computes its new buffer size using an unchecked size_t addition of the requested size and the amount already used. The doubling loop guards its own multiplication against overflow, but the addition in the loop condition is unguarded, so a request near SIZE_MAX wraps: the loop condition is already satisfied, the allocation is performed at the small pre-wrap size, and the function returns true. The caller is told the requested capacity was reserved when it was not, so a subsequent write of the requested length overflows the heap buffer. The library's own example/lib_buffer_unpack.c demonstrates the reserve-then-write pattern, and its defensive assert comparing capacity against the request is compiled out under NDEBUG. msgpack-c's own decode entry points do not derive the reservation size from untrusted input, so reaching this requires an integration that passes an attacker-influenced length to the reservation API, such as a length-prefixed streaming transport.Show less

hank-ai/darknet sizes a convolutional layer's weight and output heap buffers by multiplying configuration fields taken from a .cfg file in unchecked 32-bit int arithmetic. In src-lib/convolutional_layer.cpp, l.nweights is computed as (c / groups) * n * size * size and l.outputs as l.out_h * l.out_w * l.out_c, and both feed xcalloc directly. A .cfg whose true dimension product exceeds INT_MAX wraps to a small or zero value, so the allocation is undersized; for example width and height of 256 with filters of 65536 gives 2^32, which wraps to 0. forward_convolutional_layer then re-derives the GEMM dimensions with a different operand order, computing k as l.size*l.size*l.c / l.groups where the allocation divided before multiplying, and reads and writes through the undersized buffer. Loading the crafted .cfg for inference or training is sufficient and no valid .weights file is required. The reported proof of concept observed a heap buffer overflow read in gemm_nn_fast under AddressSanitizer and glibc allocator metadata corruption in a release build of the same input, indicating an out-of-bounds write.Show less

Libevent is an event notification library. Prior to 2.1.13 and 2.2.2-alpha, libevent has a heap out-of-bounds write in bufferevent_sock.c when bufferevent_socket_set_conn_address_ copies a kernel-supplied AF_UNIX peer address into bufferevent_private.conn_address. Release builds compiled with NDEBUG disable the EVUTIL_ASSERT length guard, and the evhttp accept path can pass a 110-byte sockaddr from accept() into the 28-byte field. An unauthenticated local peer able to connect to an AF_UNIX listener can overwrite the adjacent dns_request pointer and heap data, causing memory corruption with confidentiality, integrity, and availability impact. This issue is fixed in versions 2.1.13 and 2.2.2-alpha.Show less

Libevent is an event notification library. Prior to 2.1.13 and 2.2.2-alpha, libevent has an off-by-one stack buffer overflow in evdns.c when dnsname_to_labels formats a name-bearing DNS record at the end of the 64 KB stack buffer allocated by evdns_server_request_format_response. The final-label check permits j plus label_len plus one to equal buf_len, after which the terminating null byte is written to buf[buf_len]. A crafted DNS server response containing PTR, CNAME, MX, NS, or SOA data can trigger the one-byte out-of-bounds write and crash or corrupt the process. This issue is fixed in versions 2.1.13 and 2.2.2-alpha.Show less