fix: resolve server crash and hang causes (signal handler, buffer overflow, race conditions)

[m1ngsama]

Closes #10

Summary

Five bugs confirmed by static analysis of the full source, all causing crashes or permanent hangs in production.

Changes

1. Signal handler deadlock — main.c

signal_handler called room_destroy() (which acquires pthread_rwlock_wrlock) and printf() — neither is async-signal-safe. If SIGTERM arrived while any thread held g_room->lock, the process deadlocked.

Also: close(g_listen_fd) was closing stdin (fd 0) because ssh_server_init returns 0 on success, not a real fd.

Fix: handler writes via write(2) (async-signal-safe) and calls _exit(0). Remove the bogus close(0).

2. NULL dereference in room_broadcast — chat_room.c

client_t **clients_copy = calloc(room->client_count, sizeof(client_t*));
memcpy(clients_copy, ...);  // crash when client_count==0 and calloc returns NULL

POSIX allows calloc(0, n) to return NULL. No NULL check for the OOM case either.

Fix: early return when count == 0; check calloc return.

3. Stack buffer overflow in tui_render_screen — tui.c

char buffer[8192] is overflowed with tall terminals:
197 lines × ~1031 bytes/msg ≈ 203 KiB. Title padding loop also had unchecked buffer[pos++].

Fix: malloc(65536) with buf_size used consistently throughout. Bounds check on padding loop.

4. sleep(2) inside libssh auth callback — ssh_server.c

auth_password is invoked from ssh_event_dopoll in the main thread. Sleeping there blocked the entire accept loop — one attacker with repeated wrong passwords stalled all connections for 2s per attempt. IP blocking via record_auth_failure already handles brute force.

Fix: remove sleep(2) from auth_password.

5. Spurious sleep() in accept loop error paths — ssh_server.c

sleep(1)/sleep(2) after rejecting rate-limited connections blocked accepting the next legitimate connection with no benefit.

Fix: remove all sleep() from accept loop error paths.

Test plan

• Build passes with no new warnings: make clean && make
• Server starts and stays up under systemctl
• Multiple simultaneous connections work correctly
• SIGTERM causes clean exit (no deadlock): kill -TERM $(pidof tnt)
• Auth failures don't stall other connections