[WizzardSK]

On quit, retroarch_main_quit_finish() freed the SMB and NFS connection pools before main_exit() deinitialised the core. A core that closes files it has open on a share when it unloads then closed them on a freed connection and RetroArch crashed on the way out.

Seen with the Vita3K core on Android, which reads its games through FILE* streams over the VFS (fopencookie) and closes them in its deinit:

signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0xffffffff8000007c
  #00 libretroarch-activity.so (retro_vfs_file_close_smb+64)
  #01 libretroarch-activity.so (retro_vfs_file_close_impl+44)
  #02 vita3k_libretro_android.so
  #03 libc.so (__FILE_close(__sFILE*)+248)
  ...
  #09 libretroarch-activity.so (runloop_event_deinit_core+500)
  #11 libretroarch-activity.so (retroarch_main_deinit_finish+16)
  #12 libretroarch-activity.so (main_exit+376)

The shutdown now runs in main_exit() right after RARCH_CTL_MAIN_DEINIT, when no core is left to use the pools. Closing content without quitting is unchanged: the pools stayed up then before as well.

[LibretroAdmin]

Thanks! Merged into master with your authorship kept.

I added a small follow-up on top: smb_shutdown() and nfs_shutdown() now unhook the pool first and only free it when no connection is still in use. Two things can still hold one at that point: an SMB file nobody closed (an open SMB stream keeps its slot for its whole lifetime), and a background task the exit's 3-second drain left stuck on a dead share. Either would have hit the same freed-connection crash. If something is still holding the pool it's left alone, since the process is exiting anyway. The shutdown stays where you put it, before frontend_driver_deinit(), because the Switch tears its sockets down there.