Crash in usbmux_packet_alloc assert

[rsms]

Crash in usbmux_packet_alloc assert

[rsms]

Are you sure it's not the second assertion? I.e. assert(sizeof(usbmux_packet_t) == 16);

I'm working on a new version of peertalk where I fixed an issue pertaining to this, namely that depending on compiler version and settings, the actual size of a struct might be larger than what's needed (usually an optimization to align it to words.) This might cause sizeof to report not the logical size, but the effective size of the struct.

Can you repro the crash? If so, run the app through lldb (e.g. lldb /path/to/your.app/Contents/MacOS/your) and see where it actually aborts from an assertion. It's also possible you've got a stack overflow issue in which case the stack trace might be incorrect.

BTW, the fix to the struct size is this:

typedef struct usbmux_packet {
  ...
} __attribute__((__packed__)) usbmux_packet_t;

[ramonpoca]

The second assert should crash on the first call, as sizeof(usbmux_packet_t) is known at compile time (and might be even optimized out). We are getting the traces through Crashlytics, and the issue repeats quite frequently.

I tried to replicate the call sequence to check for any unexpected value of payloadLength causing the crash, but couldn't. But note that nowhere is checked that the received upacket_len is bigger than sizeof(usbmux_packet_t), so a corrupt/unexpected packet of zero or less than 16 size would cause payloadLength to have a value near to UINT32_MAX (adding back 16 in usbmux_packet_alloc will cause this to allocate undersized packets).

[rsms]

b3090d7 makes some changes that should help in preventing getting into these situations. usbmux_packets are only used during negotiation with the USB mux hub, but isn't used for user communication (i.e. PTChannel doesn't use these packets and doesn't have these limitations.)