Crash on "bt" with a recording of a SIGSEGV (from stack exhaustion)

When debugging with rr version 5.6.0 and GNU gdb (GDB) Fedora Linux 12.1-7.fc37, I encounter the following crash while generating a backtrace:

[FATAL src/AutoRemoteSyscalls.cc:551:child_recvmsg() errno: EIO]
 (task 609 (rec:423) at time 0)
 -> Assertion `errno == ESRCH' failed to hold.
=== Start rr backtrace:
rr(_ZN2rr9GdbServer15emergency_debugEPNS_4TaskE+0x11e)[0x4a1fee]
rr[0x4b484d]
rr[0x4b54ab]
rr[0x462839]
rr(_ZN2rr18AutoRemoteSyscalls27infallible_send_fd_if_aliveERKNS_8ScopedFdE+0xfd)[0x45c0fd]
rr(_ZN2rr7Session18create_shared_mmapERNS_18AutoRemoteSyscallsEmNS_10remote_ptrIvEEPKciiOSt10shared_ptrINS_21MonitoredSharedMemoryEE+0x14a)[0x5705fa]
rr(_ZN2rr12AddressSpace13post_vm_cloneEPNS_4TaskE+0x89)[0x45d379]
rr(_ZN2rr4Task13post_vm_cloneENS0_11CloneReasonEiPS0_+0x4f)[0x5882ef]
rr(_ZN2rr4Task5cloneENS0_11CloneReasonEiNS_10remote_ptrIvEES3_NS2_IiEEiijPNS_7SessionESt10shared_ptrINS_7FdTableEES7_INS_11ThreadGroupEE+0x2bc)[0x58ce2c]
rr(_ZN2rr4Task8os_cloneENS0_11CloneReasonEPNS_7SessionERNS_18AutoRemoteSyscallsEijjSt10shared_ptrINS_7FdTableEES6_INS_11ThreadGroupEENS_10remote_ptrIvEENSB_IiEESC_SD_+0x2cc)[0x58f06c]
rr(_ZN2rr4Task12os_fork_intoEPNS_7SessionESt10shared_ptrINS_7FdTableEE+0x8b)[0x58b25b]
rr(_ZN2rr7Session13copy_state_toERS0_RNS_5EmuFsES3_+0x6a5)[0x576665]
rr(_ZN2rr13ReplaySession15clone_diversionEv+0x25a)[0x53987a]
rr(_ZN2rr9GdbServer6divertERNS_13ReplaySessionE+0x145)[0x49a5f5]
rr(_ZN2rr9GdbServer25process_debugger_requestsENS0_11ReportStateE+0x23f)[0x49c69f]
rr(_ZN2rr9GdbServer14debug_one_stepERNS_10GdbRequestE+0x1a1)[0x49e751]
rr(_ZN2rr9GdbServer12serve_replayERKNS0_15ConnectionFlagsE+0x97b)[0x4a011b]
rr[0x534a3c]
rr(_ZN2rr13ReplayCommand3runERSt6vectorINSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEESaIS7_EE+0x672)[0x5359f2]
rr(main+0x1d7)[0x446527]
/lib64/libc.so.6(+0x27510)[0x7f56c20ea510]
/lib64/libc.so.6(__libc_start_main+0x89)[0x7f56c20ea5c9]
rr(_start+0x25)[0x4466a5]
=== End rr backtrace

Similar crashes occur attempting to go up. "Vanilla" gdb does not experience this issue.

Start with src/AutoRemoteSyscalls.cc:551 and the child_recvmsg() failure shown in the report, then trace how bt and up reach GdbServer::divert(). Reproduce with rr 5.6.0, Fedora GDB 12.1-7.fc37, and the affected recording if available. Done means bt and up no longer crash while vanilla GDB behavior remains unaffected.