`mmap.mmap.__setitem__` crashes when concurrently resized

[djarfluka]

Bug report

Bug description:

import mmap
mm = mmap.mmap(-1, 100000)         
class mid:
    def __index__(self):
        mm.resize(8)           
        return 0
mm[90000] = mid()          

AddressSanitizer:DEADLYSIGNAL

==1==ERROR: AddressSanitizer: SEGV on unknown address 0xffffa1a0cf90 (pc 0xffffa1a2aff0 bp 0xffffd2152970 sp 0xffffd2152830 T0)
==1==The signal is caused by a WRITE memory access.
#0 0xffffa1a2aff0 in safe_byte_copy Modules/mmapmodule.c:406
#1 0xffffa1a2aff0 in mmap_ass_subscript_lock_held Modules/mmapmodule.c:1692
#2 0xffffa1a2aff0 in mmap_ass_subscript Modules/mmapmodule.c:1750
#3 0xaaaadf0a9860 in _PyEval_EvalFrameDefault Python/generated_cases.c.h:12634
#4 0xaaaadf970b1c in _PyEval_EvalFrame Include/internal/pycore_ceval.h:122
#5 0xaaaadf970b1c in _PyEval_Vector Python/ceval.c:2174
#6 0xaaaadf970b1c in PyEval_EvalCode Python/ceval.c:679
#7 0xaaaadfb6feb4 in run_mod Python/pythonrun.c:1509
#8 0xaaaadfb70d38 in _PyRun_File Python/pythonrun.c:1332
#9 0xaaaadfb74ffc in _PyRun_SimpleFile Python/pythonrun.c:544
#10 0xaaaadfb78dd0 in _PyRun_AnyFile Python/pythonrun.c:92
#11 0xaaaadfc4390c in pymain_run_file_obj Modules/main.c:478
#12 0xaaaadfc4390c in pymain_run_file Modules/main.c:494
#13 0xaaaadfc4390c in pymain_run_python Modules/main.c:804
#14 0xaaaadfc4390c in Py_RunMain Modules/main.c:891
#15 0xaaaadfc45c38 in pymain_main Modules/main.c:921
#16 0xaaaadfc45c38 in Py_BytesMain Modules/main.c:945
#17 0xffffa46f84c0 (/lib/aarch64-linux-gnu/libc.so.6+0x284c0) (BuildId: 27027b96e5b8c475fc327aa445bea1c71d37b4e2)
#18 0xffffa46f8594 in __libc_start_main (/lib/aarch64-linux-gnu/libc.so.6+0x28594) (BuildId: 27027b96e5b8c475fc327aa445bea1c71d37b4e2)
#19 0xaaaadf0ebeec in _start (/src/cpython/python+0x183beec) (BuildId: eba909a3db0adabacf151d7dc3081dda7999191a)

AddressSanitizer can not provide additional info.
SUMMARY: AddressSanitizer: SEGV Modules/mmapmodule.c:406 in safe_byte_copy
==1==ABORTING

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-157335: Fix out-of-bounds write in mmap.mmap.__setitem__ #157438
• [3.15] gh-157335: Fix out-of-bounds write in mmap.mmap.__setitem__ (GH-157438) #158026
• [3.14] gh-157335: Fix out-of-bounds write in mmap.mmap.__setitem__ (GH-157438) #158027
• [3.13] gh-157335: Fix out-of-bounds write in mmap.mmap.__setitem__ (#157438) #158028

[picnixz]

I don't remember whether this was considered legitimate or not (that is, messing up with mmap like that).
I feel that this is close to #148650 (maybe not the same cause, but similar issue, that is concurrent mutation).

cc @serhiy-storchaka @vstinner

[picnixz]

@djarfluka Btw, is it a core dump when built without ASAN or is it a silent out-of-bound write?

[djarfluka]

@picnixz sorry for the messy bug report. will give more descriptive titles in the future. I'm seeing a silent oob write

[vstinner]

The reproducer now fails with:

Traceback (most recent call last):
  File "script.py", line 7, in <module>
    mm[90000] = mid()
    ~~^^^^^^^
IndexError: mmap index out of range