[ruby-core:26257] [Bug #2258] Kernel#require inside rb_require() inside rb_protect() inside SysV context fails

Issue #2258 has been updated by Suraj Kurapati.


Hi,

I ran the ruby-ucontext-full.tgz example on Ruby @ SVN trunk
in a debugger and found that the real problem was being hidden
by a call to rb_longjmp().  Below is the stack trace of the
real problem, just before the long jump is executed.

The real problem is that the "stdin" value passed from
Inito_IO() to prep_stdio() is becoming corrupted somehow:

(gdb) f
#1  0x0808b2e4 in Init_IO () at io.c:9785
(gdb) p stdin
$3 = (struct _IO_FILE *) 0xb7f3b420
(gdb) down
#0  prep_stdio (f=0x8213008, fmode=1, klass=136569860, path=0x81860bd "<STDIN>") at io.c:6267

Notice that stdin is 0xb7f3b420 inside Init_IO(),
but inside prep_stdio() its value is 0x8213008.

This can only mean *one thing*: the stack pointer in Init_IO()
is radically different from the stack pointer in prep_stdio(),
and the arguments for the function call to prep_stdio() are
being placed at a different stack address than what prep_stdio()
is reading from, as illustrated in this pseudo-assembly code:

  Init_IO():

    stack.push stdin # 0xb7f3b420 (correct value)
    stack.push fmode
    stack.push klass
    stack.push path
    stack.push program_counter + size_of_machine_instruction
    jump prep_stdio

  prep_stdio():

    # PROBLEM: stack pointer is wrong!

    return_address = stack.pop
    path           = stack.pop
    klass          = stack.pop
    fmode          = stack.pop
    stdin          = stack.pop # 0x8213008 (wrong value!)

Any suggestions?

Thanks for your consideration.


###############################################################
# The full backtrace before longjmp() is executed:
###############################################################

(gdb) bt full
#0  0xb7e23201 in siglongjmp () from /lib/libc.so.6
No symbol table info available.
#1  0x08063e73 in rb_longjmp (tag=6, mesg=136568440) at eval.c:425
        at = 136588340
        e = 136568440
        th = 0x8213290
        file = 0x0
        line = 0
#2  0x08063eaa in rb_exc_raise (mesg=136568440) at eval.c:434
No locals.
#3  0x08061866 in rb_raise (exc=136570260, fmt=0x8185657 "closed stream") at error.c:1150
        args = 0x8205108 "h\275'\b\264\336#\b\b0!\b\001"
        mesg = 136568460
#4  0x0807a6e2 in rb_io_check_closed (fptr=0x827bd68) at io.c:248
No locals.
#5  0x08085044 in prep_stdio (f=0x8213008, fmode=1, klass=136569860, path=0x81860bd "<STDIN>") at io.c:6267
        fptr = 0x827bd68
        io = 136568500
#6  0x0808b2e4 in Init_IO () at io.c:9785
        rb_cARGF = 136570780
#7  0x0807a453 in rb_call_inits () at inits.c:43
No locals.
#8  0x08063401 in ruby_init () at eval.c:59
        _th = 0x8213290
        _tag = {buf = {{__jmpbuf = {136323376, 0, 0, 136335960, -891095361, -2044316993}, __mask_was_saved = 0, __saved_mask = {__val = {134563877, 1, 136335872, 3086940374, 3087005720, 3084878200, 1, 1, 0, 1, 136335864, 3086200820, 136335880, 3085687571, 3086200820, 136335896, 3085687736, 3, 136335880, 8388608, 4294967295, 0, 0, 136335960, 135598991, 8388608, 0, 5, 0, 136335960, 3085266387, 136335960}}}}, tag = 0, retval = 0, prev = 0x0}
        initialized = 1
        state = 0
#9  0x0805d235 in ruby_context_body () at main.c:84
        variable_in_this_stack_frame = 0
        i = 2
        argc = 0
        argv = 0x8182bc9
#10 0xb7e3303b in makecontext () from /lib/libc.so.6
No symbol table info available.
#11 0x08203140 in ?? ()
No symbol table info available.
#12 0x00000000 in ?? ()
No symbol table info available.
(gdb)
----------------------------------------
http://redmine.ruby-lang.org/issues/show/2258

----------------------------------------
http://redmine.ruby-lang.org