[Caml-list] memory corruption using C stub

[Caml-list] memory corruption using C stub

[Alexey Egorov]

Hello,

I have an OCaml application with some C code which (I believe) is the
reason of some random crashes.

Here is the code - https://pastebin.com/FVtLphZu
This function reads file at given offset, divides data into chunks and
compute checksums and compression ratio:
  external compute_data_props
    : string -> int -> int -> int -> (int * int * float) list =
"compute_data_props"

The problem is, after some calls to this stub, application is crashing
at random places in OCaml code.
I can't figure out what's going wrong, but replacing this stub with
dummy function (which does nothing but returns some predefined list)
eliminates the problem.

What can we do to debug it? We are using OCaml 4.04.1 and Windows 10,
and there is no other C stubs in our codebase.

Thanks!

Re: [Caml-list] memory corruption using C stub

[Viet Le]

[-- Attachment #1: Type: text/plain, Size: 1188 bytes --]

Do you have stack frame to look at?


On Sat, 10 Jun 2017 at 16:33, Alexey Egorov <alex.only.d@gmail.com> wrote:

> Hello,
>
> I have an OCaml application with some C code which (I believe) is the
> reason of some random crashes.
>
> Here is the code - https://pastebin.com/FVtLphZu
> This function reads file at given offset, divides data into chunks and
> compute checksums and compression ratio:
>   external compute_data_props
>     : string -> int -> int -> int -> (int * int * float) list =
> "compute_data_props"
>
> The problem is, after some calls to this stub, application is crashing
> at random places in OCaml code.
> I can't figure out what's going wrong, but replacing this stub with
> dummy function (which does nothing but returns some predefined list)
> eliminates the problem.
>
> What can we do to debug it? We are using OCaml 4.04.1 and Windows 10,
> and there is no other C stubs in our codebase.
>
> Thanks!
>
> --
> Caml-list mailing list.  Subscription management and archives:
> https://sympa.inria.fr/sympa/arc/caml-list
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
>
-- 
Kind regards,
Viet

[-- Attachment #2: Type: text/html, Size: 2017 bytes --]

Re: [Caml-list] memory corruption using C stub

[Daniel Bünzli]

My C FFI may be a bit rusty and I only had a quick look but I think that on line 75 you need to put that caml_copy_double into a CAMLLocal variable since you are allocating afterwards on line 77.

See Rule 3 in http://caml.inria.fr/pub/docs/manual-ocaml/intfc.html

Best,

Daniel

P.S. A quick way of checking if you do everything correctly is to call Gc.full_major () after the call to C usually this crashes immediately if you did something wrong.

Re: [Caml-list] memory corruption using C stub

[Daniel Bünzli]

On 10 June 2017 at 16:44:50, Daniel Bünzli (daniel.buenzli@erratique.ch) wrote:

> My C FFI may be a bit rusty and I only had a quick look but I think that on line 75 you need to  
> put that caml_copy_double into a CAMLLocal variable since you are allocating afterwards  
> on line 77.

(Or rearrange your code so that caml_copy_double is the last allocation you perform in the fun.)

D

Re: [Caml-list] memory corruption using C stub

[Daniel Bünzli]

On 10 June 2017 at 16:44:50, Daniel Bünzli (daniel.buenzli@erratique.ch) wrote:
> My C FFI may be a bit rusty

It is. In fact forget what I said it's the first argument of Store_field you need to make sure to keep a root on. 

D

Re: [Caml-list] memory corruption using C stub

[Alexey Egorov]

Stack traces is not very helpful; it crashes at random places in OCaml code:

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 4204.0xb4c]
0x0000000000414fac in camlSource__fun_6516 ()
(gdb)
(gdb) bt
#0  0x0000000000414fac in camlSource__fun_6516 ()
#1  0x0000000000000000 in ?? ()
Backtrace stopped: previous frame inner to this frame (corrupt stack?)


2017-06-10 20:40 GMT+05:00 Viet Le <vietlq85@gmail.com>:
> Do you have stack frame to look at?
>
>
> On Sat, 10 Jun 2017 at 16:33, Alexey Egorov <alex.only.d@gmail.com> wrote:
>>
>> Hello,
>>
>> I have an OCaml application with some C code which (I believe) is the
>> reason of some random crashes.
>>
>> Here is the code - https://pastebin.com/FVtLphZu
>> This function reads file at given offset, divides data into chunks and
>> compute checksums and compression ratio:
>>   external compute_data_props
>>     : string -> int -> int -> int -> (int * int * float) list =
>> "compute_data_props"
>>
>> The problem is, after some calls to this stub, application is crashing
>> at random places in OCaml code.
>> I can't figure out what's going wrong, but replacing this stub with
>> dummy function (which does nothing but returns some predefined list)
>> eliminates the problem.
>>
>> What can we do to debug it? We are using OCaml 4.04.1 and Windows 10,
>> and there is no other C stubs in our codebase.
>>
>> Thanks!
>>
>> --
>> Caml-list mailing list.  Subscription management and archives:
>> https://sympa.inria.fr/sympa/arc/caml-list
>> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>> Bug reports: http://caml.inria.fr/bin/caml-bugs
>
> --
> Kind regards,
> Viet

Re: [Caml-list] memory corruption using C stub

[Alexey Egorov]

OK, I'm minimized this function as much as possible and it's still
crashing - https://pastebin.com/MZ0Qkh9B
Now I'm thinking that is compiler's bug on windows...

2017-06-10 20:33 GMT+05:00 Alexey Egorov <alex.only.d@gmail.com>:
> Hello,
>
> I have an OCaml application with some C code which (I believe) is the
> reason of some random crashes.
>
> Here is the code - https://pastebin.com/FVtLphZu
> This function reads file at given offset, divides data into chunks and
> compute checksums and compression ratio:
>   external compute_data_props
>     : string -> int -> int -> int -> (int * int * float) list =
> "compute_data_props"
>
> The problem is, after some calls to this stub, application is crashing
> at random places in OCaml code.
> I can't figure out what's going wrong, but replacing this stub with
> dummy function (which does nothing but returns some predefined list)
> eliminates the problem.
>
> What can we do to debug it? We are using OCaml 4.04.1 and Windows 10,
> and there is no other C stubs in our codebase.
>
> Thanks!

Re: [Caml-list] memory corruption using C stub

[David Allsopp]

> On 11 Jun 2017, at 12:38, Alexey Egorov <alex.only.d@gmail.com> wrote:
> 
> OK, I'm minimized this function as much as possible and it's still
> crashing - https://pastebin.com/MZ0Qkh9B
> Now I'm thinking that is compiler's bug on windows...

Are you able to post the OCaml side of your minimised code? Which port of OCaml are you using?


David


> 2017-06-10 20:33 GMT+05:00 Alexey Egorov <alex.only.d@gmail.com>:
>> Hello,
>> 
>> I have an OCaml application with some C code which (I believe) is the
>> reason of some random crashes.
>> 
>> Here is the code - https://pastebin.com/FVtLphZu
>> This function reads file at given offset, divides data into chunks and
>> compute checksums and compression ratio:
>>  external compute_data_props
>>    : string -> int -> int -> int -> (int * int * float) list =
>> "compute_data_props"
>> 
>> The problem is, after some calls to this stub, application is crashing
>> at random places in OCaml code.
>> I can't figure out what's going wrong, but replacing this stub with
>> dummy function (which does nothing but returns some predefined list)
>> eliminates the problem.
>> 
>> What can we do to debug it? We are using OCaml 4.04.1 and Windows 10,
>> and there is no other C stubs in our codebase.
>> 
>> Thanks!
> 
> -- 
> Caml-list mailing list.  Subscription management and archives:
> https://sympa.inria.fr/sympa/arc/caml-list
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: [Caml-list] memory corruption using C stub

[Alexey Egorov]

OCaml side is a complex server-side application; unfortunately, it's
closed-source and when I'm trying to minimize OCaml side too it stops
crashing...
I'm using fdopen's opam repository
(https://github.com/fdopen/opam-repository-mingw) with compiler's
versions from 4.03 to 4.05-beta3.


2017-06-11 16:45 GMT+05:00 David Allsopp <dra-news@metastack.com>:
>> On 11 Jun 2017, at 12:38, Alexey Egorov <alex.only.d@gmail.com> wrote:
>>
>> OK, I'm minimized this function as much as possible and it's still
>> crashing - https://pastebin.com/MZ0Qkh9B
>> Now I'm thinking that is compiler's bug on windows...
>
> Are you able to post the OCaml side of your minimised code? Which port of OCaml are you using?
>
>
> David
>
>
>> 2017-06-10 20:33 GMT+05:00 Alexey Egorov <alex.only.d@gmail.com>:
>>> Hello,
>>>
>>> I have an OCaml application with some C code which (I believe) is the
>>> reason of some random crashes.
>>>
>>> Here is the code - https://pastebin.com/FVtLphZu
>>> This function reads file at given offset, divides data into chunks and
>>> compute checksums and compression ratio:
>>>  external compute_data_props
>>>    : string -> int -> int -> int -> (int * int * float) list =
>>> "compute_data_props"
>>>
>>> The problem is, after some calls to this stub, application is crashing
>>> at random places in OCaml code.
>>> I can't figure out what's going wrong, but replacing this stub with
>>> dummy function (which does nothing but returns some predefined list)
>>> eliminates the problem.
>>>
>>> What can we do to debug it? We are using OCaml 4.04.1 and Windows 10,
>>> and there is no other C stubs in our codebase.
>>>
>>> Thanks!
>>
>> --
>> Caml-list mailing list.  Subscription management and archives:
>> https://sympa.inria.fr/sympa/arc/caml-list
>> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: [Caml-list] memory corruption using C stub

[David Allsopp]

On 11 Jun 2017, at 12:54, Alexey Egorov <alex.only.d@gmail.com> wrote:
> 
> OCaml side is a complex server-side application; unfortunately, it's
> closed-source and when I'm trying to minimize OCaml side too it stops
> crashing...

Ah well! Are there other packages you are using which themselves have C stubs? If minimising the OCaml application is also eliminating the problem, then it could be that it's just the allocations in your (correct) C stub which are revealing the crash elsewhere. Unfortunately, incorrect C stub can include OCaml unix functions...

Spattering Gc.full_major on the OCaml can help try to find corruptions earlier - have you tried that too?

> I'm using fdopen's opam repository
> (https://github.com/fdopen/opam-repository-mingw) with compiler's
> versions from 4.03 to 4.05-beta3.

I presume you're seeing the crash from all versions? Are you able to do a 4.02.3 test, or is that impractical?


David


> 2017-06-11 16:45 GMT+05:00 David Allsopp <dra-news@metastack.com>:
>>> On 11 Jun 2017, at 12:38, Alexey Egorov <alex.only.d@gmail.com> wrote:
>>> 
>>> OK, I'm minimized this function as much as possible and it's still
>>> crashing - https://pastebin.com/MZ0Qkh9B
>>> Now I'm thinking that is compiler's bug on windows...
>> 
>> Are you able to post the OCaml side of your minimised code? Which port of OCaml are you using?
>> 
>> 
>> David
>> 
>> 
>>> 2017-06-10 20:33 GMT+05:00 Alexey Egorov <alex.only.d@gmail.com>:
>>>> Hello,
>>>> 
>>>> I have an OCaml application with some C code which (I believe) is the
>>>> reason of some random crashes.
>>>> 
>>>> Here is the code - https://pastebin.com/FVtLphZu
>>>> This function reads file at given offset, divides data into chunks and
>>>> compute checksums and compression ratio:
>>>> external compute_data_props
>>>>   : string -> int -> int -> int -> (int * int * float) list =
>>>> "compute_data_props"
>>>> 
>>>> The problem is, after some calls to this stub, application is crashing
>>>> at random places in OCaml code.
>>>> I can't figure out what's going wrong, but replacing this stub with
>>>> dummy function (which does nothing but returns some predefined list)
>>>> eliminates the problem.
>>>> 
>>>> What can we do to debug it? We are using OCaml 4.04.1 and Windows 10,
>>>> and there is no other C stubs in our codebase.
>>>> 
>>>> Thanks!
>>> 
>>> --
>>> Caml-list mailing list.  Subscription management and archives:
>>> https://sympa.inria.fr/sympa/arc/caml-list
>>> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>>> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: [Caml-list] memory corruption using C stub

[Alexey Egorov]

> Are there other packages you are using which themselves have C stubs?
No, except maybe stubs in Array and Unix modules (we are using lots of
Unix.create_process and related functions).
You are right, our C stub is correct - I increased a load on
application and it crashes even without calling this C function...

> I presume you're seeing the crash from all versions? Are you able to do a 4.02.3 test, or is that impractical?
Yes, app crashes in all versions. I tried 4.02.3 - still crashing :(


2017-06-11 17:07 GMT+05:00 David Allsopp <dra-news@metastack.com>:
> On 11 Jun 2017, at 12:54, Alexey Egorov <alex.only.d@gmail.com> wrote:
>>
>> OCaml side is a complex server-side application; unfortunately, it's
>> closed-source and when I'm trying to minimize OCaml side too it stops
>> crashing...
>
> Ah well! Are there other packages you are using which themselves have C stubs? If minimising the OCaml application is also eliminating the problem, then it could be that it's just the allocations in your (correct) C stub which are revealing the crash elsewhere. Unfortunately, incorrect C stub can include OCaml unix functions...
>
> Spattering Gc.full_major on the OCaml can help try to find corruptions earlier - have you tried that too?
>
>> I'm using fdopen's opam repository
>> (https://github.com/fdopen/opam-repository-mingw) with compiler's
>> versions from 4.03 to 4.05-beta3.
>
> I presume you're seeing the crash from all versions? Are you able to do a 4.02.3 test, or is that impractical?
>
>
> David
>
>
>> 2017-06-11 16:45 GMT+05:00 David Allsopp <dra-news@metastack.com>:
>>>> On 11 Jun 2017, at 12:38, Alexey Egorov <alex.only.d@gmail.com> wrote:
>>>>
>>>> OK, I'm minimized this function as much as possible and it's still
>>>> crashing - https://pastebin.com/MZ0Qkh9B
>>>> Now I'm thinking that is compiler's bug on windows...
>>>
>>> Are you able to post the OCaml side of your minimised code? Which port of OCaml are you using?
>>>
>>>
>>> David
>>>
>>>
>>>> 2017-06-10 20:33 GMT+05:00 Alexey Egorov <alex.only.d@gmail.com>:
>>>>> Hello,
>>>>>
>>>>> I have an OCaml application with some C code which (I believe) is the
>>>>> reason of some random crashes.
>>>>>
>>>>> Here is the code - https://pastebin.com/FVtLphZu
>>>>> This function reads file at given offset, divides data into chunks and
>>>>> compute checksums and compression ratio:
>>>>> external compute_data_props
>>>>>   : string -> int -> int -> int -> (int * int * float) list =
>>>>> "compute_data_props"
>>>>>
>>>>> The problem is, after some calls to this stub, application is crashing
>>>>> at random places in OCaml code.
>>>>> I can't figure out what's going wrong, but replacing this stub with
>>>>> dummy function (which does nothing but returns some predefined list)
>>>>> eliminates the problem.
>>>>>
>>>>> What can we do to debug it? We are using OCaml 4.04.1 and Windows 10,
>>>>> and there is no other C stubs in our codebase.
>>>>>
>>>>> Thanks!
>>>>
>>>> --
>>>> Caml-list mailing list.  Subscription management and archives:
>>>> https://sympa.inria.fr/sympa/arc/caml-list
>>>> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>>>> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: [Caml-list] memory corruption using C stub

[Alexey Egorov]

I think I found a reason.
That was just stack overflow from non-tail-recursive List.map, and
OCaml on windows just crashes (C style) instead of throwing
exception...

2017-06-10 20:33 GMT+05:00 Alexey Egorov <alex.only.d@gmail.com>:
> Hello,
>
> I have an OCaml application with some C code which (I believe) is the
> reason of some random crashes.
>
> Here is the code - https://pastebin.com/FVtLphZu
> This function reads file at given offset, divides data into chunks and
> compute checksums and compression ratio:
>   external compute_data_props
>     : string -> int -> int -> int -> (int * int * float) list =
> "compute_data_props"
>
> The problem is, after some calls to this stub, application is crashing
> at random places in OCaml code.
> I can't figure out what's going wrong, but replacing this stub with
> dummy function (which does nothing but returns some predefined list)
> eliminates the problem.
>
> What can we do to debug it? We are using OCaml 4.04.1 and Windows 10,
> and there is no other C stubs in our codebase.
>
> Thanks!

RE: [Caml-list] memory corruption using C stub

[David Allsopp]

Alexey Egorov wrote:
> I think I found a reason.
> That was just stack overflow from non-tail-recursive List.map, and OCaml
> on windows just crashes (C style) instead of throwing exception...

You'll be pleased to know that will be fixed in 4.06.0! https://github.com/ocaml/ocaml/pull/938. 


David