Skip to contentSkip to conversation

whole system lock-up on low memory

DISCUSSION Ulrich Lukas 2 participants 3 messages May 10, 2009 – May 12, 2009

lore
Share & tools
lore
#0 lore raw

Hi,

vanilla linux 2.6.29.3, AMD64, tested on openSUSE 11.1 and Kubuntu 9.04.

I observe the following behaviour:

When any user application (non-kernel, non-root) consumes all the available system memory, the system freezes completely instead of any application being killed by the oom-killer.

The mouse pointer stalls, and there is a seemingly endless loop of hard-disc access, even when no swap space on any harddisc is activated.

IMO, this is severe, because any application can practically crash the system (e.g. in case of a memory-leak), causing data loss in case of unsaved data.

To reproduce this, I've attached a small C++ utility (compiles with g++ memory_overcommit.cc -o memory_overcommit.bin) which allocates chunks of memory of user defined size.

On the Ubuntu system, the system freeze can be observed with swap enabled on a cryptographic swap partition (dm-crypt; /etc/crypttab). With openSUSE the lock-up also occurs with deactivated swap. (swapoff -a).

Tested from a KDE terminal window as regular urser, the steps to reproduce are:

* use cryptographic swap partition or disable swap
(maybe also reproducible with normal swap; but apparently not on Suse)

* compile and invoke the attached code:
./memory_overcommit.bin

* enter a number (in MiB) of memory that is slightly smaller than the available memory and press "enter" key once.

* enter a smaller number (minimum 1 MiB) and confirm again, do the same again,..., successively approaching the limit of available memory with smaller chunks.

* finally, when most of the memory/buffers/cache are used-up, the system becomes unresponsive and constant, heavy harddisk-access commences.

Sometimes, killing the X-server or shutting down via hotkeys works after several minutes of waiting, but this is not consistent.

I've reported this issue on the Ubuntu bug tracker before:
https://bugs.launchpad.net/ubuntu/+bug/283420

but as stated above, I also had the problem on another system.

I'd be glad if I could help if you need further information.

Attachement: one file, memory_overcommit.cc
(also here: http://datenparkplatz.de/DiesUndDas/memory_overcommit.cc)
---
#include <cassert>
#include <cerrno>
#include <clocale>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <limits>

using std::exit;
using std::numeric_limits;
using std::printf;
using std::size_t;
using std::strlen;

void read_answer(char* buffer, size_t buffer_size)

{
        assert(buffer_size > 1);

        char const* ret = std::fgets(buffer, buffer_size, stdin);
        if(ret == 0)
        {
                printf("\nSorry, there was an error while reading the

answer. This program will now terminate.\n");

                exit(EXIT_FAILURE);

        }

        else if(strlen(buffer) == buffer_size - 1 && buffer[buffer_size

- 2] != '\n')
{

printf("Sorry, you answer is too long (possibly out of
range). This program will now terminate.\n");

                exit(EXIT_FAILURE);

        }

}

size_t determine_size(char const* s)

{
        assert(strlen(s) > 0);

        char* endptr = 0;
        errno = 0;
        unsigned long v = std::strtoul(s, &endptr, 10);
        if(strlen(s) == 1 || endptr != s + strlen(s) - 1)
        {
                printf("\nSorry, your answer does not appear to be

valid. This program will now terminate.\n");

                exit(EXIT_FAILURE);

        }

        else if((v == numeric_limits<unsigned long>::max() && errno ==
ERANGE) || v == 0
                || numeric_limits<unsigned long>::max() / (1024 * 1024)

< v)
{

printf("\nSorry, that value is out of range. This

program will now terminate.\n");
                exit(EXIT_FAILURE);

        }

        v *= 1024 * 1024;


        assert(numeric_limits<size_t>::is_specialized);
        if(v > numeric_limits<size_t>::max() )
        {
                printf("\nSorry, that value is out of range. This
program will now terminate.\n");
                exit(EXIT_FAILURE);

        }


        return v;
}

bool inquire_repeat(size_t& v)

{
        printf("\nShould a new allocation be made? You can enter:\n"
                        "==> \"No\" to quit,\n"
                        "==> Any Number of MiB to change the chunk size
and continue, or\n"
                        "==> Hit return to continue with last chunk

size.\n\nYour answer: ");

        char answer[20];
        read_answer(answer, sizeof(answer) );
        if(std::strcmp(answer, "\n") == 0) {
                return true;
        }
        else if(std::strcmp(answer, "No\n") == 0) {
                return false;
        }

        v = determine_size(answer);
        return true;
}

int main() {
#ifdef _POSIX_C_SOURCE
std::setlocale(LC_ALL, ""); // Prepare thousands grouping in output. #endif

printf("\n\nThis program allows to repeatedly allocate chunks of
memory of user-specified size. "
"\nAfter each allocation the user can choose to repeat or
to quit the program.\n\nFirst, please enter now "

"the amount of memory in MiB (1024 * 1024 bytes) \nto

allocate in each round: ");
        char answer[20];

        read_answer(answer, sizeof(answer) );


        size_t alloc_size = determine_size(answer);

        size_t total = 0;
        do {
                printf("\n   >>> Starting to allocate chunk...\n");
                std::fflush(stdout);
                void* p = std::malloc(alloc_size);
                if(!p) {
                        printf("\n   > The last memory allocation failed."
                               "\n   > This means the system reported
the out-of-memory condition orderly."
                               "\nThis program will now terminate.\n");

                        exit(0);

                }

                std::memset(p, 0, alloc_size);

                printf("   >>> A chunk was just allocated! <<<\n");
                if(numeric_limits<size_t>::max() - total >= alloc_size)
                {
                        total += alloc_size;

                        #ifdef _POSIX_C_SOURCE
                        char const* fmt_string = "   The total number of
bytes allocated is now %'zu.\n";
                        #else
                        char const* fmt_string = "   The total number of
bytes allocated is now %zu.\n";
                        #endif

                        printf(fmt_string, total);
                }
                else {
                        #ifdef _POSIX_C_SOURCE
                        char const* fmt_string = "   More than %'zu

bytes have been allocated in total by now.\n";
#else
char const* fmt_string = " More than %zu bytes
have been allocated in total by now.\n";

                        #endif

                        printf(fmt_string, total);
                        total = numeric_limits<size_t>::max();
                }
        } while(inquire_repeat(alloc_size) );

        printf("\nQuit was requested. This program will now terminate.\n");
}
#1 lore raw
Quoted 41 lines

Hi,

vanilla linux 2.6.29.3, AMD64, tested on openSUSE 11.1 and Kubuntu 9.04.

I observe the following behaviour:

When any user application (non-kernel, non-root) consumes all the available system memory, the system freezes completely instead of any application being killed by the oom-killer.

The mouse pointer stalls, and there is a seemingly endless loop of hard-disc access, even when no swap space on any harddisc is activated.

IMO, this is severe, because any application can practically crash the system (e.g. in case of a memory-leak), causing data loss in case of unsaved data.

To reproduce this, I've attached a small C++ utility (compiles with g++ memory_overcommit.cc -o memory_overcommit.bin) which allocates chunks of memory of user defined size.

On the Ubuntu system, the system freeze can be observed with swap enabled on a cryptographic swap partition (dm-crypt; /etc/crypttab). With openSUSE the lock-up also occurs with deactivated swap. (swapoff -a).

Tested from a KDE terminal window as regular urser, the steps to reproduce are:

* use cryptographic swap partition or disable swap
(maybe also reproducible with normal swap; but apparently not on Suse)

* compile and invoke the attached code:
./memory_overcommit.bin

* enter a number (in MiB) of memory that is slightly smaller than the available memory and press "enter" key once.

Quoted 6 lines

* enter a smaller number (minimum 1 MiB) and confirm again, do the same again,..., successively approaching the limit of available memory with smaller chunks.

* finally, when most of the memory/buffers/cache are used-up, the system becomes unresponsive and constant, heavy harddisk-access commences.

umm, I don't reproduce this.
oom-killer kill the reproduce prgram ASAP.

system don't enter unresponsible state.

my test environment:

kernel: 2.6.30-rc4-mmotm
CPU: ia64 x 8
MEM: 8GB

Quoted 200 lines

Sometimes, killing the X-server or shutting down via hotkeys works after several minutes of waiting, but this is not consistent.

I've reported this issue on the Ubuntu bug tracker before:
https://bugs.launchpad.net/ubuntu/+bug/283420

but as stated above, I also had the problem on another system.

I'd be glad if I could help if you need further information.

Attachement: one file, memory_overcommit.cc
(also here: http://datenparkplatz.de/DiesUndDas/memory_overcommit.cc)
---
#include <cassert>
#include <cerrno>
#include <clocale>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <limits>

using std::exit;
using std::numeric_limits;
using std::printf;
using std::size_t;
using std::strlen;

void read_answer(char* buffer, size_t buffer_size)

{
        assert(buffer_size > 1);

        char const* ret = std::fgets(buffer, buffer_size, stdin);
        if(ret == 0)
        {
                printf("\nSorry, there was an error while reading the

answer. This program will now terminate.\n");

                exit(EXIT_FAILURE);

        }

        else if(strlen(buffer) == buffer_size - 1 && buffer[buffer_size

- 2] != '\n')
{

printf("Sorry, you answer is too long (possibly out of
range). This program will now terminate.\n");

                exit(EXIT_FAILURE);

        }

}

size_t determine_size(char const* s)

{
        assert(strlen(s) > 0);

        char* endptr = 0;
        errno = 0;
        unsigned long v = std::strtoul(s, &endptr, 10);
        if(strlen(s) == 1 || endptr != s + strlen(s) - 1)
        {
                printf("\nSorry, your answer does not appear to be

valid. This program will now terminate.\n");

                exit(EXIT_FAILURE);

        }

        else if((v == numeric_limits<unsigned long>::max() && errno ==
ERANGE) || v == 0
                || numeric_limits<unsigned long>::max() / (1024 * 1024)

< v)
{

printf("\nSorry, that value is out of range. This

program will now terminate.\n");
                exit(EXIT_FAILURE);

        }

        v *= 1024 * 1024;


        assert(numeric_limits<size_t>::is_specialized);
        if(v > numeric_limits<size_t>::max() )
        {
                printf("\nSorry, that value is out of range. This
program will now terminate.\n");
                exit(EXIT_FAILURE);

        }


        return v;
}

bool inquire_repeat(size_t& v)

{
        printf("\nShould a new allocation be made? You can enter:\n"
                        "==> \"No\" to quit,\n"
                        "==> Any Number of MiB to change the chunk size
and continue, or\n"
                        "==> Hit return to continue with last chunk

size.\n\nYour answer: ");

        char answer[20];
        read_answer(answer, sizeof(answer) );
        if(std::strcmp(answer, "\n") == 0) {
                return true;
        }
        else if(std::strcmp(answer, "No\n") == 0) {
                return false;
        }

        v = determine_size(answer);
        return true;
}

int main() {
#ifdef _POSIX_C_SOURCE
std::setlocale(LC_ALL, ""); // Prepare thousands grouping in output. #endif

printf("\n\nThis program allows to repeatedly allocate chunks of
memory of user-specified size. "
"\nAfter each allocation the user can choose to repeat or
to quit the program.\n\nFirst, please enter now "

"the amount of memory in MiB (1024 * 1024 bytes) \nto

allocate in each round: ");
        char answer[20];

        read_answer(answer, sizeof(answer) );


        size_t alloc_size = determine_size(answer);

        size_t total = 0;
        do {
                printf("\n   >>> Starting to allocate chunk...\n");
                std::fflush(stdout);
                void* p = std::malloc(alloc_size);
                if(!p) {
                        printf("\n   > The last memory allocation failed."
                               "\n   > This means the system reported
the out-of-memory condition orderly."
                               "\nThis program will now terminate.\n");

                        exit(0);

                }

                std::memset(p, 0, alloc_size);

                printf("   >>> A chunk was just allocated! <<<\n");
                if(numeric_limits<size_t>::max() - total >= alloc_size)
                {
                        total += alloc_size;

                        #ifdef _POSIX_C_SOURCE
                        char const* fmt_string = "   The total number of
bytes allocated is now %'zu.\n";
                        #else
                        char const* fmt_string = "   The total number of
bytes allocated is now %zu.\n";
                        #endif

                        printf(fmt_string, total);
                }
                else {
                        #ifdef _POSIX_C_SOURCE
                        char const* fmt_string = "   More than %'zu

bytes have been allocated in total by now.\n";
#else
char const* fmt_string = " More than %zu bytes
have been allocated in total by now.\n";

                        #endif

                        printf(fmt_string, total);
                        total = numeric_limits<size_t>::max();
                }
        } while(inquire_repeat(alloc_size) );

        printf("\nQuit was requested. This program will now terminate.\n");
}

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#2 lore raw
Quoted 6 lines · KOSAKI Motohiro

mm, I don't reproduce this.
my test environment:

kernel: 2.6.30-rc4-mmotm
CPU: ia64 x 8
MEM: 8GB

I've never tested on an ia64 system, only on x86 and x86-64.

Sometimes more than one try is needed to yield the behaviour.

And I only had the problem with cryptographic or disabled swap.

E.g. with an /etc/crypttab like this:
swapMapping /dev/sdxx /dev/urandom swap

and /etc/fstab:
/dev/mapper/swapMapping none swap sw 0 0