mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* 2.4.22 O_DIRECT memory leak?!?
@ 2003-10-08 16:09 Magnus Andersson
  2003-10-09 19:37 ` Marcelo Tosatti
  0 siblings, 1 reply; 3+ messages in thread
From: Magnus Andersson @ 2003-10-08 16:09 UTC (permalink / raw)
  To: linux-kernel

[-- Attachment #1: Type: text/plain, Size: 401 bytes --]

Hello!

If I open one file a lot of times using the flag O_DIRECT,
memory seems to be be lost and never given back to the system.
This is happening on some kernels, see below for which ones I tried.

Attached program will produce this behavior, also attached 
is the output from vmstat while running the program.

Affected:
2.4.22
2.4.22-ac4
2.4.23-pre6

Not affected:
2.4.22-aa1
2.6.0-test6

/Magnus

[-- Attachment #2: open_odirect.c --]
[-- Type: text/plain, Size: 531 bytes --]

#define _GNU_SOURCE

#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>

#define NFDS 1000

#define ERR(args...) fprintf(stderr, args);

int main()
{
  int i, fds[NFDS];

  for(i = 0; i < NFDS; i++) {
    if((fds[i] = open("./file", O_RDONLY | O_DIRECT)) < 0) {
      ERR("main(): open(): %s\n", strerror(errno));
      exit(1);
    }
  }
  
  for(i = 0; i < NFDS; i++) {
    if(close(fds[i]) < 0) {
      ERR("main(): close(): %s\n", strerror(errno));
      exit(1);
    }
  }
}

[-- Attachment #3: vmstat.output --]
[-- Type: text/plain, Size: 955 bytes --]

procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in    cs us sy id wa
 0  0      0 498324    916   7632    0    0    54     9  109    22  1 19 80  0
 0  0      0 498320    916   7632    0    0     0     0  101     2  0  0 100  0
 0  0      0 498320    916   7632    0    0     0     0  101     2  0  0 100  0
 0  0      0 498320    916   7632    0    0     0     0  101     2  0  0 100  0
 0  0      0 498320    916   7632    0    0     0     0  101     2  0  0 100  0
 1  0      0 485072    916   7632    0    0     0     0  104    11  0  6 94  0
 0  0      0 386276    916   7632    0    0     0     0  116     7  0 41 59  0
 0  0      0 386276    916   7632    0    0     0     0  101     2  0  0 100  0
 0  0      0 386276    916   7632    0    0     0     0  101     2  0  0 100  0
 0  0      0 386276    916   7632    0    0     0     0  101     2  0  0 100  0

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: 2.4.22 O_DIRECT memory leak?!?
  2003-10-08 16:09 2.4.22 O_DIRECT memory leak?!? Magnus Andersson
@ 2003-10-09 19:37 ` Marcelo Tosatti
  2003-10-10 11:23   ` Magnus Andersson
  0 siblings, 1 reply; 3+ messages in thread
From: Marcelo Tosatti @ 2003-10-09 19:37 UTC (permalink / raw)
  To: Magnus Andersson; +Cc: linux-kernel



On Wed, 8 Oct 2003, Magnus Andersson wrote:

> Hello!
> 
> If I open one file a lot of times using the flag O_DIRECT,
> memory seems to be be lost and never given back to the system.
> This is happening on some kernels, see below for which ones I tried.
> 
> Attached program will produce this behavior, also attached 
> is the output from vmstat while running the program.

Magnus,

That memory will be freed as soon as there's memory pressure so its not a 
memory leak.




^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: 2.4.22 O_DIRECT memory leak?!?
  2003-10-09 19:37 ` Marcelo Tosatti
@ 2003-10-10 11:23   ` Magnus Andersson
  0 siblings, 0 replies; 3+ messages in thread
From: Magnus Andersson @ 2003-10-10 11:23 UTC (permalink / raw)
  To: Marcelo Tosatti; +Cc: linux-kernel

[-- Attachment #1: Type: text/plain, Size: 1123 bytes --]

On Thu, Oct 09, 2003 at 04:37:52PM -0300, Marcelo Tosatti wrote:
> 
> 
> On Wed, 8 Oct 2003, Magnus Andersson wrote:
> 
> > Hello!
> > 
> > If I open one file a lot of times using the flag O_DIRECT,
> > memory seems to be be lost and never given back to the system.
> > This is happening on some kernels, see below for which ones I tried.
> > 
> > Attached program will produce this behavior, also attached 
> > is the output from vmstat while running the program.
> 
> Magnus,
> 
> That memory will be freed as soon as there's memory pressure so its not a 
> memory leak.

Ok. It's not a leak. But it looks like the memory never goes back to the system.
The kernel has killed some processes of mine, thats why I suspect this.

Attached are a program, malloc_mem.
Before running open_odirect, malloc_mem succeeds in allocating 450 000 000 bytes.
After running open_odirect, malloc_mem fails.

It is working on the 2.4.22-aa1 kernel, but not on 2.4.22

The test is run with this amount of memory, and no swap.

Memory: 515752k/524204k available (1593k kernel code, 8064k reserved, 334k data, 264k init, 0k highmem)

/Magnus

[-- Attachment #2: run.log --]
[-- Type: text/plain, Size: 112 bytes --]

foo:~> ./malloc_mem 
All ok!
foo:~> ./open_odirect 
foo:~> ./malloc_mem
Could not malloc 450000000 bytes
foo:~>

[-- Attachment #3: malloc_mem.c --]
[-- Type: text/plain, Size: 266 bytes --]

#include <stdio.h>

#define MEM_SIZE 450000000

int main()
{
  int i;
  char *mem;
  
  mem = (char *)malloc(MEM_SIZE);

  if(mem == NULL) {
    printf("Could not malloc %d bytes\n", MEM_SIZE);
    exit(1);
  }

  memset(mem, 1, MEM_SIZE);

  printf("All ok!\n");
}

[-- Attachment #4: vmstat.output --]
[-- Type: text/plain, Size: 2388 bytes --]

procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in    cs us sy id wa
 0  0      0 496804    940   7936    0    0     0     0  101     2  0  0 100  0
 0  0      0 496800    940   7940    0    0     0     0  101     6  0  0 100  0
 0  0      0 496800    940   7940    0    0     0     0  101     4  0  0 100  0
 0  0      0 496800    940   7940    0    0     0     0  101     2  0  0 100  0
 0  0      0 496800    940   7940    0    0     0     0  101     2  0  0 100  0
 1  0      0 438912    940   7940    0    0     0     0  104    14 71 28  1  0
 1  0      0 379896    940   7940    0    0     0     0  101     4 76 24  0  0
 1  0      0 320880    940   7940    0    0     0     0  101     4 78 22  0  0
 1  0      0 261868    940   7940    0    0     0     0  101     2 74 26  0  0
 1  0      0 202848    940   7940    0    0     0     0  101     2 77 23  0  0
 1  0      0 143836    940   7940    0    0     0     0  101     2 73 27  0  0
 1  0      0  84856    940   7940    0    0     0     0  101     8 76 24  0  0
 0  0      0 496716    940   7940    0    0     0     0  105    13 35 17 49  0
 0  0      0 496716    940   7940    0    0     0     0  101     2  0  0 100  0
 0  0      0 496716    940   7940    0    0     0     0  101     2  0  0 100  0
 0  0      0 496716    940   7940    0    0     0     0  107    16  0  0 100  0
 0  0      0 496716    940   7940    0    0     0     0  104    11  0  0 100  0
 1  0      0 461916    940   7940    0    0     0   128  137    27  0 17 83  0
 0  0      0 385028    940   7940    0    0     0     0  103     7  0 50 50  0
 0  0      0 385028    940   7940    0    0     0     0  101     2  0  0 100  0
 0  0      0 385028    940   7940    0    0     0     0  101     4  0  0 100  0
 0  0      0 385028    940   7940    0    0     0     0  107    20  0  0 100  0
 0  0      0 385028    940   7940    0    0     0    76  126    22  0  0 100  0
 0  0      0 385028    940   7940    0    0     0     0  101     2  0  0 100  0
 0  0      0 385028    940   7940    0    0     0     0  101     2  0  0 100  0
 0  0      0 385024    940   7944    0    0     0     0  101     2  0  0 100  0
 0  0      0 385024    940   7944    0    0     0     0  101     4  0  0 100  0
 0  0      0 385024    940   7944    0    0     0     0  101     4  0  0 100  0

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2003-10-10 11:24 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2003-10-08 16:09 2.4.22 O_DIRECT memory leak?!? Magnus Andersson
2003-10-09 19:37 ` Marcelo Tosatti
2003-10-10 11:23   ` Magnus Andersson

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®