I've verified this in 2.4.21-pre5 by code inspection and can trigger the
problem on 2.4.18. It appears to have been fixed in 2.5.
The folowing code vmscan.c assumes that there is available swap space.
/*
* this is the non-racy check for busy page.
*/
if (!page->mapping || !is_page_cache_freeable(page)) {
spin_unlock(&pagecache_lock);
UnlockPage(page);
page_mapped:
if (--max_mapped >= 0)
continue;
/*
* Alert! We've found too many mapped pages on the
* inactive list, so we start swapping out now!
*/
spin_unlock(&pagemap_lru_lock);
swap_out(priority, gfp_mask, classzone);
return nr_pages;
}
If there is no swap space, then unfreeable pages are left on the
inactive queue and the vmtree is walked rather than going through the
rest of the inactive queue. I believe something like
/*
* this is the non-racy check for busy page.
*/
if (!page->mapping || !is_page_cache_freeable(page)) {
spin_unlock(&pagecache_lock);
UnlockPage(page);
page_mapped:
/* If we don't have any swap space left, there
is no reason to worry about pages that do
not have swap associated with them, there
is nothing we can do about it. */
if (!page->mapping && !swap_avail()) {
/* Let's make the page active since we
cannot swap it out. It get's it off
the inactive list. */
spin_unlock(&pagemap_lru_lock);
activate_page(page);
ClearPageReferenced(page);
spin_lock(&pagemap_lru_lock);
continue;
}
if (--max_mapped >= 0)
continue;
/*
* Alert! We've found too many mapped pages on the
* inactive list, so we start swapping out now!
*/
spin_unlock(&pagemap_lru_lock);
swap_out(priority, gfp_mask, classzone);
return nr_pages;
}
will work better when there is no swap space available. If this change
is made, it may also be necessary to limit refill_inactive to prevent it
from using too much cpu. This bug can be triggered with the attached
code and the correct parameters. In particular on a 3 gigabyte machine
with no swap,
for i in $(seq 0 9); do dd if=/dev/zero of=file$i bs=1024k count=512; done
killmm 1032735283 2 9
Usually causes an out of memory error when there is hundreds of
megabytes of cache.
Ross
/*
* killmm.c attempts to exploit bugs in the mm to cause a crash or
* other undesired behaviour.
* Copyright (C) 2002 Google
* Written by Ross Biro
*
*/
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <malloc.h>
#include <errno.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <asm/page.h>
/* 512 Meg */
#define BLOCKSIZE 512*1024*1024
#define USEMLOCK 1
int main(int argc, char *argv[]) {
size_t memory = atoi(argv[1]);
int blocks = atoi(argv[2]);
int maxfiles = atoi(argv[3]);
unsigned char *cptr;
int i, j, k;
void **ptrs = (void **)malloc(blocks * sizeof(*ptrs));
if (ptrs == NULL) {
fprintf (stderr, "Unable to allocate %d bytes: %s\n",
sizeof(*ptrs) * blocks,
strerror(errno));
return -1;
}
/* The first thing we do is allocate a bunch of memory. */
cptr = (unsigned char *)malloc(memory);
if (cptr == NULL) {
fprintf (stderr, "Unable to allocate %d bytes: %s\n", memory,
strerror(errno));
return -1;
}
/* now we want to make it all dirty. */
for (i = 0; i < memory; i++) {
cptr[i] = (unsigned char)(i&0xff);
if ((i & 0xffffff) == 0) {
printf ("Initializing memory: %d\n", i);
}
}
/* Now we have a bunch of dirty memory. Map in huge files. */
for (i = 0; i < maxfiles; i++) {
char filename[1024];
int fd;
int ind = i%blocks;
if (ptrs[ind] != NULL) {
printf ("Unmapping block %d @ %08X\n", ind, ptrs[ind]);
#ifdef USEMLOCK
munlock(ptrs[ind], BLOCKSIZE);
#endif
munmap(ptrs[ind], BLOCKSIZE);
}
sprintf (filename, "file%d", i);
printf ("Loading file %s into slot %d\n",
filename, ind);
fd = open (filename, O_RDONLY);
if (fd < 0) {
fprintf (stderr, "Unable to open %s: %s\n", filename, strerror(errno));
return -1;
}
ptrs[ind] = mmap (NULL, BLOCKSIZE, PROT_READ, MAP_PRIVATE, fd, 0);
if (ptrs[ind] == NULL) {
fprintf (stderr, "Unable to map file %s: %s\n",
filename, strerror(errno));
return -1;
}
#ifdef USEMLOCK
if (mlock(ptrs[ind], BLOCKSIZE) < 0) {
fprintf (stderr, "Unable to lock mem for %s: %s\n",
filename, strerror(errno));
return -1;
}
#else
// Page in the memory the old fashioned way.
for (j = 0; j <BLOCKSIZE; j+= PAGE_SIZE) {
k += ((char *)ptrs[ind])[j];
}
#endif
printf ("Block %d at %08X\n", ind, ptrs[ind]);
close(fd);
}
}
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Mar 15 2003 - 22:00:22 EST