* Testing swap
@ 2021-08-12 13:20 Matthew Wilcox
2021-08-13 6:47 ` Christoph Hellwig
0 siblings, 1 reply; 2+ messages in thread
From: Matthew Wilcox @ 2021-08-12 13:20 UTC (permalink / raw)
To: fstests; +Cc: sfrench, linux-fsdevel
All the current xfstests for swap only test setting up a swap partition,
not actually doing swap to it. That's why this bug went unnoticed for
two years and why Steve French is still under the impression that
swap-over-CIFS works (it doesn't). Could somebody adapt Dave's test
into the xfstests framework?
----- Forwarded message from David Howells <dhowells@redhat.com> -----
Date: Thu, 12 Aug 2021 12:57:50 +0100
From: David Howells <dhowells@redhat.com>
To: willy@infradead.org
Cc: dhowells@redhat.com, trond.myklebust@primarydata.com,
darrick.wong@oracle.com, hch@lst.de, jlayton@kernel.org,
sfrench@samba.org, torvalds@linux-foundation.org,
linux-nfs@vger.kernel.org, linux-mm@kvack.org,
linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: [PATCH 1/2] nfs: Fix write to swapfile failure due to
generic_write_checks()
User-Agent: StGit/0.23
Trying to use a swapfile on NFS results in every DIO write failing with
ETXTBSY because generic_write_checks(), as called by nfs_direct_write()
from nfs_direct_IO(), forbids writes to swapfiles.
Fix this by introducing a new kiocb flag, IOCB_SWAP, that's set by the swap
code to indicate that the swapper is doing this operation and so overrule
the check in generic_write_checks().
Without this patch, the following is seen:
Write error on dio swapfile (3800334336)
Altering __swap_writepage() to show the error shows:
Write error (-26) on dio swapfile (3800334336)
Tested by swapping off all swap partitions and then swapping on a prepared
NFS file (CONFIG_NFS_SWAP=y is also needed). Enough copies of the
following program then need to be run to force swapping to occur (at least
one per gigabyte of RAM):
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/mman.h>
int main()
{
unsigned int pid = getpid(), iterations = 0;
size_t i, j, size = 1024 * 1024 * 1024;
char *p;
bool mismatch;
p = malloc(size);
if (!p) {
perror("malloc");
exit(1);
}
srand(pid);
for (i = 0; i < size; i += 4)
*(unsigned int *)(p + i) = rand();
do {
for (j = 0; j < 16; j++) {
for (i = 0; i < size; i += 4096)
*(unsigned int *)(p + i) += 1;
iterations++;
}
mismatch = false;
srand(pid);
for (i = 0; i < size; i += 4) {
unsigned int r = rand();
unsigned int v = *(unsigned int *)(p + i);
if (i % 4096 == 0)
v -= iterations;
if (v != r) {
fprintf(stderr, "mismatch %zx: %x != %x (diff %x)\n",
i, v, r, v - r);
mismatch = true;
}
}
} while (!mismatch);
exit(1);
}
----- End forwarded message -----
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: Testing swap
2021-08-12 13:20 Testing swap Matthew Wilcox
@ 2021-08-13 6:47 ` Christoph Hellwig
0 siblings, 0 replies; 2+ messages in thread
From: Christoph Hellwig @ 2021-08-13 6:47 UTC (permalink / raw)
To: Matthew Wilcox, David Howells; +Cc: fstests, sfrench, linux-fsdevel
On Thu, Aug 12, 2021 at 02:20:13PM +0100, Matthew Wilcox wrote:
> All the current xfstests for swap only test setting up a swap partition,
> not actually doing swap to it. That's why this bug went unnoticed for
> two years and why Steve French is still under the impression that
> swap-over-CIFS works (it doesn't). Could somebody adapt Dave's test
> into the xfstests framework?
Dave, can you just submit your test to xfstests?
> ----- Forwarded message from David Howells <dhowells@redhat.com> -----
>
> Date: Thu, 12 Aug 2021 12:57:50 +0100
> From: David Howells <dhowells@redhat.com>
> To: willy@infradead.org
> Cc: dhowells@redhat.com, trond.myklebust@primarydata.com,
> darrick.wong@oracle.com, hch@lst.de, jlayton@kernel.org,
> sfrench@samba.org, torvalds@linux-foundation.org,
> linux-nfs@vger.kernel.org, linux-mm@kvack.org,
> linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
> Subject: [PATCH 1/2] nfs: Fix write to swapfile failure due to
> generic_write_checks()
> User-Agent: StGit/0.23
>
> Trying to use a swapfile on NFS results in every DIO write failing with
> ETXTBSY because generic_write_checks(), as called by nfs_direct_write()
> from nfs_direct_IO(), forbids writes to swapfiles.
>
> Fix this by introducing a new kiocb flag, IOCB_SWAP, that's set by the swap
> code to indicate that the swapper is doing this operation and so overrule
> the check in generic_write_checks().
>
> Without this patch, the following is seen:
>
> Write error on dio swapfile (3800334336)
>
> Altering __swap_writepage() to show the error shows:
>
> Write error (-26) on dio swapfile (3800334336)
>
> Tested by swapping off all swap partitions and then swapping on a prepared
> NFS file (CONFIG_NFS_SWAP=y is also needed). Enough copies of the
> following program then need to be run to force swapping to occur (at least
> one per gigabyte of RAM):
>
> #include <stdbool.h>
> #include <stdio.h>
> #include <stdlib.h>
> #include <unistd.h>
> #include <sys/mman.h>
> int main()
> {
> unsigned int pid = getpid(), iterations = 0;
> size_t i, j, size = 1024 * 1024 * 1024;
> char *p;
> bool mismatch;
> p = malloc(size);
> if (!p) {
> perror("malloc");
> exit(1);
> }
> srand(pid);
> for (i = 0; i < size; i += 4)
> *(unsigned int *)(p + i) = rand();
> do {
> for (j = 0; j < 16; j++) {
> for (i = 0; i < size; i += 4096)
> *(unsigned int *)(p + i) += 1;
> iterations++;
> }
> mismatch = false;
> srand(pid);
> for (i = 0; i < size; i += 4) {
> unsigned int r = rand();
> unsigned int v = *(unsigned int *)(p + i);
> if (i % 4096 == 0)
> v -= iterations;
> if (v != r) {
> fprintf(stderr, "mismatch %zx: %x != %x (diff %x)\n",
> i, v, r, v - r);
> mismatch = true;
> }
> }
> } while (!mismatch);
> exit(1);
> }
>
> ----- End forwarded message -----
---end quoted text---
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2021-08-13 6:49 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-08-12 13:20 Testing swap Matthew Wilcox
2021-08-13 6:47 ` Christoph Hellwig
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.