block: prevent multiwrite_merge from creating too large iovecs

If we go over the maximum number of iovecs support by syscall we get back EINVAL from the kernel which translate to I/O errors for the guest.

Signed-off-by Christoph Hellwig <[email protected]>

Am 19.01.2010 22:15, schrieb Christoph Hellwig:

Quoted 4 lines

If we go over the maximum number of iovecs support by syscall we get back EINVAL from the kernel which translate to I/O errors for the guest.

Signed-off-by Christoph Hellwig <[email protected]>

Is this really enough? We don't check for IOV_MAX in any other place, so can't we get a too big request directly from virtio-blk?

Quoted 4 lines

Am 19.01.2010 22:15, schrieb Christoph Hellwig:

Quoted 3 lines

If we go over the maximum number of iovecs support by syscall we get back EINVAL from the kernel which translate to I/O errors for the guest.

Quoted 4 lines

Quoted 1 line

Signed-off-by Christoph Hellwig <[email protected]>

Is this really enough? We don't check for IOV_MAX in any other place, so can't we get a too big request directly from virtio-blk?

Currently the virtqueue is limited to 1024 iovecs, but I plan to put in some better infrastructure to deal with the queue limit. For now this patch fixes an issue that we see with real life setups.

To underline that it's a backend/platform dependent thing: Your patch breaks the mingw build for me.

Quoted 3 lines · Kevin Wolf

To underline that it's a backend/platform dependent thing: Your patch breaks the mingw build for me.

Actually that's because mingw is the usual piece of crap and doesn't actually have any of the vector support you can expect from a normal Unix system.

I can either throw in an #ifdef IOV_MAX around the check or fake one up for mingw. Does any of the maintainers have a preference for either variant?

Quoted 4 lines

I can either throw in an #ifdef IOV_MAX around the check or fake one up for mingw. Does any of the maintainers have a preference for either variant?

grep for CONFIG_IOVEC in qemu-common.h and add a #define IOV_MAX.

mingw doesn't have iovec so we introduce a compat version.

If we go over the maximum number of iovecs support by syscall we get back EINVAL from the kernel which translate to I/O errors for the guest.

Signed-off-by Christoph Hellwig<[email protected]>

Index: qemu/block.c
===================================================================

block.c +4 -0

@@ -1711,6 +1711,10 @@ static int multiwrite_merge(BlockDriverS
              merge = bs->drv->bdrv_merge_requests(bs,&reqs[outidx],&reqs[i]);
          }
 
+        if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1>  IOV_MAX) {
+            merge = 0;
+        }
+
          if (merge) {
              size_t size;
              QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));