[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH 16/24] DAX/unmap: virtiofsd: Add VHOST_USER_SLAVE_FS_IO
From: |
Stefan Hajnoczi |
Subject: |
Re: [PATCH 16/24] DAX/unmap: virtiofsd: Add VHOST_USER_SLAVE_FS_IO |
Date: |
Thu, 11 Feb 2021 14:17:34 +0000 |
On Tue, Feb 09, 2021 at 07:02:16PM +0000, Dr. David Alan Gilbert (git) wrote:
> From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
>
> Define a new slave command 'VHOST_USER_SLAVE_FS_IO' for a
> client to ask qemu to perform a read/write from an fd directly
> to GPA.
>
> Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
> ---
> docs/interop/vhost-user.rst | 11 +++
> hw/virtio/trace-events | 6 ++
> hw/virtio/vhost-user-fs.c | 84 +++++++++++++++++++++++
> hw/virtio/vhost-user.c | 4 ++
> include/hw/virtio/vhost-user-fs.h | 2 +
> subprojects/libvhost-user/libvhost-user.h | 1 +
> 6 files changed, 108 insertions(+)
>
> diff --git a/docs/interop/vhost-user.rst b/docs/interop/vhost-user.rst
> index 1deedd3407..821712f4a2 100644
> --- a/docs/interop/vhost-user.rst
> +++ b/docs/interop/vhost-user.rst
> @@ -1452,6 +1452,17 @@ Slave message types
> multiple chunks can be unmapped in one command.
> A reply is generated indicating whether unmapping succeeded.
>
> +``VHOST_USER_SLAVE_FS_IO``
> + :id: 9
> + :equivalent ioctl: N/A
> + :slave payload: fd + n * (offset + address + len)
Please clarify the payload representation. This is not enough for
someone to implement the spec.
> + :master payload: N/A
> +
> + Requests that the QEMU performs IO directly from an fd to guest memory
To avoid naming a particular VMM:
s/the QEMU performs IO/IO be performed/
> + on behalf of the daemon; this is normally for a case where a memory region
> + isn't visible to the daemon. slave payload has flags which determine
> + the direction of IO operation.
Please document the payload flags in the spec.
> +
> .. _reply_ack:
>
> VHOST_USER_PROTOCOL_F_REPLY_ACK
> diff --git a/hw/virtio/trace-events b/hw/virtio/trace-events
> index c62727f879..20557a078e 100644
> --- a/hw/virtio/trace-events
> +++ b/hw/virtio/trace-events
> @@ -53,6 +53,12 @@ vhost_vdpa_get_features(void *dev, uint64_t features)
> "dev: %p features: 0x%"PRI
> vhost_vdpa_set_owner(void *dev) "dev: %p"
> vhost_vdpa_vq_get_addr(void *dev, void *vq, uint64_t desc_user_addr,
> uint64_t avail_user_addr, uint64_t used_user_addr) "dev: %p vq: %p
> desc_user_addr: 0x%"PRIx64" avail_user_addr: 0x%"PRIx64" used_user_addr:
> 0x%"PRIx64
>
> +# vhost-user-fs.c
> +
> +vhost_user_fs_slave_io_loop(const char *name, uint64_t owr, int is_ram, int
> is_romd, size_t size) "region %s with internal offset 0x%"PRIx64 " ram=%d
> romd=%d mrs.size=%zd"
> +vhost_user_fs_slave_io_loop_res(ssize_t transferred) "%zd"
> +vhost_user_fs_slave_io_exit(int res, size_t done) "res: %d done: %zd"
> +
> # virtio.c
> virtqueue_alloc_element(void *elem, size_t sz, unsigned in_num, unsigned
> out_num) "elem %p size %zd in_num %u out_num %u"
> virtqueue_fill(void *vq, const void *elem, unsigned int len, unsigned int
> idx) "vq %p elem %p len %u idx %u"
> diff --git a/hw/virtio/vhost-user-fs.c b/hw/virtio/vhost-user-fs.c
> index 5f2fca4d82..357bc1d04e 100644
> --- a/hw/virtio/vhost-user-fs.c
> +++ b/hw/virtio/vhost-user-fs.c
> @@ -23,6 +23,8 @@
> #include "hw/virtio/vhost-user-fs.h"
> #include "monitor/monitor.h"
> #include "sysemu/sysemu.h"
> +#include "exec/address-spaces.h"
> +#include "trace.h"
>
> /*
> * The powerpc kernel code expects the memory to be accessible during
> @@ -155,6 +157,88 @@ uint64_t vhost_user_fs_slave_unmap(struct vhost_dev *dev,
> return (uint64_t)res;
> }
>
> +uint64_t vhost_user_fs_slave_io(struct vhost_dev *dev, VhostUserFSSlaveMsg
> *sm,
> + int fd)
> +{
> + VHostUserFS *fs = VHOST_USER_FS(dev->vdev);
> + if (!fs) {
> + /* Shouldn't happen - but seen it in error paths */
> + error_report("Bad fs ptr");
> + return (uint64_t)-1;
> + }
Same pointer casting issue as with map/unmap.
> +
> + unsigned int i;
> + int res = 0;
> + size_t done = 0;
> +
> + if (fd < 0) {
> + error_report("Bad fd for map");
> + return (uint64_t)-1;
> + }
> +
> + for (i = 0; i < VHOST_USER_FS_SLAVE_ENTRIES && !res; i++) {
> + if (sm->len[i] == 0) {
> + continue;
> + }
> +
> + size_t len = sm->len[i];
> + hwaddr gpa = sm->c_offset[i];
> +
> + while (len && !res) {
> + MemoryRegionSection mrs = memory_region_find(get_system_memory(),
> + gpa, len);
> + size_t mrs_size = (size_t)int128_get64(mrs.size);
If there is a vIOMMU then the vhost-user device backend should be
restricted to just areas of guest RAM that are mapped. I think this can
be achieved by using the vhost-user-fs device's address space instead of
get_system_memory(). For example, virtio_pci_get_dma_as().
> +
> + if (!mrs_size) {
> + error_report("No guest region found for 0x%" HWADDR_PRIx,
> gpa);
> + res = -EFAULT;
> + break;
> + }
> +
> + trace_vhost_user_fs_slave_io_loop(mrs.mr->name,
> + (uint64_t)mrs.offset_within_region,
> + memory_region_is_ram(mrs.mr),
> + memory_region_is_romd(mrs.mr),
> + (size_t)mrs_size);
> +
> + void *hostptr = qemu_map_ram_ptr(mrs.mr->ram_block,
> + mrs.offset_within_region);
> + ssize_t transferred;
> + if (sm->flags[i] & VHOST_USER_FS_FLAG_MAP_R) {
The flag name is specific to map requests but it's shared with the IO
request. Perhaps rename the flags?
> + /* Read from file into RAM */
> + if (mrs.mr->readonly) {
> + res = -EFAULT;
> + break;
> + }
> + transferred = pread(fd, hostptr, mrs_size, sm->fd_offset[i]);
> + } else {
> + /* Write into file from RAM */
> + assert((sm->flags[i] & VHOST_USER_FS_FLAG_MAP_W));
The vhost-user device backend must not be able to crash the VMM. Please
use an if statement and fail the request if the flags are invalid
instead of assert().
> + transferred = pwrite(fd, hostptr, mrs_size,
> sm->fd_offset[i]);
> + }
> + trace_vhost_user_fs_slave_io_loop_res(transferred);
> + if (transferred < 0) {
> + res = -errno;
> + break;
> + }
> + if (!transferred) {
> + /* EOF */
> + break;
> + }
> +
> + done += transferred;
> + len -= transferred;
Is gpa += transferred missing so that this loop can handle crossing
MemoryRegion boundaries?
sm->fd_offset[i] also needs to be put into a local variable and
incremented by transferred each time around the loop.
signature.asc
Description: PGP signature
- Re: [PATCH 15/24] DAX: virtiofsd: Perform an unmap on destroy, (continued)
- [PATCH 19/24] DAX/unmap virtiofsd: Route unmappable reads, Dr. David Alan Gilbert (git), 2021/02/09
- [PATCH 18/24] DAX/unmap virtiofsd: Parse unmappable elements, Dr. David Alan Gilbert (git), 2021/02/09
- [PATCH 17/24] DAX/unmap virtiofsd: Add wrappers for VHOST_USER_SLAVE_FS_IO, Dr. David Alan Gilbert (git), 2021/02/09
- [PATCH 20/24] DAX/unmap virtiofsd: route unmappable write to slave command, Dr. David Alan Gilbert (git), 2021/02/09
- [PATCH 21/24] DAX:virtiofsd: implement FUSE_INIT map_alignment field, Dr. David Alan Gilbert (git), 2021/02/09
- [PATCH 22/24] vhost-user-fs: Extend VhostUserFSSlaveMsg to pass additional info, Dr. David Alan Gilbert (git), 2021/02/09
- [PATCH 16/24] DAX/unmap: virtiofsd: Add VHOST_USER_SLAVE_FS_IO, Dr. David Alan Gilbert (git), 2021/02/09
- Re: [PATCH 16/24] DAX/unmap: virtiofsd: Add VHOST_USER_SLAVE_FS_IO,
Stefan Hajnoczi <=
- [PATCH 23/24] vhost-user-fs: Implement drop CAP_FSETID functionality, Dr. David Alan Gilbert (git), 2021/02/09
[PATCH 24/24] virtiofsd: Ask qemu to drop CAP_FSETID if client asked for it, Dr. David Alan Gilbert (git), 2021/02/09