[Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

[Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

[Fotis Xenakis]

Currently, during FUSE_SETUPMAPPING, virtiofsd passes the wrong file
descriptor for the file to mmap() to QEMU (specifically, it passes the
file handle as requested from the guest).

This fixes it, using lo_fi_fd() to map the file handle to the right file
descriptor on the host.

Signed-off-by: Fotis Xenakis <foxen@windowslive.com>
---
 tools/virtiofsd/passthrough_ll.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c
index e09d557d74..44bb24eb5e 100644
--- a/tools/virtiofsd/passthrough_ll.c
+++ b/tools/virtiofsd/passthrough_ll.c
@@ -2709,7 +2709,7 @@ static void lo_setupmapping(fuse_req_t req, fuse_ino_t ino, uint64_t foffset,
     msg.flags[0] = vhu_flags;
 
     if (fi) {
-        fd = fi->fh;
+        fd = lo_fi_fd(req, fi);
     } else {
         res = asprintf(&buf, "%i", lo_fd(req, ino));
         if (res == -1) {
-- 
2.26.2

Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

[Liu Bo]

On Wed, May 13, 2020 at 07:33:56PM +0300, Fotis Xenakis wrote:
> Currently, during FUSE_SETUPMAPPING, virtiofsd passes the wrong file
> descriptor for the file to mmap() to QEMU (specifically, it passes the
> file handle as requested from the guest).
> 
> This fixes it, using lo_fi_fd() to map the file handle to the right file
> descriptor on the host.
>

Looks good, guest has been setting fi as NULL though.

Reviewed-by: Liu Bo <bo.liu@linux.alibaba.com>


> Signed-off-by: Fotis Xenakis <foxen@windowslive.com>
> ---
>  tools/virtiofsd/passthrough_ll.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c
> index e09d557d74..44bb24eb5e 100644
> --- a/tools/virtiofsd/passthrough_ll.c
> +++ b/tools/virtiofsd/passthrough_ll.c
> @@ -2709,7 +2709,7 @@ static void lo_setupmapping(fuse_req_t req, fuse_ino_t ino, uint64_t foffset,
>      msg.flags[0] = vhu_flags;
>  
>      if (fi) {
> -        fd = fi->fh;
> +        fd = lo_fi_fd(req, fi);
>      } else {
>          res = asprintf(&buf, "%i", lo_fd(req, ino));
>          if (res == -1) {
> -- 
> 2.26.2
> 
> 
> _______________________________________________
> Virtio-fs mailing list
> Virtio-fs@redhat.com
> https://www.redhat.com/mailman/listinfo/virtio-fs

Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

[Fotis Xenakis]

[-- Attachment #1: Type: text/plain, Size: 1920 bytes --]

From: Liu Bo <bo.liu@linux.alibaba.com>
Sent: Friday, May 15, 2020 06:48
To: Fotis Xenakis <foxen@windowslive.com>
Cc: virtio-fs@redhat.com <virtio-fs@redhat.com>
Subject: Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

On Wed, May 13, 2020 at 07:33:56PM +0300, Fotis Xenakis wrote:
> Currently, during FUSE_SETUPMAPPING, virtiofsd passes the wrong file
> descriptor for the file to mmap() to QEMU (specifically, it passes the
> file handle as requested from the guest).
>
> This fixes it, using lo_fi_fd() to map the file handle to the right file
> descriptor on the host.
>

Looks good, guest has been setting fi as NULL though.
As far as I can see, this function is only called from [1], which passes fi as NULL only when the file handle passed from the guest is (uint64_t)-1. In my testing this was never the case though.

Reference:
[1] https://gitlab.com/virtio-fs/qemu/-/blob/21336c0f3d05a97f5c409bbc894c19d87259655c/tools/virtiofsd/fuse_lowlevel.c#L1930
Reviewed-by: Liu Bo <bo.liu@linux.alibaba.com>


> Signed-off-by: Fotis Xenakis <foxen@windowslive.com>
> ---
>  tools/virtiofsd/passthrough_ll.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c
> index e09d557d74..44bb24eb5e 100644
> --- a/tools/virtiofsd/passthrough_ll.c
> +++ b/tools/virtiofsd/passthrough_ll.c
> @@ -2709,7 +2709,7 @@ static void lo_setupmapping(fuse_req_t req, fuse_ino_t ino, uint64_t foffset,
>      msg.flags[0] = vhu_flags;
>
>      if (fi) {
> -        fd = fi->fh;
> +        fd = lo_fi_fd(req, fi);
>      } else {
>          res = asprintf(&buf, "%i", lo_fd(req, ino));
>          if (res == -1) {
> --
> 2.26.2
>
>
> _______________________________________________
> Virtio-fs mailing list
> Virtio-fs@redhat.com
> https://www.redhat.com/mailman/listinfo/virtio-fs

[-- Attachment #2: Type: text/html, Size: 4129 bytes --]

Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

[Vivek Goyal]

On Fri, May 15, 2020 at 04:50:31PM +0000, Fotis Xenakis wrote:
> From: Liu Bo <bo.liu@linux.alibaba.com>
> Sent: Friday, May 15, 2020 06:48
> To: Fotis Xenakis <foxen@windowslive.com>
> Cc: virtio-fs@redhat.com <virtio-fs@redhat.com>
> Subject: Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping
> 
> On Wed, May 13, 2020 at 07:33:56PM +0300, Fotis Xenakis wrote:
> > Currently, during FUSE_SETUPMAPPING, virtiofsd passes the wrong file
> > descriptor for the file to mmap() to QEMU (specifically, it passes the
> > file handle as requested from the guest).
> >
> > This fixes it, using lo_fi_fd() to map the file handle to the right file
> > descriptor on the host.
> >
> 
> Looks good, guest has been setting fi as NULL though.
> As far as I can see, this function is only called from [1], which passes fi as NULL only when the file handle passed from the guest is (uint64_t)-1. In my testing this was never the case though.

Hi Fotis,

Do you know what's different about your setup. This has been working for
me so clearly I am seeing fi as NULL. 

Vivek

Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

[Fotis Xenakis]

[-- Attachment #1: Type: text/plain, Size: 2424 bytes --]

From: Vivek Goyal <vgoyal@redhat.com>
Sent: Monday, May 18, 2020 15:37
To: Fotis Xenakis <foxen@windowslive.com>
Cc: bo.liu@linux.alibaba.com <bo.liu@linux.alibaba.com>; virtio-fs@redhat.com <virtio-fs@redhat.com>
Subject: Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

On Fri, May 15, 2020 at 04:50:31PM +0000, Fotis Xenakis wrote:
> From: Liu Bo <bo.liu@linux.alibaba.com>
> Sent: Friday, May 15, 2020 06:48
> To: Fotis Xenakis <foxen@windowslive.com>
> Cc: virtio-fs@redhat.com <virtio-fs@redhat.com>
> Subject: Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping
>
> On Wed, May 13, 2020 at 07:33:56PM +0300, Fotis Xenakis wrote:
> > Currently, during FUSE_SETUPMAPPING, virtiofsd passes the wrong file
> > descriptor for the file to mmap() to QEMU (specifically, it passes the
> > file handle as requested from the guest).
> >
> > This fixes it, using lo_fi_fd() to map the file handle to the right file
> > descriptor on the host.
> >
>
> Looks good, guest has been setting fi as NULL though.
> As far as I can see, this function is only called from [1], which passes fi as NULL only when the file handle passed from the guest is (uint64_t)-1. In my testing this was never the case though.

Hi Fotis,

Do you know what's different about your setup. This has been working for
me so clearly I am seeing fi as NULL.
Hello,

This was on my mind as well: if others were affected by this bug, it wouldn't have gone unnoticed. Yet, I can't say I know what's different in my setup.

Unfortunately, I have a less-than-basic understanding of virtiofsd, but as far as I can see:
​- virtiofsd differentiates on the file handler passed by the guest [1]. In my case fh != (uint64_t)-1 (which makes sense tbh). fh is whatever FUSE_OPEN returned.
​- Guest-side, the open() flags are passed through to FUSE_OPEN [2]. In my case that's O_RDONLY.
- The only flag set for FUSE_INIT is FUSE_MAP_ALIGNMENT [3].
​-​ For what it's worth, virtiofsd runs with "-o cache=always".

References:
[1] https://gitlab.com/virtio-fs/qemu/-/blob/21336c0f3d05a97f5c409bbc894c19d87259655c/tools/virtiofsd/fuse_lowlevel.c#L1910
[2] https://github.com/foxeng/osv/blob/46255086015a4c801baaf61d431e3aa7eda9e64b/fs/virtiofs/virtiofs_vnops.cc#L111
[3] https://github.com/foxeng/osv/blob/46255086015a4c801baaf61d431e3aa7eda9e64b/fs/virtiofs/virtiofs_vfsops.cc#L91
Vivek


[-- Attachment #2: Type: text/html, Size: 6731 bytes --]

Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

[Vivek Goyal]

On Wed, May 13, 2020 at 07:33:56PM +0300, Fotis Xenakis wrote:
> Currently, during FUSE_SETUPMAPPING, virtiofsd passes the wrong file
> descriptor for the file to mmap() to QEMU (specifically, it passes the
> file handle as requested from the guest).
> 
> This fixes it, using lo_fi_fd() to map the file handle to the right file
> descriptor on the host.
> 
> Signed-off-by: Fotis Xenakis <foxen@windowslive.com>

Acked-by: Vivek Goyal <vgoyal@redhat.com>

Vivek
> ---
>  tools/virtiofsd/passthrough_ll.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c
> index e09d557d74..44bb24eb5e 100644
> --- a/tools/virtiofsd/passthrough_ll.c
> +++ b/tools/virtiofsd/passthrough_ll.c
> @@ -2709,7 +2709,7 @@ static void lo_setupmapping(fuse_req_t req, fuse_ino_t ino, uint64_t foffset,
>      msg.flags[0] = vhu_flags;
>  
>      if (fi) {
> -        fd = fi->fh;
> +        fd = lo_fi_fd(req, fi);
>      } else {
>          res = asprintf(&buf, "%i", lo_fd(req, ino));
>          if (res == -1) {
> -- 
> 2.26.2
> 
> 
> _______________________________________________
> Virtio-fs mailing list
> Virtio-fs@redhat.com
> https://www.redhat.com/mailman/listinfo/virtio-fs

Re: [Virtio-fs] [PATCH] Fix file descriptor passed for setupmapping

[Dr. David Alan Gilbert]

* Fotis Xenakis (foxen@windowslive.com) wrote:
> Currently, during FUSE_SETUPMAPPING, virtiofsd passes the wrong file
> descriptor for the file to mmap() to QEMU (specifically, it passes the
> file handle as requested from the guest).
> 
> This fixes it, using lo_fi_fd() to map the file handle to the right file
> descriptor on the host.
> 
> Signed-off-by: Fotis Xenakis <foxen@windowslive.com>

Thanks!  I've merged this into:
  'DAX: virtiofsd: Make setupmapping work only with inode'

This was a fixup/merging problem - the fd indirection/hiding was done
after the setupmapping; but then the fd indirection got merged first and
the setupmapping put back on top.

Dave

> ---
>  tools/virtiofsd/passthrough_ll.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c
> index e09d557d74..44bb24eb5e 100644
> --- a/tools/virtiofsd/passthrough_ll.c
> +++ b/tools/virtiofsd/passthrough_ll.c
> @@ -2709,7 +2709,7 @@ static void lo_setupmapping(fuse_req_t req, fuse_ino_t ino, uint64_t foffset,
>      msg.flags[0] = vhu_flags;
>  
>      if (fi) {
> -        fd = fi->fh;
> +        fd = lo_fi_fd(req, fi);
>      } else {
>          res = asprintf(&buf, "%i", lo_fd(req, ino));
>          if (res == -1) {
> -- 
> 2.26.2
> 
> 
> _______________________________________________
> Virtio-fs mailing list
> Virtio-fs@redhat.com
> https://www.redhat.com/mailman/listinfo/virtio-fs
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK