On 10/1/20 2:52 PM, Bruce Fields wrote:
On Thu, Oct 01, 2020 at 02:48:07PM -0700, Dai Ngo wrote:
Thanks Bruce for your comments,
On 10/1/20 1:51 PM, Bruce Fields wrote:
On Tue, Sep 29, 2020 at 08:18:54PM -0700, Dai Ngo wrote:
Have you had chance to review this patch and if it's ok would it be
possible to include it in the 5.10 pull?
I don't think the op table approach would be that difficult, I'd really
rather see that.
I think if we do the op table approach then we should also try to solve
all other dependencies between various NFS client and server modules
and not just the SSC part.
Are there any others? I'd be very surprised. It's something we've been
quite careful not to do in the past. I apologize that it got past my
review this time.
No, I'm not sure if there is any others, I'm just being cautious.
We can always start by building the infra-structure and fix the SSC first
and if there is any others then we can use the same mechanism to fix them
too. I can work on this for long term. In the mean time can we pull in this
temporary fix or you rather just want to see the long term solution?
Thanks,
-Dai
--b.
It might be a little involved so I'd like
to take some time to research before committing to the longer solution
which I plan to do. In the mean time, this small patch allows some of
us to use the inter server copy until the long term solution is available.
Is this causing someone an immediate practical problem?
This causes inter server copy to fail with any kernel build with NFS_FS=m
which I think is a common config. And it also causes compile errors if
NFSD=y, NFS_FS=y and NFS_v4=m.
-Dai
--b.
Thanks,
-Dai
On 9/23/20 4:06 PM, Dai Ngo wrote:
This patch provides a temporarily relief for inter copy to work with
some common configs. For long term solution, I think Trond's suggestion
of using fs/nfs/nfs_common to store an op table that server can use to
access the client code is the way to go.
fs/nfsd/Kconfig | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
Below are the results of my testing of upstream mainline without and with the fix.
Upstream version used for testing: 5.9-rc5
1. Upstream mainline (existing code: NFS_FS=y)
|----------------------------------------------------------------------------------------|
| NFSD | NFS_FS | NFS_V4 | RESULTS |
|----------------------------------------------------------------------------------------|
| y | y | m | Build errors: nfs42_ssc_open/close |
|----------------------------------------------------------------------------------------|
| y | m | m | Build OK, inter server copy failed with NFS4ERR_STALE |
| | | | See NOTE1. |
|----------------------------------------------------------------------------------------|
| y | m | y (m) | Build OK, inter server copy failed with NFS4ERR_STALE |
| | | | See NOTE2. |
|----------------------------------------------------------------------------------------|
| y | y | y | Build OK, inter server copy OK |
|----------------------------------------------------------------------------------------|
|----------------------------------------------------------------------------------------|
| NFSD | NFS_FS | NFS_V4 | RESULTS |
|----------------------------------------------------------------------------------------|
| m | y | m | Build OK, inter server copy OK |
|----------------------------------------------------------------------------------------|
| m | m | m | Build OK, inter server copy failed with NFS4ERR_STALE |
|----------------------------------------------------------------------------------------|
| m | m | y (m) | Build OK, inter server copy failed with NFS4ERR_STALE |
|----------------------------------------------------------------------------------------|
| m | y | y | Build OK, inter server copy OK |
|----------------------------------------------------------------------------------------|
2. Upstream mainline (with the fix: !(NFSD=y && (NFS_FS=m || NFS_V4=m))
|----------------------------------------------------------------------------------------|
| NFSD | NFS_FS | NFS_V4 | RESULTS |
|----------------------------------------------------------------------------------------|
| m | y | m | Build OK, inter server copy OK |
|----------------------------------------------------------------------------------------|
| m | m | m | Build OK, inter server copy OK |
|----------------------------------------------------------------------------------------|
| m | m | y (m) | Build OK, inter server copy OK |
|----------------------------------------------------------------------------------------|
| m | y | y | Build OK, inter server copy OK |
|----------------------------------------------------------------------------------------|
|----------------------------------------------------------------------------------------|
| NFSD | NFS_FS | NFS_V4 | RESULTS |
|----------------------------------------------------------------------------------------|
| y | y | m | Build OK, inter server copy failed with NFS4ERR_STALE |
|----------------------------------------------------------------------------------------|
| y | m | m | Build OK, inter server copy failed with NFS4ERR_STALE |
|----------------------------------------------------------------------------------------|
| y | m | y (m) | Build OK, inter server copy failed with NFS4ERR_STALE |
|----------------------------------------------------------------------------------------|
| y | y | y | Build OK, inter server copy OK |
|----------------------------------------------------------------------------------------|
NOTE1:
BUG: When inter server copy fails with NFS4ERR_STALE, it left the file
created with size of 0!
NOTE2:
When NFS_V4=y and NFS_FS=m, the build process automatically builds with NFS_V4=m
and ignores the setting NFS_V4=y in the config file.
This probably due to NFS_V4 in fs/nfs/Kconfig is configured to depend on NFS_FS.
