Skip to contentSkip to conversation

pynfs updates

DISCUSSION Frank Filz 2 participants 11 messages Sep 30, 2013 – Oct 2, 2013

Subsystems (2)
lore
Share & tools
lore
#1 lore raw
Quoted 6 lines · Frank Filz

Bruce,

Please pull the following branch that contains the pynfs updates I made at BAT:

https://github.com/ffilz/pynfs/commits/master

Thanks! A few questions:

- "4.1 server tests: Fix some exception handling": could you include in the changelog and explanation of what problem this fixes? (And ditto, maybe, for the following commit?)

- "Fix SEQ9d to work in home directory instead of root": did you intend to include the chunk in nfs4.1/nfs4lib.py? It looks irrelevant.

- "Add two SECINFO_NO_NAME tests for SECINFO_STYLE4_PARENT":
- SECNN3: is / required to have no parent? (I'd assumed here that it would also be OK to follow the convention that / is its own parent, but I'll admit to not having thought about this much.)
- SECNN4: is env.home necessarily unequal to "/"? Would seem better to do the lookup in a subdirectory just to be certain.

--b.

author
#2 lore raw
Quoted 5 lines

Thanks! A few questions:

- "4.1 server tests: Fix some exception handling": could you include in the changelog and explanation of what problem this fixes? (And ditto, maybe, for the following commit?)

Ok, I'll re-examine those. They were some things I stumbled on as I tried to resolve some other issues with pynfs.

Quoted 3 lines

- "Fix SEQ9d to work in home directory instead of root": did you intend to include the chunk in nfs4.1/nfs4lib.py? It looks irrelevant.

Hmm, will fix that.

Quoted 8 lines

- "Add two SECINFO_NO_NAME tests for
SECINFO_STYLE4_PARENT":
- SECNN3: is / required to have no parent? (I'd assumed here that it would also be OK to follow the convention that / is its own parent, but I'll admit to not having thought about this much.)

From LOOKUPP:

18.14.3. DESCRIPTION
The current filehandle is assumed to refer to a regular directory or a named attribute directory. LOOKUPP assigns the filehandle for its parent directory to be the current filehandle. If there is no parent directory, an NFS4ERR_NOENT error must be returned. Therefore, NFS4ERR_NOENT will be returned by the server when the current filehandle is at the root or top of the server's file tree.

Quoted 1 line

From SECINFO_NO_NAME:

18.45.3. DESCRIPTION

...

If the style selected is SECINFO_STYLE4_PARENT,
then SECINFO should apply the same access methodology used for LOOKUPP when evaluating the traversal to the parent directory.

...

If SECINFO_STYLE4_PARENT is specified and there is no parent directory, SECINFO_NO_NAME MUST return NFS4ERR_NOENT.

Quoted 3 lines

- SECNN4: is env.home necessarily unequal to "/"? Would seem better to do the lookup in a subdirectory just to be certain.

Env.home is the directory you specify on the command line, I think the presumption is that it is a writeable file system. Pynfs creates tmp and tree directories in home (and maybe some files also?). Guess if / was writeable, you could specify /, so yea, maybe it should go into tmp.

A better test might actually be to do LOOKUP down to home and even into tmp, looking for a junction, and then do the SECINFO_NO_NAME(parent) on the directory handle just across the junction if one was found.

Thanks for the review.

Frank

#3 lore raw
Quoted 9 lines · Frank Filz
Quoted 8 lines

- "Add two SECINFO_NO_NAME tests for
SECINFO_STYLE4_PARENT":
- SECNN3: is / required to have no parent? (I'd assumed here that it would also be OK to follow the convention that / is its own parent, but I'll admit to not having thought about this much.)

From LOOKUPP:

Quoted 7 lines

18.14.3. DESCRIPTION
The current filehandle is assumed to refer to a regular directory or a named attribute directory. LOOKUPP assigns the filehandle for its parent directory to be the current filehandle. If there is no parent directory, an NFS4ERR_NOENT error must be returned. Therefore, NFS4ERR_NOENT will be returned by the server when the current filehandle is at the root or top of the server's file tree.

OK, fine.

Quoted 8 lines
Quoted 3 lines

- SECNN4: is env.home necessarily unequal to "/"? Would seem better to do the lookup in a subdirectory just to be certain.

Env.home is the directory you specify on the command line, I think the presumption is that it is a writeable file system. Pynfs creates tmp and tree directories in home (and maybe some files also?). Guess if / was writeable, you could specify /, so yea, maybe it should go into tmp.

Sounds good.

Quoted 4 lines

A better test might actually be to do LOOKUP down to home and even into tmp, looking for a junction, and then do the SECINFO_NO_NAME(parent) on the directory handle just across the junction if one was found.

Yeah it'd be nice to check that cross-filesystem case but I don't think it's necessary (and you still have to deal with the case where a mountpoint's not found).

If tests at mountpoints were useful perhaps we could pass in a mountpoint on the commandline. Or add some sort of export-configuration interface to the serverhelper script and let pynfs setup exports itself.

--b.

#4 lore raw

One more problem: CSID10 is failing against the Linux server with NFS4ERR_TOO_MANY_OPS, because each of those lookups is actually a full lookup from PUTROOTFH to /, resulting in 17 ops on my setup. Could we maybe work relative to the parent directory instead?

--b.

author
#5 lore raw
Quoted 4 lines

One more problem: CSID10 is failing against the Linux server with NFS4ERR_TOO_MANY_OPS, because each of those lookups is actually a full lookup from PUTROOTFH to /, resulting in 17 ops on my setup. Could we maybe work relative to the parent directory instead?

Sure, I'll rework that one.

Quoted 6 lines
Quoted 3 lines

A better test might actually be to do LOOKUP down to home and even into tmp, looking for a junction, and then do the SECINFO_NO_NAME(parent) on the directory handle just across the

junction if one was found.

Yeah it'd be nice to check that cross-filesystem case but I don't think

it's

Quoted 1 line

necessary (and you still have to deal with the case where a mountpoint's

not

Quoted 1 line

found).

Well, the cross file system case is actually where you would need to use SECINFO_NO_NAME. For some reason, you just have a handle to a directory inside the export and want to navigate back up the tree. In doing so, you cross back over a junction to a file system that is exported with a different security flavor.

On the other hand, generally that higher level file system should include all the security flavors used by the lower level file systems. Unless SECINFO_NO_NAME lets you cross a junction where the new file system doesn't have security flavors in common with the upper level file system, but I don't think it does. Does anyone know the rationale of SECINFO_NO_NAME (parent)? In fact is there really any use of SECINFO_NO_NAME other than to get the secinfo for the root or public file handle? I guess it does also allow a client to recover from the security flavors for a given file system being changed on the fly (or perhaps after a migration event).

Quoted 3 lines

If tests at mountpoints were useful perhaps we could pass in a mountpoint on the commandline. Or add some sort of export-configuration interface to the serverhelper script and let pynfs setup exports itself.

Frank

author
#9 lore raw
In reply to 'Bruce Fields'
Quoted 10 lines
Quoted 8 lines
Quoted 3 lines

- SECNN4: is env.home necessarily unequal to "/"? Would seem better to do the lookup in a subdirectory just to be certain.

Env.home is the directory you specify on the command line, I think the presumption is that it is a writeable file system. Pynfs creates tmp and tree directories in home (and maybe some files also?). Guess if / was writeable, you could specify /, so yea, maybe it should go into tmp.

Sounds good.

env.home does actually include traversing into tmp, so I will leave this test alone.

Quoted 6 lines
Quoted 3 lines

A better test might actually be to do LOOKUP down to home and even into tmp, looking for a junction, and then do the SECINFO_NO_NAME(parent) on the directory handle just across the

junction if one was found.

Yeah it'd be nice to check that cross-filesystem case but I don't think

it's

Quoted 1 line

necessary (and you still have to deal with the case where a mountpoint's

not

Quoted 5 lines

found).

If tests at mountpoints were useful perhaps we could pass in a mountpoint on the commandline. Or add some sort of export-configuration interface to the serverhelper script and let pynfs setup exports itself.

Yea, that might be something interesting to explore at some point.

Frank

#10 lore raw
Quoted 9 lines · Frank Filz
Quoted 9 lines
Quoted 8 lines
Quoted 3 lines

- SECNN4: is env.home necessarily unequal to "/"? Would seem better to do the lookup in a subdirectory just to be certain.

Env.home is the directory you specify on the command line, I think the presumption is that it is a writeable file system. Pynfs creates tmp and tree directories in home (and maybe some files also?). Guess if / was writeable, you could specify /, so yea, maybe it should go into tmp.

Quoted 4 lines
Quoted 1 line

Sounds good.

env.home does actually include traversing into tmp, so I will leave this test alone.

OK!

--b.

author
#6 lore raw
In reply to 'Bruce Fields'
Quoted 4 lines

One more problem: CSID10 is failing against the Linux server with NFS4ERR_TOO_MANY_OPS, because each of those lookups is actually a full lookup from PUTROOTFH to /, resulting in 17 ops on my setup. Could we maybe work relative to the parent directory instead?

Ok, was able to fix this by doing a LOOKUP sequence in a separate compound followed by GETFH then in the compound that tests SAVEFH/RESTOREFH, just do PUTFH that saved FH.

Things could be a lot smoother as discussed on IRC if the initialization stored away env.home_fh.

Then with some work, this test could be simplified to:

SEQUENCE, PUTFH(env.home_fh), OPEN, GETFH, SAVEFH, PUTFH(env.home_fh), RESTOREFH, CLOSE

Note that that GETFH is not actually used by this test, but presumably open_create_op() would produce:

PUTFH(env.home_fh), OPEN, GETFH

Instead of what it currently does:

PUTROOTFH, LOOKUP..., OPEN, GETFH

The new branch is here:

https://github.com/ffilz/pynfs/commits/master

Frank

#7 lore raw
Quoted 22 lines · Frank Filz
Quoted 4 lines

One more problem: CSID10 is failing against the Linux server with NFS4ERR_TOO_MANY_OPS, because each of those lookups is actually a full lookup from PUTROOTFH to /, resulting in 17 ops on my setup. Could we maybe work relative to the parent directory instead?

Ok, was able to fix this by doing a LOOKUP sequence in a separate compound followed by GETFH then in the compound that tests SAVEFH/RESTOREFH, just do PUTFH that saved FH.

Things could be a lot smoother as discussed on IRC if the initialization stored away env.home_fh.

Then with some work, this test could be simplified to:

SEQUENCE, PUTFH(env.home_fh), OPEN, GETFH, SAVEFH, PUTFH(env.home_fh), RESTOREFH, CLOSE

Note that that GETFH is not actually used by this test, but presumably open_create_op() would produce:

PUTFH(env.home_fh), OPEN, GETFH

Quoted 7 lines

Instead of what it currently does:

PUTROOTFH, LOOKUP..., OPEN, GETFH

The new branch is here:

https://github.com/ffilz/pynfs/commits/master

Pulled, thanks!

--b.