nfsd: allow SC_STATUS_FREEABLE when searching via nfs4_lookup_stateid()
Subsystems (2)
Jeff Layton
maintainerPATCH 1/1 nfsd: allow SC_STATUS_FREEABLE when searching via nfs4_lookup_stateid()
When a delegation is revoked, it's initially marked with SC_STATUS_REVOKED, or SC_STATUS_ADMIN_REVOKED and later, it's marked with the SC_STATUS_FREEABLE flag, which denotes that it is waiting for s FREE_STATEID call.
nfs4_lookup_stateid() accepts a statusmask that includes the status flags that a found stateid is allowed to have. Currently, that mask never includes SC_STATUS_FREEABLE, which means that revoked delegations are (almost) never found.
Add SC_STATUS_FREEABLE to the always-allowed status flags.
---
This fixes the pynfs DELEG8 test.
--- fs/nfsd/nfs4state.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
fs/nfsd/nfs4state.c +1 -1
@@ -7051,7 +7051,7 @@ nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,*/ statusmask |=SC_STATUS_REVOKED ; - statusmask |=SC_STATUS_ADMIN_REVOKED ; + statusmask |=SC_STATUS_ADMIN_REVOKED |SC_STATUS_FREEABLE ;if (ZERO_STATEID (stateid) ||ONE_STATEID (stateid) ||CLOSE_STATEID (stateid))
---
base-commit: 4990d098433db18c854e75fb0f90d941eb7d479e
change-id: 20250212-nfsd-fixes-fa8047082335
Best regards,
Signature · 1 line
Jeff Layton <[email protected]>
Quoted 10 lines · Jeff Layton
When a delegation is revoked, it's initially marked with
SC_STATUS_REVOKED, orSC_STATUS_ADMIN_REVOKEDand later, it's marked with theSC_STATUS_FREEABLEflag, which denotes that it is waiting for sFREE_STATEIDcall.
nfs4_lookup_stateid()accepts a statusmask that includes the status flags that a found stateid is allowed to have. Currently, that mask never includesSC_STATUS_FREEABLE, which means that revoked delegations are (almost) never found.
Quoted 5 lines
Add
SC_STATUS_FREEABLEto the always-allowed status flags.Signed-off-by Jeff Layton <[email protected]>---
This fixes the pynfs DELEG8 test.
A little release engineering:
- The DELEG8 mention IMO should be the lede in the patch description
- Do we want Fixes: and/or Cc: stable tags?
Quoted 23 lines
--- fs/nfsd/nfs4state.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)fs/nfsd/nfs4state.c +1
-2@@ -7051,7 +7051,7 @@ nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,*/ statusmask |=SC_STATUS_REVOKED ; - statusmask |=SC_STATUS_ADMIN_REVOKED ; + statusmask |=SC_STATUS_ADMIN_REVOKED |SC_STATUS_FREEABLE ;if (ZERO_STATEID (stateid) ||ONE_STATEID (stateid) ||CLOSE_STATEID (stateid)) ---base-commit:
4990d098433db18c854e75fb0f90d941eb7d479e
change-id: 20250212-nfsd-fixes-fa8047082335Best regards,
Signature · 1 line
Chuck Lever
Jeff Layton
maintainerQuoted 37 lines · Jeff Layton
When a delegation is revoked, it's initially marked with
SC_STATUS_REVOKED, orSC_STATUS_ADMIN_REVOKEDand later, it's marked with theSC_STATUS_FREEABLEflag, which denotes that it is waiting for sFREE_STATEIDcall.
nfs4_lookup_stateid()accepts a statusmask that includes the status flags that a found stateid is allowed to have. Currently, that mask never includesSC_STATUS_FREEABLE, which means that revoked delegations are (almost) never found.Add
SC_STATUS_FREEABLEto the always-allowed status flags.Signed-off-by Jeff Layton <[email protected]>---
This fixes the pynfs DELEG8 test.--- fs/nfsd/nfs4state.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)fs/nfsd/nfs4state.c +1
-2@@ -7051,7 +7051,7 @@ nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,*/ statusmask |=SC_STATUS_REVOKED ; - statusmask |=SC_STATUS_ADMIN_REVOKED ; + statusmask |=SC_STATUS_ADMIN_REVOKED |SC_STATUS_FREEABLE ;if (ZERO_STATEID (stateid) ||ONE_STATEID (stateid) ||CLOSE_STATEID (stateid)) ---base-commit:
4990d098433db18c854e75fb0f90d941eb7d479e
change-id: 20250212-nfsd-fixes-fa8047082335
Quoted 1 line
Best regards,
I didn't add this, but this seems like the likely place where it was broken, since this was where SC_STATUS_FREEABLE was added:
Signature · 1 line
Jeff Layton <[email protected]>
Quoted 11 lines · Jeff Layton
When a delegation is revoked, it's initially marked with
SC_STATUS_REVOKED, orSC_STATUS_ADMIN_REVOKEDand later, it's marked with theSC_STATUS_FREEABLEflag, which denotes that it is waiting for sFREE_STATEIDcall.
nfs4_lookup_stateid()accepts a statusmask that includes the status flags that a found stateid is allowed to have. Currently, that mask never includesSC_STATUS_FREEABLE, which means that revoked delegations are (almost) never found.Add
SC_STATUS_FREEABLEto the always-allowed status flags.
There are 4 calls to nfsd4_lookup_stateid(). One already has SC_STATUS_FREEABLE passed. Which of the others need it?
If all of them, then this patch is sensible but should also remove the flag in the one place it is already passed.
If only one other call needs it, then maybe we should just pass it there?
Could you at least include in the description some detail of what request is failing and which particular nfsd4_lookup_stateid() call is relevant in that case?
Thanks,
NeilBrown
Quoted 31 lines
Signed-off-by Jeff Layton <[email protected]>---
This fixes the pynfs DELEG8 test.--- fs/nfsd/nfs4state.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)fs/nfsd/nfs4state.c +1
-2@@ -7051,7 +7051,7 @@ nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,*/ statusmask |=SC_STATUS_REVOKED ; - statusmask |=SC_STATUS_ADMIN_REVOKED ; + statusmask |=SC_STATUS_ADMIN_REVOKED |SC_STATUS_FREEABLE ;if (ZERO_STATEID (stateid) ||ONE_STATEID (stateid) ||CLOSE_STATEID (stateid)) ---base-commit:
4990d098433db18c854e75fb0f90d941eb7d479e
change-id: 20250212-nfsd-fixes-fa8047082335Best regards,
Signature · 1 line
Jeff Layton <[email protected]>
Jeff Layton
maintainerQuoted 21 lines · NeilBrown
Quoted 11 lines · Jeff Layton
When a delegation is revoked, it's initially marked with
SC_STATUS_REVOKED, orSC_STATUS_ADMIN_REVOKEDand later, it's marked with theSC_STATUS_FREEABLEflag, which denotes that it is waiting for sFREE_STATEIDcall.
nfs4_lookup_stateid()accepts a statusmask that includes the status flags that a found stateid is allowed to have. Currently, that mask never includesSC_STATUS_FREEABLE, which means that revoked delegations are (almost) never found.Add
SC_STATUS_FREEABLEto the always-allowed status flags.There are 4 calls to
nfsd4_lookup_stateid(). One already hasSC_STATUS_FREEABLEpassed. Which of the others need it?If all of them, then this patch is sensible but should also remove the flag in the one place it is already passed.
If only one other call needs it, then maybe we should just pass it there?
Quoted 7 lines
Could you at least include in the description some detail of what request is failing and which particular
nfsd4_lookup_stateid()call is relevant in that case?Thanks,
NeilBrown
I think they all need it.
The DELEG8 test gets a delegation, and then lets the lease time out. It then does a READ op and tests that it gets back NFS4ERR_DELEG_REVOKED. The read path calls nfs4_preprocess_stateid_op(), so it's needed there specifically to fix that bug.
1ac3629bf0125 ensures that SC_STATUS_ADMIN_REVOKED is always set in the allowed mask. If that's always allowed, then FREEABLE must be too. There is only a very narrow window of time when ADMIN_REVOKED or REVOKED is set, and FREEABLE is not.
I'll respin and remove the other FREEABLE flag, since I think it should be implied everywhere.
Quoted 33 lines
Quoted 32 lines
Signed-off-by Jeff Layton <[email protected]>---
This fixes the pynfs DELEG8 test.--- fs/nfsd/nfs4state.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)fs/nfsd/nfs4state.c +1
-2@@ -7051,7 +7051,7 @@ nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,*/ statusmask |=SC_STATUS_REVOKED ; - statusmask |=SC_STATUS_ADMIN_REVOKED ; + statusmask |=SC_STATUS_ADMIN_REVOKED |SC_STATUS_FREEABLE ;if (ZERO_STATEID (stateid) ||ONE_STATEID (stateid) ||CLOSE_STATEID (stateid)) ---base-commit:
4990d098433db18c854e75fb0f90d941eb7d479e
change-id: 20250212-nfsd-fixes-fa8047082335Best regards,
Signature · 1 line
Jeff Layton <[email protected]>
Signature · 1 line
Jeff Layton <[email protected]>
Dai Ngo
reviewerQuoted 24 lines · Jeff Layton
When a delegation is revoked, it's initially marked with
SC_STATUS_REVOKED, orSC_STATUS_ADMIN_REVOKEDand later, it's marked with theSC_STATUS_FREEABLEflag, which denotes that it is waiting for sFREE_STATEIDcall.
nfs4_lookup_stateid()accepts a statusmask that includes the status flags that a found stateid is allowed to have. Currently, that mask never includesSC_STATUS_FREEABLE, which means that revoked delegations are (almost) never found.Add
SC_STATUS_FREEABLEto the always-allowed status flags.Signed-off-by Jeff Layton <[email protected]>---
This fixes the pynfs DELEG8 test.--- fs/nfsd/nfs4state.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)fs/nfsd/nfs4state.c +0
-0
fs/nfsd/nfs4state.c nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate, · line 7054
Quoted 5 lines
fs/nfsd/nfs4state.c +1
-1@@ -7051,7 +7051,7 @@ nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,*/ statusmask |=SC_STATUS_REVOKED ; - statusmask |=SC_STATUS_ADMIN_REVOKED ; + statusmask |=SC_STATUS_ADMIN_REVOKED |SC_STATUS_FREEABLE ;
I think it's safer to add the SC_STATUS_FREEABLE inside the if statement that checks for SC_TYPE_DELEG above.
Also, don't we also need to mark the delegation with SC_STATUS_REVOKED in revoke_delegation()?
-Dai
Quoted 9 lines
fs/nfsd/nfs4state.c +0
-1if (ZERO_STATEID (stateid) ||ONE_STATEID (stateid) ||CLOSE_STATEID (stateid)) ---base-commit:
4990d098433db18c854e75fb0f90d941eb7d479e
change-id: 20250212-nfsd-fixes-fa8047082335Best regards,
Jeff Layton
maintainerQuoted 28 lines · Dai Ngo
Quoted 27 lines · Jeff Layton
When a delegation is revoked, it's initially marked with
SC_STATUS_REVOKED, orSC_STATUS_ADMIN_REVOKEDand later, it's marked with theSC_STATUS_FREEABLEflag, which denotes that it is waiting for sFREE_STATEIDcall.
nfs4_lookup_stateid()accepts a statusmask that includes the status flags that a found stateid is allowed to have. Currently, that mask never includesSC_STATUS_FREEABLE, which means that revoked delegations are (almost) never found.Add
SC_STATUS_FREEABLEto the always-allowed status flags.Signed-off-by Jeff Layton <[email protected]>---
This fixes the pynfs DELEG8 test.--- fs/nfsd/nfs4state.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)fs/nfsd/nfs4state.c +0
-0@@ -7051,7 +7051,7 @@ nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,*/ statusmask |=SC_STATUS_REVOKED ;
Quoted 6 lines
fs/nfsd/nfs4state.c nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate, · line 7054
Quoted 2 lines
fs/nfsd/nfs4state.c +1
-1- statusmask |=SC_STATUS_ADMIN_REVOKED ; + statusmask |=SC_STATUS_ADMIN_REVOKED |SC_STATUS_FREEABLE ;I think it's safer to add the
SC_STATUS_FREEABLEinside the if statement that checks forSC_TYPE_DELEGabove.
That's not enough. SC_STATUS_ADMIN_REVOKED is always allowed, and for that to work properly, we need to always allow SC_STATUS_FREEABLE too.
Quoted 3 lines
Also, don't we also need to mark the delegation with
SC_STATUS_REVOKEDinrevoke_delegation()?
No. The idea is that the REVOKED flag should be set before calling revoke_delegation(). Typically they get marked REVOKED or ADMIN_REVOKED under spinlock, and then we call revoke_delegation() afterward to finish the job.
Quoted 9 lines
Quoted 9 lines
fs/nfsd/nfs4state.c +0
-1if (ZERO_STATEID (stateid) ||ONE_STATEID (stateid) ||CLOSE_STATEID (stateid)) ---base-commit:
4990d098433db18c854e75fb0f90d941eb7d479e
change-id: 20250212-nfsd-fixes-fa8047082335Best regards,
Signature · 1 line
Jeff Layton <[email protected]>
Quoted 10 lines · Jeff Layton
When a delegation is revoked, it's initially marked with
SC_STATUS_REVOKED, orSC_STATUS_ADMIN_REVOKEDand later, it's marked with theSC_STATUS_FREEABLEflag, which denotes that it is waiting for sFREE_STATEIDcall.
nfs4_lookup_stateid()accepts a statusmask that includes the status flags that a found stateid is allowed to have. Currently, that mask never includesSC_STATUS_FREEABLE, which means that revoked delegations are (almost) never found.
Quoted 5 lines
Add
SC_STATUS_FREEABLEto the always-allowed status flags.Signed-off-by Jeff Layton <[email protected]>---
This fixes the pynfs DELEG8 test.
I want to follow-up this patch with the report that this will fix quite severe TEST_STATEID/FREE_STATEID storms from NFS clients that can never satisfy the server's setting SEQ4_STATUS_*_STATE_REVOKED status bits.
Correct accounting on the server's cl_revoked counter is pretty important, I do wish the protocol had something like a per-client global window of valid state.
Thanks Jeff (and thanks Olga for the original fix!),
Ben