Skip to contentSkip to conversation

generic: skip a couple more tests on NFS

PATCH In review Jeff Layton 2 participants 8 messages Jan 10, 2024 – Jan 12, 2024

Subsystems (2)
lore
Share & tools
lore

Jeff Layton

author maintainer
#0 lore raw

cover generic: skip a couple more tests on NFS

Both of these tests don't and aren't expected to ever pass on NFS. Remove them from their _supported_fs lists.

---
Jeff Layton (2):
generic/465: don't run it on NFS
generic/732: don't run it on NFS

 tests/generic/465 | 2 +-
 tests/generic/732 | 4 +---
 2 files changed, 2 insertions(+), 4 deletions(-)

---
base-commit: f814a0d8b89c84055b4351f8b9655c5868db08ba
change-id: 20240110-fixes-701e439421c1

Best regards,

Signature · 1 line
Jeff Layton <[email protected]>
author maintainer
#1 lore raw

PATCH 1/2 generic/465: don't run it on NFS

This test kicks off a thread that issues a read against a file, while writing to the file in 1M chunks. It expects that the reader will see either the written data or a short read.

NFS allows DIO reads and writes to run in parallel. That means that it's possible for them to race and the reader to see NULLs in the file if things get reordered.

Just skip this test on NFS, since we can't guarantee that it will reliably pass.

---
 tests/generic/465 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/generic/465 +1 -1
index 73fdfb5548af..0745d6a1dd3a 100755
@@ -21,7 +21,7 @@ _cleanup()
 . ./common/filter
 
 # real QA test starts here
-_supported_fs generic
+_supported_fs ^nfs
 
 _require_aiodio aio-dio-append-write-read-race
 _require_test_program "feature"
Signature · 1 line
2.43.0
author maintainer
#3 lore raw

PATCH 2/2 generic/732: don't run it on NFS

In reply to Jeff Layton

This test sets up two independent superblocks with the same backend server, and then does RENAMES of the same files in the two servers. This is basically trying to simulate the case where two clients are competing to rename files in the same directory on the same server.

This test would usually pass vs. an NFSv4 server that doesn't have dfdd2630a7398 ("nfsd: fix change_info in NFSv4 RENAME replies"), because the client would end up improperly invalidating the dcache for the whole dir after most RENAMEs.

However, this test doesn't (and shouldn't) pass on NFS, because the client has no idea that a rename has happened on the second mount. The expected behavior for the NFS client is for it to use the cache timeouts in this case, which is what it now does with the above server bug fixed.

Exempt NFS from running this test, since we don't expect it to pass.

Cc Yongcheng Yang <[email protected]>
---
 tests/generic/732 | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
tests/generic/732 +1 -3
index 785aac58f361..5b5087d5accd 100755
@@ -22,9 +22,7 @@ _cleanup()
 }
 
 # real QA test starts here
-_supported_fs generic
-[ "$FSTYP" = "nfs" ] && _fixed_by_kernel_commit fdd2630a739819 \
-	"nfsd: fix change_info in NFSv4 RENAME replies"
+_supported_fs ^nfs
 
 _require_test
 _require_scratch
Signature · 1 line
2.43.0
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH fstests 1/2] generic/465: don't run it on NFS

In reply to Jeff Layton
Quoted 11 lines · Jeff Layton

This test kicks off a thread that issues a read against a file, while writing to the file in 1M chunks. It expects that the reader will see either the written data or a short read.

NFS allows DIO reads and writes to run in parallel. That means that it's possible for them to race and the reader to see NULLs in the file if things get reordered.

Just skip this test on NFS, since we can't guarantee that it will reliably pass.

Quoted 6 lines

---

I've seen several nfs folks agree to skip it on nfs, and no any objection, so I'll merge this change.

Quoted 21 lines
 tests/generic/465 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/generic/465 +1 -1
index 73fdfb5548af..0745d6a1dd3a 100755
@@ -21,7 +21,7 @@ _cleanup()
 . ./common/filter
 
 # real QA test starts here
-_supported_fs generic
+_supported_fs ^nfs
 
 _require_aiodio aio-dio-append-write-read-race
 _require_test_program "feature"
Signature · 2 lines
-- 
2.43.0
#4 lore raw

Re: [PATCH fstests 2/2] generic/732: don't run it on NFS

In reply to Jeff Layton
Quoted 15 lines · Jeff Layton

This test sets up two independent superblocks with the same backend server, and then does RENAMES of the same files in the two servers. This is basically trying to simulate the case where two clients are competing to rename files in the same directory on the same server.

This test would usually pass vs. an NFSv4 server that doesn't have dfdd2630a7398 ("nfsd: fix change_info in NFSv4 RENAME replies"), because the client would end up improperly invalidating the dcache for the whole dir after most RENAMEs.

However, this test doesn't (and shouldn't) pass on NFS, because the client has no idea that a rename has happened on the second mount. The expected behavior for the NFS client is for it to use the cache timeouts in this case, which is what it now does with the above server bug fixed.

Quoted 5 lines

Exempt NFS from running this test, since we don't expect it to pass.

Cc Yongcheng Yang <[email protected]>

---

This case is written for a nfs fix at first. If nfs would like to skip this test, I don't know if it makes sense to keep it in fstests?

Thanks,
Zorro

Quoted 23 lines
 tests/generic/732 | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
tests/generic/732 +1 -3
index 785aac58f361..5b5087d5accd 100755
@@ -22,9 +22,7 @@ _cleanup()
 }
 
 # real QA test starts here
-_supported_fs generic
-[ "$FSTYP" = "nfs" ] && _fixed_by_kernel_commit fdd2630a739819 \
-	"nfsd: fix change_info in NFSv4 RENAME replies"
+_supported_fs ^nfs
 
 _require_test
 _require_scratch
Signature · 2 lines
-- 
2.43.0
author maintainer
#5 lore raw

Re: [PATCH fstests 2/2] generic/732: don't run it on NFS

Quoted 16 lines · Zorro Lang
Quoted 15 lines · Jeff Layton

This test sets up two independent superblocks with the same backend server, and then does RENAMES of the same files in the two servers. This is basically trying to simulate the case where two clients are competing to rename files in the same directory on the same server.

This test would usually pass vs. an NFSv4 server that doesn't have dfdd2630a7398 ("nfsd: fix change_info in NFSv4 RENAME replies"), because the client would end up improperly invalidating the dcache for the whole dir after most RENAMEs.

However, this test doesn't (and shouldn't) pass on NFS, because the client has no idea that a rename has happened on the second mount. The expected behavior for the NFS client is for it to use the cache timeouts in this case, which is what it now does with the above server bug fixed.

Quoted 10 lines
Quoted 5 lines

Exempt NFS from running this test, since we don't expect it to pass.

Cc Yongcheng Yang <[email protected]>

---

This case is written for a nfs fix at first. If nfs would like to skip this test, I don't know if it makes sense to keep it in fstests?

It might make sense to keep this test in place for stuff like cephfs, but if dropping it altogether is best, then that's fine with me.

Quoted 25 lines
Quoted 24 lines
 tests/generic/732 | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
tests/generic/732 +1 -3
index 785aac58f361..5b5087d5accd 100755
@@ -22,9 +22,7 @@ _cleanup()
 }
 
 # real QA test starts here
-_supported_fs generic
-[ "$FSTYP" = "nfs" ] && _fixed_by_kernel_commit fdd2630a739819 \
-	"nfsd: fix change_info in NFSv4 RENAME replies"
+_supported_fs ^nfs
 
 _require_test
 _require_scratch
Signature · 2 lines
-- 
2.43.0
Signature · 1 line
Jeff Layton <[email protected]>
#6 lore raw

Re: [PATCH fstests 2/2] generic/732: don't run it on NFS

Quoted 23 lines · Jeff Layton
Quoted 22 lines · Zorro Lang
Quoted 20 lines · Jeff Layton

This test sets up two independent superblocks with the same backend server, and then does RENAMES of the same files in the two servers. This is basically trying to simulate the case where two clients are competing to rename files in the same directory on the same server.

This test would usually pass vs. an NFSv4 server that doesn't have dfdd2630a7398 ("nfsd: fix change_info in NFSv4 RENAME replies"), because the client would end up improperly invalidating the dcache for the whole dir after most RENAMEs.

However, this test doesn't (and shouldn't) pass on NFS, because the client has no idea that a rename has happened on the second mount. The expected behavior for the NFS client is for it to use the cache timeouts in this case, which is what it now does with the above server bug fixed.

Exempt NFS from running this test, since we don't expect it to pass.

Cc Yongcheng Yang <[email protected]>

---

Quoted 7 lines
Quoted 4 lines

This case is written for a nfs fix at first. If nfs would like to skip this test, I don't know if it makes sense to keep it in fstests?

It might make sense to keep this test in place for stuff like cephfs, but if dropping it altogether is best, then that's fine with me.

OK, thanks for this feedback, if it still makes sense for any one fs, we can keep it.

Thanks,
Zorro

Quoted 30 lines
Quoted 26 lines
Quoted 25 lines

 tests/generic/732 | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
tests/generic/732 +1 -3
index 785aac58f361..5b5087d5accd 100755
@@ -22,9 +22,7 @@ _cleanup()
 }
 
 # real QA test starts here
-_supported_fs generic
-[ "$FSTYP" = "nfs" ] && _fixed_by_kernel_commit fdd2630a739819 \
-	"nfsd: fix change_info in NFSv4 RENAME replies"
+_supported_fs ^nfs
 
 _require_test
 _require_scratch
Signature · 2 lines
-- 
2.43.0
Signature · 1 line
Jeff Layton <[email protected]>
Reviewed-by
#7 lore raw

Re: [PATCH fstests 2/2] generic/732: don't run it on NFS

In reply to Jeff Layton
Quoted 23 lines · Jeff Layton
Quoted 22 lines · Zorro Lang
Quoted 20 lines · Jeff Layton

This test sets up two independent superblocks with the same backend server, and then does RENAMES of the same files in the two servers. This is basically trying to simulate the case where two clients are competing to rename files in the same directory on the same server.

This test would usually pass vs. an NFSv4 server that doesn't have dfdd2630a7398 ("nfsd: fix change_info in NFSv4 RENAME replies"), because the client would end up improperly invalidating the dcache for the whole dir after most RENAMEs.

However, this test doesn't (and shouldn't) pass on NFS, because the client has no idea that a rename has happened on the second mount. The expected behavior for the NFS client is for it to use the cache timeouts in this case, which is what it now does with the above server bug fixed.

Exempt NFS from running this test, since we don't expect it to pass.

Cc Yongcheng Yang <[email protected]>

---

Quoted 7 lines
Quoted 4 lines

This case is written for a nfs fix at first. If nfs would like to skip this test, I don't know if it makes sense to keep it in fstests?

It might make sense to keep this test in place for stuff like cephfs, but if dropping it altogether is best, then that's fine with me.

Oh, I forgot this:

Quoted 30 lines
Quoted 26 lines
Quoted 25 lines

 tests/generic/732 | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
tests/generic/732 +1 -3
index 785aac58f361..5b5087d5accd 100755
@@ -22,9 +22,7 @@ _cleanup()
 }
 
 # real QA test starts here
-_supported_fs generic
-[ "$FSTYP" = "nfs" ] && _fixed_by_kernel_commit fdd2630a739819 \
-	"nfsd: fix change_info in NFSv4 RENAME replies"
+_supported_fs ^nfs
 
 _require_test
 _require_scratch
Signature · 2 lines
-- 
2.43.0
Signature · 1 line
Jeff Layton <[email protected]>