Skip to contentSkip to conversation

syscalls/fanotify10: Make evictable marks test more reliable

PATCH In review Jan Kara 3 participants 7 messages Aug 25, 2022 – Aug 27, 2022

lore
Share & tools
lore

Jan Kara

author
#0 lore raw

PATCH 1/1 syscalls/fanotify10: Make evictable marks test more reliable

In some setups evictable marks tests are failing because the inode with evictable mark does not get evicted. Make sure we sync the filesystem before we try to drop caches to increase likelyhood the inode will get evicted.

---
 testcases/kernel/syscalls/fanotify/fanotify10.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
testcases/kernel/syscalls/fanotify/fanotify10.c +10 -1
index 19e43d2c2762..54636ce2ddd4 100644
@@ -342,6 +342,15 @@ static void show_fanotify_marks(int fd)
 	}
 }
 
+static void drop_caches(char *path)
+{
+	int fd = SAFE_OPEN(path, O_RDONLY);
+	if (syncfs(fd) < 0)
+		tst_brk(TBROK | TERRNO, "Unexpected error when syncing filesystem");
+	SAFE_CLOSE(fd);
+	SAFE_FILE_PRINTF(DROP_CACHES_FILE, "3");
+}
+
 static int create_fanotify_groups(unsigned int n)
 {
 	struct tcase *tc = &tcases[n];
@@ -402,7 +411,7 @@ add_mark:
 	 * drop_caches should evict inode from cache and remove evictable marks
 	 */
 	if (evictable_ignored) {
-		SAFE_FILE_PRINTF(DROP_CACHES_FILE, "3");
+		drop_caches(tc->mark_path);
 		for (p = 0; p < num_classes; p++) {
 			for (i = 0; i < GROUPS_PER_PRIO; i++) {
 				if (fd_notify[p][i] > 0)
Signature · 5 lines
2.35.3


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#1 lore raw

Hi Jan, all,

Quoted 4 lines

In some setups evictable marks tests are failing because the inode with evictable mark does not get evicted. Make sure we sync the filesystem before we try to drop caches to increase likelyhood the inode will get evicted.

Merged with minor changes to keep checkpatch.pl happy.

Thanks for fixing this and Cc me on previous discussion.

Given on previous discussion the behavior on ext2 vs. xfs:
would it make sense to transform the test to use .all_filesystems = 1 ?

Kind regards,
Petr

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
author
#2 lore raw
Quoted 8 lines · Petr Vorel

Hi Jan, all,

Quoted 4 lines

In some setups evictable marks tests are failing because the inode with evictable mark does not get evicted. Make sure we sync the filesystem before we try to drop caches to increase likelyhood the inode will get evicted.

Merged with minor changes to keep checkpatch.pl happy.

Thanks!

Quoted 2 lines

Given on previous discussion the behavior on ext2 vs. xfs:
would it make sense to transform the test to use .all_filesystems = 1 ?

Well, I don't think it would improve test coverage in any interesting way. This test tests stuff in fsnotify layer & VFS. The differences in filesystem inode reclaim are not target of this test - we are just trying to check that fsnotify does not block inode reclaim by holding inode references and for that any filesystem works. Or did you mean something else?

Honza

Signature · 5 lines
Jan Kara <[email protected]>
SUSE Labs, CR

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#3 lore raw
Quoted 4 lines · Jan Kara
Quoted 2 lines · Petr Vorel

Hi Jan, all,

Quoted 11 lines
Quoted 6 lines
Quoted 4 lines

In some setups evictable marks tests are failing because the inode with evictable mark does not get evicted. Make sure we sync the filesystem before we try to drop caches to increase likelyhood the inode will get evicted.

Merged with minor changes to keep checkpatch.pl happy.

Thanks!

Quoted 2 lines

Given on previous discussion the behavior on ext2 vs. xfs:
would it make sense to transform the test to use .all_filesystems = 1 ?

On the contrary.
We want the inode reclaim to be as predictable as possible. That is why I suggested to force the test to use ext2 because xfs has some specialized inode reclaim

Quoted 8 lines

Well, I don't think it would improve test coverage in any interesting way. This test tests stuff in fsnotify layer & VFS. The differences in filesystem inode reclaim are not target of this test - we are just trying to check that fsnotify does not block inode reclaim by holding inode references and for that any filesystem works. Or did you mean something else?

Agree. I see no reason to change that.

Thanks,
Amir.

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#4 lore raw
Quoted 14 lines · Jan Kara
Quoted 12 lines · Petr Vorel
Quoted 8 lines

Hi Jan, all,

Quoted 4 lines

In some setups evictable marks tests are failing because the inode with evictable mark does not get evicted. Make sure we sync the filesystem before we try to drop caches to increase likelyhood the inode will get evicted.

Merged with minor changes to keep checkpatch.pl happy.

Thanks!

Quoted 7 lines
Quoted 2 lines
Quoted 2 lines

Given on previous discussion the behavior on ext2 vs. xfs:
would it make sense to transform the test to use .all_filesystems = 1 ?

On the contrary.
We want the inode reclaim to be as predictable as possible. That is why I suggested to force the test to use ext2 because xfs has some specialized inode reclaim

Ah right! Could you please send a patch with .dev_fs_type = "ext2" That should be enough I guess.

Quoted 9 lines
Quoted 6 lines

Well, I don't think it would improve test coverage in any interesting way. This test tests stuff in fsnotify layer & VFS. The differences in filesystem inode reclaim are not target of this test - we are just trying to check that fsnotify does not block inode reclaim by holding inode references and for that any filesystem works. Or did you mean something else?

Agree. I see no reason to change that.

Thank you both for info!

Kind regards,
Petr

Quoted 2 lines

Thanks,
Amir.

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#5 lore raw
Quoted 18 lines · Petr Vorel
Quoted 17 lines · Jan Kara
Quoted 15 lines · Petr Vorel
Quoted 8 lines

Hi Jan, all,

Quoted 4 lines

In some setups evictable marks tests are failing because the inode with evictable mark does not get evicted. Make sure we sync the filesystem before we try to drop caches to increase likelyhood the inode will get evicted.

Merged with minor changes to keep checkpatch.pl happy.

Thanks!

Quoted 2 lines

Given on previous discussion the behavior on ext2 vs. xfs:
would it make sense to transform the test to use .all_filesystems = 1 ?

Quoted 8 lines
Quoted 4 lines

On the contrary.
We want the inode reclaim to be as predictable as possible. That is why I suggested to force the test to use ext2 because xfs has some specialized inode reclaim

Ah right! Could you please send a patch with .dev_fs_type = "ext2" That should be enough I guess.

Petr, I don't think there is a need for that.
I was only pointing out the different fs can have different inode reclaim behaviors.
I had not encountered issues when testing fanotify10 on xfs myself.

Jan's patch should be enough and then the test can run on the user's choice of filesystem, which is always better than unneeded restrictions.

If we see issues on xfs we can reconsider .dev_fs_type = "ext2".

Thanks,
Amir.

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#6 lore raw
Quoted 33 lines · Petr Vorel
Quoted 18 lines
Quoted 17 lines · Jan Kara
Quoted 15 lines · Petr Vorel
Quoted 8 lines

Hi Jan, all,

Quoted 4 lines

In some setups evictable marks tests are failing because the inode with evictable mark does not get evicted. Make sure we sync the filesystem before we try to drop caches to increase likelyhood the inode will get evicted.

Merged with minor changes to keep checkpatch.pl happy.

Thanks!

Quoted 2 lines

Given on previous discussion the behavior on ext2 vs. xfs:
would it make sense to transform the test to use .all_filesystems = 1 ?

Quoted 7 lines
Quoted 4 lines

On the contrary.
We want the inode reclaim to be as predictable as possible. That is why I suggested to force the test to use ext2 because xfs has some specialized inode reclaim

Ah right! Could you please send a patch with .dev_fs_type = "ext2" That should be enough I guess.

Petr, I don't think there is a need for that.
I was only pointing out the different fs can have different inode reclaim behaviors.
I had not encountered issues when testing fanotify10 on xfs myself.

Quoted 5 lines

Jan's patch should be enough and then the test can run on the user's choice of filesystem, which is always better than unneeded restrictions.

If we see issues on xfs we can reconsider .dev_fs_type = "ext2".

Amir, thanks for explanation. Makes sense.

Kind regards,
Petr

Quoted 2 lines

Thanks,
Amir.

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp