Skip to contentSkip to conversation

syscalls/fanotify: Fix running tests in a loop

PATCH New Amir Goldstein 2 participants 5 messages Jul 14, 2021 – Jul 15, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 syscalls/fanotify: Fix running tests in a loop

Both fanotify19 and fanotify18 leaked fds when run in a loop. fanotify19 was not checking child process exit status correctly which resulted in random failures.

---
 testcases/kernel/syscalls/fanotify/fanotify18.c | 5 ++++-
 testcases/kernel/syscalls/fanotify/fanotify19.c | 4 +++-
 2 files changed, 7 insertions(+), 2 deletions(-)
testcases/kernel/syscalls/fanotify/fanotify18.c +4 -1
index 8a7eebba3..54a4b8ba0 100644
@@ -136,7 +136,7 @@ static void test_fanotify(unsigned int n)
 			(tc->mark_flags & DISALLOWED_MARK_FLAGS ||
 			 tc->mark_mask & FAN_ALL_PERM_EVENTS)) {
 			tst_res(TPASS, "Received result EPERM, as expected");
-			return;
+			goto out;
 		}
 
 		tst_brk(TBROK | TERRNO,
@@ -151,6 +151,9 @@ static void test_fanotify(unsigned int n)
 	tst_res(TPASS,
 		"fanotify_init() and fanotify_mark() returned successfully, "
 		"as expected");
+
+out:
+	SAFE_CLOSE(fd_notify);
 }
 
 static void setup(void)
testcases/kernel/syscalls/fanotify/fanotify19.c +3 -1
index 60714408d..fac257bb4 100644
@@ -132,7 +132,7 @@ static void do_fork(void)
 
 	SAFE_WAITPID(child, &status, 0);
 
-	if (WIFEXITED(child) && WEXITSTATUS(child) != 0)
+	if (WIFEXITED(status) && WEXITSTATUS(status) != 0)
 		tst_brk(TBROK, "Child process terminated incorrectly. Aborting");
 }
 
@@ -255,6 +255,8 @@ static void test_fanotify(unsigned int n)
 			event = FAN_EVENT_NEXT(event, len);
 		}
 	}
+
+	SAFE_CLOSE(fd_notify);
 }
 
 static void setup(void)
Signature · 1 line
2.32.0
#1 lore raw

PATCH 1/1 syscalls/fanotify: Fix running tests in a loop

Hi Amir,

Quoted 3 lines

Both fanotify19 and fanotify18 leaked fds when run in a loop. fanotify19 was not checking child process exit status correctly which resulted in random failures.

Thanks for a quick fix!

FYI both fixes prolonged running for about 100% (comparing with runs on master which don't fail). Of course, that's not a problem at all, it's just interesting for me how simple closing of file descriptor can be expensive.

Kind regards,
Petr

# time ./fanotify19 -i40

Summary:

passed   360
failed   0
broken   0
skipped  0

warnings 0

real 0m0,871s
user 0m0,028s
sys 0m0,116s

vs.

Summary:

passed   360
failed   0
broken   0
skipped  0

warnings 0

real 0m0,387s
user 0m0,016s
sys 0m0,122s

# time ./fanotify19 -i10
Summary:
passed   160
failed   0
broken   0
skipped  0
warnings 0

real 0m0,436s
user 0m0,032s
sys 0m0,128s

vs.

Summary:

passed   160
failed   0
broken   0
skipped  0

warnings 0

real 0m0,876s
user 0m0,052s
sys 0m0,105s

#2 lore raw

PATCH 1/1 syscalls/fanotify: Fix running tests in a loop

Quoted 7 lines · Petr Vorel

Hi Amir,

Quoted 3 lines

Both fanotify19 and fanotify18 leaked fds when run in a loop. fanotify19 was not checking child process exit status correctly which resulted in random failures.

Quoted 7 lines

Thanks for a quick fix!

FYI both fixes prolonged running for about 100% (comparing with runs on master which don't fail). Of course, that's not a problem at all, it's just interesting for me how simple closing of file descriptor can be expensive.

Nice observation! LTP is not the only test suite that observed this issue. See:
https://lore.kernel.org/linux-fsdevel/CAC-ggsFLmFpz5Y=-9MMLwxuO2LOS9rhpewDp_-u2hrT9J79ryg@mail.gmail.com/

In order to fix that issue we will need some LTP helpers to take care of closing fanotify/inotify fds in batches for loop runs, which is kind of what happened before
the fd leak fix only the batch was infinite...

Attached sample patch for fanotify18.
You will need to run some more experiments to see if it is worth it or not and then if it is worth it, probably generalize the helpers.

Thanks,
Amir.

Quoted 48 lines

Kind regards,
Petr

# time ./fanotify19 -i40

Summary:

passed   360
failed   0
broken   0
skipped  0

warnings 0

real    0m0,871s
user    0m0,028s
sys     0m0,116s

vs.

Summary:

passed   360
failed   0
broken   0
skipped  0

warnings 0

real    0m0,387s
user    0m0,016s
sys     0m0,122s
# time ./fanotify19 -i10
Summary:
passed   160
failed   0
broken   0
skipped  0
warnings 0
real    0m0,436s
user    0m0,032s
sys     0m0,128s

vs.

Summary:

passed   160
failed   0
broken   0
skipped  0
Quoted 5 lines

warnings 0

real    0m0,876s
user    0m0,052s
sys     0m0,105s

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-syscalls-fanotify-Closing-group-fd-in-batches.patch
Type: text/x-patch
Size: 3862 bytes
Desc: not available
URL: <http://lists.linux.it/pipermail/ltp/attachments/20210715/3b5dbd3e/attachment.bin>

#3 lore raw

PATCH 1/1 syscalls/fanotify: Fix running tests in a loop

Hi Amir,

...

Quoted 7 lines
Quoted 3 lines

FYI both fixes prolonged running for about 100% (comparing with runs on master which don't fail). Of course, that's not a problem at all, it's just interesting for me how simple closing of file descriptor can be expensive.

Nice observation! LTP is not the only test suite that observed this issue. See:
https://lore.kernel.org/linux-fsdevel/CAC-ggsFLmFpz5Y=-9MMLwxuO2LOS9rhpewDp_-u2hrT9J79ryg@mail.gmail.com/

Quoted 8 lines

In order to fix that issue we will need some LTP helpers to take care of closing fanotify/inotify fds in batches for loop runs, which is kind of what happened before
the fd leak fix only the batch was infinite...

Attached sample patch for fanotify18.
You will need to run some more experiments to see if it is worth it or not and then if it is worth it, probably generalize the helpers.

Thanks for detailed info and patch. I'm going to merge the original patch as it's not an issue to run tests slower (more important is to actually fix the issue) and have look into your patchset whether it's worth of adding it.

Kind regards,
Petr