Skip to contentSkip to conversation

pathconf01: Allow EINVAL where POSIX leaves support unspecified

PATCH Applied Wake Liu via ltp 2 participants 6 messages Oct 5 – Oct 13

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pathconf01: Allow EINVAL where POSIX leaves support unspecified

POSIX lists two EINVAL cases for pathconf(). It shall fail with EINVAL when the value of name is not valid. It may also fail with EINVAL when "the implementation does not support an association of the variable name with the specified file".

The test always calls pathconf() on a directory. For a directory, POSIX leaves it unspecified whether _PC_MAX_CANON, _PC_MAX_INPUT and _PC_VDISABLE (note 2), _PC_ASYNC_IO, _PC_PRIO_IO and _PC_SYNC_IO (note 8) and the _PC_REC_* names (note 10) are supported, so EINVAL is allowed for them.

Bionic before Android 15 returns -1 and sets errno to EINVAL for _PC_REC_INCR_XFER_SIZE and _PC_REC_MAX_XFER_SIZE, which makes the test fail.

Report EINVAL as TCONF for these names. EINVAL for the other names is still a failure.

Also print the name and errno on failure, so it is clear which name failed.

---
 .../kernel/syscalls/pathconf/pathconf01.c     | 36 ++++++++++++-------
 1 file changed, 23 insertions(+), 13 deletions(-)
testcases/kernel/syscalls/pathconf/pathconf01.c +23 -13
index 5af45e417..635c356bc 100644
@@ -19,24 +19,26 @@ static char *path;
 static struct tcase {
 	int value;
 	char *name;
+	/* POSIX leaves support for a directory unspecified */
+	int einval_ok;
 } tcases[] = {
 	{NAME_DESC(_PC_LINK_MAX)},
-	{NAME_DESC(_PC_MAX_CANON)},
-	{NAME_DESC(_PC_MAX_INPUT)},
+	{NAME_DESC(_PC_MAX_CANON), .einval_ok = 1},
+	{NAME_DESC(_PC_MAX_INPUT), .einval_ok = 1},
 	{NAME_DESC(_PC_NAME_MAX)},
 	{NAME_DESC(_PC_PATH_MAX)},
 	{NAME_DESC(_PC_PIPE_BUF)},
 	{NAME_DESC(_PC_CHOWN_RESTRICTED)},
 	{NAME_DESC(_PC_NO_TRUNC)},
-	{NAME_DESC(_PC_VDISABLE)},
-	{NAME_DESC(_PC_SYNC_IO)},
-	{NAME_DESC(_PC_ASYNC_IO)},
-	{NAME_DESC(_PC_PRIO_IO)},
+	{NAME_DESC(_PC_VDISABLE), .einval_ok = 1},
+	{NAME_DESC(_PC_SYNC_IO), .einval_ok = 1},
+	{NAME_DESC(_PC_ASYNC_IO), .einval_ok = 1},
+	{NAME_DESC(_PC_PRIO_IO), .einval_ok = 1},
 	{NAME_DESC(_PC_FILESIZEBITS)},
-	{NAME_DESC(_PC_REC_INCR_XFER_SIZE)},
-	{NAME_DESC(_PC_REC_MAX_XFER_SIZE)},
-	{NAME_DESC(_PC_REC_MIN_XFER_SIZE)},
-	{NAME_DESC(_PC_REC_XFER_ALIGN)},
+	{NAME_DESC(_PC_REC_INCR_XFER_SIZE), .einval_ok = 1},
+	{NAME_DESC(_PC_REC_MAX_XFER_SIZE), .einval_ok = 1},
+	{NAME_DESC(_PC_REC_MIN_XFER_SIZE), .einval_ok = 1},
+	{NAME_DESC(_PC_REC_XFER_ALIGN), .einval_ok = 1},
 };
 
 static void verify_pathconf(unsigned int i)
@@ -47,10 +49,18 @@ static void verify_pathconf(unsigned int i)
 
 	TEST(pathconf(path, tc->value));
 
-	if (TST_RET == -1 && errno != 0)
-		tst_res(TFAIL, "pathconf Failed, errno = %d", TST_ERR);
-	else
+	if (TST_RET != -1 || !TST_ERR) {
 		tst_res(TPASS, "pathconf(%s, %s)", path, tc->name);
+		return;
+	}
+
+	if (tc->einval_ok && TST_ERR == EINVAL) {
+		tst_res(TCONF, "pathconf(%s, %s) not supported for a directory",
+			path, tc->name);
+		return;
+	}
+
+	tst_res(TFAIL | TTERRNO, "pathconf(%s, %s)", path, tc->name);
 }
 
 static struct tst_test test = {
Signature · 5 lines
2.56.0.rc1.315.gc6ed9934b7-goog


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

Hi Wake,

Quoted 1 line · Wake Liu via ltp

[PATCH] pathconf01: Allow EINVAL where POSIX leaves support unspecified

Verdict - Reviewed

---
Note:

The agent can sometimes produce false positives although often its findings are genuine. If you find issues with the review, please comment this email or ignore the suggestions.

Regards,
LTP AI Reviewer

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#2 lore raw
In reply to Wake Liu via ltp

Hi Wake,

Quoted 31 lines

POSIX lists two EINVAL cases for pathconf(). It shall fail with EINVAL when the value of name is not valid. It may also fail with EINVAL when "the implementation does not support an association of the variable name with the specified file".

The test always calls pathconf() on a directory. For a directory, POSIX leaves it unspecified whether _PC_MAX_CANON, _PC_MAX_INPUT and _PC_VDISABLE (note 2), _PC_ASYNC_IO, _PC_PRIO_IO and _PC_SYNC_IO (note 8) and the _PC_REC_* names (note 10) are supported, so EINVAL is allowed for them.

Bionic before Android 15 returns -1 and sets errno to EINVAL for _PC_REC_INCR_XFER_SIZE and _PC_REC_MAX_XFER_SIZE, which makes the test fail.

Report EINVAL as TCONF for these names. EINVAL for the other names is still a failure.

Also print the name and errno on failure, so it is clear which name failed.

---
 .../kernel/syscalls/pathconf/pathconf01.c     | 36 ++++++++++++-------
 1 file changed, 23 insertions(+), 13 deletions(-)
testcases/kernel/syscalls/pathconf/pathconf01.c +0 -0
index 5af45e417..635c356bc 100644

testcases/kernel/syscalls/pathconf/pathconf01.c static char *path; · line 23

Quoted 5 lines
testcases/kernel/syscalls/pathconf/pathconf01.c +2 -0
@@ -19,24 +19,26 @@ static char *path;
 static struct tcase {
 	int value;
 	char *name;
+	/* POSIX leaves support for a directory unspecified */
+	int einval_ok;

Usually, in these cases, it's better to define an array of errno we might have when we trigger the syscall in a certain way. And then we use TST_EXP_ARR_FAIL() on it.

--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

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

Hi Andrea,

Quoted 6 lines
Quoted 2 lines

+ /* POSIX leaves support for a directory unspecified */
+ int einval_ok;

Usually, in these cases, it's better to define an array of errno we might have when we trigger the syscall in a certain way. And then we use TST_EXP_ARR_FAIL() on it.

Thanks for the review!

pathconf01 is a positive test, and the einval_ok names actually succeed on glibc, musl and Bionic since Android 15: they return >= 0, or -1 without changing errno.

TST_EXP_FAIL_ARR() and TST_EXP_FAIL2_ARR() expect the call to fail, so they report TFAIL wherever these names are supported for directories. I tried TST_EXP_FAIL2_ARR() and all 10 einval_ok names fail on glibc ("succeeded" or "expected EINVAL: SUCCESS (0)").

Here we still expect pathconf() to succeed, and only want to treat EINVAL as TCONF for the names where POSIX leaves directory support unspecified (POSIX only defines EINVAL for this case).

Does keeping the einval_ok check make sense to you, or would you still prefer storing an exp_errs array in struct tcase and checking it with tst_errno_in_set(), the helper behind TST_EXP_FAIL_ARR()?

Best regards,
Wake

Quoted 48 lines · Andrea Cervesato

Hi Wake,

Quoted 31 lines

POSIX lists two EINVAL cases for pathconf(). It shall fail with EINVAL when the value of name is not valid. It may also fail with EINVAL when "the implementation does not support an association of the variable name with the specified file".

The test always calls pathconf() on a directory. For a directory, POSIX leaves it unspecified whether _PC_MAX_CANON, _PC_MAX_INPUT and _PC_VDISABLE (note 2), _PC_ASYNC_IO, _PC_PRIO_IO and _PC_SYNC_IO (note 8) and the _PC_REC_* names (note 10) are supported, so EINVAL is allowed for them.

Bionic before Android 15 returns -1 and sets errno to EINVAL for _PC_REC_INCR_XFER_SIZE and _PC_REC_MAX_XFER_SIZE, which makes the test fail.

Report EINVAL as TCONF for these names. EINVAL for the other names is still a failure.

Also print the name and errno on failure, so it is clear which name failed.

---
 .../kernel/syscalls/pathconf/pathconf01.c     | 36 ++++++++++++-------
 1 file changed, 23 insertions(+), 13 deletions(-)
testcases/kernel/syscalls/pathconf/pathconf01.c +0 -0
index 5af45e417..635c356bc 100644

testcases/kernel/syscalls/pathconf/pathconf01.c static char *path; · line 23

Quoted 5 lines
testcases/kernel/syscalls/pathconf/pathconf01.c +2 -0
@@ -19,24 +19,26 @@ static char *path;
 static struct tcase {
      int value;
      char *name;
+     /* POSIX leaves support for a directory unspecified */
+     int einval_ok;

Usually, in these cases, it's better to define an array of errno we might have when we trigger the syscall in a certain way. And then we use TST_EXP_ARR_FAIL() on it.

--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Signature · 5 lines
Best Regards,
Wake Liu

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
author Reviewed-by Andrea Cervesato
#4 lore raw

Hi Wake,

Quoted 9 lines

pathconf01 is a positive test, and the einval_ok names actually succeed on glibc, musl and Bionic since Android 15: they return >= 0, or -1 without changing errno.

TST_EXP_FAIL_ARR() and TST_EXP_FAIL2_ARR() expect the call to fail, so they report TFAIL wherever these names are supported for directories. I tried TST_EXP_FAIL2_ARR() and all 10 einval_ok names fail on glibc ("succeeded" or "expected EINVAL: SUCCESS (0)").

Quoted 7 lines

Here we still expect pathconf() to succeed, and only want to treat EINVAL as TCONF for the names where POSIX leaves directory support unspecified (POSIX only defines EINVAL for this case).

Does keeping the einval_ok check make sense to you, or would you still prefer storing an exp_errs array in struct tcase and checking it with tst_errno_in_set(), the helper behind TST_EXP_FAIL_ARR()?

Indeed, my review was biased by unrelated usage of TST_EXP_FAIL_ARR(). The patch is correct as is and we don't have any macro hich could substitute the manual check we are applying in this patch.

Thanks for pointing it out.

--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

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