Skip to contentSkip to conversation

xfstests: add d_type checking to fsstress

RFC PATCH In review Eric Sandeen 4 participants 5 messages Sep 16, 2013 – Sep 17, 2013

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 xfstests: add d_type checking to fsstress

This patch adds a "-D" switch to fsstress so that every time we call readdir, we stat the dentry & compare it's st_mode to the d_type.

If -D is specified only once, it ignores DT_UNKNOWN. If specified twice, it considers DT_UNKNOWN to be an error.

It skips paths of "./." and "./.." so that we only look at files newly created within the filesystem.

This could be used in an xfstest; it's noisy on a failures so would break expected output.

---

fsstress doesn't usually do validation, but it's such a handy framework for creating a ton of random files, this seems like an ok place to put it. What do folks think?

ltp/fsstress.c +69 -2
index 5d5611f..711ea9a 100644
@@ -246,6 +246,7 @@ int		rtpct;
 unsigned long	seed = 0;
 ino_t		top_ino;
 int		verbose = 0;
+int		verify_d_type = 0;
 sig_atomic_t	should_stop = 0;
 char		*execute_cmd = NULL;
 int		execute_freq = 1;
@@ -315,7 +316,7 @@ int main(int argc, char **argv)
 	int             nousage = 0;
 	xfs_error_injection_t	        err_inj;
 	struct sigaction action;
-	const char	*allopts = "d:e:f:i:m:M:n:o:p:rs:S:vwx:X:zH";
+	const char	*allopts = "d:De:f:i:m:M:n:o:p:rs:S:vwx:X:zH";
 
 	errrange = errtag = 0;
 	umask(0);
@@ -327,6 +328,9 @@ int main(int argc, char **argv)
 		case 'd':
 			dirname = optarg;
 			break;
+		case 'D':
+			verify_d_type++;
+			break;
 		case 'e':
 			sscanf(optarg, "%d", &errtag);
 			if (errtag < 0) {
@@ -1491,6 +1495,7 @@ usage(void)
 	printf("          [-p nproc][-r len][-s seed][-v][-w][-x cmd][-z][-S][-X ncmd]\n");
 	printf("where\n");
 	printf("   -d dir           specifies the base directory for operations\n");
+	printf("   -D               verify d_type in any readdir operations, 2x to disallow DT_UNKNOWN\n");
 	printf("   -e errtg         specifies error injection stuff\n");
 	printf("   -f op_name=freq  changes the frequency of option name to freq\n");
 	printf("                    the valid operation names are:\n");
@@ -2338,12 +2343,71 @@ getattr_f(int opno, long r)
 	close(fd);
 }
 
+void test_d_type(int opno, pathname_t *f, struct dirent64 *de)
+{
+	struct stat64 sb;
+	char path[PATH_MAX];
+
+	snprintf(path, PATH_MAX, "%s/%s", f->path, de->d_name);
+
+	/* Don't check ./. or ./.. */
+	if (!strncmp(path, "./.", 3))
+		return;
+
+	if (lstat64(path, &sb)) {
+		printf("%d/%d: getdents - can't stat %s\n",
+			procid, opno, path);
+	} else {
+		int bad_d_type = 0;
+
+		switch (de->d_type) {
+			case DT_BLK:
+				if (!S_ISBLK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_CHR:
+				if (!S_ISCHR(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_DIR:
+				if (!S_ISDIR(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_FIFO:
+				if (!S_ISFIFO(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_LNK:
+				if (!S_ISLNK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_REG:
+				if (!S_ISREG(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_SOCK:
+				if (!S_ISSOCK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_UNKNOWN:
+				if (verify_d_type > 1)
+					bad_d_type++;
+			break;
+		}
+		if (bad_d_type)
+			printf("%d/%d: getdents - bad d_type %d for %s\n",
+				procid, opno, de->d_type, path);
+			
+	}
+}
+
 void
 getdents_f(int opno, long r)
 {
 	DIR		*dir;
 	pathname_t	f;
 	int		v;
+	struct dirent64	*de;
 
 	init_pathname(&f);
 	if (!get_fname(FT_DIRm, r, &f, NULL, NULL, &v))
@@ -2357,8 +2421,11 @@ getdents_f(int opno, long r)
 		free_pathname(&f);
 		return;
 	}
-	while (readdir64(dir) != NULL)
+	while ((de = readdir64(dir)) != NULL) {
+		if (verify_d_type)
+			test_d_type(opno, &f, de);
 		continue;
+	}
 	if (v)
 		printf("%d/%d: getdents %s 0\n", procid, opno, f.path);
 	free_pathname(&f);

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#1 lore raw
Quoted 13 lines · Eric Sandeen

This patch adds a "-D" switch to fsstress so that every time we call readdir, we stat the dentry& compare it's st_mode to the d_type.

If -D is specified only once, it ignores DT_UNKNOWN. If specified twice, it considers DT_UNKNOWN to be an error.

It skips paths of "./." and "./.." so that we only look at files newly created within the filesystem.

This could be used in an xfstest; it's noisy on a failures so would break expected output.

Quoted 6 lines

---

fsstress doesn't usually do validation, but it's such a handy framework for creating a ton of random files, this seems like an ok place to put it. What do folks think?

I tried v5 and v4 with and without inode fields in the directory. This looks good to me. Like the 2 levels of the test.

--Mark.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#2 lore raw
In reply to Eric Sandeen
Quoted 6 lines · Eric Sandeen

This patch adds a "-D" switch to fsstress so that every time we call readdir, we stat the dentry & compare it's st_mode to the d_type.

If -D is specified only once, it ignores DT_UNKNOWN. If specified twice, it considers DT_UNKNOWN to be an error.

Hmmmm. DT_UNKNOWN is actually a valid type on disk right through to the userspace interface. I can't think of why we'd want to consider it invalid, especially as right now xfs_repair siply zeros the field when recreating directory entries...

Cheers,

Dave.

ltp/fsstress.c getattr_f(int opno, long r) · line 2355

Quoted 10 lines
ltp/fsstress.c +10 -0
+void test_d_type(int opno, pathname_t *f, struct dirent64 *de)
+{
+	struct stat64 sb;
+	char path[PATH_MAX];
+
+	snprintf(path, PATH_MAX, "%s/%s", f->path, de->d_name);
+
+	/* Don't check ./. or ./.. */
+	if (!strncmp(path, "./.", 3))
+		return;

. and .. should have the values of DT_UNKNOWN or DT_DIR. They are the only valid values for these entries.

Quoted 31 lines
ltp/fsstress.c +31 -0
+
+	if (lstat64(path, &sb)) {
+		printf("%d/%d: getdents - can't stat %s\n",
+			procid, opno, path);
+	} else {
+		int bad_d_type = 0;
+
+		switch (de->d_type) {
+			case DT_BLK:
+				if (!S_ISBLK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_CHR:
+				if (!S_ISCHR(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_DIR:
+				if (!S_ISDIR(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_FIFO:
+				if (!S_ISFIFO(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_LNK:
+				if (!S_ISLNK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_REG:
+				if (!S_ISREG(sb.st_mode))
+					bad_d_type++;

ltp/fsstress.c getattr_f(int opno, long r) · line 2396

Quoted 10 lines
ltp/fsstress.c +10 -0
+				break;
+			case DT_SOCK:
+				if (!S_ISSOCK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_UNKNOWN:
+				if (verify_d_type > 1)
+					bad_d_type++;
+			break;
+		}

And DT_WHT? That's defined on disk and in the user interface ;)

i.e. this will not do the right thing with an unknown de->d_type.

Cheers,

Dave.

Signature · 7 lines
Dave Chinner
[email protected]

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
#3 lore raw
Quoted 5 lines · Dave Chinner
Quoted 4 lines · Eric Sandeen

This patch adds a "-D" switch to fsstress so that every time we call readdir, we stat the dentry & compare it's st_mode to the d_type.

Quoted 7 lines
Quoted 2 lines

If -D is specified only once, it ignores DT_UNKNOWN. If specified twice, it considers DT_UNKNOWN to be an error.

Hmmmm. DT_UNKNOWN is actually a valid type on disk right through to the userspace interface. I can't think of why we'd want to consider it invalid, especially as right now xfs_repair siply zeros the field when recreating directory entries...

well, I didn't know that last part. (but why do that?)

I was thinking along the lines of DT_UNKNOWN returns meaning "this fs doesn't support d_type."

"If the file type could not be determined, the value DT_UNKNOWN is returned in d_type."

What d_type-grokking fs can't determine its file types?

-Eric

Quoted 62 lines

Cheers,

Dave.

ltp/fsstress.c getattr_f(int opno, long r) · line 2355

Quoted 10 lines
ltp/fsstress.c +10 -0
+void test_d_type(int opno, pathname_t *f, struct dirent64 *de)
+{
+	struct stat64 sb;
+	char path[PATH_MAX];
+
+	snprintf(path, PATH_MAX, "%s/%s", f->path, de->d_name);
+
+	/* Don't check ./. or ./.. */
+	if (!strncmp(path, "./.", 3))
+		return;

. and .. should have the values of DT_UNKNOWN or DT_DIR. They are the only valid values for these entries.

Quoted 31 lines
ltp/fsstress.c +31 -0
+
+	if (lstat64(path, &sb)) {
+		printf("%d/%d: getdents - can't stat %s\n",
+			procid, opno, path);
+	} else {
+		int bad_d_type = 0;
+
+		switch (de->d_type) {
+			case DT_BLK:
+				if (!S_ISBLK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_CHR:
+				if (!S_ISCHR(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_DIR:
+				if (!S_ISDIR(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_FIFO:
+				if (!S_ISFIFO(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_LNK:
+				if (!S_ISLNK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_REG:
+				if (!S_ISREG(sb.st_mode))
+					bad_d_type++;

ltp/fsstress.c getattr_f(int opno, long r) · line 2396

Quoted 10 lines
ltp/fsstress.c +10 -0
+				break;
+			case DT_SOCK:
+				if (!S_ISSOCK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_UNKNOWN:
+				if (verify_d_type > 1)
+					bad_d_type++;
+			break;
+		}

And DT_WHT? That's defined on disk and in the user interface ;)

Quoted 6 lines

i.e. this will not do the right thing with an unknown de->d_type.

Cheers,

Dave.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#4 lore raw
In reply to Dave Chinner
Quoted 8 lines · Dave Chinner
Quoted 6 lines · Eric Sandeen

This patch adds a "-D" switch to fsstress so that every time we call readdir, we stat the dentry & compare it's st_mode to the d_type.

If -D is specified only once, it ignores DT_UNKNOWN. If specified twice, it considers DT_UNKNOWN to be an error.

Quoted 8 lines

Hmmmm. DT_UNKNOWN is actually a valid type on disk right through to the userspace interface. I can't think of why we'd want to consider it invalid, especially as right now xfs_repair siply zeros the field when recreating directory entries...

Cheers,

Dave.

no fair signing off w/ more text below ;)

Quoted 5 lines
Quoted 5 lines
ltp/fsstress.c +5 -0
+void test_d_type(int opno, pathname_t *f, struct dirent64 *de)
+{
+	struct stat64 sb;
+	char path[PATH_MAX];
+
Quoted 8 lines

ltp/fsstress.c getattr_f(int opno, long r) · line 2355

Quoted 5 lines
ltp/fsstress.c +5 -0
+	snprintf(path, PATH_MAX, "%s/%s", f->path, de->d_name);
+
+	/* Don't check ./. or ./.. */
+	if (!strncmp(path, "./.", 3))
+		return;

. and .. should have the values of DT_UNKNOWN or DT_DIR. They are the only valid values for these entries.

Hm let me look at something, I saw something that prompted this but now that I think about it maybe it's a bug.

Quoted 35 lines
Quoted 35 lines
ltp/fsstress.c +35 -0
+
+	if (lstat64(path, &sb)) {
+		printf("%d/%d: getdents - can't stat %s\n",
+			procid, opno, path);
+	} else {
+		int bad_d_type = 0;
+
+		switch (de->d_type) {
+			case DT_BLK:
+				if (!S_ISBLK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_CHR:
+				if (!S_ISCHR(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_DIR:
+				if (!S_ISDIR(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_FIFO:
+				if (!S_ISFIFO(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_LNK:
+				if (!S_ISLNK(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_REG:
+				if (!S_ISREG(sb.st_mode))
+					bad_d_type++;
+				break;
+			case DT_SOCK:
+				if (!S_ISSOCK(sb.st_mode))
+					bad_d_type++;
Quoted 8 lines

ltp/fsstress.c getattr_f(int opno, long r) · line 2396

Quoted 6 lines
ltp/fsstress.c +6 -0
+				break;
+			case DT_UNKNOWN:
+				if (verify_d_type > 1)
+					bad_d_type++;
+			break;
+		}

And DT_WHT? That's defined on disk and in the user interface ;)

but fsstress won't create it, will it?

Quoted 1 line

i.e. this will not do the right thing with an unknown de->d_type.

... but we know what fsstress can possibly create, right, so testing those created types should be safe, I'd think.

I'll give this some thought & send V2.

-Eric

Quoted 4 lines

Cheers,

Dave.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs