fs: warn in case userspace lied about modprobe return

PATCH Superseded Luis R. Rodriguez 2 participants 3 messages May 26, 2017 – Jun 1, 2017 VFS

Share & tools

Luis R. Rodriguez

author

2017-05-26 00:44 UTC #0 lore raw

PATCH 1/1 fs: warn in case userspace lied about modprobe return

kmod <= v19 was broken -- it could return 0 to modprobe calls, incorrectly assuming that a kernel module was built-in, whereas in reality the module was just forming in the kernel. The reason for this is an incorrect userspace heuristics. A userspace kmod fix is available for it [0], however should userspace break again we could go on with an failed get_fs_type() which is hard to debug as the request_module() is detected as returning 0. The first suspect would be that there is something worth with the kernel's module loader and obviously in this case that is not the issue.

Since these issues are painful to debug complain when we know userspace has outright lied to us.

[0] http://git.kernel.org/cgit/utils/kernel/kmod/kmod.git/commit/libkmod/libkmod-module.c?id=fd44a98ae2eb5eb32161088954ab21e58e19dfc4

Cc Jessica Yu <[email protected]>

Signed-off-by Luis R. Rodriguez <[email protected]>

---
 fs/filesystems.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

fs/filesystems.c +3 -1

index cac75547d35c..0f477a5de6ea 100644
@@ -275,8 +275,10 @@ struct file_system_type *get_fs_type(const char *name)
 	int len = dot ? dot - name : strlen(name);
 
 	fs = __get_fs_type(name, len);
-	if (!fs && (request_module("fs-%.*s", len, name) == 0))
+	if (!fs && (request_module("fs-%.*s", len, name) == 0)) {
+		WARN_ONCE(!fs, "request_module fs-%.*s succeeded, but still no fs?\n", len, name);
 		fs = __get_fs_type(name, len);
+	}
 
 	if (dot && fs && !(fs->fs_flags & FS_HAS_SUBTYPE)) {
 		put_filesystem(fs);

Signature · 1 line

2.11.0

Jessica Yu

2017-05-28 02:01 UTC #1 lore raw

+++ Luis R. Rodriguez [25/05/17 17:44 -0700]:

Quoted 27 lines

kmod <= v19 was broken -- it could return 0 to modprobe calls, incorrectly assuming that a kernel module was built-in, whereas in reality the module was just forming in the kernel. The reason for this is an incorrect userspace heuristics. A userspace kmod fix is available for it [0], however should userspace break again we could go on with an failed get_fs_type() which is hard to debug as the request_module() is detected as returning 0. The first suspect would be that there is something worth with the kernel's module loader and obviously in this case that is not the issue.
Since these issues are painful to debug complain when we know userspace has outright lied to us.
[0] http://git.kernel.org/cgit/utils/kernel/kmod/kmod.git/commit/libkmod/libkmod-module.c?id=fd44a98ae2eb5eb32161088954ab21e58e19dfc4
Suggested-by Rusty Russell <[email protected]>
Cc Jessica Yu <[email protected]>
Signed-off-by Luis R. Rodriguez <[email protected]>
---
fs/filesystems.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
fs/filesystems.c +0 -0
index cac75547d35c..0f477a5de6ea 100644

fs/filesystems.c struct file_system_type *get_fs_type(const char *name) · line 279

Quoted 6 lines

fs/filesystems.c +2 -1

@@ -275,8 +275,10 @@ struct file_system_type *get_fs_type(const char *name)
 	int len = dot ? dot - name : strlen(name);
 
 	fs = __get_fs_type(name, len);
-	if (!fs && (request_module("fs-%.*s", len, name) == 0))
+	if (!fs && (request_module("fs-%.*s", len, name) == 0)) {
+		WARN_ONCE(!fs, "request_module fs-%.*s succeeded, but still no fs?\n", len, name);

The WARN needs to go below the second __get_fs_type() attempt, no? Because we want to try __get_fs_type() again right after request_module(), to see if the fs loaded, and _then_ WARN if it doesn't appear to be loaded.

Jessica

Quoted 7 lines

fs/filesystems.c +1 -0

 		fs = __get_fs_type(name, len);
+	}
 
 	if (dot && fs && !(fs->fs_flags & FS_HAS_SUBTYPE)) {
 		put_filesystem(fs);

Signature · 1 line

2.11.0

Luis R. Rodriguez

author

2017-06-01 18:03 UTC #2 lore raw

Quoted 30 lines · Jessica Yu

+++ Luis R. Rodriguez [25/05/17 17:44 -0700]:
Quoted 29 lines
kmod <= v19 was broken -- it could return 0 to modprobe calls, incorrectly assuming that a kernel module was built-in, whereas in reality the module was just forming in the kernel. The reason for this is an incorrect userspace heuristics. A userspace kmod fix is available for it [0], however should userspace break again we could go on with an failed get_fs_type() which is hard to debug as the request_module() is detected as returning 0. The first suspect would be that there is something worth with the kernel's module loader and obviously in this case that is not the issue.
Since these issues are painful to debug complain when we know userspace has outright lied to us.
[0] http://git.kernel.org/cgit/utils/kernel/kmod/kmod.git/commit/libkmod/libkmod-module.c?id=fd44a98ae2eb5eb32161088954ab21e58e19dfc4
Suggested-by Rusty Russell <[email protected]>
Cc Jessica Yu <[email protected]>
Signed-off-by Luis R. Rodriguez <[email protected]>
---
fs/filesystems.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
fs/filesystems.c +0 -0
index cac75547d35c..0f477a5de6ea 100644
@@ -275,8 +275,10 @@ struct file_system_type *get_fs_type(const char *name)
 	int len = dot ? dot - name : strlen(name);
 

Quoted 9 lines

fs/filesystems.c struct file_system_type *get_fs_type(const char *name) · line 279
Quoted 4 lines
fs/filesystems.c +2 -1
 	fs = __get_fs_type(name, len);
-	if (!fs && (request_module("fs-%.*s", len, name) == 0))
+	if (!fs && (request_module("fs-%.*s", len, name) == 0)) {
+		WARN_ONCE(!fs, "request_module fs-%.*s succeeded, but still no fs?\n", len, name);
The WARN needs to go below the second __get_fs_type() attempt, no? Because we want to try __get_fs_type() again right after request_module(), to see if the fs loaded, and _then_ WARN if it doesn't appear to be loaded.

Doh, yes, sorry will send v2.

Luis