Skip to contentSkip to conversation

fs: kill old ms_* flags for internal sb

RFC In review Luis Chamberlain 2 participants 7 messages Jan 10, 2023

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover fs: kill old ms_* flags for internal sb

David had started the sb flag split for internal flags through commit e462ec50cb5 ("VFS: Differentiate mount flags (MS_*) from internal superblock flags") but it seems we just never axed out the old flag usage.

I found this while inspecting adding a new temporary flag for the superblock for filesystem freeze support. This doesn't go even build tested, hence RFC.

Luis Chamberlain (3):
apparmor: use SB_* flags for private sb flags
fs: use SB_NOUSER on path_mount() instead of deprecated MS_NOUSER
fs: remove old MS_* internal flags for the superblock

 fs/namespace.c                    | 2 +-
 include/uapi/linux/mount.h        | 8 --------
 security/apparmor/include/mount.h | 3 ++-
 security/apparmor/lsm.c           | 1 +
 security/apparmor/mount.c         | 2 +-
 tools/include/uapi/linux/mount.h  | 8 --------
 6 files changed, 5 insertions(+), 19 deletions(-)
Signature · 1 line
2.35.1
#1 lore raw

RFC 1/3 apparmor: use SB_* flags for private sb flags

Commit 2ea3ffb7782 ("apparmor: add mount mediation") John Johansen added mount mediation support. However just the day before this commit David Howells modified the internal sb flags through commit e462ec50cb5 ("VFS: Differentiate mount flags (MS_*) from internal superblock flags").

Use the modified sb flags to make things clear and avoid further uses of the old MS_* flags for superblock internal flags. This will let us later remove the MS_* sb internal flags as userspace should not be using them.

This commit does not fix anything as the old flags used map to the same bitmask, this just tidies things up. I split up the flags to make it clearer which ones are for the superblock and used internally.

Cc John Johansen <[email protected]>
Cc David Howells <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Al Viro <[email protected]>
---
 security/apparmor/include/mount.h | 3 ++-
 security/apparmor/lsm.c           | 1 +
 security/apparmor/mount.c         | 2 +-
 3 files changed, 4 insertions(+), 2 deletions(-)
security/apparmor/include/mount.h +2 -1
index a710683b2496..f90e03405e38 100644
@@ -23,7 +23,8 @@
 #define AA_AUDIT_DATA		0x40
 #define AA_MNT_CONT_MATCH	0x40
 
-#define AA_MS_IGNORE_MASK (MS_KERNMOUNT | MS_NOSEC | MS_ACTIVE | MS_BORN)
+#define AA_MS_IGNORE_MASK (MS_KERNMOUNT)
+#define AA_SB_IGNORE_MASK (SB_NOSEC | SB_ACTIVE | SB_BORN)
 
 int aa_remount(struct aa_label *label, const struct path *path,
 	       unsigned long flags, void *data);
security/apparmor/lsm.c +1 -0
index c6728a629437..f3880956bffd 100644
@@ -583,6 +583,7 @@ static int apparmor_sb_mount(const char *dev_name, const struct path *path,
 		flags &= ~MS_MGC_MSK;
 
 	flags &= ~AA_MS_IGNORE_MASK;
+	flags &= ~AA_SB_IGNORE_MASK;
 
 	label = __begin_current_label_crit_section();
 	if (!unconfined(label)) {
security/apparmor/mount.c +1 -1
index cdfa430ae216..c37c451e8226 100644
@@ -74,7 +74,7 @@ static void audit_mnt_flags(struct audit_buffer *ab, unsigned long flags)
 		audit_log_format(ab, ", iversion");
 	if (flags & MS_STRICTATIME)
 		audit_log_format(ab, ", strictatime");
-	if (flags & MS_NOUSER)
+	if (flags & SB_NOUSER)
 		audit_log_format(ab, ", nouser");
 }
 
Signature · 1 line
2.35.1
maintainer
#2 lore raw

Re: [RFC 1/3] apparmor: use SB_* flags for private sb flags

Quoted 5 lines · Luis Chamberlain

Commit 2ea3ffb7782 ("apparmor: add mount mediation") John Johansen added mount mediation support. However just the day before this commit David Howells modified the internal sb flags through commit e462ec50cb5 ("VFS: Differentiate mount flags (MS_*) from internal superblock flags").

Quoted 8 lines

Use the modified sb flags to make things clear and avoid further uses of the old MS_* flags for superblock internal flags. This will let us later remove the MS_* sb internal flags as userspace should not be using them.

This commit does not fix anything as the old flags used map to the same bitmask, this just tidies things up. I split up the flags to make it clearer which ones are for the superblock and used internally.

I don't think that's right. apparmor_sb_mount() gets (almost) raw flags from mount(2); incidentally, MS_MGC_MSK removal directly above the modified line is BS since _that_ has already been done by the caller.

Note that the same function explicitly checks for MS_MOVE, etc. in the same argument.

security/apparmor/mount.c static void audit_mnt_flags(struct audit_buffer *ab, unsigned long flags) · line 79

Quoted 8 lines
security/apparmor/mount.c +1 -1
@@ -74,7 +74,7 @@ static void audit_mnt_flags(struct audit_buffer *ab, unsigned long flags)
 		audit_log_format(ab, ", iversion");
 	if (flags & MS_STRICTATIME)
 		audit_log_format(ab, ", strictatime");
-	if (flags & MS_NOUSER)
+	if (flags & SB_NOUSER)
 		audit_log_format(ab, ", nouser");
 }

Umm... How does one trigger that one?

#3 lore raw

RFC 2/3 fs: use SB_NOUSER on path_mount() instead of deprecated MS_NOUSER

In reply to Luis Chamberlain

The goal behind 462ec50cb5 ("VFS: Differentiate mount flags (MS_*) from internal superblock flags") was to phase out MS_* users for internal uses. But we can't remove the old MS_* until we have all users out so just use the SB_* helper for this check.

---
 fs/namespace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/namespace.c +1 -1
index ab467ee58341..bf1cc8527057 100644
@@ -3411,7 +3411,7 @@ int path_mount(const char *dev_name, struct path *path,
 	if (data_page)
 		((char *)data_page)[PAGE_SIZE - 1] = 0;
 
-	if (flags & MS_NOUSER)
+	if (flags & SB_NOUSER)
 		return -EINVAL;
 
 	ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
Signature · 1 line
2.35.1
maintainer
#4 lore raw

Re: [RFC 2/3] fs: use SB_NOUSER on path_mount() instead of deprecated MS_NOUSER

Quoted 4 lines · Luis Chamberlain

The goal behind 462ec50cb5 ("VFS: Differentiate mount flags (MS_*) from internal superblock flags") was to phase out MS_* users for internal uses. But we can't remove the old MS_* until we have all users out so just use the SB_* helper for this check.

No. The goal had been to separate the places where we deal with mount(2) argument encoding from those where we are deal with superblock flags. path_mount() is very much in the former class.

#5 lore raw

RFC 3/3 fs: remove old MS_* internal flags for the superblock

In reply to Luis Chamberlain

During commit e462ec50cb5 ("VFS: Differentiate mount flags (MS_*) from internal superblock flags") Christoph had suggested we should eventually remove these old flags which were exposed to userspace but could not be used as they were internal-only.

Nuke them.

---
 include/uapi/linux/mount.h       | 8 --------
 tools/include/uapi/linux/mount.h | 8 --------
 2 files changed, 16 deletions(-)
include/uapi/linux/mount.h +0 -8
index 4d93967f8aea..eb6617a5426b 100644
@@ -38,14 +38,6 @@
 #define MS_STRICTATIME	(1<<24) /* Always perform atime updates */
 #define MS_LAZYTIME	(1<<25) /* Update the on-disk [acm]times lazily */
 
-/* These sb flags are internal to the kernel */
-#define MS_SUBMOUNT     (1<<26)
-#define MS_NOREMOTELOCK	(1<<27)
-#define MS_NOSEC	(1<<28)
-#define MS_BORN		(1<<29)
-#define MS_ACTIVE	(1<<30)
-#define MS_NOUSER	(1<<31)
-
 /*
  * Superblock flags that can be altered by MS_REMOUNT
  */
tools/include/uapi/linux/mount.h +0 -8
index 4d93967f8aea..eb6617a5426b 100644
@@ -38,14 +38,6 @@
 #define MS_STRICTATIME	(1<<24) /* Always perform atime updates */
 #define MS_LAZYTIME	(1<<25) /* Update the on-disk [acm]times lazily */
 
-/* These sb flags are internal to the kernel */
-#define MS_SUBMOUNT     (1<<26)
-#define MS_NOREMOTELOCK	(1<<27)
-#define MS_NOSEC	(1<<28)
-#define MS_BORN		(1<<29)
-#define MS_ACTIVE	(1<<30)
-#define MS_NOUSER	(1<<31)
-
 /*
  * Superblock flags that can be altered by MS_REMOUNT
  */
Signature · 1 line
2.35.1
maintainer
#6 lore raw

Re: [RFC 3/3] fs: remove old MS_* internal flags for the superblock

Quoted 6 lines · Luis Chamberlain

During commit e462ec50cb5 ("VFS: Differentiate mount flags (MS_*) from internal superblock flags") Christoph had suggested we should eventually remove these old flags which were exposed to userspace but could not be used as they were internal-only.

Nuke them.

Umm... They are still exposed to userland after your series, though. IOW, we can't change the bit assignment of e.g. SB_ACTIVE - the apparmor part will break, etc.

If anything, it would be more honest to lift the "we ignore those bits" logics to path_mount()...