Skip to contentSkip to conversation

locks: Allow disabling mandatory locking at compile time

RFC PATCH In review Eric W. Biederman 8 participants 15 messages Nov 11, 2015 – Nov 16, 2015 6 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 locks: Allow disabling mandatory locking at compile time

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>
Cc J. Bruce Fields <[email protected]>

---

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue and I am wondering if there are enough people actually using and interested in mandatory locking that it makes sense to push people to support it, or if mandatory locking should be confined to it's own little corner of existence where it can wither and die.

From what little I can glean we want to discourage people from using mandatory locking and to let it wither and die. A Kconfig option that allows mandatory locking to be disabled at compile time seems like the first step in making that happen. Perhaps in a decade or so when all linux distributions are setting the option we can remove the code.

Does anyone know of any real world use cases of mandatory locking?

 fs/Kconfig         | 10 ++++++++
 fs/locks.c         |  2 ++
 fs/namespace.c     | 10 ++++++++
 include/linux/fs.h | 74 +++++++++++++++++++++++++++++-------------------------
 4 files changed, 62 insertions(+), 34 deletions(-)
fs/Kconfig +10 -0
index da3f32f1a4e4..59322e6e76f4 100644
@@ -67,6 +67,16 @@ config FILE_LOCKING
           for filesystems like NFS and for the flock() system
           call. Disabling this option saves about 11k.
 
+config MANDATORY_FILE_LOCKING
+	bool "Enable Mandatory file locking"
+	depends on FILE_LOCKING
+	default y
+	help
+	  This option enables files appropriately marked files on appropriely
+	  mounted filesystems to support mandatory locking.
+
+	  To the best of my knowledge this is dead code that no one cares about.
+
 source "fs/notify/Kconfig"
 
 source "fs/quota/Kconfig"
fs/locks.c +2 -0
index 0d2b3267e2a3..86c94674ab22 100644
@@ -1191,6 +1191,7 @@ static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
 	return error;
 }
 
+#ifdef CONFIG_MANDATORY_FILE_LOCKING
 /**
  * locks_mandatory_locked - Check for an active lock
  * @file: the file to check
@@ -1289,6 +1290,7 @@ int locks_mandatory_area(int read_write, struct inode *inode,
 }
 
 EXPORT_SYMBOL(locks_mandatory_area);
+#endif /* CONFIG_MANDATORY_FILE_LOCKING */
 
 static void lease_clear_pending(struct file_lock *fl, int arg)
 {
fs/namespace.c +10 -0
index da70f7c4ece1..dee44b4791f0 100644
@@ -1584,6 +1584,14 @@ static inline bool may_mount(void)
 	return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
 }
 
+static inline bool may_mandlock(void)
+{
+#ifndef	CONFIG_MANDATORY_FILE_LOCKING
+	return false;
+#endif
+	return true;
+}
+
 /*
  * Now umount can handle mount points as well as block devices.
  * This is important for filesystems which use unnamed block devices.
@@ -2667,6 +2675,8 @@ long do_mount(const char *dev_name, const char __user *dir_name,
 				   type_page, flags, data_page);
 	if (!retval && !may_mount())
 		retval = -EPERM;
+	if (!retval && (flags & MS_MANDLOCK) && !may_mandlock())
+		retval = -EPERM;
 	if (retval)
 		goto dput_out;
 
include/linux/fs.h +40 -34

92 lines not on this page · raw message

Signature · 1 line
2.2.1

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw
Quoted 15 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>
Cc J. Bruce Fields <[email protected]>
Quoted 4 lines

---

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Quoted 4 lines

and I am wondering if there are enough people actually using and interested in mandatory locking that it makes sense to push people to support it, or if mandatory locking should be confined to it's own little corner of existence where it can wither and die.

I hate mandatory locking and would be delighted, but my opinion probably shouldn't get too much weight.

Quoted 7 lines

From what little I can glean we want to discourage people from using mandatory locking and to let it wither and die. A Kconfig option that allows mandatory locking to be disabled at compile time seems like the first step in making that happen. Perhaps in a decade or so when all linux distributions are setting the option we can remove the code.

Does anyone know of any real world use cases of mandatory locking?

Isn't byte-range locking on Windows mandatory? So Samba people might be the ones to talk to. (Or Wine? Or anyone else doing Windows interoperability.)

My suspicion would be that the semantics they need are different enough from what we support that we'd be better off ignoring it and starting over from scratch anyway. But I could be wrong.

--b.

#3 lore raw
In reply to Eric W. Biederman
Quoted 15 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <[email protected]>
Cc J. Bruce Fields <[email protected]>
Quoted 4 lines

---

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Quoted 4 lines

and I am wondering if there are enough people actually using and interested in mandatory locking that it makes sense to push people to support it, or if mandatory locking should be confined to it's own little corner of existence where it can wither and die.

I hate mandatory locking and would be delighted, but my opinion probably shouldn't get too much weight.

Quoted 7 lines

From what little I can glean we want to discourage people from using mandatory locking and to let it wither and die. A Kconfig option that allows mandatory locking to be disabled at compile time seems like the first step in making that happen. Perhaps in a decade or so when all linux distributions are setting the option we can remove the code.

Does anyone know of any real world use cases of mandatory locking?

Isn't byte-range locking on Windows mandatory? So Samba people might be the ones to talk to. (Or Wine? Or anyone else doing Windows interoperability.)

My suspicion would be that the semantics they need are different enough from what we support that we'd be better off ignoring it and starting over from scratch anyway. But I could be wrong.

--b.

A message that is not in the archive (sent off-list or not yet received): [email protected]

#5 lore raw
Quoted 23 lines · On Wed, 11 Nov 2015 15:26:07 -0500 "J. Bruce Fields"…
Quoted 15 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>
Cc J. Bruce Fields <[email protected]>
Quoted 4 lines

---

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Quoted 8 lines
Quoted 4 lines

and I am wondering if there are enough people actually using and interested in mandatory locking that it makes sense to push people to support it, or if mandatory locking should be confined to it's own little corner of existence where it can wither and die.

I hate mandatory locking and would be delighted, but my opinion probably shouldn't get too much weight.

Ditto...It's really hard to believe that anyone uses them, given the well documented races in the Linux implementation.

Quoted 8 lines
Quoted 7 lines

From what little I can glean we want to discourage people from using mandatory locking and to let it wither and die. A Kconfig option that allows mandatory locking to be disabled at compile time seems like the first step in making that happen. Perhaps in a decade or so when all linux distributions are setting the option we can remove the code.

Does anyone know of any real world use cases of mandatory locking?

Quoted 8 lines

Isn't byte-range locking on Windows mandatory? So Samba people might be the ones to talk to. (Or Wine? Or anyone else doing Windows interoperability.)

My suspicion would be that the semantics they need are different enough from what we support that we'd be better off ignoring it and starting over from scratch anyway. But I could be wrong.

Windows BRLs are mandatory but they have totally different semantics.

I think there is little reason to keep POSIX mandatory locks for windows emulation purposes. I'm pretty sure Samba doesn't rely on them, for instance, given that you have to use a funky mode bit combo to enable them.

This patch seems like a logical step to me. I'll review it soon and will plan to queue it up for v4.5 unless there are objections between now and the next merge window.

Thanks!

Signature · 1 line
Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>
#6 lore raw
In reply to J. Bruce Fields
Quoted 23 lines · On Wed, 11 Nov 2015 15:26:07 -0500 "J. Bruce Fields" <[email protected]>…
Quoted 15 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <[email protected]>
Cc J. Bruce Fields <[email protected]>
Quoted 4 lines

---

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Quoted 8 lines
Quoted 4 lines

and I am wondering if there are enough people actually using and interested in mandatory locking that it makes sense to push people to support it, or if mandatory locking should be confined to it's own little corner of existence where it can wither and die.

I hate mandatory locking and would be delighted, but my opinion probably shouldn't get too much weight.

Ditto...It's really hard to believe that anyone uses them, given the well documented races in the Linux implementation.

Quoted 8 lines
Quoted 7 lines

From what little I can glean we want to discourage people from using mandatory locking and to let it wither and die. A Kconfig option that allows mandatory locking to be disabled at compile time seems like the first step in making that happen. Perhaps in a decade or so when all linux distributions are setting the option we can remove the code.

Does anyone know of any real world use cases of mandatory locking?

Quoted 8 lines

Isn't byte-range locking on Windows mandatory? So Samba people might be the ones to talk to. (Or Wine? Or anyone else doing Windows interoperability.)

My suspicion would be that the semantics they need are different enough from what we support that we'd be better off ignoring it and starting over from scratch anyway. But I could be wrong.

Windows BRLs are mandatory but they have totally different semantics.

I think there is little reason to keep POSIX mandatory locks for windows emulation purposes. I'm pretty sure Samba doesn't rely on them, for instance, given that you have to use a funky mode bit combo to enable them.

This patch seems like a logical step to me. I'll review it soon and will plan to queue it up for v4.5 unless there are objections between now and the next merge window.

Thanks!

Signature · 1 line
Jeff Layton <[email protected]>

A message that is not in the archive (sent off-list or not yet received): 20151111174401.5778153c-9yPaYZwiELC+kQycOl6kW4xkIHaj4LzF@public.gmane.org

#8 lore raw
Quoted 6 lines · Jeff Layton

Windows BRLs are mandatory but they have totally different semantics.

I think there is little reason to keep POSIX mandatory locks for windows emulation purposes. I'm pretty sure Samba doesn't rely on them, for instance, given that you have to use a funky mode bit combo to enable them.

Nope. We emulate Windows mandatory locks on top of POSIX advisory locks. We don't use POSIX mandatory locks at all.

#9 lore raw
In reply to a message not in the archive
Quoted 21 lines · Jeff Layton
Quoted 18 lines · On Wed, 11 Nov 2015 15:26:07 -0500 "J. Bruce Fields"…
Quoted 17 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>
Cc J. Bruce Fields <[email protected]>

---

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Only that I overlooked them when implementing user namespace support for mounting filesystems so it is currently possible to without privilege to mount tmpfs with mandatory locking enabled and pass a file descriptor to a daemon that was not expecting them. Causing nice denial of service attacks.

So I need to decide what to do with mandatory locking in user namespaces.

As the consensus of this thread is that users of mandatory locking are as rare as hen's teeth I can just not allow mandatory locking if you something is being mounted just user namespace permissions. If users would be more wide spread I would need to figure out how to avoid breaking users.

Quoted 31 lines
Quoted 8 lines
Quoted 4 lines

and I am wondering if there are enough people actually using and interested in mandatory locking that it makes sense to push people to support it, or if mandatory locking should be confined to it's own little corner of existence where it can wither and die.

I hate mandatory locking and would be delighted, but my opinion probably shouldn't get too much weight.

Ditto...It's really hard to believe that anyone uses them, given the well documented races in the Linux implementation.

Quoted 8 lines
Quoted 7 lines

From what little I can glean we want to discourage people from using mandatory locking and to let it wither and die. A Kconfig option that allows mandatory locking to be disabled at compile time seems like the first step in making that happen. Perhaps in a decade or so when all linux distributions are setting the option we can remove the code.

Does anyone know of any real world use cases of mandatory locking?

Quoted 8 lines

Isn't byte-range locking on Windows mandatory? So Samba people might be the ones to talk to. (Or Wine? Or anyone else doing Windows interoperability.)

My suspicion would be that the semantics they need are different enough from what we support that we'd be better off ignoring it and starting over from scratch anyway. But I could be wrong.

Windows BRLs are mandatory but they have totally different semantics.

Quoted 8 lines

I think there is little reason to keep POSIX mandatory locks for windows emulation purposes. I'm pretty sure Samba doesn't rely on them, for instance, given that you have to use a funky mode bit combo to enable them.

This patch seems like a logical step to me. I'll review it soon and will plan to queue it up for v4.5 unless there are objections between now and the next merge window.

Thanks.

Given the general concensus of this thread it looks like we will also want this incremental patch to deal with the user namespace case.

From: "Eric W. Biederman" <ebiederm-aS9lmoZGLiVWk0Htik3J/[email protected]>
Date: Wed, 11 Nov 2015 17:18:07 -0600
Subject: [PATCH] locks: Don't allow mounts in user namespaces to enable mandatory locking

Since no one uses mandatory locking and files with mandatory locks can cause problems don't allow them in user namespaces.

---
 fs/namespace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/namespace.c +1 -1
index dee44b4791f0..95a349d5003d 100644
@@ -1589,7 +1589,7 @@ static inline bool may_mandlock(void)
 #ifndef	CONFIG_MANDATORY_FILE_LOCKING
 	return false;
 #endif
-	return true;
+	return capable(CAP_SYS_ADMIN);
 }
 
 /*
Signature · 1 line
2.2.1
#10 lore raw
In reply to Jeff Layton
Quoted 6 lines · Jeff Layton

Windows BRLs are mandatory but they have totally different semantics.

I think there is little reason to keep POSIX mandatory locks for windows emulation purposes. I'm pretty sure Samba doesn't rely on them, for instance, given that you have to use a funky mode bit combo to enable them.

Nope. We emulate Windows mandatory locks on top of POSIX advisory locks. We don't use POSIX mandatory locks at all.

#11 lore raw
In reply to Jeff Layton
Quoted 21 lines · Jeff Layton
Quoted 18 lines · On Wed, 11 Nov 2015 15:26:07 -0500 "J. Bruce Fields" <[email protected]>…
Quoted 17 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <[email protected]>
Cc J. Bruce Fields <[email protected]>

---

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Only that I overlooked them when implementing user namespace support for mounting filesystems so it is currently possible to without privilege to mount tmpfs with mandatory locking enabled and pass a file descriptor to a daemon that was not expecting them. Causing nice denial of service attacks.

So I need to decide what to do with mandatory locking in user namespaces.

As the consensus of this thread is that users of mandatory locking are as rare as hen's teeth I can just not allow mandatory locking if you something is being mounted just user namespace permissions. If users would be more wide spread I would need to figure out how to avoid breaking users.

Quoted 31 lines
Quoted 8 lines
Quoted 4 lines

and I am wondering if there are enough people actually using and interested in mandatory locking that it makes sense to push people to support it, or if mandatory locking should be confined to it's own little corner of existence where it can wither and die.

I hate mandatory locking and would be delighted, but my opinion probably shouldn't get too much weight.

Ditto...It's really hard to believe that anyone uses them, given the well documented races in the Linux implementation.

Quoted 8 lines
Quoted 7 lines

From what little I can glean we want to discourage people from using mandatory locking and to let it wither and die. A Kconfig option that allows mandatory locking to be disabled at compile time seems like the first step in making that happen. Perhaps in a decade or so when all linux distributions are setting the option we can remove the code.

Does anyone know of any real world use cases of mandatory locking?

Quoted 8 lines

Isn't byte-range locking on Windows mandatory? So Samba people might be the ones to talk to. (Or Wine? Or anyone else doing Windows interoperability.)

My suspicion would be that the semantics they need are different enough from what we support that we'd be better off ignoring it and starting over from scratch anyway. But I could be wrong.

Windows BRLs are mandatory but they have totally different semantics.

Quoted 8 lines

I think there is little reason to keep POSIX mandatory locks for windows emulation purposes. I'm pretty sure Samba doesn't rely on them, for instance, given that you have to use a funky mode bit combo to enable them.

This patch seems like a logical step to me. I'll review it soon and will plan to queue it up for v4.5 unless there are objections between now and the next merge window.

Thanks.

Given the general concensus of this thread it looks like we will also want this incremental patch to deal with the user namespace case.

From: "Eric W. Biederman" <[email protected]>
Date: Wed, 11 Nov 2015 17:18:07 -0600
Subject: [PATCH] locks: Don't allow mounts in user namespaces to enable mandatory locking

Since no one uses mandatory locking and files with mandatory locks can cause problems don't allow them in user namespaces.

---
 fs/namespace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/namespace.c +1 -1
index dee44b4791f0..95a349d5003d 100644
@@ -1589,7 +1589,7 @@ static inline bool may_mandlock(void)
 #ifndef	CONFIG_MANDATORY_FILE_LOCKING
 	return false;
 #endif
-	return true;
+	return capable(CAP_SYS_ADMIN);
 }
 
 /*
Signature · 1 line
2.2.1

A message that is not in the archive (sent off-list or not yet received): [email protected]

#13 lore raw
Quoted 34 lines · Eric W. Biederman
Quoted 21 lines · Jeff Layton
Quoted 18 lines · On Wed, 11 Nov 2015 15:26:07 -0500 "J. Bruce Fields"…
Quoted 17 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>
Cc J. Bruce Fields <[email protected]>

---

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Only that I overlooked them when implementing user namespace support for mounting filesystems so it is currently possible to without privilege to mount tmpfs with mandatory locking enabled and pass a file descriptor to a daemon that was not expecting them. Causing nice denial of service attacks.

Quoted 6 lines

So I need to decide what to do with mandatory locking in user namespaces.

As the consensus of this thread is that users of mandatory locking are as rare as hen's teeth I can just not allow mandatory locking if you something is being mounted just user namespace permissions.

Sounds like a plan. If nobody notices this limitation then that's further evidence that we might be able to get away with deprecating it eventually.

(Well, I wouldn't be surprised if there's some test suite somewhere that includes a simple test for mandatory lock enforcement. So, any user other than that....)

--b.

#14 lore raw
In reply to Eric W. Biederman
Quoted 34 lines · Eric W. Biederman
Quoted 21 lines · Jeff Layton
Quoted 18 lines · On Wed, 11 Nov 2015 15:26:07 -0500 "J. Bruce Fields" <[email protected]>…
Quoted 17 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <[email protected]>
Cc J. Bruce Fields <[email protected]>

---

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Only that I overlooked them when implementing user namespace support for mounting filesystems so it is currently possible to without privilege to mount tmpfs with mandatory locking enabled and pass a file descriptor to a daemon that was not expecting them. Causing nice denial of service attacks.

Quoted 6 lines

So I need to decide what to do with mandatory locking in user namespaces.

As the consensus of this thread is that users of mandatory locking are as rare as hen's teeth I can just not allow mandatory locking if you something is being mounted just user namespace permissions.

Sounds like a plan. If nobody notices this limitation then that's further evidence that we might be able to get away with deprecating it eventually.

(Well, I wouldn't be surprised if there's some test suite somewhere that includes a simple test for mandatory lock enforcement. So, any user other than that....)

--b.

A message that is not in the archive (sent off-list or not yet received): [email protected]

#16 lore raw
Quoted 42 lines · On Wed, 11 Nov 2015 20:33:11 -0500 "J. Bruce Fields"…
Quoted 34 lines · Eric W. Biederman
Quoted 21 lines · Jeff Layton
Quoted 18 lines · On Wed, 11 Nov 2015 15:26:07 -0500 "J. Bruce Fields"…
Quoted 17 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>
Cc J. Bruce Fields <[email protected]>

---

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Only that I overlooked them when implementing user namespace support for mounting filesystems so it is currently possible to without privilege to mount tmpfs with mandatory locking enabled and pass a file descriptor to a daemon that was not expecting them. Causing nice denial of service attacks.

Quoted 6 lines

So I need to decide what to do with mandatory locking in user namespaces.

As the consensus of this thread is that users of mandatory locking are as rare as hen's teeth I can just not allow mandatory locking if you something is being mounted just user namespace permissions.

Quoted 8 lines

Sounds like a plan. If nobody notices this limitation then that's further evidence that we might be able to get away with deprecating it eventually.

(Well, I wouldn't be surprised if there's some test suite somewhere that includes a simple test for mandatory lock enforcement. So, any user other than that....)

Yeah, ISTR that there are some (maybe in LTP)?

Still, we have all sorts of options to disable pieces of the kernel these days for the tinification effort. This is just another.

Heck, we've already had CONFIG_FILE_LOCKING forever, so allowing people to disable just mandatory locking seems harmless enough.

Signature · 1 line
Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>
#17 lore raw
In reply to Eric W. Biederman
Quoted 105 lines · On Wed, 11 Nov 2015 17:22:33 -0600 ebiederm-aS9lmoZGLiVWk0Htik3J/[email protected]…
Quoted 21 lines · Jeff Layton
Quoted 18 lines · On Wed, 11 Nov 2015 15:26:07 -0500 "J. Bruce Fields"…
Quoted 17 lines · Eric W. Biederman

Mandatory locking appears to be almost unused and buggy and there appears no real interest in doing anything with it. Since effectively no one uses the code and since the code is buggy let's allow it to be disabled at compile time. I would just suggest removing the code but undoubtedly that will break some piece of userspace code somewhere.

For the distributions that don't care about this piece of code this gives a nice starting point to make mandatory locking go away.

Cc Benjamin Coddington <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>
Cc J. Bruce Fields <[email protected]>

---

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

A piece of userspace software having problematic interactions with mandatory locking recently came up as an issue

Is there any more interesting story there?

Only that I overlooked them when implementing user namespace support for mounting filesystems so it is currently possible to without privilege to mount tmpfs with mandatory locking enabled and pass a file descriptor to a daemon that was not expecting them. Causing nice denial of service attacks.

So I need to decide what to do with mandatory locking in user namespaces.

As the consensus of this thread is that users of mandatory locking are as rare as hen's teeth I can just not allow mandatory locking if you something is being mounted just user namespace permissions. If users would be more wide spread I would need to figure out how to avoid breaking users.

Quoted 31 lines
Quoted 8 lines
Quoted 4 lines

and I am wondering if there are enough people actually using and interested in mandatory locking that it makes sense to push people to support it, or if mandatory locking should be confined to it's own little corner of existence where it can wither and die.

I hate mandatory locking and would be delighted, but my opinion probably shouldn't get too much weight.

Ditto...It's really hard to believe that anyone uses them, given the well documented races in the Linux implementation.

Quoted 8 lines
Quoted 7 lines

From what little I can glean we want to discourage people from using mandatory locking and to let it wither and die. A Kconfig option that allows mandatory locking to be disabled at compile time seems like the first step in making that happen. Perhaps in a decade or so when all linux distributions are setting the option we can remove the code.

Does anyone know of any real world use cases of mandatory locking?

Quoted 8 lines

Isn't byte-range locking on Windows mandatory? So Samba people might be the ones to talk to. (Or Wine? Or anyone else doing Windows interoperability.)

My suspicion would be that the semantics they need are different enough from what we support that we'd be better off ignoring it and starting over from scratch anyway. But I could be wrong.

Windows BRLs are mandatory but they have totally different semantics.

Quoted 8 lines

I think there is little reason to keep POSIX mandatory locks for windows emulation purposes. I'm pretty sure Samba doesn't rely on them, for instance, given that you have to use a funky mode bit combo to enable them.

This patch seems like a logical step to me. I'll review it soon and will plan to queue it up for v4.5 unless there are objections between now and the next merge window.

Thanks.

Given the general concensus of this thread it looks like we will also want this incremental patch to deal with the user namespace case.

From: "Eric W. Biederman" <ebiederm-aS9lmoZGLiVWk0Htik3J/[email protected]>
Date: Wed, 11 Nov 2015 17:18:07 -0600
Subject: [PATCH] locks: Don't allow mounts in user namespaces to enable mandatory locking

Since no one uses mandatory locking and files with mandatory locks can cause problems don't allow them in user namespaces.

---
 fs/namespace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/namespace.c +0 -0
index dee44b4791f0..95a349d5003d 100644

fs/namespace.c static inline bool may_mandlock(void) · line 1595

Quoted 8 lines
fs/namespace.c +1 -1
@@ -1589,7 +1589,7 @@ static inline bool may_mandlock(void)
 #ifndef	CONFIG_MANDATORY_FILE_LOCKING
 	return false;
 #endif
-	return true;
+	return capable(CAP_SYS_ADMIN);
 }
 
 /*

Fair enough. I'll merge that as well and we'll see what breaks.

Signature · 1 line
Jeff Layton <jeff.layton-7I+n7zu2hftEKMMhf/[email protected]>

A message that is not in the archive (sent off-list or not yet received): 20151116095841.0f620a5c-9yPaYZwiELC+kQycOl6kW4xkIHaj4LzF@public.gmane.org

#19 lore raw
Quoted 27 lines · Jeff Layton
Quoted 27 lines

Given the general concensus of this thread it looks like we will also want this incremental patch to deal with the user namespace case.

From: "Eric W. Biederman" <ebiederm-aS9lmoZGLiVWk0Htik3J/[email protected]>
Date: Wed, 11 Nov 2015 17:18:07 -0600
Subject: [PATCH] locks: Don't allow mounts in user namespaces to enable mandatory locking

Since no one uses mandatory locking and files with mandatory locks can cause problems don't allow them in user namespaces.

---
 fs/namespace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/namespace.c +1 -1
index dee44b4791f0..95a349d5003d 100644
@@ -1589,7 +1589,7 @@ static inline bool may_mandlock(void)
 #ifndef	CONFIG_MANDATORY_FILE_LOCKING
 	return false;
 #endif
-	return true;
+	return capable(CAP_SYS_ADMIN);
 }
 
Quoted 3 lines

fs/namespace.c static inline bool may_mandlock(void) · line 1595

Quoted 1 line
fs/namespace.c +0 -0
 /*

Fair enough. I'll merge that as well and we'll see what breaks.

Sounds good. If something breaks just revert this bit.

The ways things work today is not an issue if people are actually using mandatory file locking.

The fundamental issues that triggered my looking into this was a single threaded server that accepts input from unprivileged users and hangs if reading that input takes too long. Mandatory locking just made it easier to supply an input that hangs.

There are a countless other ways to make reading lots of data from a file descriptor slow, the obvious ones being cpu load, disk load (for local filesystems), and network load (for network filesystems).

Passing file descriptors appears to be the only path where mandatory locking from files in a user namespace can affect processes outside of the user namespace. Given that there are many other ways to trigger a slow file descriptor, that mandatory locking problems are easy to avoid, I just can not see the existing kernel behavior as being wrong.

The current state of affairs is a problem in that we have a feature in
the kernel that apparently no one uses and everyone ignores. Turning
off the feature is the right solution for that.

There is a privileged operation involved with enabling "-o mand" on a filesystem. The intent and the only value I can see in the "-o mand" flag is that it determines if a filesystem will interpret the a specific combinationation of file mode bits as specifying mandatory locking on a file. Allowing that on files accessible only to users inside of the user namespace will not cause problems for the rest of the system except potentially with passed file descriptors.

Eric

#20 lore raw
In reply to Jeff Layton
Quoted 27 lines · Jeff Layton
Quoted 27 lines

Given the general concensus of this thread it looks like we will also want this incremental patch to deal with the user namespace case.

From: "Eric W. Biederman" <[email protected]>
Date: Wed, 11 Nov 2015 17:18:07 -0600
Subject: [PATCH] locks: Don't allow mounts in user namespaces to enable mandatory locking

Since no one uses mandatory locking and files with mandatory locks can cause problems don't allow them in user namespaces.

---
 fs/namespace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/namespace.c +1 -1
index dee44b4791f0..95a349d5003d 100644
@@ -1589,7 +1589,7 @@ static inline bool may_mandlock(void)
 #ifndef	CONFIG_MANDATORY_FILE_LOCKING
 	return false;
 #endif
-	return true;
+	return capable(CAP_SYS_ADMIN);
 }
 
Quoted 3 lines

fs/namespace.c static inline bool may_mandlock(void) · line 1595

Quoted 1 line
fs/namespace.c +0 -0
 /*

Fair enough. I'll merge that as well and we'll see what breaks.

Sounds good. If something breaks just revert this bit.

The ways things work today is not an issue if people are actually using mandatory file locking.

The fundamental issues that triggered my looking into this was a single threaded server that accepts input from unprivileged users and hangs if reading that input takes too long. Mandatory locking just made it easier to supply an input that hangs.

There are a countless other ways to make reading lots of data from a file descriptor slow, the obvious ones being cpu load, disk load (for local filesystems), and network load (for network filesystems).

Passing file descriptors appears to be the only path where mandatory locking from files in a user namespace can affect processes outside of the user namespace. Given that there are many other ways to trigger a slow file descriptor, that mandatory locking problems are easy to avoid, I just can not see the existing kernel behavior as being wrong.

The current state of affairs is a problem in that we have a feature in
the kernel that apparently no one uses and everyone ignores. Turning
off the feature is the right solution for that.

There is a privileged operation involved with enabling "-o mand" on a filesystem. The intent and the only value I can see in the "-o mand" flag is that it determines if a filesystem will interpret the a specific combinationation of file mode bits as specifying mandatory locking on a file. Allowing that on files accessible only to users inside of the user namespace will not cause problems for the rest of the system except potentially with passed file descriptors.

Eric