Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1593798 > unrolled thread

[RFC] Add option to mount only a pids subset

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [RFC] Add option to mount only a pids subset Alexey Gladkov <gladkov.alexey@gmail.com> - 2017-03-07 00:40 +0100
    Re: [RFC] Add option to mount only a pids subset Andy Lutomirski <luto@amacapital.net> - 2017-03-07 17:30 +0100
      Re: [RFC] Add option to mount only a pids subset Djalal Harouni <tixxdz@gmail.com> - 2017-03-09 12:30 +0100
        Re: [RFC] Add option to mount only a pids subset ebiederm@xmission.com (Eric W. Biederman) - 2017-03-09 22:00 +0100
        Re: [RFC] Add option to mount only a pids subset Alexey Gladkov <gladkov.alexey@gmail.com> - 2017-03-11 22:50 +0100
      Re: [RFC] Add option to mount only a pids subset Alexey Gladkov <gladkov.alexey@gmail.com> - 2017-03-11 01:00 +0100
    Re: [RFC] Add option to mount only a pids subset Oleg Nesterov <oleg@redhat.com> - 2017-03-07 19:00 +0100
      Re: [RFC] Add option to mount only a pids subset Alexey Gladkov <gladkov.alexey@gmail.com> - 2017-03-11 00:40 +0100
    Re: [RFC] Add option to mount only a pids subset Al Viro <viro@ZenIV.linux.org.uk> - 2017-03-12 03:00 +0100
      Re: [RFC] Add option to mount only a pids subset Al Viro <viro@ZenIV.linux.org.uk> - 2017-03-12 03:20 +0100
        Re: [RFC] Add option to mount only a pids subset Andy Lutomirski <luto@amacapital.net> - 2017-03-13 04:30 +0100
          Re: [RFC] Add option to mount only a pids subset Al Viro <viro@ZenIV.linux.org.uk> - 2017-03-13 14:30 +0100
            Re: [RFC] Add option to mount only a pids subset Andy Lutomirski <luto@kernel.org> - 2017-03-13 16:30 +0100

#1593798 — [RFC] Add option to mount only a pids subset

After discussion with Oleg Nesterov I reimplement my patch as an additional
option for /proc. This option affects the mountpoint. It means that in one
pid namespace it possible to have both the whole traditional /proc and
/proc with only pids subset.

However, it remains an open question about overlayfs support in the /proc.

== Overview ==

Some of the container virtualization systems are mounted /proc inside
the container. This is done in most cases to operate with information
about the processes. Knowing that /proc filesystem is not fully
virtualized they are mounted on top of dangerous places empty files or
directories (for exmaple /proc/sys, /proc/kcore, /sys/firmware, etc.).

The structure of this filesystem is dynamic and any module can create a
new object which will not necessarily be virtualized. There are
proprietary modules that aren't in the mainline whose work we can not
verify.

This opens up a potential threat to the system. The developers of the
virtualization system can't predict all dangerous places in /proc by
definition.

A more effective solution would be to mount into the container only what
is necessary and ignore the rest.

Right now there is the opportunity to pass in the container any port of
the /proc filesystem using mount --bind expect the pids.

This patch allows to mount only the part of /proc related to pids
without rest objects. Since this is an option for /proc, flags applied to
/proc have an effect on this subset of filesystem.

Originally the idea was that the container will be mounted only pid sunset
and additional required files will be mounted on top using the overlayfs.
But I found out that /proc does not support overlayfs and does not allow
to mount anything on top or under it.

== TODO ==

There is still work to do:

 * Show pidonly via proc_show_options.
 * Add overlayfs support.

---
 fs/internal.h         |  1 +
 fs/namespace.c        |  9 ++++++++
 fs/proc/generic.c     |  4 ++++
 fs/proc/inode.c       |  7 +++++-
 fs/proc/internal.h    |  8 +++++++
 fs/proc/root.c        | 62 +++++++++++++++++++++++++++++++++++++++++++++++----
 fs/stat.c             |  4 ++++
 fs/super.c            | 20 +++++++++++++++++
 include/linux/fs.h    |  2 ++
 include/linux/mount.h |  1 +
 10 files changed, 113 insertions(+), 5 deletions(-)

diff --git a/fs/internal.h b/fs/internal.h
index 4fcf517..cb44ca7 100644
--- a/fs/internal.h
+++ b/fs/internal.h
@@ -88,6 +88,7 @@ extern struct file *get_empty_filp(void);
  * super.c
  */
 extern int do_remount_sb(struct super_block *, int, void *, int);
+extern int do_mount_sb(struct vfsmount *mnt, int flags, void *data);
 extern bool trylock_super(struct super_block *sb);
 extern struct dentry *mount_fs(struct file_system_type *,
 			       int, const char *, void *);
diff --git a/fs/namespace.c b/fs/namespace.c
index e6c234b..6cb2bcb 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -938,6 +938,7 @@ static struct mount *skip_mnt_tree(struct mount *p)
 struct vfsmount *
 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
 {
+	int err;
 	struct mount *mnt;
 	struct dentry *root;
 
@@ -962,6 +963,14 @@ vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void
 	mnt->mnt.mnt_sb = root->d_sb;
 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 	mnt->mnt_parent = mnt;
+
+	err = do_mount_sb(&mnt->mnt, flags, data);
+	if(err) {
+		mnt_free_id(mnt);
+		free_vfsmnt(mnt);
+		return ERR_PTR(err);
+	}
+
 	lock_mount_hash();
 	list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
 	unlock_mount_hash();
diff --git a/fs/proc/generic.c b/fs/proc/generic.c
index 7eb3cef..dd3da60 100644
--- a/fs/proc/generic.c
+++ b/fs/proc/generic.c
@@ -306,6 +306,10 @@ int proc_readdir_de(struct proc_dir_entry *de, struct file *file,
 int proc_readdir(struct file *file, struct dir_context *ctx)
 {
 	struct inode *inode = file_inode(file);
+	struct proc_options *opts = file->f_path.mnt->fs_data;
+
+	if (opts && opts->pid_only)
+		return 1;
 
 	return proc_readdir_de(PDE(inode), file, ctx);
 }
diff --git a/fs/proc/inode.c b/fs/proc/inode.c
index 783bc19..51b1712 100644
--- a/fs/proc/inode.c
+++ b/fs/proc/inode.c
@@ -108,7 +108,6 @@ static int proc_show_options(struct seq_file *seq, struct dentry *root)
 		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
 	if (pid->hide_pid != 0)
 		seq_printf(seq, ",hidepid=%u", pid->hide_pid);
-
 	return 0;
 }
 
@@ -118,6 +117,8 @@ static const struct super_operations proc_sops = {
 	.drop_inode	= generic_delete_inode,
 	.evict_inode	= proc_evict_inode,
 	.statfs		= simple_statfs,
+	.getattr_fs	= proc_getattrfs,
+	.mount_fs	= proc_mount_cb,
 	.remount_fs	= proc_remount,
 	.show_options	= proc_show_options,
 };
@@ -323,6 +324,10 @@ static int proc_reg_open(struct inode *inode, struct file *file)
 	int (*open)(struct inode *, struct file *);
 	int (*release)(struct inode *, struct file *);
 	struct pde_opener *pdeo;
+	struct proc_options *opts = file->f_path.mnt->fs_data;
+
+	if (opts && opts->pid_only)
+		return -ENOENT;
 
 	/*
 	 * Ensure that
diff --git a/fs/proc/internal.h b/fs/proc/internal.h
index 2de5194..7fdbfee 100644
--- a/fs/proc/internal.h
+++ b/fs/proc/internal.h
@@ -267,11 +267,19 @@ static inline void proc_tty_init(void) {}
 /*
  * root.c
  */
+struct proc_options {
+	kgid_t pid_gid;
+	int hide_pid;
+	int pid_only;
+};
+
 extern struct proc_dir_entry proc_root;
 extern int proc_parse_options(char *options, struct pid_namespace *pid);
 
 extern void proc_self_init(void);
 extern int proc_remount(struct super_block *, int *, char *);
+extern int proc_mount_cb(struct vfsmount *mnt, int *flags, char *data);
+extern int proc_getattrfs(struct vfsmount *, struct dentry *, struct kstat *);
 
 /*
  * task_[no]mmu.c
diff --git a/fs/proc/root.c b/fs/proc/root.c
index 4bd0373..e07f37a 100644
--- a/fs/proc/root.c
+++ b/fs/proc/root.c
@@ -14,6 +14,7 @@
 #include <linux/stat.h>
 #include <linux/init.h>
 #include <linux/sched.h>
+#include <linux/slab.h>
 #include <linux/module.h>
 #include <linux/bitops.h>
 #include <linux/user_namespace.h>
@@ -24,16 +25,17 @@
 #include "internal.h"
 
 enum {
-	Opt_gid, Opt_hidepid, Opt_err,
+	Opt_gid, Opt_hidepid, Opt_pidonly, Opt_err,
 };
 
 static const match_table_t tokens = {
 	{Opt_hidepid, "hidepid=%u"},
 	{Opt_gid, "gid=%u"},
+	{Opt_pidonly, "pidonly"},
 	{Opt_err, NULL},
 };
 
-int proc_parse_options(char *options, struct pid_namespace *pid)
+static int proc_fill_options(char *options, struct proc_options *fs_opts)
 {
 	char *p;
 	substring_t args[MAX_OPT_ARGS];
@@ -53,7 +55,7 @@ int proc_parse_options(char *options, struct pid_namespace *pid)
 		case Opt_gid:
 			if (match_int(&args[0], &option))
 				return 0;
-			pid->pid_gid = make_kgid(current_user_ns(), option);
+			fs_opts->pid_gid = make_kgid(current_user_ns(), option);
 			break;
 		case Opt_hidepid:
 			if (match_int(&args[0], &option))
@@ -62,7 +64,10 @@ int proc_parse_options(char *options, struct pid_namespace *pid)
 				pr_err("proc: hidepid value must be between 0 and 2.\n");
 				return 0;
 			}
-			pid->hide_pid = option;
+			fs_opts->hide_pid = option;
+			break;
+		case Opt_pidonly:
+			fs_opts->pid_only = 1;
 			break;
 		default:
 			pr_err("proc: unrecognized mount option \"%s\" "
@@ -74,6 +79,19 @@ int proc_parse_options(char *options, struct pid_namespace *pid)
 	return 1;
 }
 
+int proc_parse_options(char *options, struct pid_namespace *pid)
+{
+	struct proc_options opts;
+
+	if (!proc_fill_options(options, &opts))
+		return 0;
+
+	pid->pid_gid = opts.pid_gid;
+	pid->hide_pid = opts.hide_pid;
+
+	return 1;
+}
+
 int proc_remount(struct super_block *sb, int *flags, char *data)
 {
 	struct pid_namespace *pid = sb->s_fs_info;
@@ -82,6 +100,42 @@ int proc_remount(struct super_block *sb, int *flags, char *data)
 	return !proc_parse_options(data, pid);
 }
 
+int proc_getattrfs(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
+{
+	struct inode *inode = d_inode(dentry);
+	struct pid *pid = proc_pid(dentry->d_inode);
+	struct proc_options *opts = mnt->fs_data;
+
+	if (opts && opts->pid_only && mnt->mnt_root != dentry && !pid)
+		return -ENOENT;
+
+	if (!inode->i_op->getattr) {
+		generic_fillattr(inode, stat);
+		return 0;
+	}
+
+	return inode->i_op->getattr(mnt, dentry, stat);
+}
+
+int proc_mount_cb(struct vfsmount *mnt, int *flags, char *data)
+{
+	struct proc_options *opts;
+
+	if (!data || *flags & MS_KERNMOUNT)
+		return 0;
+
+	opts = kzalloc(sizeof(struct proc_options), GFP_KERNEL);
+	if (!opts)
+		return -ENOMEM;
+
+	if (!proc_fill_options(data, opts))
+		return -EINVAL;
+
+	mnt->fs_data = opts;
+
+	return 0;
+}
+
 static struct dentry *proc_mount(struct file_system_type *fs_type,
 	int flags, const char *dev_name, void *data)
 {
diff --git a/fs/stat.c b/fs/stat.c
index bc045c7..1e26308 100644
--- a/fs/stat.c
+++ b/fs/stat.c
@@ -10,6 +10,7 @@
 #include <linux/file.h>
 #include <linux/highuid.h>
 #include <linux/fs.h>
+#include <linux/mount.h>
 #include <linux/namei.h>
 #include <linux/security.h>
 #include <linux/syscalls.h>
@@ -53,6 +54,9 @@ int vfs_getattr_nosec(struct path *path, struct kstat *stat)
 {
 	struct inode *inode = d_backing_inode(path->dentry);
 
+	if (path->mnt->mnt_sb->s_op->getattr_fs)
+		return path->mnt->mnt_sb->s_op->getattr_fs(path->mnt, path->dentry, stat);
+
 	if (inode->i_op->getattr)
 		return inode->i_op->getattr(path->mnt, path->dentry, stat);
 
diff --git a/fs/super.c b/fs/super.c
index c183835..478dd5b 100644
--- a/fs/super.c
+++ b/fs/super.c
@@ -832,6 +832,26 @@ int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
 	return retval;
 }
 
+/**
+ *	do_mount_sb - asks filesystem to finish mount with options.
+ *	@mnt:	vfsmount in question
+ *	@flags:	numeric part of options
+ *	@data:	the rest of options
+ *
+ *	Alters the mount options of a mounted filesystem.
+ */
+int do_mount_sb(struct vfsmount *mnt, int flags, void *data)
+{
+	if (mnt->mnt_sb->s_writers.frozen != SB_UNFROZEN)
+		return -EBUSY;
+
+	if (mnt->mnt_sb->s_op->mount_fs) {
+		return mnt->mnt_sb->s_op->mount_fs(mnt, &flags, data);
+	}
+
+	return 0;
+}
+
 static void do_emergency_remount(struct work_struct *work)
 {
 	struct super_block *sb, *p = NULL;
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 83de8b6..5bd1b84 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -1759,6 +1759,8 @@ struct super_operations {
 	int (*thaw_super) (struct super_block *);
 	int (*unfreeze_fs) (struct super_block *);
 	int (*statfs) (struct dentry *, struct kstatfs *);
+	int (*getattr_fs) (struct vfsmount *, struct dentry *, struct kstat *);
+	int (*mount_fs) (struct vfsmount *, int *, char *);
 	int (*remount_fs) (struct super_block *, int *, char *);
 	void (*umount_begin) (struct super_block *);
 
diff --git a/include/linux/mount.h b/include/linux/mount.h
index 1172cce..4bd364e 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -67,6 +67,7 @@ struct vfsmount {
 	struct dentry *mnt_root;	/* root of the mounted tree */
 	struct super_block *mnt_sb;	/* pointer to superblock */
 	int mnt_flags;
+	void *fs_data;			/* fs-specific data */
 };
 
 struct file; /* forward dec */
-- 
2.10.2

[toc] | [next] | [standalone]

#1594403

On Mon, Mar 6, 2017 at 3:05 PM, Alexey Gladkov <gladkov.alexey@gmail.com> wrote:
>
> After discussion with Oleg Nesterov I reimplement my patch as an additional
> option for /proc. This option affects the mountpoint. It means that in one
> pid namespace it possible to have both the whole traditional /proc and
> /proc with only pids subset.
>

I like this.  I think you should split it into two patches, though:
one that reworks how procfs gets mounted and one that makes adds the
new functionality.

Djajal had some concerns about the first part breaking applications
that use stat and expect certain behavior.  This should be manageable,
though, but making stat work appropriately.

[toc] | [prev] | [next] | [standalone]

#1595948

On Tue, Mar 7, 2017 at 5:24 PM, Andy Lutomirski <luto@amacapital.net> wrote:
>
> On Mon, Mar 6, 2017 at 3:05 PM, Alexey Gladkov <gladkov.alexey@gmail.com> wrote:
> >
> > After discussion with Oleg Nesterov I reimplement my patch as an additional
> > option for /proc. This option affects the mountpoint. It means that in one
> > pid namespace it possible to have both the whole traditional /proc and
> > /proc with only pids subset.
> >
>
> I like this.  I think you should split it into two patches, though:
> one that reworks how procfs gets mounted and one that makes adds the
> new functionality.
>
> Djajal had some concerns about the first part breaking applications
> that use stat and expect certain behavior.  This should be manageable,
> though, but making stat work appropriately.

I'm bit lost in the two discussion, however the main concern I was
discussing with Andy was if you have per superblock proc mounts then
each mount will end up with its own device ID st_dev, right now they
share the same ID if they are in the same pid namespace, but if we
change that then we may break the following:
http://man7.org/linux/man-pages/man7/namespaces.7.html

Both new NS_GET_PARENT and NS_GET_USERNS ioctl() that return an fd,
suggests to follow up with fstat() to identify the namespaces..
"By applying fstat(2) to the returned file descriptor, one obtains a
stat structure whose st_dev (resident device) and st_ino (inode
number) fields together identify the owning/parent namespace."

Other /proc/self/ns/* comparison and stat() logic...

Andy suggested that we may have the same st_dev for mounts in the same
pid namespace... I'm not sure which side effect this may bring!

Thanks!

-- 
tixxdz

[toc] | [prev] | [next] | [standalone]

#1596386

Djalal Harouni <tixxdz@gmail.com> writes:

> On Tue, Mar 7, 2017 at 5:24 PM, Andy Lutomirski <luto@amacapital.net> wrote:
>>
>> On Mon, Mar 6, 2017 at 3:05 PM, Alexey Gladkov <gladkov.alexey@gmail.com> wrote:
>> >
>> > After discussion with Oleg Nesterov I reimplement my patch as an additional
>> > option for /proc. This option affects the mountpoint. It means that in one
>> > pid namespace it possible to have both the whole traditional /proc and
>> > /proc with only pids subset.
>> >
>>
>> I like this.  I think you should split it into two patches, though:
>> one that reworks how procfs gets mounted and one that makes adds the
>> new functionality.
>>
>> Djajal had some concerns about the first part breaking applications
>> that use stat and expect certain behavior.  This should be manageable,
>> though, but making stat work appropriately.
>
> I'm bit lost in the two discussion, however the main concern I was
> discussing with Andy was if you have per superblock proc mounts then
> each mount will end up with its own device ID st_dev, right now they
> share the same ID if they are in the same pid namespace, but if we
> change that then we may break the following:
> http://man7.org/linux/man-pages/man7/namespaces.7.html
>
> Both new NS_GET_PARENT and NS_GET_USERNS ioctl() that return an fd,
> suggests to follow up with fstat() to identify the namespaces..
> "By applying fstat(2) to the returned file descriptor, one obtains a
> stat structure whose st_dev (resident device) and st_ino (inode
> number) fields together identify the owning/parent namespace."
>
> Other /proc/self/ns/* comparison and stat() logic...
>
> Andy suggested that we may have the same st_dev for mounts in the same
> pid namespace... I'm not sure which side effect this may bring!

Well right now it is less of an issue than you image.  I suggest you
stat /proc and /proc/self/ns/* and look at st_dev.

That said anything that changes today's proc needs to be tested with a
range of distro's especially those using selinux and apparmor as they
tend to have policies that are very sensitive to the implementation
details.  Such that seemingly innocent changes can cause systems to stop
booting.

Eric

[toc] | [prev] | [next] | [standalone]

#1598496

On Thu, Mar 09, 2017 at 12:26:49PM +0100, Djalal Harouni wrote:
> I'm bit lost in the two discussion, however the main concern I was
> discussing with Andy was if you have per superblock proc mounts then
> each mount will end up with its own device ID st_dev, right now they
> share the same ID if they are in the same pid namespace, but if we
> change that then we may break the following:
> http://man7.org/linux/man-pages/man7/namespaces.7.html

In fact, nothing has changed. I added a parameter that affects
the mountpoint, not the entire pid namespace.

The procfs will still be global. The device ID will be the same as before.

> Both new NS_GET_PARENT and NS_GET_USERNS ioctl() that return an fd,
> suggests to follow up with fstat() to identify the namespaces..
> "By applying fstat(2) to the returned file descriptor, one obtains a
> stat structure whose st_dev (resident device) and st_ino (inode
> number) fields together identify the owning/parent namespace."
> 
> Other /proc/self/ns/* comparison and stat() logic...
> 
> Andy suggested that we may have the same st_dev for mounts in the same
> pid namespace... I'm not sure which side effect this may bring!

Basically we have here a issue because other proc options (hidepid for
example) affect the entire pid namespace, but, I guess, have to affect
the mountpoint.

# grep ^proc /proc/mounts 
proc /proc proc rw,relatime 0 0

# mount -t proc proc /tmp/proc
# mount -o remount,hidepid=2 -t proc proc /tmp/proc

# grep ^proc /proc/mounts 
proc /proc proc rw,relatime,hidepid=2 0 0
proc /tmp/proc proc rw,relatime,hidepid=2 0 0

-- 
Rgrds, legion

[toc] | [prev] | [next] | [standalone]

#1598186

On Tue, Mar 07, 2017 at 08:24:12AM -0800, Andy Lutomirski wrote:
> On Mon, Mar 6, 2017 at 3:05 PM, Alexey Gladkov <gladkov.alexey@gmail.com> wrote:
> >
> > After discussion with Oleg Nesterov I reimplement my patch as an additional
> > option for /proc. This option affects the mountpoint. It means that in one
> > pid namespace it possible to have both the whole traditional /proc and
> > /proc with only pids subset.
> >
> 
> I like this.  I think you should split it into two patches, though:
> one that reworks how procfs gets mounted and one that makes adds the
> new functionality.

Sure, but first I wanted to discuss the idea. My patch isn't very useful
without the ability to add additional files.

I will split it into parts in the new version.

-- 
Rgrds, legion

[toc] | [prev] | [next] | [standalone]

#1594479

I can't really review this... but in any case I think you should split
this patch to separate the vfs and proc changes.

On 03/07, Alexey Gladkov wrote:
>
> @@ -962,6 +963,14 @@ vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void
>  	mnt->mnt.mnt_sb = root->d_sb;
>  	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
>  	mnt->mnt_parent = mnt;
> +
> +	err = do_mount_sb(&mnt->mnt, flags, data);
> +	if(err) {
> +		mnt_free_id(mnt);
> +		free_vfsmnt(mnt);
> +		return ERR_PTR(err);
> +	}

This duplicates the error handling, we do the same if mount_fs() fails.
Perhaps you should move these 2 lines into cleanup block and add goto's.

> +int proc_getattrfs(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
> +{
> +	struct inode *inode = d_inode(dentry);
> +	struct pid *pid = proc_pid(dentry->d_inode);
> +	struct proc_options *opts = mnt->fs_data;
> +
> +	if (opts && opts->pid_only && mnt->mnt_root != dentry && !pid)
> +		return -ENOENT;

Hmm. I don't quite understand why do we need this, and how this should work.

Yes, "/bin/ls /pidonly-proc/sys" or opendir(/pidonly-proc/sys) should fail,
but only because they both do stat() ?

Afaics you still can do open("/pidonly-proc/sys") + getdents() and this should
work ?

I still think proc_dir_operations.open() makes more sense. Yes, as you pointed
out we also need to update proc_sys_dir_file_operations too and may be something
else...

> +
> +	if (!inode->i_op->getattr) {
> +		generic_fillattr(inode, stat);
> +		return 0;
> +	}
> +
> +	return inode->i_op->getattr(mnt, dentry, stat);
> +}

Oh, it would be nice to not duplicate the code from the caller, imo.

Oleg.

[toc] | [prev] | [next] | [standalone]

#1598181

On Tue, Mar 07, 2017 at 06:49:09PM +0100, Oleg Nesterov wrote:
> I can't really review this... but in any case I think you should split
> this patch to separate the vfs and proc changes.
> 
> On 03/07, Alexey Gladkov wrote:
> >
> > @@ -962,6 +963,14 @@ vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void
> >  	mnt->mnt.mnt_sb = root->d_sb;
> >  	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
> >  	mnt->mnt_parent = mnt;
> > +
> > +	err = do_mount_sb(&mnt->mnt, flags, data);
> > +	if(err) {
> > +		mnt_free_id(mnt);
> > +		free_vfsmnt(mnt);
> > +		return ERR_PTR(err);
> > +	}
> 
> This duplicates the error handling, we do the same if mount_fs() fails.
> Perhaps you should move these 2 lines into cleanup block and add goto's.
> 
> > +int proc_getattrfs(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
> > +{
> > +	struct inode *inode = d_inode(dentry);
> > +	struct pid *pid = proc_pid(dentry->d_inode);
> > +	struct proc_options *opts = mnt->fs_data;
> > +
> > +	if (opts && opts->pid_only && mnt->mnt_root != dentry && !pid)
> > +		return -ENOENT;
> 
> Hmm. I don't quite understand why do we need this, and how this should work.
> 
> Yes, "/bin/ls /pidonly-proc/sys" or opendir(/pidonly-proc/sys) should fail,
> but only because they both do stat() ?
> 
> Afaics you still can do open("/pidonly-proc/sys") + getdents() and this should
> work ?

Yes, you're right! I thought that getattr is called always together with
open(). I wanted to prevent all attempts open() for not-pid directories.

> I still think proc_dir_operations.open() makes more sense. Yes, as you pointed
> out we also need to update proc_sys_dir_file_operations too and may be something
> else...

My main task was to hide all possible direcitrices from the /proc
(in pidonly mode)... even those which we do not know. In this case we
can't rely on the fact that everyone will follow the rules and to
properly handle open().

My current attempt was to force filesystem level check of mountpoint flag.
This is necessary to avoid even the theoretical possibility of ignoring
"pidonly" parameter.

I guess I need to add callback to vfs_open or something to can be sure
that we will not open the wrong file or directory in pidonly mode.

-- 
Rgrds, legion

[toc] | [prev] | [next] | [standalone]

#1598526

On Tue, Mar 07, 2017 at 12:05:16AM +0100, Alexey Gladkov wrote:

> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 83de8b6..5bd1b84 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -1759,6 +1759,8 @@ struct super_operations {
>  	int (*thaw_super) (struct super_block *);
>  	int (*unfreeze_fs) (struct super_block *);
>  	int (*statfs) (struct dentry *, struct kstatfs *);
> +	int (*getattr_fs) (struct vfsmount *, struct dentry *, struct kstat *);
> +	int (*mount_fs) (struct vfsmount *, int *, char *);
>  	int (*remount_fs) (struct super_block *, int *, char *);
>  	void (*umount_begin) (struct super_block *);
>  
> diff --git a/include/linux/mount.h b/include/linux/mount.h
> index 1172cce..4bd364e 100644
> --- a/include/linux/mount.h
> +++ b/include/linux/mount.h
> @@ -67,6 +67,7 @@ struct vfsmount {
>  	struct dentry *mnt_root;	/* root of the mounted tree */
>  	struct super_block *mnt_sb;	/* pointer to superblock */
>  	int mnt_flags;
> +	void *fs_data;			/* fs-specific data */

	No.  This is really asking for trouble - you *can't* hang
fs-specific data structures off vfsmount.  Lifetimes make no
sense whatsoever.

	BTW, your patch leaks supreblock reference on failure, and
is kludgy as hell wrt to what it's doing in procfs itself, but
that's secondary - the quoted part is enough for a hard NAK on
architectural grounds.  Don't go there.

[toc] | [prev] | [next] | [standalone]

#1598531

On Sun, Mar 12, 2017 at 01:54:38AM +0000, Al Viro wrote:
> On Tue, Mar 07, 2017 at 12:05:16AM +0100, Alexey Gladkov wrote:
> 
> > diff --git a/include/linux/fs.h b/include/linux/fs.h
> > index 83de8b6..5bd1b84 100644
> > --- a/include/linux/fs.h
> > +++ b/include/linux/fs.h
> > @@ -1759,6 +1759,8 @@ struct super_operations {
> >  	int (*thaw_super) (struct super_block *);
> >  	int (*unfreeze_fs) (struct super_block *);
> >  	int (*statfs) (struct dentry *, struct kstatfs *);
> > +	int (*getattr_fs) (struct vfsmount *, struct dentry *, struct kstat *);
> > +	int (*mount_fs) (struct vfsmount *, int *, char *);
> >  	int (*remount_fs) (struct super_block *, int *, char *);
> >  	void (*umount_begin) (struct super_block *);
> >  
> > diff --git a/include/linux/mount.h b/include/linux/mount.h
> > index 1172cce..4bd364e 100644
> > --- a/include/linux/mount.h
> > +++ b/include/linux/mount.h
> > @@ -67,6 +67,7 @@ struct vfsmount {
> >  	struct dentry *mnt_root;	/* root of the mounted tree */
> >  	struct super_block *mnt_sb;	/* pointer to superblock */
> >  	int mnt_flags;
> > +	void *fs_data;			/* fs-specific data */
> 
> 	No.  This is really asking for trouble - you *can't* hang
> fs-specific data structures off vfsmount.  Lifetimes make no
> sense whatsoever.
>
> 	BTW, your patch leaks supreblock reference on failure, and
> is kludgy as hell wrt to what it's doing in procfs itself, but
> that's secondary - the quoted part is enough for a hard NAK on
> architectural grounds.  Don't go there.

PS: AFAICS, simple mount --bind of your pid-only mount will suddenly
expose the full thing.  And as for the lifetimes making no sense...
note that you are simply not freeing these structures of yours.
Try to handle that and you'll get a serious PITA all over the
place.

What are you trying to achieve, anyway?  Why not add a second vfsmount
pointer per pid_namespace and make it initialized on demand, at the
first attempt of no-pid mount?  Just have a separate no-pid instance
created for those namespaces where it had been asked for, with
separate superblock and dentry tree not containing anything other
that pid-only parts + self + thread-self...

[toc] | [prev] | [next] | [standalone]

#1598838

On Sat, Mar 11, 2017 at 6:13 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> PS: AFAICS, simple mount --bind of your pid-only mount will suddenly
> expose the full thing.  And as for the lifetimes making no sense...
> note that you are simply not freeing these structures of yours.
> Try to handle that and you'll get a serious PITA all over the
> place.
>
> What are you trying to achieve, anyway?  Why not add a second vfsmount
> pointer per pid_namespace and make it initialized on demand, at the
> first attempt of no-pid mount?  Just have a separate no-pid instance
> created for those namespaces where it had been asked for, with
> separate superblock and dentry tree not containing anything other
> that pid-only parts + self + thread-self...

Can't we just make procfs work like most other filesystems and have
each mount have its own superblock?  If we need to do something funky
to stat() output to keep existing userspace working, I think that's
okay.

As far as I can tell, proc_mnt is very nearly useless -- it seems to
be used for proc_flush_task (which claims to be purely an optimization
and could be preserved in the common case where there's only one
relevant mount) and for sysctl_binary.  For the latter, we could
create proc_mnt but make actual user-initiated mounts be new
superblocks anyway.

[toc] | [prev] | [next] | [standalone]

#1599361

On Sun, Mar 12, 2017 at 08:19:33PM -0700, Andy Lutomirski wrote:
> On Sat, Mar 11, 2017 at 6:13 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> > PS: AFAICS, simple mount --bind of your pid-only mount will suddenly
> > expose the full thing.  And as for the lifetimes making no sense...
> > note that you are simply not freeing these structures of yours.
> > Try to handle that and you'll get a serious PITA all over the
> > place.
> >
> > What are you trying to achieve, anyway?  Why not add a second vfsmount
> > pointer per pid_namespace and make it initialized on demand, at the
> > first attempt of no-pid mount?  Just have a separate no-pid instance
> > created for those namespaces where it had been asked for, with
> > separate superblock and dentry tree not containing anything other
> > that pid-only parts + self + thread-self...
> 
> Can't we just make procfs work like most other filesystems and have
> each mount have its own superblock?  If we need to do something funky
> to stat() output to keep existing userspace working, I think that's
> okay.

First of all, most of the filesystems do *NOT* guarantee anything of
that sort.  And what's the point of having more instances than
necessary, anyway?

> As far as I can tell, proc_mnt is very nearly useless -- it seems to
> be used for proc_flush_task (which claims to be purely an optimization
> and could be preserved in the common case where there's only one
> relevant mount) and for sysctl_binary.  For the latter, we could
> create proc_mnt but make actual user-initiated mounts be new
> superblocks anyway.

Again, what for?  It won't salvage that kludge...  It's not as if it
had been hard to have separate pid-only instance created when asked
for (and reused every time when we are asked for pid-only).  What's
the point of ever having more than two instances per pidns?  IDGI...

Folks, there is no one-to-one correspondence between mountpoints and
superblocks.  Not since 2000 or so.  Just don't try to shove your
per-superblock stuff into vfsmount; it simply won't work.  If you
want a separate instance for that thing, then just go ahead and
have ->mount() decide which one to use (and whether to create a new
one).  All there is to it...

[toc] | [prev] | [next] | [standalone]

#1599511

On Mon, Mar 13, 2017 at 6:27 AM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> On Sun, Mar 12, 2017 at 08:19:33PM -0700, Andy Lutomirski wrote:
>> On Sat, Mar 11, 2017 at 6:13 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
>> > PS: AFAICS, simple mount --bind of your pid-only mount will suddenly
>> > expose the full thing.  And as for the lifetimes making no sense...
>> > note that you are simply not freeing these structures of yours.
>> > Try to handle that and you'll get a serious PITA all over the
>> > place.
>> >
>> > What are you trying to achieve, anyway?  Why not add a second vfsmount
>> > pointer per pid_namespace and make it initialized on demand, at the
>> > first attempt of no-pid mount?  Just have a separate no-pid instance
>> > created for those namespaces where it had been asked for, with
>> > separate superblock and dentry tree not containing anything other
>> > that pid-only parts + self + thread-self...
>>
>> Can't we just make procfs work like most other filesystems and have
>> each mount have its own superblock?  If we need to do something funky
>> to stat() output to keep existing userspace working, I think that's
>> okay.
>
> First of all, most of the filesystems do *NOT* guarantee anything of
> that sort.  And what's the point of having more instances than
> necessary, anyway?

I mean that, if I do:

mount -t proc -o foobar none a
mount -t proc -o baz none b

Then I think that the second mount should create a whole new proc
instance rather than just a new vfsmount.  Then the options could
differ, which would solve a bunch of problems.

>
> Again, what for?  It won't salvage that kludge...  It's not as if it
> had been hard to have separate pid-only instance created when asked
> for (and reused every time when we are asked for pid-only).  What's
> the point of ever having more than two instances per pidns?  IDGI...

I can easily procfs growing more than one interesting option.
Pid-only and hidepid come to mind, and that's already six possible
combinations.  The current hidepid implementation is really awful.

>
> Folks, there is no one-to-one correspondence between mountpoints and
> superblocks.  Not since 2000 or so.  Just don't try to shove your
> per-superblock stuff into vfsmount; it simply won't work.  If you
> want a separate instance for that thing, then just go ahead and
> have ->mount() decide which one to use (and whether to create a new
> one).  All there is to it...

That's what I mean.  I just don't see the point of going all-out in
trying to reuse superblocks.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web