Skip to contentSkip to conversation

ksmbd: fix lookup on idmapped mounts

PATCH In review Christian Brauner 5 participants 15 messages Aug 16, 2021 – Aug 21, 2021

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

PATCH 1/1 ksmbd: fix lookup on idmapped mounts

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped mounts out of the box! However, lookup is currently broken. Lookup helpers such as lookup_one_len() call inode_permission() internally to ensure that the caller is privileged over the inode of the base dentry they are trying to lookup under. So the permission checking here is currently wrong.

Linux v5.15 will gain a new lookup helper lookup_one() that does take idmappings into account. I've added it as part of my patch series to make btrfs support idmapped mounts. The new helper is in linux-next as part of David's (Sterba) btrfs for-next branch as commit c972214c133b ("namei: add mapping aware lookup helper").

I've said it before during one of my first reviews: I would very much recommend adding fstests to [1]. It already seems to have very rudimentary cifs support. There is a completely generic idmapped mount testsuite that supports idmapped mounts.

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/

Cc Steve French <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Namjae Jeon <[email protected]>
Cc Hyunchul Lee <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc David Sterba <[email protected]>

---
I merged David's for-next tree into cifsd-next to test this. I did only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

One more thing, the tree for ksmbd was very hard to find. I had to do a lot archeology to end up at:

git://git.samba.org/ksmbd.git

Would be appreciated if this tree could be reflected in MAINTAINERS or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Thanks!
Christian

---
 fs/ksmbd/smb2pdu.c | 18 +++++++++++-------
 fs/ksmbd/vfs.c     | 43 ++++++++++++++++++++++++-------------------
 fs/ksmbd/vfs.h     |  3 ++-
 3 files changed, 37 insertions(+), 27 deletions(-)
fs/ksmbd/smb2pdu.c +11 -7
index 636570ecfa31..9a1a9a024714 100644
@@ -3539,9 +3539,9 @@ static int process_query_dir_entries(struct smb2_query_dir_private *priv)
 			return -EINVAL;
 
 		lock_dir(priv->dir_fp);
-		dent = lookup_one_len(priv->d_info->name,
-				      priv->dir_fp->filp->f_path.dentry,
-				      priv->d_info->name_len);
+		dent = lookup_one(user_ns, priv->d_info->name,
+				  priv->dir_fp->filp->f_path.dentry,
+				  priv->d_info->name_len);
 		unlock_dir(priv->dir_fp);
 
 		if (IS_ERR(dent)) {
@@ -5242,7 +5242,9 @@ int smb2_echo(struct ksmbd_work *work)
 	return 0;
 }
 
-static int smb2_rename(struct ksmbd_work *work, struct ksmbd_file *fp,
+static int smb2_rename(struct ksmbd_work *work,
+		       struct ksmbd_file *fp,
+		       struct user_namespace *user_ns,
 		       struct smb2_file_rename_info *file_info,
 		       struct nls_table *local_nls)
 {
@@ -5306,7 +5308,7 @@ static int smb2_rename(struct ksmbd_work *work, struct ksmbd_file *fp,
 		if (rc)
 			goto out;
 
-		rc = ksmbd_vfs_setxattr(file_mnt_user_ns(fp->filp),
+		rc = ksmbd_vfs_setxattr(user_ns,
 					fp->filp->f_path.dentry,
 					xattr_stream_name,
 					NULL, 0, 0);
@@ -5620,6 +5622,7 @@ static int set_end_of_file_info(struct ksmbd_work *work, struct ksmbd_file *fp,
 static int set_rename_info(struct ksmbd_work *work, struct ksmbd_file *fp,
 			   char *buf)
 {
+	struct user_namespace *user_ns;
 	struct ksmbd_file *parent_fp;
 	struct dentry *parent;
 	struct dentry *dentry = fp->filp->f_path.dentry;
@@ -5633,8 +5636,9 @@ static int set_rename_info(struct ksmbd_work *work, struct ksmbd_file *fp,
 	if (ksmbd_stream_fd(fp))
 		goto next;
 
+	user_ns = file_mnt_user_ns(fp->filp);
 	parent = dget_parent(dentry);
-	ret = ksmbd_vfs_lock_parent(parent, dentry);
+	ret = ksmbd_vfs_lock_parent(user_ns, parent, dentry);
 	if (ret) {
 		dput(parent);
 		return ret;
@@ -5651,7 +5655,7 @@ static int set_rename_info(struct ksmbd_work *work, struct ksmbd_file *fp,
 		}
 	}
 next:
-	return smb2_rename(work, fp,
+	return smb2_rename(work, fp, user_ns,
 			   (struct smb2_file_rename_info *)buf,
 			   work->sess->conn->local_nls);
 }
fs/ksmbd/vfs.c +24 -19

137 lines not on this page · raw message

fs/ksmbd/vfs.h +2 -1
index cb0cba0d5d07..85db50abdb24 100644
@@ -107,7 +107,8 @@ struct ksmbd_kstat {
 	__le32			file_attributes;
 };
 
-int ksmbd_vfs_lock_parent(struct dentry *parent, struct dentry *child);
+int ksmbd_vfs_lock_parent(struct user_namespace *user_ns, struct dentry *parent,
+			  struct dentry *child);
 int ksmbd_vfs_may_delete(struct user_namespace *user_ns, struct dentry *dentry);
 int ksmbd_vfs_query_maximal_access(struct user_namespace *user_ns,
 				   struct dentry *dentry, __le32 *daccess);

base-commit: 456af438ad490bac7ed954cb929bcec1df7f0c82

Signature · 1 line
2.30.2
#1 lore raw
I love your patch! Perhaps something to improve:

Hi Christian,

I love your patch! Perhaps something to improve:

[auto build test WARNING on 456af438ad490bac7ed954cb929bcec1df7f0c82]

url:    https://github.com/0day-ci/linux/commits/Christian-Brauner/ksmbd-fix-lookup-on-idmapped-mounts/20210816-200054
base:   456af438ad490bac7ed954cb929bcec1df7f0c82
config: s390-allyesconfig (attached as .config)
compiler: s390-linux-gcc (GCC) 11.2.0

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/0day-ci/linux/commit/ebbc41b9d3f7b7f6c558cb5d5f12ecb8134c9c2e
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Christian-Brauner/ksmbd-fix-lookup-on-idmapped-mounts/20210816-200054 git checkout ebbc41b9d3f7b7f6c558cb5d5f12ecb8134c9c2e
# save the attached .config to linux build tree
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.2.0 make.cross ARCH=s390

If you fix the issue, kindly add following tag as appropriate

All warnings (new ones prefixed by >>):

fs/ksmbd/vfs.c: In function 'ksmbd_vfs_lock_parent':
fs/ksmbd/vfs.c:79:18: error: implicit declaration of function 'lookup_one'; did you mean 'lookup_bdev'? [-Werror=implicit-function-declaration]

      79 |         dentry = lookup_one(user_ns, child->d_name.name, parent,
         |                  ^~~~~~~~~~
         |                  lookup_bdev
>> fs/ksmbd/vfs.c:79:16: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
      79 |         dentry = lookup_one(user_ns, child->d_name.name, parent,
         |                ^
   fs/ksmbd/vfs.c: In function 'ksmbd_vfs_mkdir':
   fs/ksmbd/vfs.c:223:19: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
     223 |                 d = lookup_one(user_ns, dentry->d_name.name, dentry->d_parent,
         |                   ^
   fs/ksmbd/vfs.c: In function '__ksmbd_vfs_rename':
   fs/ksmbd/vfs.c:753:18: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
     753 |         dst_dent = lookup_one(dst_user_ns, dst_name, dst_dent_parent,
         |                  ^
   fs/ksmbd/vfs.c: In function 'ksmbd_vfs_fp_rename':
   fs/ksmbd/vfs.c:816:19: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
     816 |         src_child = lookup_one(user_ns, src_dent->d_name.name, src_dent_parent,
         |                   ^

cc1: some warnings being treated as errors
--
fs/ksmbd/smb2pdu.c: In function 'process_query_dir_entries':
fs/ksmbd/smb2pdu.c:3542:24: error: implicit declaration of function 'lookup_one'; did you mean 'lookup_bdev'? [-Werror=implicit-function-declaration]

    3542 |                 dent = lookup_one(user_ns, priv->d_info->name,
         |                        ^~~~~~~~~~
         |                        lookup_bdev
>> fs/ksmbd/smb2pdu.c:3542:22: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
    3542 |                 dent = lookup_one(user_ns, priv->d_info->name,
         |                      ^

cc1: some warnings being treated as errors

vim +79 fs/ksmbd/vfs.c

    62	
    63	/**
    64	 * ksmbd_vfs_lock_parent() - lock parent dentry if it is stable
    65	 *
    66	 * the parent dentry got by dget_parent or @parent could be
    67	 * unstable, we try to lock a parent inode and lookup the
    68	 * child dentry again.
    69	 *
    70	 * the reference count of @parent isn't incremented.
    71	 */
    72	int ksmbd_vfs_lock_parent(struct user_namespace *user_ns, struct dentry *parent,
    73				  struct dentry *child)
    74	{
    75		struct dentry *dentry;
    76		int ret = 0;
    77	
    78		inode_lock_nested(d_inode(parent), I_MUTEX_PARENT);
  > 79		dentry = lookup_one(user_ns, child->d_name.name, parent,
    80				    child->d_name.len);
    81		if (IS_ERR(dentry)) {
    82			ret = PTR_ERR(dentry);
    83			goto out_err;
    84		}
    85	
    86		if (dentry != child) {
    87			ret = -ESTALE;
    88			dput(dentry);
    89			goto out_err;
    90		}
    91	
    92		dput(dentry);
    93		return 0;
    94	out_err:
    95		inode_unlock(d_inode(parent));
    96		return ret;
    97	}
    98	

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

#2 lore raw
Quoted 20 lines · kernel test robot

Hi Christian,

I love your patch! Perhaps something to improve:

[auto build test WARNING on 456af438ad490bac7ed954cb929bcec1df7f0c82]

url:    https://github.com/0day-ci/linux/commits/Christian-Brauner/ksmbd-fix-lookup-on-idmapped-mounts/20210816-200054
base:   456af438ad490bac7ed954cb929bcec1df7f0c82
config: s390-allyesconfig (attached as .config)
compiler: s390-linux-gcc (GCC) 11.2.0

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/0day-ci/linux/commit/ebbc41b9d3f7b7f6c558cb5d5f12ecb8134c9c2e
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Christian-Brauner/ksmbd-fix-lookup-on-idmapped-mounts/20210816-200054 git checkout ebbc41b9d3f7b7f6c558cb5d5f12ecb8134c9c2e
# save the attached .config to linux build tree
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.2.0 make.cross ARCH=s390

Quoted 2 lines

If you fix the issue, kindly add following tag as appropriate

Note, this failure is expected as lookup_one() is in another tree. This is mentioned in the commit message which obviously the automatic tool can't be expected to figure out. But just an fyi, that there's nothing to do right now. :)

Christian

#3 lore raw
In reply to Christian Brauner
Quoted 25 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped mounts out of the box! However, lookup is currently broken. Lookup helpers such as lookup_one_len() call inode_permission() internally to ensure that the caller is privileged over the inode of the base dentry they are trying to lookup under. So the permission checking here is currently wrong.

Linux v5.15 will gain a new lookup helper lookup_one() that does take idmappings into account. I've added it as part of my patch series to make btrfs support idmapped mounts. The new helper is in linux- next as part of David's (Sterba) btrfs for-next branch as commit c972214c133b ("namei: add mapping aware lookup helper").

I've said it before during one of my first reviews: I would very much recommend adding fstests to [1]. It already seems to have very rudimentary cifs support. There is a completely generic idmapped mount testsuite that supports idmapped mounts.

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/

Cc Steve French <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Namjae Jeon <[email protected]>
Cc Hyunchul Lee <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc David Sterba <[email protected]>
Quoted 1 line

---

Hi Christian,

Quoted 3 lines

I merged David's for-next tree into cifsd-next to test this. I did only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

There is "How to run ksmbd" section in patch cover letter.
https://lkml.org/lkml/2021/8/5/54

Let me know if it doesn't work well even if you try to run it with this step. And We will also check whether your patch work fine.

Quoted 4 lines

One more thing, the tree for ksmbd was very hard to find. I had to do a lot archeology to end up at:

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section. I think that you can use :
https://github.com/namjaejeon/smb3-kernel/tree/ksmbd-v7-series

Quoted 3 lines

Would be appreciated if this tree could be reflected in MAINTAINERS or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Quoted 237 lines
---
 fs/ksmbd/smb2pdu.c | 18 +++++++++++-------
 fs/ksmbd/vfs.c     | 43 ++++++++++++++++++++++++-------------------
 fs/ksmbd/vfs.h     |  3 ++-
 3 files changed, 37 insertions(+), 27 deletions(-)
fs/ksmbd/smb2pdu.c +8 -5
@@ -3539,9 +3539,9 @@ static int process_query_dir_entries(struct smb2_query_dir_private *priv)
 			return -EINVAL;
 
 		lock_dir(priv->dir_fp);
-		dent = lookup_one_len(priv->d_info->name,
-				      priv->dir_fp->filp->f_path.dentry,
-				      priv->d_info->name_len);
+		dent = lookup_one(user_ns, priv->d_info->name,
+				  priv->dir_fp->filp->f_path.dentry,
+				  priv->d_info->name_len);
 		unlock_dir(priv->dir_fp);
 
 		if (IS_ERR(dent)) {
@@ -5242,7 +5242,9 @@ int smb2_echo(struct ksmbd_work *work)
 	return 0;
 }
 
-static int smb2_rename(struct ksmbd_work *work, struct ksmbd_file *fp,
+static int smb2_rename(struct ksmbd_work *work,
+		       struct ksmbd_file *fp,
+		       struct user_namespace *user_ns,
 		       struct smb2_file_rename_info *file_info,
 		       struct nls_table *local_nls)
 {
@@ -5306,7 +5308,7 @@ static int smb2_rename(struct ksmbd_work *work, struct ksmbd_file *fp,
 		if (rc)
 			goto out;
 
-		rc = ksmbd_vfs_setxattr(file_mnt_user_ns(fp->filp),
+		rc = ksmbd_vfs_setxattr(user_ns,
 					fp->filp->f_path.dentry,
 					xattr_stream_name,
 					NULL, 0, 0);
@@ -5620,6 +5622,7 @@ static int set_end_of_file_info(struct ksmbd_work *work, struct ksmbd_file *fp,
 static int set_rename_info(struct ksmbd_work *work, struct ksmbd_file *fp,
 			   char *buf)
 {
+	struct user_namespace *user_ns;
 	struct ksmbd_file *parent_fp;
 	struct dentry *parent;
 	struct dentry *dentry = fp->filp->f_path.dentry; @@ -5633,8 +5636,9 @@ static int

set_rename_info(struct ksmbd_work *work, struct ksmbd_file *fp,

fs/ksmbd/smb2pdu.c +3 -2
 	if (ksmbd_stream_fd(fp))
 		goto next;
 
+	user_ns = file_mnt_user_ns(fp->filp);
 	parent = dget_parent(dentry);
-	ret = ksmbd_vfs_lock_parent(parent, dentry);
+	ret = ksmbd_vfs_lock_parent(user_ns, parent, dentry);
 	if (ret) {
 		dput(parent);
 		return ret;
@@ -5651,7 +5655,7 @@ static int set_rename_info(struct ksmbd_work *work, struct ksmbd_file *fp,
 		}
 	}
 next:
-	return smb2_rename(work, fp,
+	return smb2_rename(work, fp, user_ns,
 			   (struct smb2_file_rename_info *)buf,
 			   work->sess->conn->local_nls);
 }
fs/ksmbd/vfs.c +18 -15
@@ -69,14 +69,15 @@ static void ksmbd_vfs_inherit_owner(struct ksmbd_work *work,
  *
  * the reference count of @parent isn't incremented.
  */
-int ksmbd_vfs_lock_parent(struct dentry *parent, struct dentry *child)
+int ksmbd_vfs_lock_parent(struct user_namespace *user_ns, struct dentry *parent,
+			  struct dentry *child)
 {
 	struct dentry *dentry;
 	int ret = 0;
 
 	inode_lock_nested(d_inode(parent), I_MUTEX_PARENT);
-	dentry = lookup_one_len(child->d_name.name, parent,
-				child->d_name.len);
+	dentry = lookup_one(user_ns, child->d_name.name, parent,
+			    child->d_name.len);
 	if (IS_ERR(dentry)) {
 		ret = PTR_ERR(dentry);
 		goto out_err;
@@ -102,7 +103,7 @@ int ksmbd_vfs_may_delete(struct user_namespace *user_ns,
 	int ret;
 
 	parent = dget_parent(dentry);
-	ret = ksmbd_vfs_lock_parent(parent, dentry);
+	ret = ksmbd_vfs_lock_parent(user_ns, parent, dentry);
 	if (ret) {
 		dput(parent);
 		return ret;
@@ -137,7 +138,7 @@ int ksmbd_vfs_query_maximal_access(struct user_namespace *user_ns,
 		*daccess |= FILE_EXECUTE_LE;
 
 	parent = dget_parent(dentry);
-	ret = ksmbd_vfs_lock_parent(parent, dentry);
+	ret = ksmbd_vfs_lock_parent(user_ns, parent, dentry);
 	if (ret) {
 		dput(parent);
 		return ret;
@@ -197,6 +198,7 @@ int ksmbd_vfs_create(struct ksmbd_work *work, const char *name, umode_t mode)
  */
 int ksmbd_vfs_mkdir(struct ksmbd_work *work, const char *name, umode_t mode)  {
+	struct user_namespace *user_ns;
 	struct path path;
 	struct dentry *dentry;
 	int err;
@@ -210,16 +212,16 @@ int ksmbd_vfs_mkdir(struct ksmbd_work *work, const char *name, umode_t mode)
 		return err;
 	}
 
+	user_ns = mnt_user_ns(path.mnt);
 	mode |= S_IFDIR;
-	err = vfs_mkdir(mnt_user_ns(path.mnt), d_inode(path.dentry),
-			dentry, mode);
+	err = vfs_mkdir(user_ns, d_inode(path.dentry), dentry, mode);
 	if (err) {
 		goto out;
 	} else if (d_unhashed(dentry)) {
 		struct dentry *d;
 
-		d = lookup_one_len(dentry->d_name.name, dentry->d_parent,
-				   dentry->d_name.len);
+		d = lookup_one(user_ns, dentry->d_name.name, dentry->d_parent,
+			       dentry->d_name.len);
 		if (IS_ERR(d)) {
 			err = PTR_ERR(d);
 			goto out;
@@ -582,6 +584,7 @@ int ksmbd_vfs_fsync(struct ksmbd_work *work, u64 fid, u64 p_id)
  */
 int ksmbd_vfs_remove_file(struct ksmbd_work *work, char *name)  {
+	struct user_namespace *user_ns;
 	struct path path;
 	struct dentry *parent;
 	int err;
@@ -601,8 +604,9 @@ int ksmbd_vfs_remove_file(struct ksmbd_work *work, char *name)
 		return err;
 	}
 
+	user_ns = mnt_user_ns(path.mnt);
 	parent = dget_parent(path.dentry);
-	err = ksmbd_vfs_lock_parent(parent, path.dentry);
+	err = ksmbd_vfs_lock_parent(user_ns, parent, path.dentry);
 	if (err) {
 		dput(parent);
 		path_put(&path);
@@ -616,14 +620,12 @@ int ksmbd_vfs_remove_file(struct ksmbd_work *work, char *name)
 	}
 
 	if (S_ISDIR(d_inode(path.dentry)->i_mode)) {
-		err = vfs_rmdir(mnt_user_ns(path.mnt), d_inode(parent),
-				path.dentry);
+		err = vfs_rmdir(user_ns, d_inode(parent), path.dentry);
 		if (err && err != -ENOTEMPTY)
 			ksmbd_debug(VFS, "%s: rmdir failed, err %d\n", name,
 				    err);
 	} else {
-		err = vfs_unlink(mnt_user_ns(path.mnt), d_inode(parent),
-				 path.dentry, NULL);
+		err = vfs_unlink(user_ns, d_inode(parent), path.dentry, NULL);
 		if (err)
 			ksmbd_debug(VFS, "%s: unlink failed, err %d\n", name,
 				    err);
@@ -748,7 +750,8 @@ static int __ksmbd_vfs_rename(struct ksmbd_work *work,
 	if (ksmbd_override_fsids(work))
 		return -ENOMEM;
 
-	dst_dent = lookup_one_len(dst_name, dst_dent_parent, strlen(dst_name));
+	dst_dent = lookup_one(dst_user_ns, dst_name, dst_dent_parent,
+			      strlen(dst_name));
 	err = PTR_ERR(dst_dent);
 	if (IS_ERR(dst_dent)) {
 		pr_err("lookup failed %s [%d]\n", dst_name, err); @@ -779,6 +782,7 @@ static int

__ksmbd_vfs_rename(struct ksmbd_work *work, int ksmbd_vfs_fp_rename(struct ksmbd_work *work, struct ksmbd_file *fp,

fs/ksmbd/vfs.c +1 -0
 			char *newname)
 {
+	struct user_namespace *user_ns;
 	struct path dst_path;
 	struct dentry *src_dent_parent, *dst_dent_parent;
 	struct dentry *src_dent, *trap_dent, *src_child; @@ -808,8 +812,9 @@ int

ksmbd_vfs_fp_rename(struct ksmbd_work *work, struct ksmbd_file *fp,

fs/ksmbd/vfs.c +5 -4
 	trap_dent = lock_rename(src_dent_parent, dst_dent_parent);
 	dget(src_dent);
 	dget(dst_dent_parent);
-	src_child = lookup_one_len(src_dent->d_name.name, src_dent_parent,
-				   src_dent->d_name.len);
+	user_ns = file_mnt_user_ns(fp->filp);
+	src_child = lookup_one(user_ns, src_dent->d_name.name, src_dent_parent,
+			       src_dent->d_name.len);
 	if (IS_ERR(src_child)) {
 		err = PTR_ERR(src_child);
 		goto out_lock;
@@ -823,7 +828,7 @@ int ksmbd_vfs_fp_rename(struct ksmbd_work *work, struct ksmbd_file *fp,
 	dput(src_child);
 
 	err = __ksmbd_vfs_rename(work,
-				 file_mnt_user_ns(fp->filp),
+				 user_ns,
 				 src_dent_parent,
 				 src_dent,
 				 mnt_user_ns(dst_path.mnt),
@@ -1109,7 +1114,7 @@ int ksmbd_vfs_unlink(struct user_namespace *user_ns,  {
 	int err = 0;
 
-	err = ksmbd_vfs_lock_parent(dir, dentry);
+	err = ksmbd_vfs_lock_parent(user_ns, dir, dentry);
 	if (err)
 		return err;
 	dget(dentry);
fs/ksmbd/vfs.h +2 -1
@@ -107,7 +107,8 @@ struct ksmbd_kstat {
 	__le32			file_attributes;
 };
 
-int ksmbd_vfs_lock_parent(struct dentry *parent, struct dentry *child);
+int ksmbd_vfs_lock_parent(struct user_namespace *user_ns, struct dentry *parent,
+			  struct dentry *child);
 int ksmbd_vfs_may_delete(struct user_namespace *user_ns, struct dentry *dentry);  int
 ksmbd_vfs_query_maximal_access(struct user_namespace *user_ns,
 				   struct dentry *dentry, __le32 *daccess);

base-commit: 456af438ad490bac7ed954cb929bcec1df7f0c82 --
2.30.2

#4 lore raw
Quoted 47 lines · Namjae Jeon
Quoted 25 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped mounts out of the box! However, lookup is currently broken. Lookup helpers such as lookup_one_len() call inode_permission() internally to ensure that the caller is privileged over the inode of the base dentry they are trying to lookup under. So the permission checking here is currently wrong.

Linux v5.15 will gain a new lookup helper lookup_one() that does take idmappings into account. I've added it as part of my patch series to make btrfs support idmapped mounts. The new helper is in linux- next as part of David's (Sterba) btrfs for-next branch as commit c972214c133b ("namei: add mapping aware lookup helper").

I've said it before during one of my first reviews: I would very much recommend adding fstests to [1]. It already seems to have very rudimentary cifs support. There is a completely generic idmapped mount testsuite that supports idmapped mounts.

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/

Cc Steve French <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Namjae Jeon <[email protected]>
Cc Hyunchul Lee <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc David Sterba <[email protected]>
Quoted 1 line

---

Hi Christian,

Quoted 3 lines

I merged David's for-next tree into cifsd-next to test this. I did only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

There is "How to run ksmbd" section in patch cover letter.
https://lkml.org/lkml/2021/8/5/54

Let me know if it doesn't work well even if you try to run it with this step. And We will also check whether your patch work fine.

Quoted 4 lines

One more thing, the tree for ksmbd was very hard to find. I had to do a lot archeology to end up at:

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section. I think that you can use :
https://github.com/namjaejeon/smb3-kernel/tree/ksmbd-v7-series

Quoted 1 line
Quoted 8 lines
Quoted 2 lines

Would be appreciated if this tree could be reflected in MAINTAINERS or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Thank your for the pointers.

Ok, so I've been taking the time to look into cifs and ksmbd today. My mental model was wrong. There are two things to consider here:

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.
Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no

So /opt can be an idmapped mount and ksmb would know how to deal with that correctly, i.e. you could do:

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

That however is different from 2. which is cifs itself being idmappable. Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand your earlier confusion. In another mail you pointed out that cifs reports that idmapped mounts are not supported. That is correct insofar as it means 2. is not supported, i.e. you can't do:

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What you want is to test 1.

So your test setup would require you to setup an idmapped mount and have ksmbd use that which can then be mounted by a client.

With your instructions I was at least able to get a ksmb instance running and be able to mount a client with -t cifs. All on the same machine, i.e. my server is localhost.

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Are unix extension supported with ksmb? Everytime I try to use "posix" as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get "uid=0" and "gid=0" and "noposix". I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

Christian

#5 lore raw
Quoted 72 lines · Namjae Jeon
Quoted 6 lines
Quoted 6 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped
mounts out of the box! However, lookup is currently broken. Lookup
helpers such as lookup_one_len() call inode_permission() internally
to ensure that the caller is privileged over the inode of the base dentry they are trying to

lookup under. So the permission checking here is currently wrong.

Quoted 5 lines
Quoted 5 lines

Linux v5.15 will gain a new lookup helper lookup_one() that does
take idmappings into account. I've added it as part of my patch
series to make btrfs support idmapped mounts. The new helper is in
linux- next as part of David's (Sterba) btrfs for-next branch as commit c972214c133b ("namei: add

mapping aware lookup helper").

Quoted 2 lines
Quoted 2 lines

I've said it before during one of my first reviews: I would very much recommend adding fstests to

[1].

Quoted 21 lines
Quoted 12 lines

It already seems to have very rudimentary cifs support. There is a completely generic idmapped mount testsuite that supports idmapped mounts.

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/

Cc Steve French <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Namjae Jeon <[email protected]>
Cc Hyunchul Lee <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc David Sterba <[email protected]>
Quoted 1 line

---

Hi Christian,

Quoted 5 lines

I merged David's for-next tree into cifsd-next to test this. I did only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

Quoted 11 lines

There is "How to run ksmbd" section in patch cover letter.

https://protect2.fireeye.com/v1/url?k=65ecaaf0-3a779239-65ed21bf-0cc47
a336fae-53bc47005a1a97a9&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Flkml.org%2Flkml%2F2021%2F8%2F5%2F54

Let me know if it doesn't work well even if you try to run it with this step. And We will also check whether your patch work fine.

Quoted 2 lines

One more thing, the tree for ksmbd was very hard to find. I had to do a lot archeology to end up

at:

Quoted 11 lines
Quoted 2 lines

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section. I think that you can use :

https://protect2.fireeye.com/v1/url?k=8af83a5d-d5630294-8af9b112-0cc47
a336fae-e471ffbdb93d05b7&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Fgithub.com%2Fnamjaejeon%2Fsmb3-kernel%2Ftree%2Fksmbd-v7- series

Quoted 1 line
Quoted 8 lines
Quoted 2 lines

Would be appreciated if this tree could be reflected in MAINTAINERS
or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Thank your for the pointers.

Quoted 7 lines

Ok, so I've been taking the time to look into cifs and ksmbd today. My mental model was wrong. There are two things to consider here:

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.

Right.

Quoted 12 lines

Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no
Quoted 8 lines

So /opt can be an idmapped mount and ksmb would know how to deal with that correctly, i.e. you could do:

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

Right.

Quoted 2 lines

That however is different from 2. which is cifs itself being idmappable.

Right.

Quoted 5 lines

Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand your earlier confusion. In another mail you pointed out that cifs reports that idmapped mounts are not supported. That is correct insofar as it means 2. is not supported, i.e. you can't do:

Right.

Quoted 8 lines

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What you want is to test 1.

Right. So we have manually tested it, not xfstests.

Quoted 6 lines

So your test setup would require you to setup an idmapped mount and have ksmbd use that which can then be mounted by a client.

With your instructions I was at least able to get a ksmb instance running and be able to mount a client with -t cifs. All on the same machine, i.e. my server is localhost.

Okay.

Quoted 2 lines

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Okay. And I have applied your patch to ksmbd.

Quoted 4 lines

Are unix extension supported with ksmb? Everytime I try to use "posix"
as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get "uid=0" and "gid=0" and "noposix". I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

With posix mount option, It should work. It worked well before but it is strange now.

I'm not sure this is the correct fix, But could you please try to mount with the below change ?

fs/cifs/fs_context.c +3 -1
index eed59bc1d913..5fd0b0ddcc57 100644
@@ -1268,8 +1268,10 @@ static int smb3_fs_context_parse_param(struct fs_context *fc,
        case Opt_unix:
                if (result.negated)
                        ctx->linux_ext = 0;
-               else
+               else {
+                       ctx->linux_ext = 1;
                        ctx->no_linux_ext = 1;
+               }
                break;
        case Opt_nocase:
                ctx->nocase = 1;

Thanks for your work!

Quoted 2 lines

Christian

#6 lore raw
Quoted 150 lines · Namjae Jeon
Quoted 72 lines · Namjae Jeon
Quoted 6 lines
Quoted 6 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped
mounts out of the box! However, lookup is currently broken. Lookup
helpers such as lookup_one_len() call inode_permission() internally
to ensure that the caller is privileged over the inode of the base dentry they are trying to

lookup under. So the permission checking here is currently wrong.

Quoted 5 lines
Quoted 5 lines

Linux v5.15 will gain a new lookup helper lookup_one() that does
take idmappings into account. I've added it as part of my patch
series to make btrfs support idmapped mounts. The new helper is in
linux- next as part of David's (Sterba) btrfs for-next branch as commit c972214c133b ("namei: add

mapping aware lookup helper").

Quoted 2 lines
Quoted 2 lines

I've said it before during one of my first reviews: I would very much recommend adding fstests to

[1].

Quoted 21 lines
Quoted 12 lines

It already seems to have very rudimentary cifs support. There is a completely generic idmapped mount testsuite that supports idmapped mounts.

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/

Cc Steve French <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Namjae Jeon <[email protected]>
Cc Hyunchul Lee <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc David Sterba <[email protected]>
Quoted 1 line

---

Hi Christian,

Quoted 5 lines

I merged David's for-next tree into cifsd-next to test this. I did only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

Quoted 11 lines

There is "How to run ksmbd" section in patch cover letter.

https://protect2.fireeye.com/v1/url?k=65ecaaf0-3a779239-65ed21bf-0cc47
a336fae-53bc47005a1a97a9&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Flkml.org%2Flkml%2F2021%2F8%2F5%2F54

Let me know if it doesn't work well even if you try to run it with this step. And We will also check whether your patch work fine.

Quoted 2 lines

One more thing, the tree for ksmbd was very hard to find. I had to do a lot archeology to end up

at:

Quoted 11 lines
Quoted 2 lines

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section. I think that you can use :

https://protect2.fireeye.com/v1/url?k=8af83a5d-d5630294-8af9b112-0cc47
a336fae-e471ffbdb93d05b7&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Fgithub.com%2Fnamjaejeon%2Fsmb3-kernel%2Ftree%2Fksmbd-v7- series

Quoted 1 line
Quoted 8 lines
Quoted 2 lines

Would be appreciated if this tree could be reflected in MAINTAINERS
or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Thank your for the pointers.

Quoted 7 lines

Ok, so I've been taking the time to look into cifs and ksmbd today. My mental model was wrong. There are two things to consider here:

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.

Right.

Quoted 12 lines

Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no
Quoted 8 lines

So /opt can be an idmapped mount and ksmb would know how to deal with that correctly, i.e. you could do:

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

Right.

Quoted 2 lines

That however is different from 2. which is cifs itself being idmappable.

Right.

Quoted 5 lines

Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand your earlier confusion. In another mail you pointed out that cifs reports that idmapped mounts are not supported. That is correct insofar as it means 2. is not supported, i.e. you can't do:

Right.

Quoted 8 lines

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What you want is to test 1.

Right. So we have manually tested it, not xfstests.

Quoted 6 lines

So your test setup would require you to setup an idmapped mount and have ksmbd use that which can then be mounted by a client.

With your instructions I was at least able to get a ksmb instance running and be able to mount a client with -t cifs. All on the same machine, i.e. my server is localhost.

Okay.

Quoted 2 lines

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Okay. And I have applied your patch to ksmbd.

Quoted 4 lines

Are unix extension supported with ksmb? Everytime I try to use "posix"
as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get "uid=0" and "gid=0" and "noposix". I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

With posix mount option, It should work. It worked well before but it is strange now.

I'm not sure this is the correct fix, But could you please try to mount with the below change ?

fs/cifs/fs_context.c +0 -0
index eed59bc1d913..5fd0b0ddcc57 100644
@@ -1268,8 +1268,10 @@ static int smb3_fs_context_parse_param(struct fs_context *fc,
        case Opt_unix:

fs/cifs/fs_context.c static int smb3_fs_context_parse_param(struct fs_context *fc, · line 1277

Quoted 10 lines
fs/cifs/fs_context.c +3 -1
                if (result.negated)
                        ctx->linux_ext = 0;
-               else
+               else {
+                       ctx->linux_ext = 1;
                        ctx->no_linux_ext = 1;
+               }
                break;
        case Opt_nocase:
                ctx->nocase = 1;

That stops the bleeding indeed. :)
I think a slightly nicer fix might be:

fs/cifs/fs_context.c +2 -1
index eed59bc1d913..424b8dc2232e 100644
@@ -1269,7 +1269,8 @@ static int smb3_fs_context_parse_param(struct fs_context *fc,
                if (result.negated)
                        ctx->linux_ext = 0;
                else
-                       ctx->no_linux_ext = 1;
+                       ctx->linux_ext = 1;
+               ctx->no_linux_ext = !ctx->linux_ext;
                break;
        case Opt_nocase:
                ctx->nocase = 1;

So with this patch applied I got unix permissions working after all. So I was able to do some more testing.

Just a few questions (independent of idmapped mounts):

1. Are the "uid=" and "gid=" mount options ignored when "username=" is specified and "posix" is specified?

It seems that "uid=" and "gid=" have are silently ignored when "posix' is set. They are neither used to report file ownership under the cifs mountpoint nor are they relevant when determining ownership on disk?

As an example, assume I have added a user "foo" with uid 1000 to ksmbd via:

ksmbd.adduser -a foo

And I mounted a share via:

           mount -t cifs -o username=foo,password=bar,posix,uid=1234,gid=1234,forceuid,forcegid //127.0.0.1/test /mnt

i) Ownership in /mnt appears posix-correct, i.e. "uid=" and "gid=" have no effect on the reported ownership.

ii) Assume I'm logged in as the root user with uid 0. If I create
file or directory in /mnt it will be owned by user foo, i.e. uid 1000, i.e., the "uid=1234" and "gid=1234" mount option have zero effect on the ownership of the files?

2. Are the "uid=" and "gid=" options ignored for permission checking when "posix" is specified?

3. Am I correct in assuming that there is no mount option that makes chown() or chmod() have an actual effect.

cifs seems to have support for it but the ksmbd server doesn't seem to?

Both with "perm" and "noperm" chown() and chmod() report success but without actually changing ownership or mode. That seems to be in line with what I can gather from various mangpages.

Christian

maintainer
#7 lore raw

2021-08-19 22:01 GMT+09:00, Christian Brauner <[email protected]>:

Quoted 179 lines · Namjae Jeon
Quoted 178 lines
Quoted 82 lines · Namjae Jeon
Quoted 8 lines
Quoted 8 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped
mounts out of the box! However, lookup is currently broken. Lookup helpers such as lookup_one_len() call inode_permission() internally to ensure that the caller is privileged over the inode of the base dentry they are trying to

lookup under. So the permission checking here is currently wrong.

Quoted 6 lines
Quoted 6 lines

Linux v5.15 will gain a new lookup helper lookup_one() that does take idmappings into account. I've added it as part of my patch series to make btrfs support idmapped mounts. The new helper is in linux- next as part of David's (Sterba) btrfs for-next branch as commit c972214c133b ("namei: add

mapping aware lookup helper").

Quoted 3 lines
Quoted 3 lines

I've said it before during one of my first reviews: I would very much recommend adding fstests to

[1].

Quoted 24 lines
Quoted 13 lines

It already seems to have very rudimentary cifs support. There is a completely generic idmapped mount testsuite that supports idmapped mounts.

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/

Cc Steve French <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Namjae Jeon <[email protected]>
Cc Hyunchul Lee <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc David Sterba <[email protected]>
Quoted 1 line

---

Hi Christian,

Quoted 5 lines

I merged David's for-next tree into cifsd-next to test this. I did only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

There is "How to run ksmbd" section in patch cover letter.

Quoted 11 lines

https://protect2.fireeye.com/v1/url?k=65ecaaf0-3a779239-65ed21bf-0cc47
a336fae-53bc47005a1a97a9&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Flkml.org%2Flkml%2F2021%2F8%2F5%2F54

Let me know if it doesn't work well even if you try to run it with this step.
And We will also check whether your patch work fine.

Quoted 3 lines

One more thing, the tree for ksmbd was very hard to find. I had to do a lot archeology to end up

at:

Quoted 15 lines
Quoted 2 lines

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section.
I think that you can use :

https://protect2.fireeye.com/v1/url?k=8af83a5d-d5630294-8af9b112-0cc47
a336fae-e471ffbdb93d05b7&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Fgithub.com%2Fnamjaejeon%2Fsmb3-kernel%2Ftree%2Fksmbd-v7- series

Quoted 4 lines

Would be appreciated if this tree could be reflected in MAINTAINERS or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Quoted 7 lines

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Thank your for the pointers.

Quoted 8 lines

Ok, so I've been taking the time to look into cifs and ksmbd today. My mental model was wrong. There
are two things to consider here:

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.

Right.

Quoted 16 lines

Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no

So /opt can be an idmapped mount and ksmb would know how to deal with that correctly, i.e. you could
do:

Quoted 5 lines

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

Right.

Quoted 3 lines

That however is different from 2. which is cifs itself being idmappable.

Right.

Quoted 7 lines

Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand your earlier confusion. In
another mail you pointed out that cifs reports that idmapped mounts are not supported. That is correct
insofar as it means 2. is not supported, i.e. you can't do:

Right.

Quoted 9 lines

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What you want is to test 1.

Right. So we have manually tested it, not xfstests.

Quoted 9 lines

So your test setup would require you to setup an idmapped mount and have ksmbd use that which can then
be mounted by a client.

With your instructions I was at least able to get a ksmb instance running and be able to mount a
client with -t cifs. All on the same machine, i.e. my server is localhost.

Okay.

Quoted 3 lines

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Okay. And I have applied your patch to ksmbd.

Quoted 5 lines

Are unix extension supported with ksmb? Everytime I try to use "posix" as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get "uid=0" and "gid=0" and "noposix".
I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

With posix mount option, It should work. It worked well before but it is strange now.

I'm not sure this is the correct fix, But could you please try to mount with the below change ?

fs/cifs/fs_context.c +1 -1
index eed59bc1d913..5fd0b0ddcc57 100644
@@ -1268,8 +1268,10 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
        case Opt_unix:
                if (result.negated)
                        ctx->linux_ext = 0;
-               else
+               else {
Quoted 8 lines
Quoted 6 lines
fs/cifs/fs_context.c +2 -0
+                       ctx->linux_ext = 1;
                        ctx->no_linux_ext = 1;
+               }
                break;
        case Opt_nocase:
                ctx->nocase = 1;

That stops the bleeding indeed. :)

Okay, sorry for late response.

Quoted 18 lines

I think a slightly nicer fix might be:

fs/cifs/fs_context.c +2 -1
index eed59bc1d913..424b8dc2232e 100644
@@ -1269,7 +1269,8 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
                if (result.negated)
                        ctx->linux_ext = 0;
                else
-                       ctx->no_linux_ext = 1;
+                       ctx->linux_ext = 1;
+               ctx->no_linux_ext = !ctx->linux_ext;
                break;
        case Opt_nocase:
                ctx->nocase = 1;
Quoted 2 lines

So with this patch applied I got unix permissions working after all. So I was able to do some more testing.

Okay.

Quoted 30 lines

Just a few questions (independent of idmapped mounts):

1. Are the "uid=" and "gid=" mount options ignored when "username=" is specified and "posix" is specified?

It seems that "uid=" and "gid=" have are silently ignored when "posix' is set. They are neither used to report file ownership under the cifs mountpoint nor are they relevant when determining ownership on disk?

As an example, assume I have added a user "foo" with uid 1000 to ksmbd via:

ksmbd.adduser -a foo

And I mounted a share via:

mount -t cifs -o
username=foo,password=bar,posix,uid=1234,gid=1234,forceuid,forcegid //127.0.0.1/test /mnt

i) Ownership in /mnt appears posix-correct, i.e. "uid=" and "gid=" have no effect on the reported ownership.

ii) Assume I'm logged in as the root user with uid 0. If I create
file or directory in /mnt it will be owned by user foo, i.e. uid 1000, i.e., the "uid=1234" and "gid=1234" mount option have zero effect on the ownership of the files?

Quoted 5 lines

2. Are the "uid=" and "gid=" options ignored for permission checking when "posix" is specified?

3. Am I correct in assuming that there is no mount option that makes chown() or chmod() have an actual effect.

That will be an answer for 1,2, 3 question. There are mount options for this.
1. modefromsid
2. idsfromsid

You can use this mount option and please check it.

Quoted 3 lines

cifs seems to have support for it but the ksmbd server doesn't seem to?

No, you need to use mount options for this as I said.
ksmbd have supported it but I found an issue related to chown and have fixed.

Could you please check the below git branch (pulled David'tree + ksmbd fixes) ?

  git clone --branch=for-christian https://github.com/namjaejeon/smb3-kernel

Thanks!

Quoted 7 lines

Both with "perm" and "noperm" chown() and chmod() report success but without actually changing ownership or mode. That seems to be in line with what I can gather from various mangpages.

Christian

#8 lore raw
Quoted 243 lines · Namjae Jeon

2021-08-19 22:01 GMT+09:00, Christian Brauner <[email protected]>:

Quoted 179 lines · Namjae Jeon
Quoted 178 lines
Quoted 82 lines · Namjae Jeon
Quoted 8 lines
Quoted 8 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped
mounts out of the box! However, lookup is currently broken. Lookup helpers such as lookup_one_len() call inode_permission() internally to ensure that the caller is privileged over the inode of the base dentry they are trying to

lookup under. So the permission checking here is currently wrong.

Quoted 6 lines
Quoted 6 lines

Linux v5.15 will gain a new lookup helper lookup_one() that does take idmappings into account. I've added it as part of my patch series to make btrfs support idmapped mounts. The new helper is in linux- next as part of David's (Sterba) btrfs for-next branch as commit c972214c133b ("namei: add

mapping aware lookup helper").

Quoted 3 lines
Quoted 3 lines

I've said it before during one of my first reviews: I would very much recommend adding fstests to

[1].

Quoted 24 lines
Quoted 13 lines

It already seems to have very rudimentary cifs support. There is a completely generic idmapped mount testsuite that supports idmapped mounts.

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/

Cc Steve French <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Namjae Jeon <[email protected]>
Cc Hyunchul Lee <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc David Sterba <[email protected]>
Quoted 1 line

---

Hi Christian,

Quoted 5 lines

I merged David's for-next tree into cifsd-next to test this. I did only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

There is "How to run ksmbd" section in patch cover letter.

Quoted 11 lines

https://protect2.fireeye.com/v1/url?k=65ecaaf0-3a779239-65ed21bf-0cc47
a336fae-53bc47005a1a97a9&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Flkml.org%2Flkml%2F2021%2F8%2F5%2F54

Let me know if it doesn't work well even if you try to run it with this step.
And We will also check whether your patch work fine.

Quoted 3 lines

One more thing, the tree for ksmbd was very hard to find. I had to do a lot archeology to end up

at:

Quoted 15 lines
Quoted 2 lines

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section.
I think that you can use :

https://protect2.fireeye.com/v1/url?k=8af83a5d-d5630294-8af9b112-0cc47
a336fae-e471ffbdb93d05b7&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Fgithub.com%2Fnamjaejeon%2Fsmb3-kernel%2Ftree%2Fksmbd-v7- series

Quoted 4 lines

Would be appreciated if this tree could be reflected in MAINTAINERS or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Quoted 7 lines

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Thank your for the pointers.

Quoted 8 lines

Ok, so I've been taking the time to look into cifs and ksmbd today. My mental model was wrong. There
are two things to consider here:

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.

Right.

Quoted 16 lines

Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no

So /opt can be an idmapped mount and ksmb would know how to deal with that correctly, i.e. you could
do:

Quoted 5 lines

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

Right.

Quoted 3 lines

That however is different from 2. which is cifs itself being idmappable.

Right.

Quoted 7 lines

Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand your earlier confusion. In
another mail you pointed out that cifs reports that idmapped mounts are not supported. That is correct
insofar as it means 2. is not supported, i.e. you can't do:

Right.

Quoted 9 lines

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What you want is to test 1.

Right. So we have manually tested it, not xfstests.

Quoted 9 lines

So your test setup would require you to setup an idmapped mount and have ksmbd use that which can then
be mounted by a client.

With your instructions I was at least able to get a ksmb instance running and be able to mount a
client with -t cifs. All on the same machine, i.e. my server is localhost.

Okay.

Quoted 3 lines

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Okay. And I have applied your patch to ksmbd.

Quoted 5 lines

Are unix extension supported with ksmb? Everytime I try to use "posix" as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get "uid=0" and "gid=0" and "noposix".
I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

With posix mount option, It should work. It worked well before but it is strange now.

I'm not sure this is the correct fix, But could you please try to mount with the below change ?

fs/cifs/fs_context.c +1 -1
index eed59bc1d913..5fd0b0ddcc57 100644
@@ -1268,8 +1268,10 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
        case Opt_unix:
                if (result.negated)
                        ctx->linux_ext = 0;
-               else
+               else {
Quoted 8 lines
Quoted 6 lines
fs/cifs/fs_context.c +2 -0
+                       ctx->linux_ext = 1;
                        ctx->no_linux_ext = 1;
+               }
                break;
        case Opt_nocase:
                ctx->nocase = 1;

That stops the bleeding indeed. :)

Okay, sorry for late response.

Quoted 18 lines

I think a slightly nicer fix might be:

fs/cifs/fs_context.c +2 -1
index eed59bc1d913..424b8dc2232e 100644
@@ -1269,7 +1269,8 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
                if (result.negated)
                        ctx->linux_ext = 0;
                else
-                       ctx->no_linux_ext = 1;
+                       ctx->linux_ext = 1;
+               ctx->no_linux_ext = !ctx->linux_ext;
                break;
        case Opt_nocase:
                ctx->nocase = 1;
Quoted 2 lines

So with this patch applied I got unix permissions working after all. So I was able to do some more testing.

Okay.

Quoted 33 lines

Just a few questions (independent of idmapped mounts):

1. Are the "uid=" and "gid=" mount options ignored when "username=" is specified and "posix" is specified?

It seems that "uid=" and "gid=" have are silently ignored when "posix' is set. They are neither used to report file ownership under the cifs mountpoint nor are they relevant when determining ownership on disk?

As an example, assume I have added a user "foo" with uid 1000 to ksmbd via:

ksmbd.adduser -a foo

And I mounted a share via:

mount -t cifs -o
username=foo,password=bar,posix,uid=1234,gid=1234,forceuid,forcegid //127.0.0.1/test /mnt

i) Ownership in /mnt appears posix-correct, i.e. "uid=" and "gid=" have no effect on the reported ownership.

ii) Assume I'm logged in as the root user with uid 0. If I create
file or directory in /mnt it will be owned by user foo, i.e. uid 1000, i.e., the "uid=1234" and "gid=1234" mount option have zero effect on the ownership of the files?

2. Are the "uid=" and "gid=" options ignored for permission checking when "posix" is specified?

Quoted 7 lines
Quoted 2 lines

3. Am I correct in assuming that there is no mount option that makes chown() or chmod() have an actual effect.

That will be an answer for 1,2, 3 question. There are mount options for this.
1. modefromsid
2. idsfromsid

You can use this mount option and please check it.

Thank you! This works and finally I can hit some codepaths I wasn't able to until now.

Quoted 9 lines
Quoted 3 lines

cifs seems to have support for it but the ksmbd server doesn't seem to?

No, you need to use mount options for this as I said.
ksmbd have supported it but I found an issue related to chown and have fixed.

Could you please check the below git branch (pulled David'tree + ksmbd fixes) ?

  git clone --branch=for-christian https://github.com/namjaejeon/smb3-kernel

Thanks, I've pulled that branch.

I have a some patches for ksmb that I'll be sending out next week. I just need to test the changes and verify that it all makes sense.

Christian

#9 lore raw
Quoted 258 lines · Christian Brauner
Quoted 243 lines · Namjae Jeon

2021-08-19 22:01 GMT+09:00, Christian Brauner <[email protected]>:

Quoted 179 lines · Namjae Jeon
Quoted 178 lines
Quoted 82 lines · Namjae Jeon
Quoted 8 lines
Quoted 8 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped
mounts out of the box! However, lookup is currently broken. Lookup helpers such as lookup_one_len() call inode_permission() internally to ensure that the caller is privileged over the inode of the base dentry they are trying to

lookup under. So the permission checking here is currently wrong.

Quoted 6 lines
Quoted 6 lines

Linux v5.15 will gain a new lookup helper lookup_one() that does take idmappings into account. I've added it as part of my patch series to make btrfs support idmapped mounts. The new helper is in linux- next as part of David's (Sterba) btrfs for-next branch as commit c972214c133b ("namei: add

mapping aware lookup helper").

Quoted 3 lines
Quoted 3 lines

I've said it before during one of my first reviews: I would very much recommend adding fstests to

[1].

Quoted 24 lines
Quoted 13 lines

It already seems to have very rudimentary cifs support. There is a completely generic idmapped mount testsuite that supports idmapped mounts.

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/

Cc Steve French <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Namjae Jeon <[email protected]>
Cc Hyunchul Lee <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc David Sterba <[email protected]>
Quoted 1 line

---

Hi Christian,

Quoted 5 lines

I merged David's for-next tree into cifsd-next to test this. I did only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

There is "How to run ksmbd" section in patch cover letter.

Quoted 11 lines

https://protect2.fireeye.com/v1/url?k=65ecaaf0-3a779239-65ed21bf-0cc47
a336fae-53bc47005a1a97a9&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Flkml.org%2Flkml%2F2021%2F8%2F5%2F54

Let me know if it doesn't work well even if you try to run it with this step.
And We will also check whether your patch work fine.

Quoted 3 lines

One more thing, the tree for ksmbd was very hard to find. I had to do a lot archeology to end up

at:

Quoted 15 lines
Quoted 2 lines

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section.
I think that you can use :

https://protect2.fireeye.com/v1/url?k=8af83a5d-d5630294-8af9b112-0cc47
a336fae-e471ffbdb93d05b7&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Fgithub.com%2Fnamjaejeon%2Fsmb3-kernel%2Ftree%2Fksmbd-v7- series

Quoted 4 lines

Would be appreciated if this tree could be reflected in MAINTAINERS or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Quoted 7 lines

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Thank your for the pointers.

Quoted 8 lines

Ok, so I've been taking the time to look into cifs and ksmbd today. My mental model was wrong. There
are two things to consider here:

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.

Right.

Quoted 16 lines

Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no

So /opt can be an idmapped mount and ksmb would know how to deal with that correctly, i.e. you could
do:

Quoted 5 lines

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

Right.

Quoted 3 lines

That however is different from 2. which is cifs itself being idmappable.

Right.

Quoted 7 lines

Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand your earlier confusion. In
another mail you pointed out that cifs reports that idmapped mounts are not supported. That is correct
insofar as it means 2. is not supported, i.e. you can't do:

Right.

Quoted 9 lines

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What you want is to test 1.

Right. So we have manually tested it, not xfstests.

Quoted 9 lines

So your test setup would require you to setup an idmapped mount and have ksmbd use that which can then
be mounted by a client.

With your instructions I was at least able to get a ksmb instance running and be able to mount a
client with -t cifs. All on the same machine, i.e. my server is localhost.

Okay.

Quoted 3 lines

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Okay. And I have applied your patch to ksmbd.

Quoted 5 lines

Are unix extension supported with ksmb? Everytime I try to use "posix" as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get "uid=0" and "gid=0" and "noposix".
I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

With posix mount option, It should work. It worked well before but it is strange now.

I'm not sure this is the correct fix, But could you please try to mount with the below change ?

fs/cifs/fs_context.c +1 -1
index eed59bc1d913..5fd0b0ddcc57 100644
@@ -1268,8 +1268,10 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
        case Opt_unix:
                if (result.negated)
                        ctx->linux_ext = 0;
-               else
+               else {
Quoted 8 lines
Quoted 6 lines
fs/cifs/fs_context.c +2 -0
+                       ctx->linux_ext = 1;
                        ctx->no_linux_ext = 1;
+               }
                break;
        case Opt_nocase:
                ctx->nocase = 1;

That stops the bleeding indeed. :)

Okay, sorry for late response.

Quoted 18 lines

I think a slightly nicer fix might be:

fs/cifs/fs_context.c +2 -1
index eed59bc1d913..424b8dc2232e 100644
@@ -1269,7 +1269,8 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
                if (result.negated)
                        ctx->linux_ext = 0;
                else
-                       ctx->no_linux_ext = 1;
+                       ctx->linux_ext = 1;
+               ctx->no_linux_ext = !ctx->linux_ext;
                break;
        case Opt_nocase:
                ctx->nocase = 1;
Quoted 2 lines

So with this patch applied I got unix permissions working after all. So I was able to do some more testing.

Okay.

Quoted 33 lines

Just a few questions (independent of idmapped mounts):

1. Are the "uid=" and "gid=" mount options ignored when "username=" is specified and "posix" is specified?

It seems that "uid=" and "gid=" have are silently ignored when "posix' is set. They are neither used to report file ownership under the cifs mountpoint nor are they relevant when determining ownership on disk?

As an example, assume I have added a user "foo" with uid 1000 to ksmbd via:

ksmbd.adduser -a foo

And I mounted a share via:

mount -t cifs -o
username=foo,password=bar,posix,uid=1234,gid=1234,forceuid,forcegid //127.0.0.1/test /mnt

i) Ownership in /mnt appears posix-correct, i.e. "uid=" and "gid=" have no effect on the reported ownership.

ii) Assume I'm logged in as the root user with uid 0. If I create
file or directory in /mnt it will be owned by user foo, i.e. uid 1000, i.e., the "uid=1234" and "gid=1234" mount option have zero effect on the ownership of the files?

2. Are the "uid=" and "gid=" options ignored for permission checking when "posix" is specified?

Quoted 7 lines
Quoted 2 lines

3. Am I correct in assuming that there is no mount option that makes chown() or chmod() have an actual effect.

That will be an answer for 1,2, 3 question. There are mount options for this.
1. modefromsid
2. idsfromsid

You can use this mount option and please check it.

Thank you! This works and finally I can hit some codepaths I wasn't able to until now.

Quoted 3 lines
Quoted 3 lines

cifs seems to have support for it but the ksmbd server doesn't seem to?

Quoted 8 lines
Quoted 6 lines

No, you need to use mount options for this as I said.
ksmbd have supported it but I found an issue related to chown and have fixed.

Could you please check the below git branch (pulled David'tree + ksmbd fixes) ?

  git clone --branch=for-christian https://github.com/namjaejeon/smb3-kernel

Thanks, I've pulled that branch.

There's two problems in this patch. I'll post and point out here quickly, if you don't mind:

commit fd7d13c387798cbc3abd68ecc07b2c868c6d96cb

Author: Namjae Jeon <[email protected]>
Date:   Sat Aug 21 14:39:43 2021 +0900

ksmbd: fix permission check issue on chown and chmod

When commanding chmod and chown on cifs&ksmbd, ksmbd allows it without file permissions check. There is code to check it in settattr_prepare. Instead of setting the inode directly, update the mode and uid/gid through notify_change.

fs/ksmbd/smb2pdu.c +5 -0
index 0131997c2177..d329ea49fa14 100644
@@ -5861,10 +5861,15 @@ int smb2_set_info(struct ksmbd_work *work)
                break;
        case SMB2_O_INFO_SECURITY:
                ksmbd_debug(SMB, "GOT SMB2_O_INFO_SECURITY\n");
+               if (ksmbd_override_fsids(work)) {
+                       rc = -ENOMEM;
+                       goto err_out;
+               }
                rc = smb2_set_info_sec(fp,
                                       le32_to_cpu(req->AdditionalInformation),
                                       req->Buffer,
                                       le32_to_cpu(req->BufferLength));
+               ksmbd_revert_fsids(work);
                break;
        default:
                rc = -EOPNOTSUPP;
fs/ksmbd/smbacl.c +7 -4
index 20455d810523..f28af33c0bd5 100644
@@ -1300,6 +1300,7 @@ int set_info_sec(struct ksmbd_conn *conn, struct ksmbd_tree_connect *tcon,
        struct smb_fattr fattr = {{0}};
        struct inode *inode = d_inode(path->dentry);
        struct user_namespace *user_ns = mnt_user_ns(path->mnt);
+       struct iattr newattrs;
 
        fattr.cf_uid = INVALID_UID;
        fattr.cf_gid = INVALID_GID;
@@ -1309,12 +1310,28 @@ int set_info_sec(struct ksmbd_conn *conn, struct ksmbd_tree_connect *tcon,
        if (rc)
                goto out;
 
-       inode->i_mode = (inode->i_mode & ~0777) | (fattr.cf_mode & 0777);
-       if (!uid_eq(fattr.cf_uid, INVALID_UID))
-               inode->i_uid = fattr.cf_uid;
-       if (!gid_eq(fattr.cf_gid, INVALID_GID))
+       newattrs.ia_valid = ATTR_CTIME;
+       if (!uid_eq(fattr.cf_uid, INVALID_UID)) {
+               newattrs.ia_valid |= ATTR_UID;
+               newattrs.ia_uid = fattr.cf_uid;
+       }
+       if (!gid_eq(fattr.cf_gid, INVALID_GID)) {
                inode->i_gid = fattr.cf_gid;
 

This needs to be removed. If setattr_prepare() fails you will still end up with inode->i_gid changed, i.e. you're changing group ownership even if you fail the subsequent setattr_prepare().

- mark_inode_dirty(inode);
+ newattrs.ia_valid |= ATTR_GID;
+ newattrs.ia_gid = fattr.cf_gid;

+       }
+       newattrs.ia_valid |= ATTR_MODE;
+       newattrs.ia_mode = (inode->i_mode & ~0777) | (fattr.cf_mode & 0777);
+       rc = setattr_prepare(user_ns, path->dentry, &newattrs);
+       if (rc)

+ goto out;
+
+ inode_lock(inode);
+ setattr_copy(user_ns, inode, &newattrs);

This needs to be removed. setattr_copy() will be called in the i_io->setattr()
method of the underlying filesystem which notify_change() will all.

Right now, with the setattr_copy() will commit the changes to the inode. If the following notify_change() call fails you end up with a change in ownership that should've failed.

Christian

maintainer
#10 lore raw

2021-08-21 20:39 GMT+09:00, Christian Brauner <[email protected]>:

Quoted 305 lines · Christian Brauner
Quoted 304 lines · Namjae Jeon
Quoted 280 lines

2021-08-19 22:01 GMT+09:00, Christian Brauner
<[email protected]>:

Quoted 208 lines · Namjae Jeon
Quoted 207 lines
Quoted 104 lines · Namjae Jeon
Quoted 11 lines
Quoted 11 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped
mounts out of the box! However, lookup is currently broken. Lookup
helpers such as lookup_one_len() call inode_permission() internally
to ensure that the caller is privileged over the inode of the base
dentry they are trying to

lookup under. So the permission checking here is currently wrong.

Quoted 10 lines
Quoted 10 lines

Linux v5.15 will gain a new lookup helper lookup_one() that does
take idmappings into account. I've added it as part of my patch
series to make btrfs support idmapped mounts. The new helper is in
linux- next as part of David's (Sterba) btrfs for-next branch as
commit c972214c133b ("namei: add

mapping aware lookup helper").

Quoted 4 lines
Quoted 4 lines

I've said it before during one of my first reviews: I would very
much recommend adding fstests to

[1].

Quoted 29 lines
Quoted 6 lines

It already seems to have very rudimentary cifs support. There is a
completely generic idmapped mount testsuite that supports idmapped
mounts.

Quoted 11 lines

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/
Cc: Steve French <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Namjae Jeon <[email protected]>
Cc: Hyunchul Lee <[email protected]>
Cc: Sergey Senozhatsky <[email protected]>
Cc: David Sterba <[email protected]>
Cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
---

Hi Christian,

Quoted 7 lines

I merged David's for-next tree into cifsd-next to test this. I did
only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least
try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

There is "How to run ksmbd" section in patch cover letter.

Quoted 13 lines

https://protect2.fireeye.com/v1/url?k=65ecaaf0-3a779239-65ed21bf-0cc47
a336fae-53bc47005a1a97a9&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Flkml.org%2Flkml%2F2021%2F8%2F5%2F54

Let me know if it doesn't work well even if you try to run it with
this step.
And We will also check whether your patch work fine.

Quoted 4 lines

One more thing, the tree for ksmbd was very hard to find. I had to
do a lot archeology to end up

at:

Quoted 16 lines
Quoted 2 lines

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section.
I think that you can use :

https://protect2.fireeye.com/v1/url?k=8af83a5d-d5630294-8af9b112-0cc47
a336fae-e471ffbdb93d05b7&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Fgithub.com%2Fnamjaejeon%2Fsmb3-kernel%2Ftree%2Fksmbd-v7- series

Quoted 5 lines

Would be appreciated if this tree could be reflected in MAINTAINERS
or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Quoted 8 lines

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Thank your for the pointers.

Ok, so I've been taking the time to look into cifs and ksmbd today. My
mental model was wrong. There
are two things to consider here:

Quoted 4 lines

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.

Right.

Quoted 17 lines

Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no

So /opt can be an idmapped mount and ksmb would know how to deal with
that correctly, i.e. you could
do:

Quoted 5 lines

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

Right.

Quoted 3 lines

That however is different from 2. which is cifs itself being idmappable.

Right.

Quoted 10 lines

Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand
your earlier confusion. In
another mail you pointed out that cifs reports that idmapped mounts are
not supported. That is correct
insofar as it means 2. is not supported, i.e. you can't do:

Right.

Quoted 10 lines

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What you
want is to test 1.

Right. So we have manually tested it, not xfstests.

Quoted 10 lines

So your test setup would require you to setup an idmapped mount and have
ksmbd use that which can then
be mounted by a client.

With your instructions I was at least able to get a ksmb instance running and be able to mount a
client with -t cifs. All on the same machine, i.e. my server is localhost.

Okay.

Quoted 3 lines

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Okay. And I have applied your patch to ksmbd.

Quoted 8 lines

Are unix extension supported with ksmb? Everytime I try to use "posix"
as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get
"uid=0" and "gid=0" and "noposix".
I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

With posix mount option, It should work. It worked well before but it is
strange now.

I'm not sure this is the correct fix, But could you please try to mount
with the below change ?

fs/cifs/fs_context.c +1 -1
index eed59bc1d913..5fd0b0ddcc57 100644
@@ -1268,8 +1268,10 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
        case Opt_unix:
                if (result.negated)
                        ctx->linux_ext = 0;
-               else
+               else {
Quoted 8 lines
Quoted 6 lines
fs/cifs/fs_context.c +2 -0
+                       ctx->linux_ext = 1;
                        ctx->no_linux_ext = 1;
+               }
                break;
        case Opt_nocase:
                ctx->nocase = 1;

That stops the bleeding indeed. :)

Okay, sorry for late response.

Quoted 18 lines

I think a slightly nicer fix might be:

fs/cifs/fs_context.c +2 -1
index eed59bc1d913..424b8dc2232e 100644
@@ -1269,7 +1269,8 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
                if (result.negated)
                        ctx->linux_ext = 0;
                else
-                       ctx->no_linux_ext = 1;
+                       ctx->linux_ext = 1;
+               ctx->no_linux_ext = !ctx->linux_ext;
                break;
        case Opt_nocase:
                ctx->nocase = 1;
Quoted 3 lines

So with this patch applied I got unix permissions working after all. So
I was able to do some more testing.

Okay.

Quoted 39 lines

Just a few questions (independent of idmapped mounts):

1. Are the "uid=" and "gid=" mount options ignored when "username=" is
specified and "posix" is specified?

It seems that "uid=" and "gid=" have are silently ignored when "posix' is set. They are neither used to report file ownership
under
the cifs mountpoint nor are they relevant when determining
ownership
on disk?

As an example, assume I have added a user "foo" with uid 1000 to ksmbd via:

ksmbd.adduser -a foo

And I mounted a share via:

mount -t cifs -o
username=foo,password=bar,posix,uid=1234,gid=1234,forceuid,forcegid //127.0.0.1/test /mnt

i) Ownership in /mnt appears posix-correct, i.e. "uid=" and "gid="
have
no effect on the reported ownership.

ii) Assume I'm logged in as the root user with uid 0. If I create
file or directory in /mnt it will be owned by user foo, i.e.
uid

       1000, i.e., the "uid=1234" and "gid=1234" mount option have

zero
effect on the ownership of the files?

2. Are the "uid=" and "gid=" options ignored for permission checking when "posix" is specified?

Quoted 8 lines
Quoted 2 lines

3. Am I correct in assuming that there is no mount option that makes chown() or chmod() have an actual effect.

That will be an answer for 1,2, 3 question. There are mount options for this.
1. modefromsid
2. idsfromsid

You can use this mount option and please check it.

Thank you! This works and finally I can hit some codepaths I wasn't able to until now.

Quoted 11 lines
Quoted 4 lines

cifs seems to have support for it but the ksmbd server doesn't
seem
to?

No, you need to use mount options for this as I said.
ksmbd have supported it but I found an issue related to chown and have fixed.

Could you please check the below git branch (pulled David'tree + ksmbd fixes) ?

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

git clone --branch=for-christian
https://github.com/namjaejeon/smb3-kernel

Thanks, I've pulled that branch.

There's two problems in this patch. I'll post and point out here quickly, if you don't mind:

Thanks for your reiview!

Quoted 67 lines

commit fd7d13c387798cbc3abd68ecc07b2c868c6d96cb

Author: Namjae Jeon <[email protected]>
Date:   Sat Aug 21 14:39:43 2021 +0900

ksmbd: fix permission check issue on chown and chmod

When commanding chmod and chown on cifs&ksmbd, ksmbd allows it without
file
permissions check. There is code to check it in settattr_prepare. Instead of setting the inode directly, update the mode and uid/gid through notify_change.

fs/ksmbd/smb2pdu.c +5 -0
index 0131997c2177..d329ea49fa14 100644
@@ -5861,10 +5861,15 @@ int smb2_set_info(struct ksmbd_work *work)
                break;
        case SMB2_O_INFO_SECURITY:
                ksmbd_debug(SMB, "GOT SMB2_O_INFO_SECURITY\n");
+               if (ksmbd_override_fsids(work)) {
+                       rc = -ENOMEM;
+                       goto err_out;
+               }
                rc = smb2_set_info_sec(fp,
 
 le32_to_cpu(req->AdditionalInformation),
                                       req->Buffer,
                                       le32_to_cpu(req->BufferLength));
+               ksmbd_revert_fsids(work);
                break;
        default:
                rc = -EOPNOTSUPP;
fs/ksmbd/smbacl.c +1 -0
index 20455d810523..f28af33c0bd5 100644
@@ -1300,6 +1300,7 @@ int set_info_sec(struct ksmbd_conn *conn, struct
 ksmbd_tree_connect *tcon,
        struct smb_fattr fattr = {{0}};
        struct inode *inode = d_inode(path->dentry);
        struct user_namespace *user_ns = mnt_user_ns(path->mnt);
+       struct iattr newattrs;
 
        fattr.cf_uid = INVALID_UID;
        fattr.cf_gid = INVALID_GID;
fs/ksmbd/smbacl.c +6 -4
@@ -1309,12 +1310,28 @@ int set_info_sec(struct ksmbd_conn *conn, struct
 ksmbd_tree_connect *tcon,
        if (rc)
                goto out;
 
-       inode->i_mode = (inode->i_mode & ~0777) | (fattr.cf_mode & 0777);
-       if (!uid_eq(fattr.cf_uid, INVALID_UID))
-               inode->i_uid = fattr.cf_uid;
-       if (!gid_eq(fattr.cf_gid, INVALID_GID))
+       newattrs.ia_valid = ATTR_CTIME;
+       if (!uid_eq(fattr.cf_uid, INVALID_UID)) {
+               newattrs.ia_valid |= ATTR_UID;
+               newattrs.ia_uid = fattr.cf_uid;
+       }
+       if (!gid_eq(fattr.cf_gid, INVALID_GID)) {
                inode->i_gid = fattr.cf_gid;
 
Quoted 3 lines

This needs to be removed. If setattr_prepare() fails you will still end up with inode->i_gid changed, i.e. you're changing group ownership even if you fail the subsequent setattr_prepare().

Oops, My mistake, I am missing remove this line.

Quoted 15 lines
fs/ksmbd/smbacl.c +5 -1
-       mark_inode_dirty(inode);
+               newattrs.ia_valid |= ATTR_GID;
+               newattrs.ia_gid = fattr.cf_gid;
+       }
+       newattrs.ia_valid |= ATTR_MODE;
+       newattrs.ia_mode = (inode->i_mode & ~0777) | (fattr.cf_mode &

0777);

fs/ksmbd/smbacl.c +6 -0
+       rc = setattr_prepare(user_ns, path->dentry, &newattrs);
+       if (rc)
+               goto out;
+
+       inode_lock(inode);
+       setattr_copy(user_ns, inode, &newattrs);
Quoted 7 lines

This needs to be removed. setattr_copy() will be called in the i_io->setattr()
method of the underlying filesystem which notify_change() will all.

Right now, with the setattr_copy() will commit the changes to the inode. If the following notify_change() call fails you end up with a change in ownership that should've failed.

Okay, I will remove it.

Thanks!

Quoted 3 lines

Christian

#11 lore raw
Quoted 381 lines · Namjae Jeon

2021-08-21 20:39 GMT+09:00, Christian Brauner <[email protected]>:

Quoted 305 lines · Christian Brauner
Quoted 304 lines · Namjae Jeon
Quoted 280 lines

2021-08-19 22:01 GMT+09:00, Christian Brauner
<[email protected]>:

Quoted 208 lines · Namjae Jeon
Quoted 207 lines
Quoted 104 lines · Namjae Jeon
Quoted 11 lines
Quoted 11 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped
mounts out of the box! However, lookup is currently broken. Lookup
helpers such as lookup_one_len() call inode_permission() internally
to ensure that the caller is privileged over the inode of the base
dentry they are trying to

lookup under. So the permission checking here is currently wrong.

Quoted 10 lines
Quoted 10 lines

Linux v5.15 will gain a new lookup helper lookup_one() that does
take idmappings into account. I've added it as part of my patch
series to make btrfs support idmapped mounts. The new helper is in
linux- next as part of David's (Sterba) btrfs for-next branch as
commit c972214c133b ("namei: add

mapping aware lookup helper").

Quoted 4 lines
Quoted 4 lines

I've said it before during one of my first reviews: I would very
much recommend adding fstests to

[1].

Quoted 29 lines
Quoted 6 lines

It already seems to have very rudimentary cifs support. There is a
completely generic idmapped mount testsuite that supports idmapped
mounts.

Quoted 11 lines

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/
Cc: Steve French <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Namjae Jeon <[email protected]>
Cc: Hyunchul Lee <[email protected]>
Cc: Sergey Senozhatsky <[email protected]>
Cc: David Sterba <[email protected]>
Cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
---

Hi Christian,

Quoted 7 lines

I merged David's for-next tree into cifsd-next to test this. I did
only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least
try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

There is "How to run ksmbd" section in patch cover letter.

Quoted 13 lines

https://protect2.fireeye.com/v1/url?k=65ecaaf0-3a779239-65ed21bf-0cc47
a336fae-53bc47005a1a97a9&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Flkml.org%2Flkml%2F2021%2F8%2F5%2F54

Let me know if it doesn't work well even if you try to run it with
this step.
And We will also check whether your patch work fine.

Quoted 4 lines

One more thing, the tree for ksmbd was very hard to find. I had to
do a lot archeology to end up

at:

Quoted 16 lines
Quoted 2 lines

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section.
I think that you can use :

https://protect2.fireeye.com/v1/url?k=8af83a5d-d5630294-8af9b112-0cc47
a336fae-e471ffbdb93d05b7&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Fgithub.com%2Fnamjaejeon%2Fsmb3-kernel%2Ftree%2Fksmbd-v7- series

Quoted 5 lines

Would be appreciated if this tree could be reflected in MAINTAINERS
or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Quoted 8 lines

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Thank your for the pointers.

Ok, so I've been taking the time to look into cifs and ksmbd today. My
mental model was wrong. There
are two things to consider here:

Quoted 4 lines

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.

Right.

Quoted 17 lines

Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no

So /opt can be an idmapped mount and ksmb would know how to deal with
that correctly, i.e. you could
do:

Quoted 5 lines

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

Right.

Quoted 3 lines

That however is different from 2. which is cifs itself being idmappable.

Right.

Quoted 10 lines

Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand
your earlier confusion. In
another mail you pointed out that cifs reports that idmapped mounts are
not supported. That is correct
insofar as it means 2. is not supported, i.e. you can't do:

Right.

Quoted 10 lines

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What you
want is to test 1.

Right. So we have manually tested it, not xfstests.

Quoted 10 lines

So your test setup would require you to setup an idmapped mount and have
ksmbd use that which can then
be mounted by a client.

With your instructions I was at least able to get a ksmb instance running and be able to mount a
client with -t cifs. All on the same machine, i.e. my server is localhost.

Okay.

Quoted 3 lines

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Okay. And I have applied your patch to ksmbd.

Quoted 8 lines

Are unix extension supported with ksmb? Everytime I try to use "posix"
as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get
"uid=0" and "gid=0" and "noposix".
I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

With posix mount option, It should work. It worked well before but it is
strange now.

I'm not sure this is the correct fix, But could you please try to mount
with the below change ?

fs/cifs/fs_context.c +1 -1
index eed59bc1d913..5fd0b0ddcc57 100644
@@ -1268,8 +1268,10 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
        case Opt_unix:
                if (result.negated)
                        ctx->linux_ext = 0;
-               else
+               else {
Quoted 8 lines
Quoted 6 lines
fs/cifs/fs_context.c +2 -0
+                       ctx->linux_ext = 1;
                        ctx->no_linux_ext = 1;
+               }
                break;
        case Opt_nocase:
                ctx->nocase = 1;

That stops the bleeding indeed. :)

Okay, sorry for late response.

Quoted 18 lines

I think a slightly nicer fix might be:

fs/cifs/fs_context.c +2 -1
index eed59bc1d913..424b8dc2232e 100644
@@ -1269,7 +1269,8 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
                if (result.negated)
                        ctx->linux_ext = 0;
                else
-                       ctx->no_linux_ext = 1;
+                       ctx->linux_ext = 1;
+               ctx->no_linux_ext = !ctx->linux_ext;
                break;
        case Opt_nocase:
                ctx->nocase = 1;
Quoted 3 lines

So with this patch applied I got unix permissions working after all. So
I was able to do some more testing.

Okay.

Quoted 39 lines

Just a few questions (independent of idmapped mounts):

1. Are the "uid=" and "gid=" mount options ignored when "username=" is
specified and "posix" is specified?

It seems that "uid=" and "gid=" have are silently ignored when "posix' is set. They are neither used to report file ownership
under
the cifs mountpoint nor are they relevant when determining
ownership
on disk?

As an example, assume I have added a user "foo" with uid 1000 to ksmbd via:

ksmbd.adduser -a foo

And I mounted a share via:

mount -t cifs -o
username=foo,password=bar,posix,uid=1234,gid=1234,forceuid,forcegid //127.0.0.1/test /mnt

i) Ownership in /mnt appears posix-correct, i.e. "uid=" and "gid="
have
no effect on the reported ownership.

ii) Assume I'm logged in as the root user with uid 0. If I create
file or directory in /mnt it will be owned by user foo, i.e.
uid

       1000, i.e., the "uid=1234" and "gid=1234" mount option have

zero
effect on the ownership of the files?

2. Are the "uid=" and "gid=" options ignored for permission checking when "posix" is specified?

Quoted 8 lines
Quoted 2 lines

3. Am I correct in assuming that there is no mount option that makes chown() or chmod() have an actual effect.

That will be an answer for 1,2, 3 question. There are mount options for this.
1. modefromsid
2. idsfromsid

You can use this mount option and please check it.

Thank you! This works and finally I can hit some codepaths I wasn't able to until now.

Quoted 11 lines
Quoted 4 lines

cifs seems to have support for it but the ksmbd server doesn't
seem
to?

No, you need to use mount options for this as I said.
ksmbd have supported it but I found an issue related to chown and have fixed.

Could you please check the below git branch (pulled David'tree + ksmbd fixes) ?

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

git clone --branch=for-christian
https://github.com/namjaejeon/smb3-kernel

Thanks, I've pulled that branch.

There's two problems in this patch. I'll post and point out here quickly, if you don't mind:

Thanks for your reiview!

Quoted 67 lines

commit fd7d13c387798cbc3abd68ecc07b2c868c6d96cb

Author: Namjae Jeon <[email protected]>
Date:   Sat Aug 21 14:39:43 2021 +0900

ksmbd: fix permission check issue on chown and chmod

When commanding chmod and chown on cifs&ksmbd, ksmbd allows it without
file
permissions check. There is code to check it in settattr_prepare. Instead of setting the inode directly, update the mode and uid/gid through notify_change.

fs/ksmbd/smb2pdu.c +5 -0
index 0131997c2177..d329ea49fa14 100644
@@ -5861,10 +5861,15 @@ int smb2_set_info(struct ksmbd_work *work)
                break;
        case SMB2_O_INFO_SECURITY:
                ksmbd_debug(SMB, "GOT SMB2_O_INFO_SECURITY\n");
+               if (ksmbd_override_fsids(work)) {
+                       rc = -ENOMEM;
+                       goto err_out;
+               }
                rc = smb2_set_info_sec(fp,
 
 le32_to_cpu(req->AdditionalInformation),
                                       req->Buffer,
                                       le32_to_cpu(req->BufferLength));
+               ksmbd_revert_fsids(work);
                break;
        default:
                rc = -EOPNOTSUPP;
fs/ksmbd/smbacl.c +1 -0
index 20455d810523..f28af33c0bd5 100644
@@ -1300,6 +1300,7 @@ int set_info_sec(struct ksmbd_conn *conn, struct
 ksmbd_tree_connect *tcon,
        struct smb_fattr fattr = {{0}};
        struct inode *inode = d_inode(path->dentry);
        struct user_namespace *user_ns = mnt_user_ns(path->mnt);
+       struct iattr newattrs;
 
        fattr.cf_uid = INVALID_UID;
        fattr.cf_gid = INVALID_GID;
fs/ksmbd/smbacl.c +6 -4
@@ -1309,12 +1310,28 @@ int set_info_sec(struct ksmbd_conn *conn, struct
 ksmbd_tree_connect *tcon,
        if (rc)
                goto out;
 
-       inode->i_mode = (inode->i_mode & ~0777) | (fattr.cf_mode & 0777);
-       if (!uid_eq(fattr.cf_uid, INVALID_UID))
-               inode->i_uid = fattr.cf_uid;
-       if (!gid_eq(fattr.cf_gid, INVALID_GID))
+       newattrs.ia_valid = ATTR_CTIME;
+       if (!uid_eq(fattr.cf_uid, INVALID_UID)) {
+               newattrs.ia_valid |= ATTR_UID;
+               newattrs.ia_uid = fattr.cf_uid;
+       }
+       if (!gid_eq(fattr.cf_gid, INVALID_GID)) {
                inode->i_gid = fattr.cf_gid;
 
Quoted 13 lines
Quoted 3 lines

This needs to be removed. If setattr_prepare() fails you will still end up with inode->i_gid changed, i.e. you're changing group ownership even if you fail the subsequent setattr_prepare().

Oops, My mistake, I am missing remove this line.

Quoted 9 lines
fs/ksmbd/smbacl.c +5 -1
-       mark_inode_dirty(inode);
+               newattrs.ia_valid |= ATTR_GID;
+               newattrs.ia_gid = fattr.cf_gid;
+       }
+       newattrs.ia_valid |= ATTR_MODE;
+       newattrs.ia_mode = (inode->i_mode & ~0777) | (fattr.cf_mode &

0777);
+ rc = setattr_prepare(user_ns, path->dentry, &newattrs);

Sorry, I also forgot to mention that you need to remove this too. All of this is handled in the underlying filesystem's ->setattr. So this just needs a notify_change().

maintainer
#12 lore raw

2021-08-21 23:10 GMT+09:00, Christian Brauner <[email protected]>:

Quoted 431 lines · Namjae Jeon
Quoted 416 lines

2021-08-21 20:39 GMT+09:00, Christian Brauner
<[email protected]>:

Quoted 336 lines · Christian Brauner
Quoted 335 lines · Namjae Jeon
Quoted 309 lines

2021-08-19 22:01 GMT+09:00, Christian Brauner
<[email protected]>:

Quoted 226 lines · Namjae Jeon
Quoted 225 lines
Quoted 115 lines · Namjae Jeon
Quoted 12 lines
Quoted 12 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped
mounts out of the box! However, lookup is currently broken. Lookup
helpers such as lookup_one_len() call inode_permission() internally
to ensure that the caller is privileged over the inode of the
base
dentry they are trying to

lookup under. So the permission checking here is currently wrong.

Quoted 12 lines
Quoted 12 lines

Linux v5.15 will gain a new lookup helper lookup_one() that does
take idmappings into account. I've added it as part of my patch
series to make btrfs support idmapped mounts. The new helper is
in
linux- next as part of David's (Sterba) btrfs for-next branch
as
commit c972214c133b ("namei: add

mapping aware lookup helper").

Quoted 4 lines
Quoted 4 lines

I've said it before during one of my first reviews: I would very
much recommend adding fstests to

[1].

Quoted 33 lines
Quoted 7 lines

It already seems to have very rudimentary cifs support. There
is a
completely generic idmapped mount testsuite that supports idmapped
mounts.

Quoted 12 lines

[1]:
https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/
Cc: Steve French <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Namjae Jeon <[email protected]>
Cc: Hyunchul Lee <[email protected]>
Cc: Sergey Senozhatsky <[email protected]>
Cc: David Sterba <[email protected]>
Cc: [email protected]
Signed-off-by: Christian Brauner
<[email protected]>
---

Hi Christian,

Quoted 9 lines

I merged David's for-next tree into cifsd-next to test this. I
did
only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least
try to cut some time out of my week to do more reviews. (I'd
especially like to see acl behavior with ksmbd.)

There is "How to run ksmbd" section in patch cover letter.

Quoted 14 lines

https://protect2.fireeye.com/v1/url?k=65ecaaf0-3a779239-65ed21bf-0cc47
a336fae-53bc47005a1a97a9&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Flkml.org%2Flkml%2F2021%2F8%2F5%2F54

Let me know if it doesn't work well even if you try to run it with
this step.
And We will also check whether your patch work fine.

Quoted 5 lines

One more thing, the tree for ksmbd was very hard to find. I had
to
do a lot archeology to end up

at:

Quoted 20 lines
Quoted 2 lines

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section.
I think that you can use :

https://protect2.fireeye.com/v1/url?k=8af83a5d-d5630294-8af9b112-0cc47
a336fae-e471ffbdb93d05b7&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Fgithub.com%2Fnamjaejeon%2Fsmb3-kernel%2Ftree%2Fksmbd-v7- series

Quoted 5 lines

Would be appreciated if this tree could be reflected in MAINTAINERS
or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 1 line
Quoted 5 lines
Quoted 2 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after
checking it.

Thank your for the pointers.

Ok, so I've been taking the time to look into cifs and ksmbd today.
My
mental model was wrong. There
are two things to consider here:

Quoted 4 lines

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.

Right.

Quoted 18 lines

Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no

So /opt can be an idmapped mount and ksmb would know how to deal
with
that correctly, i.e. you could
do:

Quoted 5 lines

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

Right.

Quoted 3 lines

That however is different from 2. which is cifs itself being idmappable.

Right.

Quoted 11 lines

Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand
your earlier confusion. In
another mail you pointed out that cifs reports that idmapped mounts
are
not supported. That is correct
insofar as it means 2. is not supported, i.e. you can't do:

Right.

Quoted 11 lines

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What
you
want is to test 1.

Right. So we have manually tested it, not xfstests.

Quoted 7 lines

So your test setup would require you to setup an idmapped mount and
have
ksmbd use that which can then
be mounted by a client.

Quoted 5 lines

With your instructions I was at least able to get a ksmb instance
running and be able to mount a
client with -t cifs. All on the same machine, i.e. my server is localhost.

Okay.

Quoted 3 lines

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Okay. And I have applied your patch to ksmbd.

Quoted 8 lines

Are unix extension supported with ksmb? Everytime I try to use "posix"
as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get
"uid=0" and "gid=0" and "noposix".
I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

With posix mount option, It should work. It worked well before but it
is
strange now.

I'm not sure this is the correct fix, But could you please try to mount
with the below change ?

fs/cifs/fs_context.c +0 -0
index eed59bc1d913..5fd0b0ddcc57 100644
@@ -1268,8 +1268,10 @@ static int

smb3_fs_context_parse_param(struct
fs_context *fc,

        case Opt_unix:
                if (result.negated)
                        ctx->linux_ext = 0;
fs/cifs/fs_context.c +1 -1
-               else
+               else {
Quoted 8 lines

fs/cifs/fs_context.c

Quoted 6 lines
fs/cifs/fs_context.c +2 -0
+                       ctx->linux_ext = 1;
                        ctx->no_linux_ext = 1;
+               }
                break;
        case Opt_nocase:
                ctx->nocase = 1;

That stops the bleeding indeed. :)

Okay, sorry for late response.

Quoted 19 lines

I think a slightly nicer fix might be:

fs/cifs/fs_context.c +0 -0
index eed59bc1d913..424b8dc2232e 100644
@@ -1269,7 +1269,8 @@ static int

smb3_fs_context_parse_param(struct
fs_context *fc,

                if (result.negated)
                        ctx->linux_ext = 0;
                else
fs/cifs/fs_context.c +2 -1
-                       ctx->no_linux_ext = 1;
+                       ctx->linux_ext = 1;
+               ctx->no_linux_ext = !ctx->linux_ext;
                break;
        case Opt_nocase:
                ctx->nocase = 1;
Quoted 4 lines

So with this patch applied I got unix permissions working after all.
So
I was able to do some more testing.

Okay.

Quoted 41 lines

Just a few questions (independent of idmapped mounts):

1. Are the "uid=" and "gid=" mount options ignored when "username="
is
specified and "posix" is specified?

It seems that "uid=" and "gid=" have are silently ignored when "posix' is set. They are neither used to report file ownership
under
the cifs mountpoint nor are they relevant when determining
ownership
on disk?

As an example, assume I have added a user "foo" with uid 1000
to
ksmbd via:

ksmbd.adduser -a foo

And I mounted a share via:

mount -t cifs -o
username=foo,password=bar,posix,uid=1234,gid=1234,forceuid,forcegid //127.0.0.1/test /mnt

i) Ownership in /mnt appears posix-correct, i.e. "uid=" and
"gid="
have
no effect on the reported ownership.

ii) Assume I'm logged in as the root user with uid 0. If I
create
file or directory in /mnt it will be owned by user foo,
i.e.
uid

       1000, i.e., the "uid=1234" and "gid=1234" mount option have

zero
effect on the ownership of the files?

Quoted 7 lines

2. Are the "uid=" and "gid=" options ignored for permission checking
when "posix" is specified?

3. Am I correct in assuming that there is no mount option that makes
chown() or chmod() have an actual effect.

Quoted 7 lines

That will be an answer for 1,2, 3 question. There are mount options for
this.
1. modefromsid
2. idsfromsid

You can use this mount option and please check it.

Thank you! This works and finally I can hit some codepaths I wasn't able
to until now.

Quoted 13 lines
Quoted 4 lines

cifs seems to have support for it but the ksmbd server doesn't
seem
to?

No, you need to use mount options for this as I said. ksmbd have supported it but I found an issue related to chown and have
fixed.

Could you please check the below git branch (pulled David'tree + ksmbd
fixes) ?

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

git clone --branch=for-christian
https://github.com/namjaejeon/smb3-kernel

Thanks, I've pulled that branch.

There's two problems in this patch. I'll post and point out here quickly, if you don't mind:

Thanks for your reiview!

Quoted 70 lines

commit fd7d13c387798cbc3abd68ecc07b2c868c6d96cb

Author: Namjae Jeon <[email protected]>
Date:   Sat Aug 21 14:39:43 2021 +0900

ksmbd: fix permission check issue on chown and chmod

When commanding chmod and chown on cifs&ksmbd, ksmbd allows it
without
file
permissions check. There is code to check it in settattr_prepare. Instead of setting the inode directly, update the mode and uid/gid through notify_change.

fs/ksmbd/smb2pdu.c +5 -0
index 0131997c2177..d329ea49fa14 100644
@@ -5861,10 +5861,15 @@ int smb2_set_info(struct ksmbd_work *work)
                break;
        case SMB2_O_INFO_SECURITY:
                ksmbd_debug(SMB, "GOT SMB2_O_INFO_SECURITY\n");
+               if (ksmbd_override_fsids(work)) {
+                       rc = -ENOMEM;
+                       goto err_out;
+               }
                rc = smb2_set_info_sec(fp,
 
 le32_to_cpu(req->AdditionalInformation),
                                       req->Buffer,
                                       le32_to_cpu(req->BufferLength));
+               ksmbd_revert_fsids(work);
                break;
        default:
                rc = -EOPNOTSUPP;
fs/ksmbd/smbacl.c +1 -0
index 20455d810523..f28af33c0bd5 100644
@@ -1300,6 +1300,7 @@ int set_info_sec(struct ksmbd_conn *conn, struct
 ksmbd_tree_connect *tcon,
        struct smb_fattr fattr = {{0}};
        struct inode *inode = d_inode(path->dentry);
        struct user_namespace *user_ns = mnt_user_ns(path->mnt);
+       struct iattr newattrs;
 
        fattr.cf_uid = INVALID_UID;
        fattr.cf_gid = INVALID_GID;
fs/ksmbd/smbacl.c +0 -0
@@ -1309,12 +1310,28 @@ int set_info_sec(struct ksmbd_conn *conn,

struct
ksmbd_tree_connect *tcon,

        if (rc)
                goto out;

- inode->i_mode = (inode->i_mode & ~0777) | (fattr.cf_mode &
0777);

fs/ksmbd/smbacl.c +6 -3
-       if (!uid_eq(fattr.cf_uid, INVALID_UID))
-               inode->i_uid = fattr.cf_uid;
-       if (!gid_eq(fattr.cf_gid, INVALID_GID))
+       newattrs.ia_valid = ATTR_CTIME;
+       if (!uid_eq(fattr.cf_uid, INVALID_UID)) {
+               newattrs.ia_valid |= ATTR_UID;
+               newattrs.ia_uid = fattr.cf_uid;
+       }
+       if (!gid_eq(fattr.cf_gid, INVALID_GID)) {
                inode->i_gid = fattr.cf_gid;
Quoted 13 lines
Quoted 3 lines

This needs to be removed. If setattr_prepare() fails you will still end up with inode->i_gid changed, i.e. you're changing group ownership even if you fail the subsequent setattr_prepare().

Oops, My mistake, I am missing remove this line.

Quoted 9 lines
fs/ksmbd/smbacl.c +5 -1
-       mark_inode_dirty(inode);
+               newattrs.ia_valid |= ATTR_GID;
+               newattrs.ia_gid = fattr.cf_gid;
+       }
+       newattrs.ia_valid |= ATTR_MODE;
+       newattrs.ia_mode = (inode->i_mode & ~0777) | (fattr.cf_mode &

0777);
+ rc = setattr_prepare(user_ns, path->dentry, &newattrs);

Quoted 3 lines

Sorry, I also forgot to mention that you need to remove this too. All of this is handled in the underlying filesystem's ->setattr. So this just needs a notify_change().

Done:)

Thanks for your review!

Quoted 1 line
maintainer
#13 lore raw
In reply to Christian Brauner

2021-08-21 20:11 GMT+09:00, Christian Brauner <[email protected]>:

Quoted 288 lines · Namjae Jeon
Quoted 265 lines

2021-08-19 22:01 GMT+09:00, Christian Brauner
<[email protected]>:

Quoted 199 lines · Namjae Jeon
Quoted 198 lines
Quoted 97 lines · Namjae Jeon
Quoted 11 lines
Quoted 11 lines

From: Christian Brauner <[email protected]>

It's great that the new in-kernel ksmbd server will support idmapped
mounts out of the box! However, lookup is currently broken. Lookup
helpers such as lookup_one_len() call inode_permission() internally
to ensure that the caller is privileged over the inode of the base
dentry they are trying to

lookup under. So the permission checking here is currently wrong.

Quoted 7 lines
Quoted 7 lines

Linux v5.15 will gain a new lookup helper lookup_one() that does take idmappings into account. I've added it as part of my patch series to make btrfs support idmapped mounts. The new helper is in
linux- next as part of David's (Sterba) btrfs for-next branch as commit c972214c133b ("namei: add

mapping aware lookup helper").

Quoted 3 lines
Quoted 3 lines

I've said it before during one of my first reviews: I would very much recommend adding fstests to

[1].

Quoted 28 lines
Quoted 15 lines

It already seems to have very rudimentary cifs support. There is a
completely generic idmapped mount testsuite that supports idmapped
mounts.

[1]: https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/

Cc Steve French <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Namjae Jeon <[email protected]>
Cc Hyunchul Lee <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc David Sterba <[email protected]>
Quoted 1 line

---

Hi Christian,

Quoted 7 lines

I merged David's for-next tree into cifsd-next to test this. I did
only compile test this. If someone gives me a clear set of instructions how to test ksmbd on my local machine I can at least
try to cut some time out of my week to do more reviews. (I'd especially like to see acl behavior with ksmbd.)

There is "How to run ksmbd" section in patch cover letter.

Quoted 12 lines

https://protect2.fireeye.com/v1/url?k=65ecaaf0-3a779239-65ed21bf-0cc47
a336fae-53bc47005a1a97a9&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Flkml.org%2Flkml%2F2021%2F8%2F5%2F54

Let me know if it doesn't work well even if you try to run it with this step.
And We will also check whether your patch work fine.

Quoted 4 lines

One more thing, the tree for ksmbd was very hard to find. I had to
do a lot archeology to end up

at:

Quoted 16 lines
Quoted 2 lines

git://git.samba.org/ksmbd.git

This is also in the patch cover letter. See "Mailing list and repositories" section.
I think that you can use :

https://protect2.fireeye.com/v1/url?k=8af83a5d-d5630294-8af9b112-0cc47
a336fae-e471ffbdb93d05b7&q=1&e=e44c9f9f-d7ae-4768-8cc2-8f02d748fc6e&u= https%3A%2F%2Fgithub.com%2Fnamjaejeon%2Fsmb3-kernel%2Ftree%2Fksmbd-v7- series

Quoted 5 lines

Would be appreciated if this tree could be reflected in MAINTAINERS
or somewhere else. The github repos with the broken out patches/module aren't really that helpful.

Quoted 7 lines

Okay, I will add git address of ksmbd in MAINTAINERS on next spin.

Quoted 3 lines

Thanks!
Christian

Really thanks for your review and I will apply this patch after checking it.

Thank your for the pointers.

Ok, so I've been taking the time to look into cifs and ksmbd today. My
mental model was wrong. There
are two things to consider here:

Quoted 4 lines

1. server: idmapped mounts with ksmbd
2. client: idmapped mounts with cifs

Your patchset adds support for 1.

Right.

Quoted 17 lines

Let's say I have the following ksmbd config:

root@f2-vm:~# cat /etc/ksmbd/smb.conf
[global]
        netbios name = SMBD
        server max protocol = SMB3
[test]
        path = /opt
        writeable = yes
        comment = TEST
        read only = no

So /opt can be an idmapped mount and ksmb would know how to deal with
that correctly, i.e. you could
do:

Quoted 5 lines

mount-idmapped --map-mount=b:1000:0:1 /opt /opt

ksmbd.mountd

and ksmbd would take the idmapping of /opt into account.

Right.

Quoted 3 lines

That however is different from 2. which is cifs itself being idmappable.

Right.

Quoted 9 lines

Whether or not that makes sense or is needed will need some thinking.

In any case, this has consequences for xfstests and now I understand your earlier confusion. In
another mail you pointed out that cifs reports that idmapped mounts are
not supported. That is correct
insofar as it means 2. is not supported, i.e. you can't do:

Right.

Quoted 10 lines

mount -t cifs -o username=foo,password=bar //server/files /mnt

and then

mount-idmapped --map-mount=b:1000:0:1 /mnt /mnt

but that's also not what you want in order to test for ksmbd. What you
want is to test 1.

Right. So we have manually tested it, not xfstests.

Quoted 10 lines

So your test setup would require you to setup an idmapped mount and have
ksmbd use that which can then
be mounted by a client.

With your instructions I was at least able to get a ksmb instance running and be able to mount a
client with -t cifs. All on the same machine, i.e. my server is localhost.

Okay.

Quoted 3 lines

However, I need to dig a bit into the semantics to make better assertions about what's going on.

Okay. And I have applied your patch to ksmbd.

Quoted 7 lines

Are unix extension supported with ksmb? Everytime I try to use "posix"
as a mount option with mount -t cifs -o //127.0.0.1/test /mnt I get "uid=0" and "gid=0" and "noposix".
I do set "unix extensions = yes" in both the samba and ksmbd smb.conf.

With posix mount option, It should work. It worked well before but it is
strange now.

I'm not sure this is the correct fix, But could you please try to mount
with the below change ?

fs/cifs/fs_context.c +1 -1
index eed59bc1d913..5fd0b0ddcc57 100644
@@ -1268,8 +1268,10 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
        case Opt_unix:
                if (result.negated)
                        ctx->linux_ext = 0;
-               else
+               else {
Quoted 8 lines
Quoted 6 lines
fs/cifs/fs_context.c +2 -0
+                       ctx->linux_ext = 1;
                        ctx->no_linux_ext = 1;
+               }
                break;
        case Opt_nocase:
                ctx->nocase = 1;

That stops the bleeding indeed. :)

Okay, sorry for late response.

Quoted 18 lines

I think a slightly nicer fix might be:

fs/cifs/fs_context.c +2 -1
index eed59bc1d913..424b8dc2232e 100644
@@ -1269,7 +1269,8 @@ static int smb3_fs_context_parse_param(struct
 fs_context *fc,
                if (result.negated)
                        ctx->linux_ext = 0;
                else
-                       ctx->no_linux_ext = 1;
+                       ctx->linux_ext = 1;
+               ctx->no_linux_ext = !ctx->linux_ext;
                break;
        case Opt_nocase:
                ctx->nocase = 1;
Quoted 2 lines

So with this patch applied I got unix permissions working after all. So I was able to do some more testing.

Okay.

Quoted 34 lines

Just a few questions (independent of idmapped mounts):

1. Are the "uid=" and "gid=" mount options ignored when "username=" is specified and "posix" is specified?

It seems that "uid=" and "gid=" have are silently ignored when "posix' is set. They are neither used to report file ownership under the cifs mountpoint nor are they relevant when determining ownership on disk?

As an example, assume I have added a user "foo" with uid 1000 to ksmbd via:

ksmbd.adduser -a foo

And I mounted a share via:

mount -t cifs -o
username=foo,password=bar,posix,uid=1234,gid=1234,forceuid,forcegid //127.0.0.1/test /mnt

i) Ownership in /mnt appears posix-correct, i.e. "uid=" and "gid="
have
no effect on the reported ownership.

ii) Assume I'm logged in as the root user with uid 0. If I create
file or directory in /mnt it will be owned by user foo, i.e. uid 1000, i.e., the "uid=1234" and "gid=1234" mount option have zero effect on the ownership of the files?

2. Are the "uid=" and "gid=" options ignored for permission checking when "posix" is specified?

Quoted 8 lines
Quoted 2 lines

3. Am I correct in assuming that there is no mount option that makes chown() or chmod() have an actual effect.

That will be an answer for 1,2, 3 question. There are mount options for this.
1. modefromsid
2. idsfromsid

You can use this mount option and please check it.

Thank you! This works and finally I can hit some codepaths I wasn't able to until now.

Quoted 10 lines
Quoted 3 lines

cifs seems to have support for it but the ksmbd server doesn't seem to?

No, you need to use mount options for this as I said.
ksmbd have supported it but I found an issue related to chown and have fixed.

Could you please check the below git branch (pulled David'tree + ksmbd fixes) ?

Quoted 7 lines
Quoted 2 lines

git clone --branch=for-christian
https://github.com/namjaejeon/smb3-kernel

Thanks, I've pulled that branch.

I have a some patches for ksmb that I'll be sending out next week. I just need to test the changes and verify that it all makes sense.

Okay, Thanks for your work!

Quoted 3 lines

Christian

#14 lore raw
In reply to Christian Brauner
I love your patch! Yet something to improve:

Hi Christian,

I love your patch! Yet something to improve:

[auto build test ERROR on 456af438ad490bac7ed954cb929bcec1df7f0c82]

url:    https://github.com/0day-ci/linux/commits/Christian-Brauner/ksmbd-fix-lookup-on-idmapped-mounts/20210816-200054
base:   456af438ad490bac7ed954cb929bcec1df7f0c82
config: s390-allyesconfig (attached as .config)
compiler: s390-linux-gcc (GCC) 11.2.0

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/0day-ci/linux/commit/ebbc41b9d3f7b7f6c558cb5d5f12ecb8134c9c2e
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Christian-Brauner/ksmbd-fix-lookup-on-idmapped-mounts/20210816-200054 git checkout ebbc41b9d3f7b7f6c558cb5d5f12ecb8134c9c2e
# save the attached .config to linux build tree
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.2.0 make.cross ARCH=s390

If you fix the issue, kindly add following tag as appropriate

All errors (new ones prefixed by >>):

fs/ksmbd/vfs.c: In function 'ksmbd_vfs_lock_parent':

>> fs/ksmbd/vfs.c:79:18: error: implicit declaration of function 'lookup_one'; did you mean 'lookup_bdev'? [-Werror=implicit-function-declaration]
      79 |         dentry = lookup_one(user_ns, child->d_name.name, parent,
         |                  ^~~~~~~~~~
         |                  lookup_bdev
   fs/ksmbd/vfs.c:79:16: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
      79 |         dentry = lookup_one(user_ns, child->d_name.name, parent,
         |                ^
   fs/ksmbd/vfs.c: In function 'ksmbd_vfs_mkdir':
   fs/ksmbd/vfs.c:223:19: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
     223 |                 d = lookup_one(user_ns, dentry->d_name.name, dentry->d_parent,
         |                   ^
   fs/ksmbd/vfs.c: In function '__ksmbd_vfs_rename':
   fs/ksmbd/vfs.c:753:18: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
     753 |         dst_dent = lookup_one(dst_user_ns, dst_name, dst_dent_parent,
         |                  ^
   fs/ksmbd/vfs.c: In function 'ksmbd_vfs_fp_rename':
   fs/ksmbd/vfs.c:816:19: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
     816 |         src_child = lookup_one(user_ns, src_dent->d_name.name, src_dent_parent,
         |                   ^
   cc1: some warnings being treated as errors
--
   fs/ksmbd/smb2pdu.c: In function 'process_query_dir_entries':
>> fs/ksmbd/smb2pdu.c:3542:24: error: implicit declaration of function 'lookup_one'; did you mean 'lookup_bdev'? [-Werror=implicit-function-declaration]
    3542 |                 dent = lookup_one(user_ns, priv->d_info->name,
         |                        ^~~~~~~~~~
         |                        lookup_bdev
   fs/ksmbd/smb2pdu.c:3542:22: warning: assignment to 'struct dentry *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
    3542 |                 dent = lookup_one(user_ns, priv->d_info->name,
         |                      ^

cc1: some warnings being treated as errors

vim +79 fs/ksmbd/vfs.c

    62	
    63	/**
    64	 * ksmbd_vfs_lock_parent() - lock parent dentry if it is stable
    65	 *
    66	 * the parent dentry got by dget_parent or @parent could be
    67	 * unstable, we try to lock a parent inode and lookup the
    68	 * child dentry again.
    69	 *
    70	 * the reference count of @parent isn't incremented.
    71	 */
    72	int ksmbd_vfs_lock_parent(struct user_namespace *user_ns, struct dentry *parent,
    73				  struct dentry *child)
    74	{
    75		struct dentry *dentry;
    76		int ret = 0;
    77	
    78		inode_lock_nested(d_inode(parent), I_MUTEX_PARENT);
  > 79		dentry = lookup_one(user_ns, child->d_name.name, parent,
    80				    child->d_name.len);
    81		if (IS_ERR(dentry)) {
    82			ret = PTR_ERR(dentry);
    83			goto out_err;
    84		}
    85	
    86		if (dentry != child) {
    87			ret = -ESTALE;
    88			dput(dentry);
    89			goto out_err;
    90		}
    91	
    92		dput(dentry);
    93		return 0;
    94	out_err:
    95		inode_unlock(d_inode(parent));
    96		return ret;
    97	}
    98	

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org