btrfs subvolume invisible on macos #487

cyayon · 2024-09-17T08:51:10Z

Hi,

All BTRFS subvolume are invisible on shares from the MacOS native smb client.

All other regulars files and directories are shown and accessible as expected.

Using another MacOS smb client (not native like Filebrowser) do not have the issue, it seems related to native MacOS smb/cifs client.

Mounting with the cli mount_smbfs have the same issue (Finder and terminal path).

Same issue with MacOS 14 (Sonoma) and 15 (Sequoia).

On Linux side, the last 6.10.10 have the issue (since a long time).

Steps to reproduce :

mount smb share exported via ksmbd (btrfs filesystem) on MacOS (with Finder or cli mount_smbfs)
you should NOT see subvolume
if there is no subvolume on the shared volume, just create on (btrfs sub create ./toto_sub)
./toto_sub should be invisible on MacOS client
create another regular directory (mkdir ./toto_dir)
./toto_dir show be visible on MacOS client

Thanks.

namjaejeon · 2024-09-19T01:16:19Z

Hm,, I am not familiar with btfs sub-volume. Can you provide me the two packet dump files using wireshark on problem situation ?

wireshark dump using another MacOS smb client (not native like Filebrowser)
wireshark dump using native MacOS smb/cifs client.

cyayon · 2024-09-19T07:04:06Z

Hi,

Here are the pcap files (I used tcpdump).

Thanks.

smb_macos_fb_OK.pcap.zip
smb_macos_native_KO.pcap.zip

namjaejeon · 2024-09-20T01:03:03Z

Thanks for your help:)
I have known MacOS client send request to look-up specific files(e.g. .DS_STORE, DCIM, etc.) that is not in your share. and ksmbd return no-such-file error to MacOS client.
Can you test it with Native MacOS client and samba ? and if it is okay, Please give me wireshark dump.

Thanks!

cyayon · 2024-09-20T05:03:36Z

Hi,Sorry but I don’t have any samba install available.I would be a complicated task in my context.Is there something else that can I do to help ?Thanks On 20 Sep 2024, at 03:03, Namjae Jeon ***@***.***> wrote: Thanks for your help:) I have known MacOS client send request to look-up specific files(e.g. .DS_STORE, DCIM, etc.) that is not in your share. and ksmbd return no-such-file error to MacOS client. Can you test it with Native MacOS client and samba ? and if it is okay, Please give me wireshark dump. Thanks! —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>

namjaejeon · 2024-09-23T05:45:36Z

Sorry for late response. Okay. I am confusing when checking packet dump (smb_macos_fb_OK.pcap.zip)

./toto_sub should be invisible on MacOS client

ksmbd return datas, 'monit', 'pkg', 'vm', 'users', 'services', 'logs' filename to client. there is no toto_sub filename in response buffer of ksmbd. You can see toto_sub file with Filebrowser() ?

cyayon · 2024-09-23T12:18:23Z

Hi,
No problem.
Yes toto_sub is a btrfs subvol and should be visible in MacOS native Finder (but it is not). It is visible in Filebrowser app (which is not native to MacOS).

'monit', 'pkg', 'vm', 'users', 'services', 'logs' are btrfs subvolumes. They are visible in Filebrowser app but NOT in MacOS client (which is an issue).

Only regular directories and files are visible in MacOS native Finder and it is definitely not normal.

gatispei · 2025-03-11T10:06:15Z

same problem here. MacOS builtin smb client not showing btrfs-subvolume "directories"

gatispei · 2025-03-11T10:42:40Z

It seems that cause of the problem is same inode number for all subvolumes and the "." entry (the btrfs mount itself)

gatispei · 2025-03-11T12:22:22Z

all btrfs subvolumes have inode number 256. Here is a ksmbd hack that seems to workaround that - map 256 to 0. Clients seem to be required to support value of zero - when they see 0, they are required to do some client side inode number generation.

diff --git a/kernel/fs/ksmbd/smb2pdu.c b/kernel/fs/ksmbd/smb2pdu.c
index 7efc3a8e282..e8d04b78918 100644
--- a/kernel/fs/ksmbd/smb2pdu.c
+++ b/kernel/fs/ksmbd/smb2pdu.c
@@ -3695,7 +3695,7 @@ static int smb2_populate_readdir_entry(struct ksmbd_conn *conn, int info_level,
                if (dinfo->EaSize)
                        dinfo->ExtFileAttributes = ATTR_REPARSE_POINT_LE;
                dinfo->Reserved = 0;
-               dinfo->UniqueId = cpu_to_le64(ksmbd_kstat->kstat->ino);
+               dinfo->UniqueId = cpu_to_le64(ksmbd_kstat->kstat->ino == 256 ? 0 : ksmbd_kstat->kstat->ino);
                if (d_info->hide_dot_file && d_info->name[0] == '.')
                        dinfo->ExtFileAttributes |= ATTR_HIDDEN_LE;
                memcpy(dinfo->FileName, conv_name, conv_len);
@@ -3712,7 +3712,7 @@ static int smb2_populate_readdir_entry(struct ksmbd_conn *conn, int info_level,
                        smb2_get_reparse_tag_special_file(ksmbd_kstat->kstat->mode);
                if (fibdinfo->EaSize)
                        fibdinfo->ExtFileAttributes = ATTR_REPARSE_POINT_LE;
-               fibdinfo->UniqueId = cpu_to_le64(ksmbd_kstat->kstat->ino);
+               fibdinfo->UniqueId = cpu_to_le64(ksmbd_kstat->kstat->ino == 256 ? 0 : ksmbd_kstat->kstat->ino);
                fibdinfo->ShortNameLength = 0;
                fibdinfo->Reserved = 0;
                fibdinfo->Reserved2 = cpu_to_le16(0);

namjaejeon · 2025-03-11T23:59:41Z

@gatispei Thanks for your check and help:)

I can see this code in btrfs. Can we use ->subvol of kstat if result_mask has STATX_SUBVOL?
All subvolumes in btrfs has same ->subvol value ?

static int btrfs_getattr(struct mnt_idmap *idmap,
                         const struct path *path, struct kstat *stat,
                         u32 request_mask, unsigned int flags)
...
        stat->subvol = BTRFS_I(inode)->root->root_key.objectid;
        stat->result_mask |= STATX_SUBVOL;
..

diff --git a/fs/smb/server/smb2pdu.c b/fs/smb/server/smb2pdu.c
index c53121538990..de3af46a838f 100644
--- a/fs/smb/server/smb2pdu.c
+++ b/fs/smb/server/smb2pdu.c
@@ -3949,6 +3949,9 @@ static int smb2_populate_readdir_entry(struct ksmbd_conn *conn, int info_level,
                if (dinfo->EaSize)
                        dinfo->ExtFileAttributes = FILE_ATTRIBUTE_REPARSE_POINT_LE;
                dinfo->Reserved = 0;
+               if (ksmbd_kstat->kstat->result_mask & STATX_SUBVOL)
+                       pr_err("kstat->subvol : %llu\n", ksmbd_kstat->kstat->subvol);
+
                dinfo->UniqueId = cpu_to_le64(ksmbd_kstat->kstat->ino);
                if (d_info->hide_dot_file && d_info->name[0] == '.')
                        dinfo->ExtFileAttributes |= FILE_ATTRIBUTE_HIDDEN_LE;

gatispei · 2025-03-12T06:55:58Z

That seems useful, you would need to xor subvol with inode or combine them some other way to get semi unique file id, as files that appear in multiple subvolumes also have same inode number in respective subvolumes.

Now for some reason non-uniuque inodes in different directories did not seem to upset the macos client, but maybe it is because it reverted to local id generation for -all- files upon seeing first zero fileid.

Not the first to hit the problem:
https://lwn.net/Articles/866582/
https://lwn.net/Articles/866709/

namjaejeon · 2025-03-12T07:27:14Z

That seems useful, you would need to xor subvol with inode or combine them some other way to get semi unique file id, as files that appear in multiple subvolumes also have same inode number in respective subvolumes.

Can you suggest the patch for this ?

Not the first to hit the problem:

I have checked it. But it seems that NFS did not apply the patch mentioned in the article.

gatispei · 2025-03-12T15:32:29Z

diff --git a/fs/smb/server/smb2pdu.c b/fs/smb/server/smb2pdu.c
index c53121538990..40e5ca64a6ae 100644
--- a/fs/smb/server/smb2pdu.c
+++ b/fs/smb/server/smb2pdu.c
@@ -3950,6 +3950,8 @@ static int smb2_populate_readdir_entry(struct ksmbd_conn *conn, int info_level,
                        dinfo->ExtFileAttributes = FILE_ATTRIBUTE_REPARSE_POINT_LE;
                dinfo->Reserved = 0;
                dinfo->UniqueId = cpu_to_le64(ksmbd_kstat->kstat->ino);
+               if (ksmbd_kstat->kstat->result_mask & STATX_SUBVOL)
+                       dinfo->UniqueId ^= cpu_to_le64(rol64(ksmbd_kstat->kstat->subvol, 32));
                if (d_info->hide_dot_file && d_info->name[0] == '.')
                        dinfo->ExtFileAttributes |= FILE_ATTRIBUTE_HIDDEN_LE;
                memcpy(dinfo->FileName, conv_name, conv_len);
@@ -3967,6 +3969,8 @@ static int smb2_populate_readdir_entry(struct ksmbd_conn *conn, int info_level,
                if (fibdinfo->EaSize)
                        fibdinfo->ExtFileAttributes = FILE_ATTRIBUTE_REPARSE_POINT_LE;
                fibdinfo->UniqueId = cpu_to_le64(ksmbd_kstat->kstat->ino);
+               if (ksmbd_kstat->kstat->result_mask & STATX_SUBVOL)
+                       fibdinfo->UniqueId ^= cpu_to_le64(rol64(ksmbd_kstat->kstat->subvol, 32));
                fibdinfo->ShortNameLength = 0;
                fibdinfo->Reserved = 0;
                fibdinfo->Reserved2 = cpu_to_le16(0);
@@ -4016,6 +4020,8 @@ static int smb2_populate_readdir_entry(struct ksmbd_conn *conn, int info_level,
                }
 
                posix_info->Inode = cpu_to_le64(ksmbd_kstat->kstat->ino);
+               if (ksmbd_kstat->kstat->result_mask & STATX_SUBVOL)
+                       posix_info->Inode ^= cpu_to_le64(rol64(ksmbd_kstat->kstat->subvol, 32));
                posix_info->DosAttributes =
                        S_ISDIR(ksmbd_kstat->kstat->mode) ?
                                FILE_ATTRIBUTE_DIRECTORY_LE : FILE_ATTRIBUTE_ARCHIVE_LE;

namjaejeon · 2025-03-12T23:45:33Z

@gatispei Looks good to me:)

Can you update the patch description and add the subject using git command ?
You can refer the following example. Then I will apply this patch.
https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/fs/smb/server?h=v6.14-rc6&id=43fb7bce8866e793275c4f9f25af6a37745f3416

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

btrfs subvolume invisible on macos #487

btrfs subvolume invisible on macos #487

cyayon commented Sep 17, 2024 •

edited

Loading

namjaejeon commented Sep 19, 2024

cyayon commented Sep 19, 2024

namjaejeon commented Sep 20, 2024

cyayon commented Sep 20, 2024 via email

namjaejeon commented Sep 23, 2024 •

edited

Loading

cyayon commented Sep 23, 2024

gatispei commented Mar 11, 2025

gatispei commented Mar 11, 2025

gatispei commented Mar 11, 2025 •

edited

Loading

namjaejeon commented Mar 11, 2025

gatispei commented Mar 12, 2025

namjaejeon commented Mar 12, 2025

gatispei commented Mar 12, 2025

namjaejeon commented Mar 12, 2025

btrfs subvolume invisible on macos #487

btrfs subvolume invisible on macos #487

Comments

cyayon commented Sep 17, 2024 • edited Loading

namjaejeon commented Sep 19, 2024

cyayon commented Sep 19, 2024

namjaejeon commented Sep 20, 2024

cyayon commented Sep 20, 2024 via email

namjaejeon commented Sep 23, 2024 • edited Loading

cyayon commented Sep 23, 2024

gatispei commented Mar 11, 2025

gatispei commented Mar 11, 2025

gatispei commented Mar 11, 2025 • edited Loading

namjaejeon commented Mar 11, 2025

gatispei commented Mar 12, 2025

namjaejeon commented Mar 12, 2025

gatispei commented Mar 12, 2025

namjaejeon commented Mar 12, 2025

cyayon commented Sep 17, 2024 •

edited

Loading

namjaejeon commented Sep 23, 2024 •

edited

Loading

gatispei commented Mar 11, 2025 •

edited

Loading