Directories with inodes below 17 are invisible on macOS mounts

[JustinTArthur]

Any inode below 17 can trigger unexpected behavior on Apple SMBClient running on macOS due to macOS treating the mount's inodes as Catalog Node IDs. In this case, reserved Catalog Node IDs: https://developer.apple.com/library/archive/technotes/tn/tn1150.html#CNID

This happens frequently for me on OpenZFS. The quick/terrible workaround is to create a new directory, hope for a higher inode number, and move the contents over to it.

This is similar to #487 except it's not an issue with uniqueness, just lowness. It looks like Samba's vfs_fruit works around this by optionally setting all conveyed inodes to 0 when it detects an Apple SMBClient session.

[namjaejeon]

It looks like Samba's vfs_fruit works around this by optionally setting all conveyed inodes to 0 when it detects an Apple SMBClient session.

Okay, How does Samba know when an Apple client is connected?

[JustinTArthur]

Okay, How does Samba know when an Apple client is connected?

The Apple client implementation adds the string "AAPL" to the CREATE context here:
https://github.com/apple-oss-distributions/SMBClient/blob/926c0c131ffcf91d95f1747af37a228d7611753f/kernel/netsmb/smb_smb_2.c#L279

I believe this is what Samba's vfs_fruit looks for. I don't know if the server needs to have given the client a hint it supports this protocol prior.

[namjaejeon]

Okay, Can you check it if I provide some patch to you ? It's because I don't have Mac device. Thanks!

[JustinTArthur]

Okay, Can you check it if I provide some patch to you ? It's because I don't have Mac device. Thanks!

Sure thing. You can point me at a branch or find my email address on my profile. It could be a day or two before I can get to it.

[namjaejeon]

This happens frequently for me on OpenZFS.

@JustinTArthur Can you check if this issue is coming from same inode numbers like BTRFS ?

[namjaejeon]

It looks like Samba's vfs_fruit works around this by optionally setting all conveyed inodes to 0 when it detects an Apple SMBClient session.

Can you tell me what code set inode numbers as zero in vfs_fruit.c ?

[JustinTArthur]

@JustinTArthur Can you check if this issue is coming from same inode numbers like BTRFS ?

With OpenZFS (and I'm assuming some other ZFS implementations), you can end up with a surprise low number after having allocated some bigger ones for a while, but they are unique as far as I know. So as I said, this is slightly different than #487 but a fix could easily apply to both scenarios. Older article with some background:
https://utcc.utoronto.ca/~cks/space/blog/solaris/ZFSDnodeIdsAllocation
I think the DMU object is used as the ino.

Can you tell me what code set inode numbers as zero in vfs_fruit.c ?

Sure. I'm not a Samba dev, so you will want to double-check. The search for the AAPL magic string in the CREATE looks to be here.

The zeroing logic is here in the module's override for fs_file_id_fn. It follows a check if the server is opted in to the zeroing via config.

[namjaejeon]

@JustinTArthur Okay. Thanks for the detailed information.

Could you please check the following branch ? I have applied the patch for this to #next branch.

git clone https://github.com/namjaejeon/ksmbd --branch=next

[JustinTArthur]

Looking pretty good, @namjaejeon. On the next branch, I was able to see a directory with inode number 2 and access its contents from macOS 15.5. That directory had been invisible when served from Linux 6.12.21's ksmbd.

I did a very brief sanity check with Samba smbclient and didn't see any issue.

[namjaejeon]

@cyayon Can you confirm if this issue is fixed on btrfs subvolume with the following branch ?

git clone https://github.com/namjaejeon/ksmbd --branch=next

[namjaejeon]

@JustinTArthur Really thanks for your report and help. I will apply this patch to ksmbd in linux kernel mainline also.

[cyayon]

@cyayon Can you confirm if this issue is fixed on btrfs subvolume with the following branch ?

git clone https://github.com/namjaejeon/ksmbd --branch=next

Hello,
Sorry, but I can not upgrade the related server to this "not stable" branch. I have to wait for a stable release included in the mainline kernel. Do you know when it will be included ?
thanks

[namjaejeon]

@cyayon It may take about a month for it to be included in the stable release. I will inform you if it is included in stable kernels.