file size incorrect briefly after hard link creation

[joeyh]

If a file is hard linked and statted immediately afterwards, it always appears 50 bytes larger than the actual file size. Within 1 second whatever is causing this corrects itself.

joey@darkstar:~/mnt>echo hi > a
joey@darkstar:~/mnt>ln a b ; stat b ; sleep 1; stat b
  File: b
  Size: 53        	Blocks: 8          IO Block: 4096   regular file
Device: 0,57	Inode: 13244313    Links: 2
Access: (0644/-rw-r--r--)  Uid: ( 1000/    joey)   Gid: ( 1000/    joey)
Access: 2023-03-28 13:59:26.326414785 -0400
Modify: 2023-03-28 13:59:26.326414785 -0400
Change: 2023-03-28 13:59:28.890424359 -0400
 Birth: -
  File: b
  Size: 3         	Blocks: 8          IO Block: 4096   regular file
Device: 0,57	Inode: 13244313    Links: 2
Access: (0644/-rw-r--r--)  Uid: ( 1000/    joey)   Gid: ( 1000/    joey)
Access: 2023-03-28 13:59:26.326414785 -0400
Modify: 2023-03-28 13:59:26.326414785 -0400
Change: 2023-03-28 13:59:28.890424359 -0400
 Birth: -
joey@darkstar:~/mnt>mount | grep /mnt
/home/joey/tmp/goc on /home/joey/mnt type fuse.gocryptfs (rw,nosuid,nodev,relatime,user_id=1000,group_id=1000,max_read=131072)

I am testing on Debian, kernel 6.0.0-6-amd64. Using gocryptfs version 2.3-1+b3 installed from Debian unstable.

An example of this breaking a program is at https://git-annex.branchable.com/forum/Can_you_use_git_annex_on_gocryptfs__63__/

[joeyh]

Note that according to the user on that page, version 1.8.0 does not have this problem.

[lechner]

@rfjakob Hi, I am excited about this filing! As your former Debian maintainer, I feel obligated to point out that @joeyh is (or perhaps was) Debian royalty. He (or perhaps they) made many significant and lasting contributions to free software, including Ikiwiki. You can read more here.

[rfjakob]

Thanks for the report. Looks like I broke this in the v2.0 rewrite.

Fixed now, and test added so this cannot regress again.

[neurolit]

Hi @rfjakob! Thank you very much for developing gocryptfs, it is really useful. As you fixed this bug, would you mind releasing a new version?

Sorry to bother you if your release planning follows some rules I am not aware of.

[dpifke]

As the person who reported the git-annex issue, I can confirm that git-annex now seems to work perfectly when using gocryptfs built from master. Tested just now with:

gocryptfs v2.3.1-3-g24b3978; go-fuse v2.1.1-0.20221117175120-915cf5413cde; 2023-04-24 go1.19.3 linux/amd64

(Apologies for the delayed response; I forgot to circle back and re-test.)

[rfjakob]

Released as https://github.com/rfjakob/gocryptfs/releases/tag/v2.3.2