back up atime and ctime

[anarcat]

after reviewing this documentation i noticed that only modification times were backed up in borg/attic (source).

is that on purpose? if so, why?

it seems it would be worthwhile to also index the access time, at the very least - and ctime can be valuable during forensics as well...

[ThomasWaldmann]

iirc, you can't restore them anyway.

[anarcat]

On 2015-10-20 13:26:08, TW wrote:

iirc, you can't restore them anyway.

atime can certainly be restored:

[994]anarcat@marcos:test$ stat foo
  Fichier : « foo »
   Taille : 0           Blocs : 0          Blocs d'E/S : 4096   fichier vide
Périphérique : fe04h/65028d     Inœud : 3707433     Liens : 1
Accès : (0644/-rw-r--r--)  UID : ( 1000/ anarcat)   GID : ( 1000/ anarcat)
 Accès : 2015-10-20 13:28:00.342574329 -0400
Modif. : 2015-10-20 13:28:00.342574329 -0400
Changt : 2015-10-20 13:28:00.342574329 -0400
  Créé : -
[995]anarcat@marcos:test$ touch -r / foo
[996]anarcat@marcos:test$ stat foo
  Fichier : « foo »
   Taille : 0           Blocs : 0          Blocs d'E/S : 4096   fichier vide
Périphérique : fe04h/65028d     Inœud : 3707433     Liens : 1
Accès : (0644/-rw-r--r--)  UID : ( 1000/ anarcat)   GID : ( 1000/ anarcat)
 Accès : 2014-05-14 19:03:45.973953996 -0400
Modif. : 2015-09-28 21:35:32.555686445 -0400
Changt : 2015-10-20 13:28:07.362656450 -0400
  Créé : -

As for ctime, it's possible to "restore" it through various hacks, but
none that seem to make sense to me. What I have read ranges from:

• changing the clock
• patching the kernel
• changing the filesystem directly

All seem a little nuts, and leaving the ctime as-is actually makes sense
on restore, as we do change the file...

However, i think it could still be useful to store it, as we could
inspect it through the FUSE mountpoint, for example... Then it can be
used to check for inconsistencies or issues...

[anarcat]

(wtf github, why isn't markdown supported by email?)

source of the ways to change the ctime: http://stackoverflow.com/questions/4537291/setting-creation-or-change-timestamps

[ThomasWaldmann]

no, we won't change the clock or patch the kernel or change fs directly.

about saving atime and ctime: yes, that seems reasonable. we just need to be careful to not change it before "reading" it.

[anarcat]

no, we won't change the clock or patch the kernel or change fs directly.

of course, i wasn't suggesting that. :)

about saving atime and ctime: yes, that seems reasonable. we just need to be careful to not change it before "reading" it.

i believe that was fixed in attic a while back, see jborg/attic#113

i guess the next step here is to add atime and ctime to the archive creation, and atime to extraction.

[ThomasWaldmann]

yup. os.utime().