Call enterDir/leaveDir if a directory's child may be selected #3000
Conversation
This allows leaveDir to set metadata properly for directories
with descendants that are selected by an include filter, even
if the directory in question isn't selected by that filter.
For example, let's say you have the following file in your repo:
/home/user/nested/data/values.txt
...and / and /home are owned by root, but /home/user and below are owned
by user.
If you run "restic restore -i /home/user/nested/data -t /tmp/restore $SNAPSHOT" as root,
/home/user/nested/ will currently be restored with root as the owner,
even though it's supposed to be owned by user.
This change makes sure that leaveDir is called to restore metadata to
nodes like /home/user/nested - enterDir is also called under the same
circumstances for purposes of symmetry.
This fixes GH restic#1402 and (I think) GH restic#2563
|
I haven't added tests for this yet, since the issue can only manifest when restic is run as root. I'm open to to ideas on how I could write a test for this, though! One behavior I'm uncertain of is |
|
Does the issue also manifest when the owner is the same, but the mode is different? In that case, backing up a directory with mode 710 (or some other unlikely value) would trigger the bug with needing root for the test. |
|
This PR spews out a large number of errors, when using a relative include pattern: How does this PR relate to #2906 which AFAIK fixes the exact same problem? |
| // #leaveDir restores dir metadata after visiting all children | ||
| return fs.MkdirAll(target, 0700) | ||
| }, | ||
| enterDir: noop, |
There was a problem hiding this comment.
This change breaks restoring empty directories.
|
@MichaelEischer Ah, I neglected to check if the target directory was created by restic before trying to restore the metadata - I fixed this, but that PR you linked to looks to be more comprehensive |
This allows leaveDir to set metadata properly for directories
with descendants that are selected by an include filter, even
if the directory in question isn't selected by that filter.
For example, let's say you have the following file in your repo:
...and / and /home are owned by root, but /home/user and below are owned
by user.
If you run "restic restore -i /home/user/nested/data -t /tmp/restore $SNAPSHOT" as root,
/home/user/nested/ will currently be restored with root as the owner,
even though it's supposed to be owned by user.
This change makes sure that leaveDir is called to restore metadata to
nodes like /home/user/nested - enterDir is also called under the same
circumstances for purposes of symmetry.
This fixes GH #1402 and (I think) GH #2563
What does this PR change? What problem does it solve?
This PR changes directory traversal logic when restoring a snapshot, and it fixes a bug around restoration of metadata.
Was the change discussed in an issue or in the forum before?
This PR closes #1402 and closes #2563
Checklist
changelog/unreleased/that describes the changes for our users (template here)gofmton the code in all commits