Bug #22529: rgw: copy_object doubles leading underscore on object names. - rgw - Ceph

Actions

Copy link

Bug #22529

closed

rgw: copy_object doubles leading underscore on object names.

Added by Marcus Watts over 6 years ago. Updated about 6 years ago.

Status:

Resolved

Priority:

Normal

Assignee:

Marcus Watts

Target version:

% Done:

Source:

Tags:

Backport:

jewel, luminous

Regression:

Severity:

3 - minor

Reviewed:

Affected Versions:

ceph-qa-suite:

Pull request ID:

Crash signature (v1):

Crash signature (v2):

Description

swift and s3 have server-side copy functions. When these are applied to an object whose name starts with a leading _, the _ is doubled.

for s3, the following copies testix/bar.txt to testix/__bar62.txt . Then s3cmd emits a bogus warning.
s3cmd cp s3://testix/bar.txt s3://testix/_bar62.txt
the warning is "WARNING: Key not found s3://testix/bar.txt". This actually means it couldn't set acls on testix/_bar62.txt .

for swift, the following results in copying container1/_file1 to container2/__file1
swift copy -d /container2 container1 _file1

This behavior does not happen if if the destination is "under" some folders. Most likely destination folders starting with _ would be wonky too, but I didn't specifically test that.

This bug is present in both master & jewel.

Related issues 2 (0 open — 2 closed)

Actions

Copy link

Updated by Marcus Watts over 6 years ago

After tracing the code, I believe the problem is that when RGWRados::copy_obj_data()
creates an instance of RGWPutObjProcessor_Atomic, it is passing dest_obj.get_oid() - that then gets used to construct a new object key inside later logic. Logically, that's a "get_oid(get_oid( object_name ) - and each call to get_oid prefixes a _ when the name starts with a _. I have an experimental patch that removes the doubled _. This seems to behave correctly. I'll post that as a PR shortly.

Actions

Copy link

Updated by Marcus Watts over 6 years ago

I should update this a bit. I made this PR, https://github.com/ceph/ceph/pull/19652 which worked with my slightly outdated master, but didn't apply against the head. That's because this PR, https://github.com/ceph/ceph/pull/18662 , conflicted, with a fix to another bug that I believe also fixes this bug.

Matt Benjamin put some PR's out to address this in jewel/luminous:
https://github.com/ceph/ceph/pull/19747
jewel
https://github.com/ceph/ceph/pull/19787
luminous, merged

Actions

Copy link

Updated by Matt Benjamin over 6 years ago

Status changed from New to 17

Actions

Copy link

Updated by Matt Benjamin over 6 years ago

Assignee set to Marcus Watts

Actions

Copy link

Updated by Nathan Cutler over 6 years ago

Status changed from 17 to Pending Backport
Backport set to jewel, luminous

If I understand this correctly, this bug is fixed in master by https://github.com/ceph/ceph/pull/18662 and this PR was backported to luminous (and merged). The jewel backport uses a different fix and is still open.

Actions

Copy link