Multipart objects are kept in non hex md5sum format.
This format doesn't comply with hex, so decoding
would fail invariably.
This is not necessary to validate and its not expected
error during a CopyObject operation.
Fixes#1519
Requires skipping necessary parts of dataBlocks during
decoding phase and requires us to properly skip the
entries as needed.
Thanks to Karthic for reproducing this important issue.
Fixes#1503
Since input to quick.New() is a pointer the unmarshalled value
internally already has the value, subsequent type assertions
are not needed.
Thanks to Bala for finding this behavior.
Fixes#1475
When server is run with multiple disks which uses xl interface where
order and count of disks are important, this patch saves such disks
configuration and compares in next run if there is a mismatch.
Fixes#1458
All transactions happen through this directory inside ".minio/temp".
Adding this allows us to remove any temporary files which were not
committed before.
Fixes#1462Fixes#1444
S3 API returns BucketAlreadyExists error when some another user has such bucket.
If user that creates the bucket already has it, s3 returns BucketAlreadyOwnedByYou.
As minio has only one user, it should behave accordingly.
Otherwise it causes failures in the applications that ignore creation of already existing bucket in the account, but fail when bucket name is used by someone else.
- PutObject()
- PutObjectPart()
- NewMultipartUpload()
- AbortMultipartUpload()
Implementations across both FS and XL object layer
share common implementation.
* XL/Multipart: Use json.NewDecoder to decode read stream.
* XL/Multipart: fix recursive and non-recursive listing.
* XL/Multipart: Create object part with md5sum later using RenameFile.
* XL/Multipart: ListObjectParts should list parts in order.
previously: uploadID.10.md5sum < uploadID.2.md5sum
fix : uploadID.00010.md5sum > uploadID.00002.md5sum
* XL/Multipart: Keep the size of each part in the multipart metadata file to avoid stats on the parts.
* XL/Multipart: fix listing bug which was showing size of the multipart uploaded objects as 0.
For files less than 'dataBlocks', erasure encoding would fail
with short data due to a bug in the implementation itself.
Relax the error return, even a single byte can be properly
erasure coded without issues.
Fixes#1413
