[gdal-dev] errors using IAM instance profile auth in s3

Even Rouault even.rouault at spatialys.com
Sun Nov 20 09:52:47 PST 2022


Mike,

I've ended up firing a EC2 instance and I did replicate with my private 
bucket too.

With a EC2 Ubuntu 22.04 instance, on a minimal GDAL master debug build, 
things work fine with IAM authentication, in the VM itself

But if I run inside a ubuntu:22.04 Docker image the same build (mounting 
my home directory), then I got exactly the same symptoms as you!

I finally figured out that the "curl -X PUT 
"http://169.254.169.254/latest/api/token" -H 
"X-aws-ec2-metadata-token-ttl-seconds: 10" request for IDMSv2 was 
failing inside Docker but not in the outside EC2 VM. Known (intended) 
issue according to 
https://community.grafana.com/t/imdsv2-is-not-working-from-docker/65944

The solution/workaround is to run docker with --network=host (I guess it 
could be possible to restrict it a bit to just 169.254.169.254)

In https://github.com/OSGeo/gdal/pull/6752, I've pushed a proper fix, 
that makes sure that the fallback to IDMSv1 works by clearing the GDAL 
error state (which was the reason for the weird behaviour that 
GDALOpen() only worked the second time), and emits a hopefully debug 
hint when it sees that IDMSv2 fails on timeout inside a Docker container.

With the fix, when trying inside a Docker with default networking rules:

# gdalinfo /vsis3/XXXXXXX/byte.tif --debug on
AWS: AWS_ROLE_ARN configuration option not defined
HTTP: Fetch(http://169.254.169.254/latest/api/token)
HTTP: libcurl/7.81.0 GnuTLS/3.7.3 zlib/1.2.11 brotli/1.0.9 zstd/1.4.8 
libidn2/2.3.2 libpsl/0.21.0 (+libidn2/2.3.2) libssh/0.9.6/openssl/zlib 
nghttp2/1.43.0 librtmp/2.3 OpenLDAP/2.5.13
HTTP: Fetch(http://169.254.169.254/latest/meta-data)
S3: /latest/api/token EC2 IDMSv2 request timed out, but /latest/metadata 
succeeded. Trying with IDMSv1. Try running your Docker container with 
--network=host.
HTTP: 
Fetch(http://169.254.169.254/latest/meta-data/iam/security-credentials/)
HTTP: 
Fetch(http://169.254.169.254/latest/meta-data/iam/security-credentials/ec2_s3)
AWS: Storing AIM credentials until 2022-11-20T23:38:10Z
S3: Switching to region eu-central-1
S3: Downloading 0-1796 (https://XXXXXXX.s3.amazonaws.com/byte.tif)...
S3: Got response_code=206
GDAL: GDALOpen(/vsis3/XXXXXXX/byte.tif, this=0x5599cfbce470) succeeds as 
GTiff.

It was an "interesting" bug...

Even

Le 20/11/2022 à 14:47, Michael Smith a écrit :
>
> Is there a reason why OpenEx would work but Open wouldn’t?
>
> In [1]: from osgeo import gdal
>
> In [2]: pszFilename = 
> "/vsis3/grid-dev-publiclidar/estonia/dtm/estonia_dtm_5m.tif"
>
> In [3]: hDataset = gdal.Open(pszFilename, gdal.GA_ReadOnly)
>
> In [4]: hDataset
>
> In [5]: hDataset = gdal.OpenEx(pszFilename, gdal.GA_ReadOnly)
>
> In [6]: hDataset
>
> Out[6]: <osgeo.gdal.Dataset; proxy of <Swig Object of type 
> 'GDALDatasetShadow *' at 0x7f827217c450> >
>
> In [7]: hDataset.GetGeoTransform()
>
> Out[7]: (365000.0, 5.0, 0.0, 6635000.0, 0.0, -5.0)
>
> Mike
>
> *From: *Even Rouault <even.rouault at spatialys.com>
> *Date: *Saturday, November 19, 2022 at 10:08 AM
> *To: *<michael.smith.erdc at gmail.com>
> *Cc: *gdal-dev <gdal-dev at lists.osgeo.org>
> *Subject: *Re: [gdal-dev] errors using IAM instance profile auth in s3
>
> Le 19/11/2022 à 16:00, michael.smith.erdc at gmail.com a écrit :
>
>     Correct, not a public bucket, which is why the IAM credentials are
>     needed. If I set them manually, it all works fine.
>
> That's super weird if the result of a range request changes depending 
> on how credentials have been set... Perhaps enable CPL_CURL_VERBOSE=ON 
> env variable and diff the logs ?
>
> You could also try the gdal_cp.py sample script at 
> https://github.com/OSGeo/gdal/blob/master/swig/python/gdal-utils/osgeo_utils/samples/gdal_cp.py 
> , which is a cp-like utility working with GDAL virtual file systems, 
> with the 2 authentication methods
>
> python gdal_cp.py 
> /vsis3/grid-dev-publiclidar/estonia/dtm/estonia_dtm_5m.tif out.tif
>
> (you can interrupt it with ctrl-c after a few seconds. that will be 
> enough to get the first bytes)
>
> you might need to run an hexadecimal editor to inspect a bit the content.
>
>     [ u02]$ export AWS_ACCESS_KEY_ID=xxxxx
>
>     Yes, a 206 response code means success here as we are requesting
>     only bytes 0-16383. So maybe the file is not a valid TIFF ?
>
>     ( "grid-dev-publiclidar" must not be so public I guess, because
>     when trying with my credentials, I get a Access Denied)
>
>     Le 19/11/2022 à 15:40, michael.smith.erdc at gmail.com a écrit :
>
>         I’m seeing that it’s getting a 206 response code, so wouldn’t
>         that indicate auth is working?
>
>          gdalinfo
>         /vsis3/grid-dev-publiclidar/estonia/dtm/estonia_dtm_5m.tif
>
>         HTTP: Fetch(http://169.254.169.254/latest/api/token)
>
>         HTTP: libcurl/7.86.0 OpenSSL/3.0.7 zlib/1.2.13 libssh2/1.10.0
>         nghttp2/1.47.0
>
>         HTTP: These HTTP headers were set:
>         X-aws-ec2-metadata-token-ttl-seconds: 10
>
>         HTTP:
>         Fetch(http://169.254.169.254/latest/meta-data/iam/security-credentials/)
>
>         HTTP:
>         Fetch(http://169.254.169.254/latest/meta-data/iam/security-credentials/iam-grid-s3)
>
>         AWS: Storing AIM credentials until 2022-11-19T20:42:58Z
>
>         S3: Downloading 0-16383
>         (https://grid-dev-publiclidar.s3.amazonaws.com/estonia/dtm/estonia_dtm_5m.tif)...
>
>         S3: Got response_code=206
>
>         gdalinfo failed - unable to open
>         '/vsis3/grid-dev-publiclidar/estonia/dtm/estonia_dtm_5m.tif'.
>
>         Mike
>
>
>
>             On Nov 19, 2022, at 9:26 AM, Even Rouault
>             <even.rouault at spatialys.com>
>             <mailto:even.rouault at spatialys.com> wrote:
>
>             Hi Mike,
>
>             could you send the output of
>
>             curl
>             http://169.254.169.254/latest/meta-data/iam/security-credentials/iam-grid-s3
>
>             Slightly redacted of course, but with the exact
>             formatting. This part of thee code currently uses a
>             "simple JSON parser"
>             (https://github.com/OSGeo/gdal/blob/c61d116a469821b769630a112dee7f1a61fed885/port/cpl_aws.cpp#L554),
>             which is actually just a non JSON-aware string tokenizer,
>             and I suspect it could be defeated by a new formatting of
>             S3 or something specific to your credentials.
>
>             It could also be that something unhandled by that parser
>             appears inside quoted strings, like an escaped double
>             quote or some other JSON escaped character (like an
>             escaped forward slash \/ )
>
>             If that was the case we should likely switch to proper
>             JSON deserialization (that part of the code must predate
>             libjson-c being a build requirement of GDAL).
>
>             Even
>
>
>             -- 
>             http://www.spatialys.com
>             My software is free, but my time generally not.
>
>     -- 
>
>     http://www.spatialys.com
>
>     My software is free, but my time generally not.
>
> -- 
> http://www.spatialys.com
> My software is free, but my time generally not.

-- 
http://www.spatialys.com
My software is free, but my time generally not.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20221120/9e32b780/attachment-0001.htm>


More information about the gdal-dev mailing list