On Mon, Aug 18, 2014 at 7:37 AM, Benedikt Fraunhofer <fraunhofer@xxxxxxxxxx> wrote: > Hi Yehuda, > > sorry to contact you in person. > I originally posted that to cephdevel, but it seems this got lost: > http://comments.gmane.org/gmane.comp.file-systems.ceph.devel/19035 > > I'm trying to replace our old radosgw with the new civetweb-frontend, > because ubuntu14.04 ships with apache2.4 and the patched fastcgi-stuff > isnt there yet for apache2.4; > > I just wanted to check with you if you currently have time available to > to get the civetweb-frontend flying, or if it's on hold due lack of time > and more important stuff. > It's also fine if you can only give me a hint at the right direction. > > Currently I'm running off tag 0.80.5 > with these fixes: > > diff --git a/src/rgw/rgw_civetweb.cc b/src/rgw/rgw_civetweb.cc > index a31177f..b44a40c 100644 > --- a/src/rgw/rgw_civetweb.cc > +++ b/src/rgw/rgw_civetweb.cc > @@ -42,7 +42,7 @@ int RGWMongoose::complete_request() > > if (0 && data.length() == 0) { > has_content_length = true; > - print("Transfer-Enconding: %s\n", "chunked"); > + print("Transfer-Enconding: %s\r\n", "chunked"); > data.append("0\r\n\r\n", sizeof("0\r\n\r\n")-1); > } else { > int r = send_content_length(data.length()); > @@ -128,7 +128,7 @@ int RGWMongoose::send_status(const char *status, > const char *status_name) > if (!status_name) > status_name = ""; > > - snprintf(buf, sizeof(buf), "HTTP/1.1 %s %s\n", status, status_name); > + snprintf(buf, sizeof(buf), "HTTP/1.1 %s %s\r\n", status, status_name); > > bufferlist bl; > bl.append(buf); > @@ -168,5 +168,5 @@ int RGWMongoose::send_content_length(uint64_t len) > has_content_length = true; > char buf[21]; > snprintf(buf, sizeof(buf), "%"PRIu64, len); > - return print("Content-Length: %s\n", buf); > + return print("Content-Length: %s\r\n", buf); > } > > This seems to only affect windows users using some .net function to make > http-calls and which seem to rely on the presence of the "\r" there. > It fixes some, but not all of the problems. > > I added this one, too: > > diff --git a/src/rgw/rgw_rest.cc b/src/rgw/rgw_rest.cc > index 69948a6..0142e21 100644 > --- a/src/rgw/rgw_rest.cc > +++ b/src/rgw/rgw_rest.cc > @@ -244,7 +244,7 @@ void dump_content_length(struct req_state *s, uint64_t len) > if (r < 0) { > ldout(s->cct, 0) << "ERROR: s->cio->print() returned err=" << r << dendl; > } > - r = s->cio->print("Accept-Ranges: %s\n", "bytes"); > + r = s->cio->print("Accept-Ranges: %s\r\n", "bytes"); > if (r < 0) { > ldout(s->cct, 0) << "ERROR: s->cio->print() returned err=" << r << dendl; > } > Looks fine, can you send a pull request via github? > but don't know if that affects the fastcgi-part. > > With these fixes, one can do some operations on some files, but e.g. > setting acls fails for me: > > here's the http-log: > ------------- > T 127.0.0.1:57854 -> 127.0.0.1:2080 [AP] > GET /b2d9709c-6a5e-4a7c-9d92-5060b241c254/50%2Dmb.bin?acl HTTP/1.1..User-Ag > ent: CloudBerryLab.Base.HttpUtil.Client 3.8.7 (http://www.cloudberrylab.com > /)..x-amz-date: Mon, 18 Aug 2014 13:06:38 GMT..Authorization: AWS 9F3CKMDV8 > 18PPXG6BYD4:5X44NdbP2b9af7Uq6e5DnKWhhLo=..Host: ceph1-entry2.ceph1.globalwa > ys.net.... > ## > T 127.0.0.1:2080 -> 127.0.0.1:57854 [AP] > HTTP/1.1 403 Forbidden..Content-Length: 78..Accept-Ranges: bytes..Content-t > ype: application/xml.... > ------------- > > and here what radosgw has to say about it: > > ------------------- > 2014-08-18 15:06:37.511655 7fc5b4ff9700 10 get_canon_resource(): > dest=/b2d9709c-6a5e-4a7c-9d92-5060b241c254/50-mb.bin?acl > 2014-08-18 15:06:37.511657 7fc5b4ff9700 10 auth_hdr: > GET > > > > x-amz-date:Mon, 18 Aug 2014 13:06:38 GMT > /b2d9709c-6a5e-4a7c-9d92-5060b241c254/50-mb.bin?acl > 2014-08-18 15:06:37.511669 7fc5b4ff9700 15 calculated > digest=ufY9cgyIXx1kYhMJMg+Ss5KJ47o= > 2014-08-18 15:06:37.511670 7fc5b4ff9700 15 > auth_sign=5X44NdbP2b9af7Uq6e5DnKWhhLo= > ------------------- > > The signatures differ because of the last escape "-" (%2D). > One side calculates it with the last "-" as "%2D" and radosgw decides > to calculate the signature using the original request, before > replacing all %XX occurences, despite logging it with the %-escaped > stuff already unescaped. I don't think it's related to your fixes, sounds like issue #8702, which will make it to the next firefly point release. Yehuda > > > If I calculate the signatures "by hand" I get the these values, too; I > just dont know who's doing it wrong, > because it works if I put the entire apache<>fastcgi<>radosgw chain into place. > (Maybe it's because the client get's the inital object-name already > wrongly encoded?) > > bf@bfc:~$ s3cmd sign 'GET > > > > x-amz-date:Mon, 18 Aug 2014 13:06:38 GMT > /b2d9709c-6a5e-4a7c-9d92-5060b241c254/50-mb.bin?acl' > Signature: ufY9cgyIXx1kYhMJMg+Ss5KJ47o= > bf@bfc:~$ s3cmd sign 'GET > > > > x-amz-date:Mon, 18 Aug 2014 13:06:38 GMT > /b2d9709c-6a5e-4a7c-9d92-5060b241c254/50%2Dmb.bin?acl' > Signature: 5X44NdbP2b9af7Uq6e5DnKWhhLo= > > > Thx in advance, > > Benedikt. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
