You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

SmartClientSmartServerTest.java 65KB

Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Do authentication re-tries on HTTP POST There is at least one git server out there (GOGS) that does not require authentication on the initial GET for info/refs?service=git-receive-pack but that _does_ require authentication for the subsequent POST to actually do the push. This occurs on GOGS with public repositories; for private repositories it wants authentication up front. Handle this behavior by adding 401 handling to our POST request. Note that this is suboptimal; we'll re-send the push data at least twice if an authentication failure on POST occurs. It would be much better if the server required authentication up-front in the GET request. Added authentication unit tests (using BASIC auth) to the SmartClientSmartServerTest: - clone with authentication - clone with authentication but lacking CredentialsProvider - clone with authentication and wrong password - clone with authentication after redirect - clone with authentication only on POST, but not on GET Also tested manually in the wild using repositories at try.gogs.io. That server offers only BASIC auth, so the other paths (DIGEST, NEGOTIATE, fall back from DIGEST to BASIC) are untested and I have no way to test them. * public repository: GET unauthenticated, POST authenticated Also tested after clearing the credentials and then entering a wrong password: correctly asks three times during the HTTP POST for user name and password, then gives up. * private repository: authentication already on GET; then gets applied correctly initially to the POST request, which succeeds. Also fix the authentication to use the credentials for the redirected URI if redirects had occurred. We must not present the credentials for the original URI in that case. Consider a malicious redirect A->B: this would allow server B to harvest the user credentials for server A. The unit test for authentication after a redirect also tests for this. Bug: 513043 Change-Id: I97ee5058569efa1545a6c6f6edfd2b357c40592a Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
7 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Client-side protocol V2 support for fetching Make all transports request protocol V2 when fetching. Depending on the transport, set the GIT_PROTOCOL environment variable (file and ssh), pass the Git-Protocol header (http), or set the hidden "\0version=2\0" (git anon). We'll fall back to V0 if the server doesn't reply with a version 2 answer. A user can control which protocol the client requests via the git config protocol.version; if not set, JGit requests protocol V2 for fetching. Pushing always uses protocol V0 still. In the API, there is only a new Transport.openFetch() version that takes a collection of RefSpecs plus additional patterns to construct the Ref prefixes for the "ls-refs" command in protocol V2. If none are given, the server will still advertise all refs, even in protocol V2. BasePackConnection.readAdvertisedRefs() handles falling back to protocol V0. It newly returns true if V0 was used and the advertised refs were read, and false if V2 is used and an explicit "ls-refs" is needed. (This can't be done transparently inside readAdvertisedRefs() because a "stateless RPC" transport like TransportHttp may need to open a new connection for writing.) BasePackFetchConnection implements the changes needed for the protocol V2 "fetch" command (stateless protocol, simplified ACK handling, delimiters, section headers). In TransportHttp, change readSmartHeaders() to also recognize the "version 2" packet line as a valid smart server indication. Adapt tests, and run all the HTTP tests not only with both HTTP connection factories (JDK and Apache HttpClient) but also with both protocol V0 and V2. The SSH tests are much slower and much more focused on the SSH protocol and SSH key handling. Factor out two very simple cloning and pulling tests and make those run with protocol V2. Bug: 553083 Change-Id: I357c7f5daa7efb2872f1c64ee6f6d54229031ae1 Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
3 years ago
Add support to follow HTTP redirects git-core follows HTTP redirects so JGit should also provide this. Implement config setting http.followRedirects with possible values "false" (= never), "true" (= always), and "initial" (only on GET, but not on POST).[1] We must do our own redirect handling and cannot rely on the support that the underlying real connection may offer. At least the JDK's HttpURLConnection has two features that get in the way: * it does not allow cross-protocol redirects and thus fails on http->https redirects (for instance, on Github). * it translates a redirect after a POST to a GET unless the system property "http.strictPostRedirect" is set to true. We don't want to manipulate that system setting nor require it. Additionally, git has its own rules about what redirects it accepts;[2] for instance, it does not allow a redirect that adds query arguments. We handle response codes 301, 302, 303, and 307 as per RFC 2616.[3] On POST we do not handle 303, and we follow redirects only if http.followRedirects == true. Redirects are followed only a certain number of times. There are two ways to control that limit: * by default, the limit is given by the http.maxRedirects system property that is also used by the JDK. If the system property is not set, the default is 5. (This is much lower than the JDK default of 20, but I don't see the value of following so many redirects.) * this can be overwritten by a http.maxRedirects git config setting. The JGit http.* git config settings are currently all global; JGit has no support yet for URI-specific settings "http.<pattern>.name". Adding support for that is well beyond the scope of this change. Like git-core, we log every redirect attempt (LOG.info) so that users may know about the redirection having occurred. Extends the test framework to configure an AppServer with HTTPS support so that we can test cloning via HTTPS and redirections involving HTTPS. [1] https://git-scm.com/docs/git-config [2] https://kernel.googlesource.com/pub/scm/git/git/+/6628eb41db5189c0cdfdced6d8697e7c813c5f0f [3] https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html CQ: 13987 Bug: 465167 Change-Id: I86518cb76842f7d326b51f8715e3bbf8ada89859 Signed-off-by: Matthias Sohn <matthias.sohn@sap.com> Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
9 years ago
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829
  1. /*
  2. * Copyright (C) 2010, 2020 Google Inc. and others
  3. *
  4. * This program and the accompanying materials are made available under the
  5. * terms of the Eclipse Distribution License v. 1.0 which is available at
  6. * https://www.eclipse.org/org/documents/edl-v10.php.
  7. *
  8. * SPDX-License-Identifier: BSD-3-Clause
  9. */
  10. package org.eclipse.jgit.http.test;
  11. import static java.nio.charset.StandardCharsets.UTF_8;
  12. import static org.eclipse.jgit.util.HttpSupport.HDR_CONTENT_ENCODING;
  13. import static org.eclipse.jgit.util.HttpSupport.HDR_CONTENT_LENGTH;
  14. import static org.eclipse.jgit.util.HttpSupport.HDR_CONTENT_TYPE;
  15. import static org.junit.Assert.assertEquals;
  16. import static org.junit.Assert.assertFalse;
  17. import static org.junit.Assert.assertNotNull;
  18. import static org.junit.Assert.assertNull;
  19. import static org.junit.Assert.assertThrows;
  20. import static org.junit.Assert.assertTrue;
  21. import static org.junit.Assert.fail;
  22. import java.io.ByteArrayOutputStream;
  23. import java.io.File;
  24. import java.io.IOException;
  25. import java.io.OutputStreamWriter;
  26. import java.io.PrintWriter;
  27. import java.io.Writer;
  28. import java.net.Proxy;
  29. import java.net.URI;
  30. import java.net.URISyntaxException;
  31. import java.net.URL;
  32. import java.nio.charset.StandardCharsets;
  33. import java.text.MessageFormat;
  34. import java.util.Collections;
  35. import java.util.EnumSet;
  36. import java.util.List;
  37. import java.util.Map;
  38. import java.util.regex.Matcher;
  39. import java.util.regex.Pattern;
  40. import javax.servlet.DispatcherType;
  41. import javax.servlet.Filter;
  42. import javax.servlet.FilterChain;
  43. import javax.servlet.FilterConfig;
  44. import javax.servlet.RequestDispatcher;
  45. import javax.servlet.ServletException;
  46. import javax.servlet.ServletRequest;
  47. import javax.servlet.ServletResponse;
  48. import javax.servlet.http.HttpServletRequest;
  49. import javax.servlet.http.HttpServletResponse;
  50. import org.eclipse.jetty.servlet.FilterHolder;
  51. import org.eclipse.jetty.servlet.ServletContextHandler;
  52. import org.eclipse.jetty.servlet.ServletHolder;
  53. import org.eclipse.jgit.api.Git;
  54. import org.eclipse.jgit.api.TransportConfigCallback;
  55. import org.eclipse.jgit.errors.RemoteRepositoryException;
  56. import org.eclipse.jgit.errors.TransportException;
  57. import org.eclipse.jgit.errors.UnsupportedCredentialItem;
  58. import org.eclipse.jgit.http.server.GitServlet;
  59. import org.eclipse.jgit.http.server.resolver.DefaultUploadPackFactory;
  60. import org.eclipse.jgit.internal.JGitText;
  61. import org.eclipse.jgit.internal.storage.dfs.DfsRepositoryDescription;
  62. import org.eclipse.jgit.junit.TestRepository;
  63. import org.eclipse.jgit.junit.TestRng;
  64. import org.eclipse.jgit.junit.http.AccessEvent;
  65. import org.eclipse.jgit.junit.http.AppServer;
  66. import org.eclipse.jgit.lib.ConfigConstants;
  67. import org.eclipse.jgit.lib.Constants;
  68. import org.eclipse.jgit.lib.NullProgressMonitor;
  69. import org.eclipse.jgit.lib.ObjectId;
  70. import org.eclipse.jgit.lib.ObjectIdRef;
  71. import org.eclipse.jgit.lib.ObjectInserter;
  72. import org.eclipse.jgit.lib.Ref;
  73. import org.eclipse.jgit.lib.ReflogEntry;
  74. import org.eclipse.jgit.lib.ReflogReader;
  75. import org.eclipse.jgit.lib.Repository;
  76. import org.eclipse.jgit.lib.StoredConfig;
  77. import org.eclipse.jgit.lib.TextProgressMonitor;
  78. import org.eclipse.jgit.revwalk.RevBlob;
  79. import org.eclipse.jgit.revwalk.RevCommit;
  80. import org.eclipse.jgit.revwalk.RevWalk;
  81. import org.eclipse.jgit.transport.AbstractAdvertiseRefsHook;
  82. import org.eclipse.jgit.transport.AdvertiseRefsHook;
  83. import org.eclipse.jgit.transport.CredentialItem;
  84. import org.eclipse.jgit.transport.CredentialsProvider;
  85. import org.eclipse.jgit.transport.FetchConnection;
  86. import org.eclipse.jgit.transport.HttpTransport;
  87. import org.eclipse.jgit.transport.RefSpec;
  88. import org.eclipse.jgit.transport.RemoteRefUpdate;
  89. import org.eclipse.jgit.transport.Transport;
  90. import org.eclipse.jgit.transport.TransportHttp;
  91. import org.eclipse.jgit.transport.URIish;
  92. import org.eclipse.jgit.transport.UploadPack;
  93. import org.eclipse.jgit.transport.UsernamePasswordCredentialsProvider;
  94. import org.eclipse.jgit.transport.http.HttpConnection;
  95. import org.eclipse.jgit.transport.http.HttpConnectionFactory;
  96. import org.eclipse.jgit.util.HttpSupport;
  97. import org.eclipse.jgit.util.SystemReader;
  98. import org.junit.Before;
  99. import org.junit.Test;
  100. public class SmartClientSmartServerTest extends AllProtocolsHttpTestCase {
  101. private static final String HDR_TRANSFER_ENCODING = "Transfer-Encoding";
  102. private AdvertiseRefsHook advertiseRefsHook;
  103. private Repository remoteRepository;
  104. private CredentialsProvider testCredentials = new UsernamePasswordCredentialsProvider(
  105. AppServer.username, AppServer.password);
  106. private URIish remoteURI;
  107. private URIish brokenURI;
  108. private URIish redirectURI;
  109. private URIish authURI;
  110. private URIish authOnPostURI;
  111. private URIish slowURI;
  112. private URIish slowAuthURI;
  113. private RevBlob A_txt;
  114. private RevCommit A, B, unreachableCommit;
  115. public SmartClientSmartServerTest(TestParameters params) {
  116. super(params);
  117. }
  118. @Override
  119. @Before
  120. public void setUp() throws Exception {
  121. super.setUp();
  122. final TestRepository<Repository> src = createTestRepository();
  123. final String srcName = src.getRepository().getDirectory().getName();
  124. StoredConfig cfg = src.getRepository().getConfig();
  125. cfg.setBoolean(ConfigConstants.CONFIG_CORE_SECTION, null,
  126. ConfigConstants.CONFIG_KEY_LOGALLREFUPDATES, true);
  127. cfg.setInt("protocol", null, "version", enableProtocolV2 ? 2 : 0);
  128. cfg.save();
  129. GitServlet gs = new GitServlet();
  130. gs.setUploadPackFactory((HttpServletRequest req, Repository db) -> {
  131. DefaultUploadPackFactory f = new DefaultUploadPackFactory();
  132. UploadPack up = f.create(req, db);
  133. if (advertiseRefsHook != null) {
  134. up.setAdvertiseRefsHook(advertiseRefsHook);
  135. }
  136. return up;
  137. });
  138. ServletContextHandler app = addNormalContext(gs, src, srcName);
  139. ServletContextHandler broken = addBrokenContext(gs, srcName);
  140. ServletContextHandler redirect = addRedirectContext(gs);
  141. ServletContextHandler auth = addAuthContext(gs, "auth");
  142. ServletContextHandler authOnPost = addAuthContext(gs, "pauth", "POST");
  143. ServletContextHandler slow = addSlowContext(gs, "slow", false);
  144. ServletContextHandler slowAuth = addSlowContext(gs, "slowAuth", true);
  145. server.setUp();
  146. remoteRepository = src.getRepository();
  147. remoteURI = toURIish(app, srcName);
  148. brokenURI = toURIish(broken, srcName);
  149. redirectURI = toURIish(redirect, srcName);
  150. authURI = toURIish(auth, srcName);
  151. authOnPostURI = toURIish(authOnPost, srcName);
  152. slowURI = toURIish(slow, srcName);
  153. slowAuthURI = toURIish(slowAuth, srcName);
  154. A_txt = src.blob("A");
  155. A = src.commit().add("A_txt", A_txt).create();
  156. B = src.commit().parent(A).add("A_txt", "C").add("B", "B").create();
  157. src.update(master, B);
  158. unreachableCommit = src.commit().add("A_txt", A_txt).create();
  159. src.update("refs/garbage/a/very/long/ref/name/to/compress", B);
  160. }
  161. private ServletContextHandler addNormalContext(GitServlet gs, TestRepository<Repository> src, String srcName) {
  162. ServletContextHandler app = server.addContext("/git");
  163. app.addFilter(new FilterHolder(new Filter() {
  164. @Override
  165. public void init(FilterConfig filterConfig)
  166. throws ServletException {
  167. // empty
  168. }
  169. // Does an internal forward for GET requests containing "/post/",
  170. // and issues a 301 redirect on POST requests for such URLs. Used
  171. // in the POST redirect tests.
  172. @Override
  173. public void doFilter(ServletRequest request,
  174. ServletResponse response, FilterChain chain)
  175. throws IOException, ServletException {
  176. final HttpServletResponse httpServletResponse = (HttpServletResponse) response;
  177. final HttpServletRequest httpServletRequest = (HttpServletRequest) request;
  178. final StringBuffer fullUrl = httpServletRequest.getRequestURL();
  179. if (httpServletRequest.getQueryString() != null) {
  180. fullUrl.append("?")
  181. .append(httpServletRequest.getQueryString());
  182. }
  183. String urlString = fullUrl.toString();
  184. if ("POST".equalsIgnoreCase(httpServletRequest.getMethod())) {
  185. httpServletResponse.setStatus(
  186. HttpServletResponse.SC_MOVED_PERMANENTLY);
  187. httpServletResponse.setHeader(HttpSupport.HDR_LOCATION,
  188. urlString.replace("/post/", "/"));
  189. } else {
  190. String path = httpServletRequest.getPathInfo();
  191. path = path.replace("/post/", "/");
  192. if (httpServletRequest.getQueryString() != null) {
  193. path += '?' + httpServletRequest.getQueryString();
  194. }
  195. RequestDispatcher dispatcher = httpServletRequest
  196. .getRequestDispatcher(path);
  197. dispatcher.forward(httpServletRequest, httpServletResponse);
  198. }
  199. }
  200. @Override
  201. public void destroy() {
  202. // empty
  203. }
  204. }), "/post/*", EnumSet.of(DispatcherType.REQUEST));
  205. gs.setRepositoryResolver(new TestRepositoryResolver(src, srcName));
  206. app.addServlet(new ServletHolder(gs), "/*");
  207. return app;
  208. }
  209. private ServletContextHandler addBrokenContext(GitServlet gs,
  210. String srcName) {
  211. ServletContextHandler broken = server.addContext("/bad");
  212. broken.addFilter(new FilterHolder(new Filter() {
  213. @Override
  214. public void doFilter(ServletRequest request,
  215. ServletResponse response, FilterChain chain)
  216. throws IOException, ServletException {
  217. final HttpServletResponse r = (HttpServletResponse) response;
  218. r.setContentType("text/plain");
  219. r.setCharacterEncoding(UTF_8.name());
  220. try (PrintWriter w = r.getWriter()) {
  221. w.print("OK");
  222. }
  223. }
  224. @Override
  225. public void init(FilterConfig filterConfig)
  226. throws ServletException {
  227. // empty
  228. }
  229. @Override
  230. public void destroy() {
  231. // empty
  232. }
  233. }), "/" + srcName + "/git-upload-pack",
  234. EnumSet.of(DispatcherType.REQUEST));
  235. broken.addServlet(new ServletHolder(gs), "/*");
  236. return broken;
  237. }
  238. private ServletContextHandler addAuthContext(GitServlet gs,
  239. String contextPath, String... methods) {
  240. ServletContextHandler auth = server.addContext('/' + contextPath);
  241. auth.addServlet(new ServletHolder(gs), "/*");
  242. return server.authBasic(auth, methods);
  243. }
  244. private ServletContextHandler addRedirectContext(GitServlet gs) {
  245. ServletContextHandler redirect = server.addContext("/redirect");
  246. redirect.addFilter(new FilterHolder(new Filter() {
  247. // Enables tests for different codes, and for multiple redirects.
  248. // First parameter is the number of redirects, second one is the
  249. // redirect status code that should be used
  250. private Pattern responsePattern = Pattern
  251. .compile("/response/(\\d+)/(30[1237])/");
  252. // Enables tests to specify the context that the request should be
  253. // redirected to in the end. If not present, redirects got to the
  254. // normal /git context.
  255. private Pattern targetPattern = Pattern.compile("/target(/\\w+)/");
  256. @Override
  257. public void init(FilterConfig filterConfig)
  258. throws ServletException {
  259. // empty
  260. }
  261. private String local(String url, boolean toLocal) {
  262. if (!toLocal) {
  263. return url;
  264. }
  265. try {
  266. URI u = new URI(url);
  267. String fragment = u.getRawFragment();
  268. if (fragment != null) {
  269. return u.getRawPath() + '#' + fragment;
  270. }
  271. return u.getRawPath();
  272. } catch (URISyntaxException e) {
  273. return url;
  274. }
  275. }
  276. @Override
  277. public void doFilter(ServletRequest request,
  278. ServletResponse response, FilterChain chain)
  279. throws IOException, ServletException {
  280. final HttpServletResponse httpServletResponse = (HttpServletResponse) response;
  281. final HttpServletRequest httpServletRequest = (HttpServletRequest) request;
  282. final StringBuffer fullUrl = httpServletRequest.getRequestURL();
  283. if (httpServletRequest.getQueryString() != null) {
  284. fullUrl.append("?")
  285. .append(httpServletRequest.getQueryString());
  286. }
  287. String urlString = fullUrl.toString();
  288. boolean localRedirect = false;
  289. if (urlString.contains("/local")) {
  290. urlString = urlString.replace("/local", "");
  291. localRedirect = true;
  292. }
  293. if (urlString.contains("/loop/")) {
  294. urlString = urlString.replace("/loop/", "/loop/x/");
  295. if (urlString.contains("/loop/x/x/x/x/x/x/x/x/")) {
  296. // Go back to initial.
  297. urlString = urlString.replace("/loop/x/x/x/x/x/x/x/x/",
  298. "/loop/");
  299. }
  300. httpServletResponse.setStatus(
  301. HttpServletResponse.SC_MOVED_TEMPORARILY);
  302. httpServletResponse.setHeader(HttpSupport.HDR_LOCATION,
  303. local(urlString, localRedirect));
  304. return;
  305. }
  306. int responseCode = HttpServletResponse.SC_MOVED_PERMANENTLY;
  307. int nofRedirects = 0;
  308. Matcher matcher = responsePattern.matcher(urlString);
  309. if (matcher.find()) {
  310. nofRedirects = Integer
  311. .parseUnsignedInt(matcher.group(1));
  312. responseCode = Integer.parseUnsignedInt(matcher.group(2));
  313. if (--nofRedirects <= 0) {
  314. urlString = urlString.substring(0, matcher.start())
  315. + '/' + urlString.substring(matcher.end());
  316. } else {
  317. urlString = urlString.substring(0, matcher.start())
  318. + "/response/" + nofRedirects + "/"
  319. + responseCode + '/'
  320. + urlString.substring(matcher.end());
  321. }
  322. }
  323. httpServletResponse.setStatus(responseCode);
  324. if (nofRedirects <= 0) {
  325. String targetContext = "/git";
  326. matcher = targetPattern.matcher(urlString);
  327. if (matcher.find()) {
  328. urlString = urlString.substring(0, matcher.start())
  329. + '/' + urlString.substring(matcher.end());
  330. targetContext = matcher.group(1);
  331. }
  332. urlString = urlString.replace("/redirect", targetContext);
  333. }
  334. httpServletResponse.setHeader(HttpSupport.HDR_LOCATION,
  335. local(urlString, localRedirect));
  336. }
  337. @Override
  338. public void destroy() {
  339. // empty
  340. }
  341. }), "/*", EnumSet.of(DispatcherType.REQUEST));
  342. redirect.addServlet(new ServletHolder(gs), "/*");
  343. return redirect;
  344. }
  345. private ServletContextHandler addSlowContext(GitServlet gs, String path,
  346. boolean auth) {
  347. ServletContextHandler slow = server.addContext('/' + path);
  348. slow.addFilter(new FilterHolder(new Filter() {
  349. @Override
  350. public void init(FilterConfig filterConfig)
  351. throws ServletException {
  352. // empty
  353. }
  354. // Simply delays the servlet for two seconds. Used for timeout
  355. // tests, which use a one-second timeout.
  356. @Override
  357. public void doFilter(ServletRequest request,
  358. ServletResponse response, FilterChain chain)
  359. throws IOException, ServletException {
  360. try {
  361. Thread.sleep(2000);
  362. } catch (InterruptedException e) {
  363. throw new IOException(e);
  364. }
  365. chain.doFilter(request, response);
  366. }
  367. @Override
  368. public void destroy() {
  369. // empty
  370. }
  371. }), "/*", EnumSet.of(DispatcherType.REQUEST));
  372. slow.addServlet(new ServletHolder(gs), "/*");
  373. if (auth) {
  374. return server.authBasic(slow);
  375. }
  376. return slow;
  377. }
  378. @Test
  379. public void testListRemote() throws IOException {
  380. assertEquals("http", remoteURI.getScheme());
  381. Map<String, Ref> map;
  382. try (Repository dst = createBareRepository();
  383. Transport t = Transport.open(dst, remoteURI)) {
  384. // I didn't make up these public interface names, I just
  385. // approved them for inclusion into the code base. Sorry.
  386. // --spearce
  387. //
  388. assertTrue("isa TransportHttp", t instanceof TransportHttp);
  389. assertTrue("isa HttpTransport", t instanceof HttpTransport);
  390. try (FetchConnection c = t.openFetch()) {
  391. map = c.getRefsMap();
  392. }
  393. }
  394. assertNotNull("have map of refs", map);
  395. assertEquals(3, map.size());
  396. assertNotNull("has " + master, map.get(master));
  397. assertEquals(B, map.get(master).getObjectId());
  398. assertNotNull("has " + Constants.HEAD, map.get(Constants.HEAD));
  399. assertEquals(B, map.get(Constants.HEAD).getObjectId());
  400. List<AccessEvent> requests = getRequests();
  401. assertEquals(enableProtocolV2 ? 2 : 1, requests.size());
  402. AccessEvent info = requests.get(0);
  403. assertEquals("GET", info.getMethod());
  404. assertEquals(join(remoteURI, "info/refs"), info.getPath());
  405. assertEquals(1, info.getParameters().size());
  406. assertEquals("git-upload-pack", info.getParameter("service"));
  407. assertEquals(200, info.getStatus());
  408. assertEquals("application/x-git-upload-pack-advertisement", info
  409. .getResponseHeader(HDR_CONTENT_TYPE));
  410. if (!enableProtocolV2) {
  411. assertEquals("gzip", info.getResponseHeader(HDR_CONTENT_ENCODING));
  412. } else {
  413. AccessEvent lsRefs = requests.get(1);
  414. assertEquals("POST", lsRefs.getMethod());
  415. assertEquals(join(remoteURI, "git-upload-pack"), lsRefs.getPath());
  416. assertEquals(0, lsRefs.getParameters().size());
  417. assertNotNull("has content-length",
  418. lsRefs.getRequestHeader(HDR_CONTENT_LENGTH));
  419. assertNull("not chunked",
  420. lsRefs.getRequestHeader(HDR_TRANSFER_ENCODING));
  421. assertEquals("version=2", lsRefs.getRequestHeader("Git-Protocol"));
  422. assertEquals(200, lsRefs.getStatus());
  423. assertEquals("application/x-git-upload-pack-result",
  424. lsRefs.getResponseHeader(HDR_CONTENT_TYPE));
  425. }
  426. }
  427. @Test
  428. public void testListRemote_BadName() throws IOException, URISyntaxException {
  429. URIish uri = new URIish(this.remoteURI.toString() + ".invalid");
  430. try (Repository dst = createBareRepository();
  431. Transport t = Transport.open(dst, uri)) {
  432. try {
  433. t.openFetch();
  434. fail("fetch connection opened");
  435. } catch (RemoteRepositoryException notFound) {
  436. assertEquals(uri + ": Git repository not found",
  437. notFound.getMessage());
  438. }
  439. }
  440. List<AccessEvent> requests = getRequests();
  441. assertEquals(1, requests.size());
  442. AccessEvent info = requests.get(0);
  443. assertEquals("GET", info.getMethod());
  444. assertEquals(join(uri, "info/refs"), info.getPath());
  445. assertEquals(1, info.getParameters().size());
  446. assertEquals("git-upload-pack", info.getParameter("service"));
  447. assertEquals(200, info.getStatus());
  448. assertEquals("application/x-git-upload-pack-advertisement",
  449. info.getResponseHeader(HDR_CONTENT_TYPE));
  450. }
  451. @Test
  452. public void testFetchBySHA1() throws Exception {
  453. try (Repository dst = createBareRepository();
  454. Transport t = Transport.open(dst, remoteURI)) {
  455. assertFalse(dst.getObjectDatabase().has(A_txt));
  456. t.fetch(NullProgressMonitor.INSTANCE,
  457. Collections.singletonList(new RefSpec(B.name())));
  458. assertTrue(dst.getObjectDatabase().has(A_txt));
  459. }
  460. }
  461. @Test
  462. public void testFetchBySHA1Unreachable() throws Exception {
  463. try (Repository dst = createBareRepository();
  464. Transport t = Transport.open(dst, remoteURI)) {
  465. assertFalse(dst.getObjectDatabase().has(A_txt));
  466. Exception e = assertThrows(TransportException.class,
  467. () -> t.fetch(NullProgressMonitor.INSTANCE,
  468. Collections.singletonList(
  469. new RefSpec(unreachableCommit.name()))));
  470. assertTrue(e.getMessage().contains(
  471. "want " + unreachableCommit.name() + " not valid"));
  472. }
  473. }
  474. @Test
  475. public void testFetchBySHA1UnreachableByAdvertiseRefsHook()
  476. throws Exception {
  477. advertiseRefsHook = new AbstractAdvertiseRefsHook() {
  478. @Override
  479. protected Map<String, Ref> getAdvertisedRefs(Repository repository,
  480. RevWalk revWalk) {
  481. return Collections.emptyMap();
  482. }
  483. };
  484. try (Repository dst = createBareRepository();
  485. Transport t = Transport.open(dst, remoteURI)) {
  486. assertFalse(dst.getObjectDatabase().has(A_txt));
  487. Exception e = assertThrows(TransportException.class,
  488. () -> t.fetch(NullProgressMonitor.INSTANCE,
  489. Collections.singletonList(new RefSpec(A.name()))));
  490. assertTrue(
  491. e.getMessage().contains("want " + A.name() + " not valid"));
  492. }
  493. }
  494. @Test
  495. public void testTimeoutExpired() throws Exception {
  496. try (Repository dst = createBareRepository();
  497. Transport t = Transport.open(dst, slowURI)) {
  498. t.setTimeout(1);
  499. TransportException expected = assertThrows(TransportException.class,
  500. () -> t.fetch(NullProgressMonitor.INSTANCE,
  501. mirror(master)));
  502. assertTrue("Unexpected exception message: " + expected.toString(),
  503. expected.getMessage().contains("time"));
  504. }
  505. }
  506. @Test
  507. public void testTimeoutExpiredWithAuth() throws Exception {
  508. try (Repository dst = createBareRepository();
  509. Transport t = Transport.open(dst, slowAuthURI)) {
  510. t.setTimeout(1);
  511. t.setCredentialsProvider(testCredentials);
  512. TransportException expected = assertThrows(TransportException.class,
  513. () -> t.fetch(NullProgressMonitor.INSTANCE,
  514. mirror(master)));
  515. assertTrue("Unexpected exception message: " + expected.toString(),
  516. expected.getMessage().contains("time"));
  517. assertFalse("Unexpected exception message: " + expected.toString(),
  518. expected.getMessage().contains("auth"));
  519. }
  520. }
  521. @Test
  522. public void testInitialClone_Small() throws Exception {
  523. try (Repository dst = createBareRepository();
  524. Transport t = Transport.open(dst, remoteURI)) {
  525. assertFalse(dst.getObjectDatabase().has(A_txt));
  526. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  527. assertTrue(dst.getObjectDatabase().has(A_txt));
  528. assertEquals(B, dst.exactRef(master).getObjectId());
  529. fsck(dst, B);
  530. }
  531. List<AccessEvent> requests = getRequests();
  532. assertEquals(enableProtocolV2 ? 3 : 2, requests.size());
  533. int requestNumber = 0;
  534. AccessEvent info = requests.get(requestNumber++);
  535. assertEquals("GET", info.getMethod());
  536. assertEquals(join(remoteURI, "info/refs"), info.getPath());
  537. assertEquals(1, info.getParameters().size());
  538. assertEquals("git-upload-pack", info.getParameter("service"));
  539. assertEquals(200, info.getStatus());
  540. assertEquals("application/x-git-upload-pack-advertisement", info
  541. .getResponseHeader(HDR_CONTENT_TYPE));
  542. if (!enableProtocolV2) {
  543. assertEquals("gzip", info.getResponseHeader(HDR_CONTENT_ENCODING));
  544. } else {
  545. AccessEvent lsRefs = requests.get(requestNumber++);
  546. assertEquals("POST", lsRefs.getMethod());
  547. assertEquals(join(remoteURI, "git-upload-pack"), lsRefs.getPath());
  548. assertEquals(0, lsRefs.getParameters().size());
  549. assertNotNull("has content-length",
  550. lsRefs.getRequestHeader(HDR_CONTENT_LENGTH));
  551. assertNull("not chunked",
  552. lsRefs.getRequestHeader(HDR_TRANSFER_ENCODING));
  553. assertEquals("version=2", lsRefs.getRequestHeader("Git-Protocol"));
  554. assertEquals(200, lsRefs.getStatus());
  555. assertEquals("application/x-git-upload-pack-result",
  556. lsRefs.getResponseHeader(HDR_CONTENT_TYPE));
  557. }
  558. AccessEvent service = requests.get(requestNumber);
  559. assertEquals("POST", service.getMethod());
  560. assertEquals(join(remoteURI, "git-upload-pack"), service.getPath());
  561. assertEquals(0, service.getParameters().size());
  562. assertNotNull("has content-length", service
  563. .getRequestHeader(HDR_CONTENT_LENGTH));
  564. assertNull("not chunked", service
  565. .getRequestHeader(HDR_TRANSFER_ENCODING));
  566. assertEquals(200, service.getStatus());
  567. assertEquals("application/x-git-upload-pack-result", service
  568. .getResponseHeader(HDR_CONTENT_TYPE));
  569. }
  570. @Test
  571. public void test_CloneWithCustomFactory() throws Exception {
  572. HttpConnectionFactory globalFactory = HttpTransport
  573. .getConnectionFactory();
  574. HttpConnectionFactory failingConnectionFactory = new HttpConnectionFactory() {
  575. @Override
  576. public HttpConnection create(URL url) throws IOException {
  577. throw new IOException("Should not be reached");
  578. }
  579. @Override
  580. public HttpConnection create(URL url, Proxy proxy)
  581. throws IOException {
  582. throw new IOException("Should not be reached");
  583. }
  584. };
  585. HttpTransport.setConnectionFactory(failingConnectionFactory);
  586. try {
  587. File tmp = createTempDirectory("cloneViaApi");
  588. boolean[] localFactoryUsed = { false };
  589. TransportConfigCallback callback = new TransportConfigCallback() {
  590. @Override
  591. public void configure(Transport transport) {
  592. if (transport instanceof TransportHttp) {
  593. ((TransportHttp) transport).setHttpConnectionFactory(
  594. new HttpConnectionFactory() {
  595. @Override
  596. public HttpConnection create(URL url)
  597. throws IOException {
  598. localFactoryUsed[0] = true;
  599. return globalFactory.create(url);
  600. }
  601. @Override
  602. public HttpConnection create(URL url,
  603. Proxy proxy) throws IOException {
  604. localFactoryUsed[0] = true;
  605. return globalFactory.create(url, proxy);
  606. }
  607. });
  608. }
  609. }
  610. };
  611. try (Git git = Git.cloneRepository().setDirectory(tmp)
  612. .setTransportConfigCallback(callback)
  613. .setURI(remoteURI.toPrivateString()).call()) {
  614. assertTrue("Should have used the local HttpConnectionFactory",
  615. localFactoryUsed[0]);
  616. }
  617. } finally {
  618. HttpTransport.setConnectionFactory(globalFactory);
  619. }
  620. }
  621. private void initialClone_Redirect(int nofRedirects, int code)
  622. throws Exception {
  623. initialClone_Redirect(nofRedirects, code, false);
  624. }
  625. private void initialClone_Redirect(int nofRedirects, int code,
  626. boolean localRedirect) throws Exception {
  627. URIish cloneFrom = redirectURI;
  628. if (localRedirect) {
  629. cloneFrom = extendPath(cloneFrom, "/local");
  630. }
  631. if (code != 301 || nofRedirects > 1) {
  632. cloneFrom = extendPath(cloneFrom,
  633. "/response/" + nofRedirects + "/" + code);
  634. }
  635. try (Repository dst = createBareRepository();
  636. Transport t = Transport.open(dst, cloneFrom)) {
  637. assertFalse(dst.getObjectDatabase().has(A_txt));
  638. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  639. assertTrue(dst.getObjectDatabase().has(A_txt));
  640. assertEquals(B, dst.exactRef(master).getObjectId());
  641. fsck(dst, B);
  642. }
  643. List<AccessEvent> requests = getRequests();
  644. assertEquals((enableProtocolV2 ? 3 : 2) + nofRedirects,
  645. requests.size());
  646. int n = 0;
  647. while (n < nofRedirects) {
  648. AccessEvent redirect = requests.get(n++);
  649. assertEquals(code, redirect.getStatus());
  650. }
  651. AccessEvent info = requests.get(n++);
  652. assertEquals("GET", info.getMethod());
  653. assertEquals(join(remoteURI, "info/refs"), info.getPath());
  654. assertEquals(1, info.getParameters().size());
  655. assertEquals("git-upload-pack", info.getParameter("service"));
  656. assertEquals(200, info.getStatus());
  657. assertEquals("application/x-git-upload-pack-advertisement",
  658. info.getResponseHeader(HDR_CONTENT_TYPE));
  659. if (!enableProtocolV2) {
  660. assertEquals("gzip", info.getResponseHeader(HDR_CONTENT_ENCODING));
  661. } else {
  662. AccessEvent lsRefs = requests.get(n++);
  663. assertEquals("POST", lsRefs.getMethod());
  664. assertEquals(join(remoteURI, "git-upload-pack"), lsRefs.getPath());
  665. assertEquals(0, lsRefs.getParameters().size());
  666. assertNotNull("has content-length",
  667. lsRefs.getRequestHeader(HDR_CONTENT_LENGTH));
  668. assertNull("not chunked",
  669. lsRefs.getRequestHeader(HDR_TRANSFER_ENCODING));
  670. assertEquals("version=2", lsRefs.getRequestHeader("Git-Protocol"));
  671. assertEquals(200, lsRefs.getStatus());
  672. assertEquals("application/x-git-upload-pack-result",
  673. lsRefs.getResponseHeader(HDR_CONTENT_TYPE));
  674. }
  675. AccessEvent service = requests.get(n++);
  676. assertEquals("POST", service.getMethod());
  677. assertEquals(join(remoteURI, "git-upload-pack"), service.getPath());
  678. assertEquals(0, service.getParameters().size());
  679. assertNotNull("has content-length",
  680. service.getRequestHeader(HDR_CONTENT_LENGTH));
  681. assertNull("not chunked",
  682. service.getRequestHeader(HDR_TRANSFER_ENCODING));
  683. assertEquals(200, service.getStatus());
  684. assertEquals("application/x-git-upload-pack-result",
  685. service.getResponseHeader(HDR_CONTENT_TYPE));
  686. }
  687. @Test
  688. public void testInitialClone_Redirect301Small() throws Exception {
  689. initialClone_Redirect(1, 301);
  690. }
  691. @Test
  692. public void testInitialClone_Redirect301Local() throws Exception {
  693. initialClone_Redirect(1, 301, true);
  694. }
  695. @Test
  696. public void testInitialClone_Redirect302Small() throws Exception {
  697. initialClone_Redirect(1, 302);
  698. }
  699. @Test
  700. public void testInitialClone_Redirect303Small() throws Exception {
  701. initialClone_Redirect(1, 303);
  702. }
  703. @Test
  704. public void testInitialClone_Redirect307Small() throws Exception {
  705. initialClone_Redirect(1, 307);
  706. }
  707. @Test
  708. public void testInitialClone_RedirectMultiple() throws Exception {
  709. initialClone_Redirect(4, 302);
  710. }
  711. @Test
  712. public void testInitialClone_RedirectMax() throws Exception {
  713. StoredConfig userConfig = SystemReader.getInstance()
  714. .getUserConfig();
  715. userConfig.setInt("http", null, "maxRedirects", 4);
  716. userConfig.save();
  717. initialClone_Redirect(4, 302);
  718. }
  719. @Test
  720. public void testInitialClone_RedirectTooOften() throws Exception {
  721. StoredConfig userConfig = SystemReader.getInstance()
  722. .getUserConfig();
  723. userConfig.setInt("http", null, "maxRedirects", 3);
  724. userConfig.save();
  725. URIish cloneFrom = extendPath(redirectURI, "/response/4/302");
  726. String remoteUri = cloneFrom.toString();
  727. try (Repository dst = createBareRepository();
  728. Transport t = Transport.open(dst, cloneFrom)) {
  729. assertFalse(dst.getObjectDatabase().has(A_txt));
  730. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  731. fail("Should have failed (too many redirects)");
  732. } catch (TransportException e) {
  733. String expectedMessageBegin = remoteUri.toString() + ": "
  734. + MessageFormat.format(JGitText.get().redirectLimitExceeded,
  735. "3", remoteUri.replace("/4/", "/1/") + '/', "");
  736. String message = e.getMessage();
  737. if (message.length() > expectedMessageBegin.length()) {
  738. message = message.substring(0, expectedMessageBegin.length());
  739. }
  740. assertEquals(expectedMessageBegin, message);
  741. }
  742. }
  743. @Test
  744. public void testInitialClone_RedirectLoop() throws Exception {
  745. URIish cloneFrom = extendPath(redirectURI, "/loop");
  746. try (Repository dst = createBareRepository();
  747. Transport t = Transport.open(dst, cloneFrom)) {
  748. assertFalse(dst.getObjectDatabase().has(A_txt));
  749. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  750. fail("Should have failed (redirect loop)");
  751. } catch (TransportException e) {
  752. assertTrue(e.getMessage().contains("Redirected more than"));
  753. }
  754. }
  755. @Test
  756. public void testInitialClone_RedirectOnPostAllowed() throws Exception {
  757. StoredConfig userConfig = SystemReader.getInstance()
  758. .getUserConfig();
  759. userConfig.setString("http", null, "followRedirects", "true");
  760. userConfig.save();
  761. URIish cloneFrom = extendPath(remoteURI, "/post");
  762. try (Repository dst = createBareRepository();
  763. Transport t = Transport.open(dst, cloneFrom)) {
  764. assertFalse(dst.getObjectDatabase().has(A_txt));
  765. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  766. assertTrue(dst.getObjectDatabase().has(A_txt));
  767. assertEquals(B, dst.exactRef(master).getObjectId());
  768. fsck(dst, B);
  769. }
  770. List<AccessEvent> requests = getRequests();
  771. assertEquals(enableProtocolV2 ? 4 : 3, requests.size());
  772. AccessEvent info = requests.get(0);
  773. assertEquals("GET", info.getMethod());
  774. assertEquals(join(cloneFrom, "info/refs"), info.getPath());
  775. assertEquals(1, info.getParameters().size());
  776. assertEquals("git-upload-pack", info.getParameter("service"));
  777. assertEquals(200, info.getStatus());
  778. assertEquals("application/x-git-upload-pack-advertisement",
  779. info.getResponseHeader(HDR_CONTENT_TYPE));
  780. if (!enableProtocolV2) {
  781. assertEquals("gzip", info.getResponseHeader(HDR_CONTENT_ENCODING));
  782. }
  783. AccessEvent redirect = requests.get(1);
  784. assertEquals("POST", redirect.getMethod());
  785. assertEquals(301, redirect.getStatus());
  786. for (int i = 2; i < requests.size(); i++) {
  787. AccessEvent service = requests.get(i);
  788. assertEquals("POST", service.getMethod());
  789. assertEquals(join(remoteURI, "git-upload-pack"), service.getPath());
  790. assertEquals(0, service.getParameters().size());
  791. assertNotNull("has content-length",
  792. service.getRequestHeader(HDR_CONTENT_LENGTH));
  793. assertNull("not chunked",
  794. service.getRequestHeader(HDR_TRANSFER_ENCODING));
  795. assertEquals(200, service.getStatus());
  796. assertEquals("application/x-git-upload-pack-result",
  797. service.getResponseHeader(HDR_CONTENT_TYPE));
  798. }
  799. }
  800. @Test
  801. public void testInitialClone_RedirectOnPostForbidden() throws Exception {
  802. URIish cloneFrom = extendPath(remoteURI, "/post");
  803. try (Repository dst = createBareRepository();
  804. Transport t = Transport.open(dst, cloneFrom)) {
  805. assertFalse(dst.getObjectDatabase().has(A_txt));
  806. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  807. fail("Should have failed (redirect on POST)");
  808. } catch (TransportException e) {
  809. assertTrue(e.getMessage().contains("301"));
  810. }
  811. }
  812. @Test
  813. public void testInitialClone_RedirectForbidden() throws Exception {
  814. StoredConfig userConfig = SystemReader.getInstance()
  815. .getUserConfig();
  816. userConfig.setString("http", null, "followRedirects", "false");
  817. userConfig.save();
  818. try (Repository dst = createBareRepository();
  819. Transport t = Transport.open(dst, redirectURI)) {
  820. assertFalse(dst.getObjectDatabase().has(A_txt));
  821. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  822. fail("Should have failed (redirects forbidden)");
  823. } catch (TransportException e) {
  824. assertTrue(
  825. e.getMessage().contains("http.followRedirects is false"));
  826. }
  827. }
  828. private void assertFetchRequests(List<AccessEvent> requests, int index) {
  829. AccessEvent info = requests.get(index++);
  830. assertEquals("GET", info.getMethod());
  831. assertEquals(join(authURI, "info/refs"), info.getPath());
  832. assertEquals(1, info.getParameters().size());
  833. assertEquals("git-upload-pack", info.getParameter("service"));
  834. assertEquals(200, info.getStatus());
  835. assertEquals("application/x-git-upload-pack-advertisement",
  836. info.getResponseHeader(HDR_CONTENT_TYPE));
  837. if (!enableProtocolV2) {
  838. assertEquals("gzip", info.getResponseHeader(HDR_CONTENT_ENCODING));
  839. }
  840. for (int i = index; i < requests.size(); i++) {
  841. AccessEvent service = requests.get(i);
  842. assertEquals("POST", service.getMethod());
  843. assertEquals(join(authURI, "git-upload-pack"), service.getPath());
  844. assertEquals(0, service.getParameters().size());
  845. assertNotNull("has content-length",
  846. service.getRequestHeader(HDR_CONTENT_LENGTH));
  847. assertNull("not chunked",
  848. service.getRequestHeader(HDR_TRANSFER_ENCODING));
  849. assertEquals(200, service.getStatus());
  850. assertEquals("application/x-git-upload-pack-result",
  851. service.getResponseHeader(HDR_CONTENT_TYPE));
  852. }
  853. }
  854. @Test
  855. public void testInitialClone_WithAuthentication() throws Exception {
  856. try (Repository dst = createBareRepository();
  857. Transport t = Transport.open(dst, authURI)) {
  858. assertFalse(dst.getObjectDatabase().has(A_txt));
  859. t.setCredentialsProvider(testCredentials);
  860. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  861. assertTrue(dst.getObjectDatabase().has(A_txt));
  862. assertEquals(B, dst.exactRef(master).getObjectId());
  863. fsck(dst, B);
  864. }
  865. List<AccessEvent> requests = getRequests();
  866. assertEquals(enableProtocolV2 ? 4 : 3, requests.size());
  867. AccessEvent info = requests.get(0);
  868. assertEquals("GET", info.getMethod());
  869. assertEquals(401, info.getStatus());
  870. assertFetchRequests(requests, 1);
  871. }
  872. @Test
  873. public void testInitialClone_WithPreAuthentication() throws Exception {
  874. try (Repository dst = createBareRepository();
  875. Transport t = Transport.open(dst, authURI)) {
  876. assertFalse(dst.getObjectDatabase().has(A_txt));
  877. ((TransportHttp) t).setPreemptiveBasicAuthentication(
  878. AppServer.username, AppServer.password);
  879. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  880. assertTrue(dst.getObjectDatabase().has(A_txt));
  881. assertEquals(B, dst.exactRef(master).getObjectId());
  882. fsck(dst, B);
  883. }
  884. List<AccessEvent> requests = getRequests();
  885. assertEquals(enableProtocolV2 ? 3 : 2, requests.size());
  886. assertFetchRequests(requests, 0);
  887. }
  888. @Test
  889. public void testInitialClone_WithPreAuthenticationCleared()
  890. throws Exception {
  891. try (Repository dst = createBareRepository();
  892. Transport t = Transport.open(dst, authURI)) {
  893. assertFalse(dst.getObjectDatabase().has(A_txt));
  894. ((TransportHttp) t).setPreemptiveBasicAuthentication(
  895. AppServer.username, AppServer.password);
  896. ((TransportHttp) t).setPreemptiveBasicAuthentication(null, null);
  897. t.setCredentialsProvider(testCredentials);
  898. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  899. assertTrue(dst.getObjectDatabase().has(A_txt));
  900. assertEquals(B, dst.exactRef(master).getObjectId());
  901. fsck(dst, B);
  902. }
  903. List<AccessEvent> requests = getRequests();
  904. assertEquals(enableProtocolV2 ? 4 : 3, requests.size());
  905. AccessEvent info = requests.get(0);
  906. assertEquals("GET", info.getMethod());
  907. assertEquals(401, info.getStatus());
  908. assertFetchRequests(requests, 1);
  909. }
  910. @Test
  911. public void testInitialClone_PreAuthenticationTooLate() throws Exception {
  912. try (Repository dst = createBareRepository();
  913. Transport t = Transport.open(dst, authURI)) {
  914. assertFalse(dst.getObjectDatabase().has(A_txt));
  915. ((TransportHttp) t).setPreemptiveBasicAuthentication(
  916. AppServer.username, AppServer.password);
  917. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  918. assertTrue(dst.getObjectDatabase().has(A_txt));
  919. assertEquals(B, dst.exactRef(master).getObjectId());
  920. fsck(dst, B);
  921. List<AccessEvent> requests = getRequests();
  922. assertEquals(enableProtocolV2 ? 3 : 2, requests.size());
  923. assertFetchRequests(requests, 0);
  924. assertThrows(IllegalStateException.class,
  925. () -> ((TransportHttp) t).setPreemptiveBasicAuthentication(
  926. AppServer.username, AppServer.password));
  927. assertThrows(IllegalStateException.class, () -> ((TransportHttp) t)
  928. .setPreemptiveBasicAuthentication(null, null));
  929. }
  930. }
  931. @Test
  932. public void testInitialClone_WithWrongPreAuthenticationAndCredentialProvider()
  933. throws Exception {
  934. try (Repository dst = createBareRepository();
  935. Transport t = Transport.open(dst, authURI)) {
  936. assertFalse(dst.getObjectDatabase().has(A_txt));
  937. ((TransportHttp) t).setPreemptiveBasicAuthentication(
  938. AppServer.username, AppServer.password + 'x');
  939. t.setCredentialsProvider(testCredentials);
  940. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  941. assertTrue(dst.getObjectDatabase().has(A_txt));
  942. assertEquals(B, dst.exactRef(master).getObjectId());
  943. fsck(dst, B);
  944. }
  945. List<AccessEvent> requests = getRequests();
  946. assertEquals(enableProtocolV2 ? 4 : 3, requests.size());
  947. AccessEvent info = requests.get(0);
  948. assertEquals("GET", info.getMethod());
  949. assertEquals(401, info.getStatus());
  950. assertFetchRequests(requests, 1);
  951. }
  952. @Test
  953. public void testInitialClone_WithWrongPreAuthentication() throws Exception {
  954. try (Repository dst = createBareRepository();
  955. Transport t = Transport.open(dst, authURI)) {
  956. assertFalse(dst.getObjectDatabase().has(A_txt));
  957. ((TransportHttp) t).setPreemptiveBasicAuthentication(
  958. AppServer.username, AppServer.password + 'x');
  959. TransportException e = assertThrows(TransportException.class,
  960. () -> t.fetch(NullProgressMonitor.INSTANCE,
  961. mirror(master)));
  962. String msg = e.getMessage();
  963. assertTrue("Unexpected exception message: " + msg,
  964. msg.contains("no CredentialsProvider"));
  965. }
  966. List<AccessEvent> requests = getRequests();
  967. assertEquals(1, requests.size());
  968. AccessEvent info = requests.get(0);
  969. assertEquals("GET", info.getMethod());
  970. assertEquals(401, info.getStatus());
  971. }
  972. @Test
  973. public void testInitialClone_WithUserInfo() throws Exception {
  974. URIish withUserInfo = authURI.setUser(AppServer.username)
  975. .setPass(AppServer.password);
  976. try (Repository dst = createBareRepository();
  977. Transport t = Transport.open(dst, withUserInfo)) {
  978. assertFalse(dst.getObjectDatabase().has(A_txt));
  979. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  980. assertTrue(dst.getObjectDatabase().has(A_txt));
  981. assertEquals(B, dst.exactRef(master).getObjectId());
  982. fsck(dst, B);
  983. }
  984. List<AccessEvent> requests = getRequests();
  985. assertEquals(enableProtocolV2 ? 3 : 2, requests.size());
  986. assertFetchRequests(requests, 0);
  987. }
  988. @Test
  989. public void testInitialClone_PreAuthOverridesUserInfo() throws Exception {
  990. URIish withUserInfo = authURI.setUser(AppServer.username)
  991. .setPass(AppServer.password + 'x');
  992. try (Repository dst = createBareRepository();
  993. Transport t = Transport.open(dst, withUserInfo)) {
  994. assertFalse(dst.getObjectDatabase().has(A_txt));
  995. ((TransportHttp) t).setPreemptiveBasicAuthentication(
  996. AppServer.username, AppServer.password);
  997. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  998. assertTrue(dst.getObjectDatabase().has(A_txt));
  999. assertEquals(B, dst.exactRef(master).getObjectId());
  1000. fsck(dst, B);
  1001. }
  1002. List<AccessEvent> requests = getRequests();
  1003. assertEquals(enableProtocolV2 ? 3 : 2, requests.size());
  1004. assertFetchRequests(requests, 0);
  1005. }
  1006. @Test
  1007. public void testInitialClone_WithAuthenticationNoCredentials()
  1008. throws Exception {
  1009. try (Repository dst = createBareRepository();
  1010. Transport t = Transport.open(dst, authURI)) {
  1011. assertFalse(dst.getObjectDatabase().has(A_txt));
  1012. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1013. fail("Should not have succeeded -- no authentication");
  1014. } catch (TransportException e) {
  1015. String msg = e.getMessage();
  1016. assertTrue("Unexpected exception message: " + msg,
  1017. msg.contains("no CredentialsProvider"));
  1018. }
  1019. List<AccessEvent> requests = getRequests();
  1020. assertEquals(1, requests.size());
  1021. AccessEvent info = requests.get(0);
  1022. assertEquals("GET", info.getMethod());
  1023. assertEquals(401, info.getStatus());
  1024. }
  1025. @Test
  1026. public void testInitialClone_WithAuthenticationWrongCredentials()
  1027. throws Exception {
  1028. try (Repository dst = createBareRepository();
  1029. Transport t = Transport.open(dst, authURI)) {
  1030. assertFalse(dst.getObjectDatabase().has(A_txt));
  1031. t.setCredentialsProvider(new UsernamePasswordCredentialsProvider(
  1032. AppServer.username, "wrongpassword"));
  1033. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1034. fail("Should not have succeeded -- wrong password");
  1035. } catch (TransportException e) {
  1036. String msg = e.getMessage();
  1037. assertTrue("Unexpected exception message: " + msg,
  1038. msg.contains("auth"));
  1039. }
  1040. List<AccessEvent> requests = getRequests();
  1041. // Once without authentication plus three re-tries with authentication
  1042. assertEquals(4, requests.size());
  1043. for (AccessEvent event : requests) {
  1044. assertEquals("GET", event.getMethod());
  1045. assertEquals(401, event.getStatus());
  1046. }
  1047. }
  1048. @Test
  1049. public void testInitialClone_WithAuthenticationAfterRedirect()
  1050. throws Exception {
  1051. URIish cloneFrom = extendPath(redirectURI, "/target/auth");
  1052. CredentialsProvider uriSpecificCredentialsProvider = new UsernamePasswordCredentialsProvider(
  1053. "unknown", "none") {
  1054. @Override
  1055. public boolean get(URIish uri, CredentialItem... items)
  1056. throws UnsupportedCredentialItem {
  1057. // Only return the true credentials if the uri path starts with
  1058. // /auth. This ensures that we do provide the correct
  1059. // credentials only for the URi after the redirect, making the
  1060. // test fail if we should be asked for the credentials for the
  1061. // original URI.
  1062. if (uri.getPath().startsWith("/auth")) {
  1063. return testCredentials.get(uri, items);
  1064. }
  1065. return super.get(uri, items);
  1066. }
  1067. };
  1068. try (Repository dst = createBareRepository();
  1069. Transport t = Transport.open(dst, cloneFrom)) {
  1070. assertFalse(dst.getObjectDatabase().has(A_txt));
  1071. t.setCredentialsProvider(uriSpecificCredentialsProvider);
  1072. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1073. assertTrue(dst.getObjectDatabase().has(A_txt));
  1074. assertEquals(B, dst.exactRef(master).getObjectId());
  1075. fsck(dst, B);
  1076. }
  1077. List<AccessEvent> requests = getRequests();
  1078. assertEquals(enableProtocolV2 ? 5 : 4, requests.size());
  1079. int requestNumber = 0;
  1080. AccessEvent redirect = requests.get(requestNumber++);
  1081. assertEquals("GET", redirect.getMethod());
  1082. assertEquals(join(cloneFrom, "info/refs"), redirect.getPath());
  1083. assertEquals(301, redirect.getStatus());
  1084. AccessEvent info = requests.get(requestNumber++);
  1085. assertEquals("GET", info.getMethod());
  1086. assertEquals(join(authURI, "info/refs"), info.getPath());
  1087. assertEquals(401, info.getStatus());
  1088. info = requests.get(requestNumber++);
  1089. assertEquals("GET", info.getMethod());
  1090. assertEquals(join(authURI, "info/refs"), info.getPath());
  1091. assertEquals(1, info.getParameters().size());
  1092. assertEquals("git-upload-pack", info.getParameter("service"));
  1093. assertEquals(200, info.getStatus());
  1094. assertEquals("application/x-git-upload-pack-advertisement",
  1095. info.getResponseHeader(HDR_CONTENT_TYPE));
  1096. if (!enableProtocolV2) {
  1097. assertEquals("gzip", info.getResponseHeader(HDR_CONTENT_ENCODING));
  1098. } else {
  1099. AccessEvent lsRefs = requests.get(requestNumber++);
  1100. assertEquals("POST", lsRefs.getMethod());
  1101. assertEquals(join(authURI, "git-upload-pack"), lsRefs.getPath());
  1102. assertEquals(0, lsRefs.getParameters().size());
  1103. assertNotNull("has content-length",
  1104. lsRefs.getRequestHeader(HDR_CONTENT_LENGTH));
  1105. assertNull("not chunked",
  1106. lsRefs.getRequestHeader(HDR_TRANSFER_ENCODING));
  1107. assertEquals("version=2", lsRefs.getRequestHeader("Git-Protocol"));
  1108. assertEquals(200, lsRefs.getStatus());
  1109. assertEquals("application/x-git-upload-pack-result",
  1110. lsRefs.getResponseHeader(HDR_CONTENT_TYPE));
  1111. }
  1112. AccessEvent service = requests.get(requestNumber);
  1113. assertEquals("POST", service.getMethod());
  1114. assertEquals(join(authURI, "git-upload-pack"), service.getPath());
  1115. assertEquals(0, service.getParameters().size());
  1116. assertNotNull("has content-length",
  1117. service.getRequestHeader(HDR_CONTENT_LENGTH));
  1118. assertNull("not chunked",
  1119. service.getRequestHeader(HDR_TRANSFER_ENCODING));
  1120. assertEquals(200, service.getStatus());
  1121. assertEquals("application/x-git-upload-pack-result",
  1122. service.getResponseHeader(HDR_CONTENT_TYPE));
  1123. }
  1124. @Test
  1125. public void testInitialClone_WithAuthenticationOnPostOnly()
  1126. throws Exception {
  1127. try (Repository dst = createBareRepository();
  1128. Transport t = Transport.open(dst, authOnPostURI)) {
  1129. assertFalse(dst.getObjectDatabase().has(A_txt));
  1130. t.setCredentialsProvider(testCredentials);
  1131. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1132. assertTrue(dst.getObjectDatabase().has(A_txt));
  1133. assertEquals(B, dst.exactRef(master).getObjectId());
  1134. fsck(dst, B);
  1135. }
  1136. List<AccessEvent> requests = getRequests();
  1137. assertEquals(enableProtocolV2 ? 4 : 3, requests.size());
  1138. AccessEvent info = requests.get(0);
  1139. assertEquals("GET", info.getMethod());
  1140. assertEquals(join(authOnPostURI, "info/refs"), info.getPath());
  1141. assertEquals(1, info.getParameters().size());
  1142. assertEquals("git-upload-pack", info.getParameter("service"));
  1143. assertEquals(200, info.getStatus());
  1144. assertEquals("application/x-git-upload-pack-advertisement",
  1145. info.getResponseHeader(HDR_CONTENT_TYPE));
  1146. if (!enableProtocolV2) {
  1147. assertEquals("gzip", info.getResponseHeader(HDR_CONTENT_ENCODING));
  1148. }
  1149. AccessEvent service = requests.get(1);
  1150. assertEquals("POST", service.getMethod());
  1151. assertEquals(join(authOnPostURI, "git-upload-pack"), service.getPath());
  1152. assertEquals(401, service.getStatus());
  1153. for (int i = 2; i < requests.size(); i++) {
  1154. service = requests.get(i);
  1155. assertEquals("POST", service.getMethod());
  1156. assertEquals(join(authOnPostURI, "git-upload-pack"),
  1157. service.getPath());
  1158. assertEquals(0, service.getParameters().size());
  1159. assertNotNull("has content-length",
  1160. service.getRequestHeader(HDR_CONTENT_LENGTH));
  1161. assertNull("not chunked",
  1162. service.getRequestHeader(HDR_TRANSFER_ENCODING));
  1163. assertEquals(200, service.getStatus());
  1164. assertEquals("application/x-git-upload-pack-result",
  1165. service.getResponseHeader(HDR_CONTENT_TYPE));
  1166. }
  1167. }
  1168. @Test
  1169. public void testFetch_FewLocalCommits() throws Exception {
  1170. // Bootstrap by doing the clone.
  1171. //
  1172. TestRepository dst = createTestRepository();
  1173. try (Transport t = Transport.open(dst.getRepository(), remoteURI)) {
  1174. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1175. }
  1176. assertEquals(B, dst.getRepository().exactRef(master).getObjectId());
  1177. List<AccessEvent> cloneRequests = getRequests();
  1178. // Only create a few new commits.
  1179. TestRepository.BranchBuilder b = dst.branch(master);
  1180. for (int i = 0; i < 4; i++)
  1181. b.commit().tick(3600 /* 1 hour */).message("c" + i).create();
  1182. // Create a new commit on the remote.
  1183. //
  1184. RevCommit Z;
  1185. try (TestRepository<Repository> tr = new TestRepository<>(
  1186. remoteRepository)) {
  1187. b = tr.branch(master);
  1188. Z = b.commit().message("Z").create();
  1189. }
  1190. // Now incrementally update.
  1191. //
  1192. try (Transport t = Transport.open(dst.getRepository(), remoteURI)) {
  1193. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1194. }
  1195. assertEquals(Z, dst.getRepository().exactRef(master).getObjectId());
  1196. List<AccessEvent> requests = getRequests();
  1197. requests.removeAll(cloneRequests);
  1198. assertEquals(enableProtocolV2 ? 3 : 2, requests.size());
  1199. int requestNumber = 0;
  1200. AccessEvent info = requests.get(requestNumber++);
  1201. assertEquals("GET", info.getMethod());
  1202. assertEquals(join(remoteURI, "info/refs"), info.getPath());
  1203. assertEquals(1, info.getParameters().size());
  1204. assertEquals("git-upload-pack", info.getParameter("service"));
  1205. assertEquals(200, info.getStatus());
  1206. assertEquals("application/x-git-upload-pack-advertisement",
  1207. info.getResponseHeader(HDR_CONTENT_TYPE));
  1208. if (enableProtocolV2) {
  1209. AccessEvent lsRefs = requests.get(requestNumber++);
  1210. assertEquals("POST", lsRefs.getMethod());
  1211. assertEquals(join(remoteURI, "git-upload-pack"), lsRefs.getPath());
  1212. assertEquals(0, lsRefs.getParameters().size());
  1213. assertNotNull("has content-length",
  1214. lsRefs.getRequestHeader(HDR_CONTENT_LENGTH));
  1215. assertNull("not chunked",
  1216. lsRefs.getRequestHeader(HDR_TRANSFER_ENCODING));
  1217. assertEquals("version=2", lsRefs.getRequestHeader("Git-Protocol"));
  1218. assertEquals(200, lsRefs.getStatus());
  1219. assertEquals("application/x-git-upload-pack-result",
  1220. lsRefs.getResponseHeader(HDR_CONTENT_TYPE));
  1221. }
  1222. // We should have needed one request to perform the fetch.
  1223. //
  1224. AccessEvent service = requests.get(requestNumber);
  1225. assertEquals("POST", service.getMethod());
  1226. assertEquals(join(remoteURI, "git-upload-pack"), service.getPath());
  1227. assertEquals(0, service.getParameters().size());
  1228. assertNotNull("has content-length",
  1229. service.getRequestHeader(HDR_CONTENT_LENGTH));
  1230. assertNull("not chunked",
  1231. service.getRequestHeader(HDR_TRANSFER_ENCODING));
  1232. assertEquals(200, service.getStatus());
  1233. assertEquals("application/x-git-upload-pack-result",
  1234. service.getResponseHeader(HDR_CONTENT_TYPE));
  1235. }
  1236. @Test
  1237. public void testFetch_TooManyLocalCommits() throws Exception {
  1238. // Bootstrap by doing the clone.
  1239. //
  1240. TestRepository dst = createTestRepository();
  1241. try (Transport t = Transport.open(dst.getRepository(), remoteURI)) {
  1242. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1243. }
  1244. assertEquals(B, dst.getRepository().exactRef(master).getObjectId());
  1245. List<AccessEvent> cloneRequests = getRequests();
  1246. // Force enough into the local client that enumeration will
  1247. // need multiple packets, but not too many to overflow and
  1248. // not pick up the ACK_COMMON message.
  1249. //
  1250. TestRepository.BranchBuilder b = dst.branch(master);
  1251. for (int i = 0; i < 32 - 1; i++)
  1252. b.commit().tick(3600 /* 1 hour */).message("c" + i).create();
  1253. // Create a new commit on the remote.
  1254. //
  1255. RevCommit Z;
  1256. try (TestRepository<Repository> tr = new TestRepository<>(
  1257. remoteRepository)) {
  1258. b = tr.branch(master);
  1259. Z = b.commit().message("Z").create();
  1260. }
  1261. // Now incrementally update.
  1262. //
  1263. try (Transport t = Transport.open(dst.getRepository(), remoteURI)) {
  1264. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1265. }
  1266. assertEquals(Z, dst.getRepository().exactRef(master).getObjectId());
  1267. List<AccessEvent> requests = getRequests();
  1268. requests.removeAll(cloneRequests);
  1269. assertEquals(enableProtocolV2 ? 4 : 3, requests.size());
  1270. int requestNumber = 0;
  1271. AccessEvent info = requests.get(requestNumber++);
  1272. assertEquals("GET", info.getMethod());
  1273. assertEquals(join(remoteURI, "info/refs"), info.getPath());
  1274. assertEquals(1, info.getParameters().size());
  1275. assertEquals("git-upload-pack", info.getParameter("service"));
  1276. assertEquals(200, info.getStatus());
  1277. assertEquals("application/x-git-upload-pack-advertisement", info
  1278. .getResponseHeader(HDR_CONTENT_TYPE));
  1279. if (enableProtocolV2) {
  1280. AccessEvent lsRefs = requests.get(requestNumber++);
  1281. assertEquals("POST", lsRefs.getMethod());
  1282. assertEquals(join(remoteURI, "git-upload-pack"), lsRefs.getPath());
  1283. assertEquals(0, lsRefs.getParameters().size());
  1284. assertNotNull("has content-length",
  1285. lsRefs.getRequestHeader(HDR_CONTENT_LENGTH));
  1286. assertNull("not chunked",
  1287. lsRefs.getRequestHeader(HDR_TRANSFER_ENCODING));
  1288. assertEquals("version=2", lsRefs.getRequestHeader("Git-Protocol"));
  1289. assertEquals(200, lsRefs.getStatus());
  1290. assertEquals("application/x-git-upload-pack-result",
  1291. lsRefs.getResponseHeader(HDR_CONTENT_TYPE));
  1292. }
  1293. // We should have needed two requests to perform the fetch
  1294. // due to the high number of local unknown commits.
  1295. //
  1296. AccessEvent service = requests.get(requestNumber++);
  1297. assertEquals("POST", service.getMethod());
  1298. assertEquals(join(remoteURI, "git-upload-pack"), service.getPath());
  1299. assertEquals(0, service.getParameters().size());
  1300. assertNotNull("has content-length", service
  1301. .getRequestHeader(HDR_CONTENT_LENGTH));
  1302. assertNull("not chunked", service
  1303. .getRequestHeader(HDR_TRANSFER_ENCODING));
  1304. assertEquals(200, service.getStatus());
  1305. assertEquals("application/x-git-upload-pack-result", service
  1306. .getResponseHeader(HDR_CONTENT_TYPE));
  1307. service = requests.get(requestNumber);
  1308. assertEquals("POST", service.getMethod());
  1309. assertEquals(join(remoteURI, "git-upload-pack"), service.getPath());
  1310. assertEquals(0, service.getParameters().size());
  1311. assertNotNull("has content-length", service
  1312. .getRequestHeader(HDR_CONTENT_LENGTH));
  1313. assertNull("not chunked", service
  1314. .getRequestHeader(HDR_TRANSFER_ENCODING));
  1315. assertEquals(200, service.getStatus());
  1316. assertEquals("application/x-git-upload-pack-result", service
  1317. .getResponseHeader(HDR_CONTENT_TYPE));
  1318. }
  1319. @Test
  1320. public void testFetch_MaxHavesCutoffAfterAckOnly() throws Exception {
  1321. // Bootstrap by doing the clone.
  1322. //
  1323. TestRepository dst = createTestRepository();
  1324. try (Transport t = Transport.open(dst.getRepository(), remoteURI)) {
  1325. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1326. }
  1327. assertEquals(B, dst.getRepository().exactRef(master).getObjectId());
  1328. // Force enough into the local client that enumeration will
  1329. // need more than MAX_HAVES (256) haves to be sent. The server
  1330. // doesn't know any of these, so it will never ACK. The client
  1331. // should keep going.
  1332. //
  1333. // If it does, client and server will find a common commit, and
  1334. // the pack file will contain exactly the one commit object Z
  1335. // we create on the remote, which we can test for via the progress
  1336. // monitor, which should have something like
  1337. // "Receiving objects: 100% (1/1)". If the client sends a "done"
  1338. // too early, the server will send more objects, and we'll have
  1339. // a line like "Receiving objects: 100% (8/8)".
  1340. TestRepository.BranchBuilder b = dst.branch(master);
  1341. // The client will send 32 + 64 + 128 + 256 + 512 haves. Only the
  1342. // last one will be a common commit. If the cutoff kicks in too
  1343. // early (after 480), we'll get too many objects in the fetch.
  1344. for (int i = 0; i < 992; i++)
  1345. b.commit().tick(3600 /* 1 hour */).message("c" + i).create();
  1346. // Create a new commit on the remote.
  1347. //
  1348. RevCommit Z;
  1349. try (TestRepository<Repository> tr = new TestRepository<>(
  1350. remoteRepository)) {
  1351. b = tr.branch(master);
  1352. Z = b.commit().message("Z").create();
  1353. }
  1354. // Now incrementally update.
  1355. //
  1356. ByteArrayOutputStream buffer = new ByteArrayOutputStream();
  1357. Writer writer = new OutputStreamWriter(buffer, StandardCharsets.UTF_8);
  1358. TextProgressMonitor monitor = new TextProgressMonitor(writer);
  1359. try (Transport t = Transport.open(dst.getRepository(), remoteURI)) {
  1360. t.fetch(monitor, mirror(master));
  1361. }
  1362. assertEquals(Z, dst.getRepository().exactRef(master).getObjectId());
  1363. String progressMessages = new String(buffer.toByteArray(),
  1364. StandardCharsets.UTF_8);
  1365. Pattern expected = Pattern
  1366. .compile("Receiving objects:\\s+100% \\(1/1\\)\n");
  1367. if (!expected.matcher(progressMessages).find()) {
  1368. System.out.println(progressMessages);
  1369. fail("Expected only one object to be sent");
  1370. }
  1371. }
  1372. @Test
  1373. public void testInitialClone_BrokenServer() throws Exception {
  1374. try (Repository dst = createBareRepository();
  1375. Transport t = Transport.open(dst, brokenURI)) {
  1376. assertFalse(dst.getObjectDatabase().has(A_txt));
  1377. try {
  1378. t.fetch(NullProgressMonitor.INSTANCE, mirror(master));
  1379. fail("fetch completed despite upload-pack being broken");
  1380. } catch (TransportException err) {
  1381. String exp = brokenURI + ": expected"
  1382. + " Content-Type application/x-git-upload-pack-result;"
  1383. + " received Content-Type text/plain;charset=utf-8";
  1384. assertEquals(exp, err.getMessage());
  1385. }
  1386. }
  1387. List<AccessEvent> requests = getRequests();
  1388. assertEquals(2, requests.size());
  1389. AccessEvent info = requests.get(0);
  1390. assertEquals("GET", info.getMethod());
  1391. assertEquals(join(brokenURI, "info/refs"), info.getPath());
  1392. assertEquals(1, info.getParameters().size());
  1393. assertEquals("git-upload-pack", info.getParameter("service"));
  1394. assertEquals(200, info.getStatus());
  1395. assertEquals("application/x-git-upload-pack-advertisement", info
  1396. .getResponseHeader(HDR_CONTENT_TYPE));
  1397. AccessEvent service = requests.get(1);
  1398. assertEquals("POST", service.getMethod());
  1399. assertEquals(join(brokenURI, "git-upload-pack"), service.getPath());
  1400. assertEquals(0, service.getParameters().size());
  1401. assertEquals(200, service.getStatus());
  1402. assertEquals("text/plain;charset=utf-8",
  1403. service.getResponseHeader(HDR_CONTENT_TYPE));
  1404. }
  1405. @Test
  1406. public void testInvalidWant() throws Exception {
  1407. ObjectId id;
  1408. try (ObjectInserter.Formatter formatter = new ObjectInserter.Formatter()) {
  1409. id = formatter.idFor(Constants.OBJ_BLOB,
  1410. "testInvalidWant".getBytes(UTF_8));
  1411. }
  1412. try (Repository dst = createBareRepository();
  1413. Transport t = Transport.open(dst, remoteURI);
  1414. FetchConnection c = t.openFetch()) {
  1415. Ref want = new ObjectIdRef.Unpeeled(Ref.Storage.NETWORK, id.name(),
  1416. id);
  1417. c.fetch(NullProgressMonitor.INSTANCE, Collections.singleton(want),
  1418. Collections.<ObjectId> emptySet());
  1419. fail("Server accepted want " + id.name());
  1420. } catch (TransportException err) {
  1421. assertTrue(err.getMessage()
  1422. .contains("want " + id.name() + " not valid"));
  1423. }
  1424. }
  1425. @Test
  1426. public void testFetch_RefsUnreadableOnUpload() throws Exception {
  1427. AppServer noRefServer = new AppServer();
  1428. try {
  1429. final String repoName = "refs-unreadable";
  1430. RefsUnreadableInMemoryRepository badRefsRepo = new RefsUnreadableInMemoryRepository(
  1431. new DfsRepositoryDescription(repoName));
  1432. final TestRepository<Repository> repo = new TestRepository<>(
  1433. badRefsRepo);
  1434. badRefsRepo.getConfig().setInt("protocol", null, "version",
  1435. enableProtocolV2 ? 2 : 0);
  1436. ServletContextHandler app = noRefServer.addContext("/git");
  1437. GitServlet gs = new GitServlet();
  1438. gs.setRepositoryResolver(new TestRepositoryResolver(repo, repoName));
  1439. app.addServlet(new ServletHolder(gs), "/*");
  1440. noRefServer.setUp();
  1441. RevBlob A2_txt = repo.blob("A2");
  1442. RevCommit A2 = repo.commit().add("A2_txt", A2_txt).create();
  1443. RevCommit B2 = repo.commit().parent(A2).add("A2_txt", "C2")
  1444. .add("B2", "B2").create();
  1445. repo.update(master, B2);
  1446. URIish badRefsURI = new URIish(noRefServer.getURI()
  1447. .resolve(app.getContextPath() + "/" + repoName).toString());
  1448. try (Repository dst = createBareRepository();
  1449. Transport t = Transport.open(dst, badRefsURI);
  1450. FetchConnection c = t.openFetch()) {
  1451. // We start failing here to exercise the post-advertisement
  1452. // upload pack handler.
  1453. badRefsRepo.startFailing();
  1454. // Need to flush caches because ref advertisement populated them.
  1455. badRefsRepo.getRefDatabase().refresh();
  1456. c.fetch(NullProgressMonitor.INSTANCE,
  1457. Collections.singleton(c.getRef(master)),
  1458. Collections.<ObjectId> emptySet());
  1459. fail("Successfully served ref with value " + c.getRef(master));
  1460. } catch (TransportException err) {
  1461. assertTrue("Unexpected exception message " + err.getMessage(),
  1462. err.getMessage().contains("Internal server error"));
  1463. }
  1464. } finally {
  1465. noRefServer.tearDown();
  1466. }
  1467. }
  1468. @Test
  1469. public void testPush_NotAuthorized() throws Exception {
  1470. final TestRepository src = createTestRepository();
  1471. final RevBlob Q_txt = src.blob("new text");
  1472. final RevCommit Q = src.commit().add("Q", Q_txt).create();
  1473. final Repository db = src.getRepository();
  1474. final String dstName = Constants.R_HEADS + "new.branch";
  1475. // push anonymous shouldn't be allowed.
  1476. //
  1477. try (Transport t = Transport.open(db, remoteURI)) {
  1478. final String srcExpr = Q.name();
  1479. final boolean forceUpdate = false;
  1480. final String localName = null;
  1481. final ObjectId oldId = null;
  1482. RemoteRefUpdate u = new RemoteRefUpdate(src.getRepository(),
  1483. srcExpr, dstName, forceUpdate, localName, oldId);
  1484. try {
  1485. t.push(NullProgressMonitor.INSTANCE, Collections.singleton(u));
  1486. fail("anonymous push incorrectly accepted without error");
  1487. } catch (TransportException e) {
  1488. final String exp = remoteURI + ": "
  1489. + JGitText.get().authenticationNotSupported;
  1490. assertEquals(exp, e.getMessage());
  1491. }
  1492. }
  1493. List<AccessEvent> requests = getRequests();
  1494. assertEquals(1, requests.size());
  1495. AccessEvent info = requests.get(0);
  1496. assertEquals("GET", info.getMethod());
  1497. assertEquals(join(remoteURI, "info/refs"), info.getPath());
  1498. assertEquals(1, info.getParameters().size());
  1499. assertEquals("git-receive-pack", info.getParameter("service"));
  1500. assertEquals(401, info.getStatus());
  1501. }
  1502. @Test
  1503. public void testPush_CreateBranch() throws Exception {
  1504. final TestRepository src = createTestRepository();
  1505. final RevBlob Q_txt = src.blob("new text");
  1506. final RevCommit Q = src.commit().add("Q", Q_txt).create();
  1507. final Repository db = src.getRepository();
  1508. final String dstName = Constants.R_HEADS + "new.branch";
  1509. enableReceivePack();
  1510. try (Transport t = Transport.open(db, remoteURI)) {
  1511. final String srcExpr = Q.name();
  1512. final boolean forceUpdate = false;
  1513. final String localName = null;
  1514. final ObjectId oldId = null;
  1515. RemoteRefUpdate u = new RemoteRefUpdate(src.getRepository(),
  1516. srcExpr, dstName, forceUpdate, localName, oldId);
  1517. t.push(NullProgressMonitor.INSTANCE, Collections.singleton(u));
  1518. }
  1519. assertTrue(remoteRepository.getObjectDatabase().has(Q_txt));
  1520. assertNotNull("has " + dstName, remoteRepository.exactRef(dstName));
  1521. assertEquals(Q, remoteRepository.exactRef(dstName).getObjectId());
  1522. fsck(remoteRepository, Q);
  1523. final ReflogReader log = remoteRepository.getReflogReader(dstName);
  1524. assertNotNull("has log for " + dstName, log);
  1525. final ReflogEntry last = log.getLastEntry();
  1526. assertNotNull("has last entry", last);
  1527. assertEquals(ObjectId.zeroId(), last.getOldId());
  1528. assertEquals(Q, last.getNewId());
  1529. assertEquals("anonymous", last.getWho().getName());
  1530. // Assumption: The host name we use to contact the server should
  1531. // be the server's own host name, because it should be the loopback
  1532. // network interface.
  1533. //
  1534. final String clientHost = remoteURI.getHost();
  1535. assertEquals("anonymous@" + clientHost, last.getWho().getEmailAddress());
  1536. assertEquals("push: created", last.getComment());
  1537. List<AccessEvent> requests = getRequests();
  1538. assertEquals(2, requests.size());
  1539. AccessEvent info = requests.get(0);
  1540. assertEquals("GET", info.getMethod());
  1541. assertEquals(join(remoteURI, "info/refs"), info.getPath());
  1542. assertEquals(1, info.getParameters().size());
  1543. assertEquals("git-receive-pack", info.getParameter("service"));
  1544. assertEquals(200, info.getStatus());
  1545. assertEquals("application/x-git-receive-pack-advertisement", info
  1546. .getResponseHeader(HDR_CONTENT_TYPE));
  1547. AccessEvent service = requests.get(1);
  1548. assertEquals("POST", service.getMethod());
  1549. assertEquals(join(remoteURI, "git-receive-pack"), service.getPath());
  1550. assertEquals(0, service.getParameters().size());
  1551. assertNotNull("has content-length", service
  1552. .getRequestHeader(HDR_CONTENT_LENGTH));
  1553. assertNull("not chunked", service
  1554. .getRequestHeader(HDR_TRANSFER_ENCODING));
  1555. assertEquals(200, service.getStatus());
  1556. assertEquals("application/x-git-receive-pack-result", service
  1557. .getResponseHeader(HDR_CONTENT_TYPE));
  1558. }
  1559. @Test
  1560. public void testPush_ChunkedEncoding() throws Exception {
  1561. final TestRepository<Repository> src = createTestRepository();
  1562. final RevBlob Q_bin = src.blob(new TestRng("Q").nextBytes(128 * 1024));
  1563. final RevCommit Q = src.commit().add("Q", Q_bin).create();
  1564. final Repository db = src.getRepository();
  1565. final String dstName = Constants.R_HEADS + "new.branch";
  1566. enableReceivePack();
  1567. final StoredConfig cfg = db.getConfig();
  1568. cfg.setInt("core", null, "compression", 0);
  1569. cfg.setInt("http", null, "postbuffer", 8 * 1024);
  1570. cfg.save();
  1571. try (Transport t = Transport.open(db, remoteURI)) {
  1572. final String srcExpr = Q.name();
  1573. final boolean forceUpdate = false;
  1574. final String localName = null;
  1575. final ObjectId oldId = null;
  1576. RemoteRefUpdate u = new RemoteRefUpdate(src.getRepository(),
  1577. srcExpr, dstName, forceUpdate, localName, oldId);
  1578. t.push(NullProgressMonitor.INSTANCE, Collections.singleton(u));
  1579. }
  1580. assertTrue(remoteRepository.getObjectDatabase().has(Q_bin));
  1581. assertNotNull("has " + dstName, remoteRepository.exactRef(dstName));
  1582. assertEquals(Q, remoteRepository.exactRef(dstName).getObjectId());
  1583. fsck(remoteRepository, Q);
  1584. List<AccessEvent> requests = getRequests();
  1585. assertEquals(2, requests.size());
  1586. AccessEvent info = requests.get(0);
  1587. assertEquals("GET", info.getMethod());
  1588. assertEquals(join(remoteURI, "info/refs"), info.getPath());
  1589. assertEquals(1, info.getParameters().size());
  1590. assertEquals("git-receive-pack", info.getParameter("service"));
  1591. assertEquals(200, info.getStatus());
  1592. assertEquals("application/x-git-receive-pack-advertisement", info
  1593. .getResponseHeader(HDR_CONTENT_TYPE));
  1594. AccessEvent service = requests.get(1);
  1595. assertEquals("POST", service.getMethod());
  1596. assertEquals(join(remoteURI, "git-receive-pack"), service.getPath());
  1597. assertEquals(0, service.getParameters().size());
  1598. assertNull("no content-length", service
  1599. .getRequestHeader(HDR_CONTENT_LENGTH));
  1600. assertEquals("chunked", service.getRequestHeader(HDR_TRANSFER_ENCODING));
  1601. assertEquals(200, service.getStatus());
  1602. assertEquals("application/x-git-receive-pack-result", service
  1603. .getResponseHeader(HDR_CONTENT_TYPE));
  1604. }
  1605. private void enableReceivePack() throws IOException {
  1606. final StoredConfig cfg = remoteRepository.getConfig();
  1607. cfg.setBoolean("http", null, "receivepack", true);
  1608. cfg.save();
  1609. }
  1610. }