patchbundle-uri: refresh packed_git if unbundle succeed

From: Xing Xin <xingxin.xx@bytedance.com>

When using the bundle-uri mechanism with a bundle list containing multiple interrelated bundles, we encountered a bug where tips from downloaded bundles were not being discovered, resulting in rather slow clones. This was particularly problematic when employing the heuristic `creationTokens`.

And this is easy to reproduce. Suppose we have a repository with a single branch `main` pointing to commit `A`, firstly we create a base bundle with

  git bundle create base.bundle main

Then let's add a new commit `B` on top of `A`, so that an incremental bundle for `main` can be created with

  git bundle create incr.bundle A..main

Now we can generate a bundle list with the following content:

  [bundle]
      version = 1
      mode = all
      heuristic = creationToken

  [bundle "base"]
      uri = base.bundle
      creationToken = 1

  [bundle "incr"]
      uri = incr.bundle
      creationToken = 2

A fresh clone with the bundle list above would give the expected `refs/bundles/main` pointing at `B` in new repository, in other words we already had everything locally from the bundles, but git would still download everything from server as if we got nothing.

So why the `refs/bundles/main` is not discovered? After some digging I found that:

1. when unbundling a downloaded bundle, a `verify_bundle` is called to
   check its prerequisites if any. The verify procedure would find oids
   so `packed_git` is initialized.

2. after unbundled all bundles, we would enter `do_fetch_pack_v2`,
   during which `mark_complete_and_common_ref` and `mark_tips` would
   find oids with `OBJECT_INFO_QUICK` flag set, so no new packs would be
   enlisted if `packed_git` has already initialized in 1.

Back to the example above, when unbunding `incr.bundle`, `base.pack` is enlisted to `packed_git` bacause of the prerequisites to verify. Then we can not find `B` for negotiation at a latter time bacause `B` exists in `incr.pack` which is not enlisted in `packed_git`.

This commit fixes this by adding a `reprepare_packed_git` call for every successfully unbundled bundle, which ensures to enlist all generated packs from bundle uri. And a set of negotiation related tests are added.

Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
    bundle-uri: refresh packed_git if unbundle succeed

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1730%2Fblanet%2Fxx%2Fbundle-uri-bug-using-bundle-list-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1730/blanet/xx/bundle-uri-bug-using-bundle-list-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/1730

 bundle-uri.c                |   3 +
 t/t5558-clone-bundle-uri.sh | 129 ++++++++++++++++++++++++++++++++++++
 2 files changed, 132 insertions(+)

On Wed, May 15, 2024 at 03:01:09AM +0000, blanet via GitGitGadget wrote:

Long time no see :)

And I assume we do not want it to not use `OBJECT_INFO_QUICK`?

Okay, the explanation feels sensible.

This makes me wonder though. Do we really need to call `reprepare_packed_git()` once for every bundle, or can't we instead call it once at the end once we have fetched all bundles? Reading on.

So what's hidden here is that `unbundle_from_file()` will also try to access the bundle's refs right away. Surprisingly, we do so by calling `refs_update_ref()` with `REF_SKIP_OID_VERIFICATION`, which has the effect that we basically accept arbitrary object IDs here even if we do not know those. That's why we didn't have to `reprepare_packed_git()` before this change.

Now there are two conflicting thoughts here:

  - Either we can now drop `REF_SKIP_OID_VERIFICATION` as the object IDs
    should now be accessible.

  - Or we can avoid calling `reprepare_packed_git()` inside the loop and
    instead call it once after we have fetched all bundles.

The second one feels a bit like premature optimization to me. But the first item does feel like it could help us to catch broken bundles because we wouldn't end up creating refs for objects that neither we nor the bundle have.

Patrick

Patrick Steinhardt <ps@pks.im> writes:

I like the way your thoughts are structured around here.

I do agree that the latter is a wrong approach---we shouldn't be trusting what came from elsewhere over the network without first checking. We should probably be running the "index-pack --fix-thin" the unbundling process runs with also the "--fsck-objects" option if we are not doing so already, and even then, we should make sure that the object we are making our ref point at have everything behind it.

At 2024-05-18 00:14:47, "Junio C Hamano" <gitster@pobox.com> wrote:

Currently `unbundle` and all its callers are not adding "--fsck-objects". There is a param `extra_index_pack_args` for `unbundle` but it is mainly used for progress related options.

Personally I think data from bundles and data received via network should be treated equally. For "fetch-pack" we now have some configs such as "fetch.fsckobjects" and "transfer.fsckobjects" to decide the behavior, these configs are invisible when we are fetching bundles.

So for bundles I think we have some different ways to go:

  - Acknowledge the configs mentioned above and behave like
    "fetch-pack".

  - Add "--fsck-objects" as a default in `unbundle`.

  - In `unbundle_from_file`, pass in "--fsck-objects" as
    `extra_index_pack_args` for `unbundle` so this only affect the
    bundle-uri related process.

What do you think?

Xing Xin

"Xing Xin" <bupt_xingxin@163.com> writes:

Yup, that is not personal ;-) but is universally accepted as a good discipline. In the case of bundle-uri, the bundle came over the network so it is even more true that they should be treated the same.

When fetching over network, transport.c:fetch_refs_via_pack() calls fetch_pack.c:fetch_pack(), which eventually calls get_pack() and the configuration variables are honored there. It appears that the transport layer is unaware of the .fsckobjects configuration knobs.

When fetching from a bundle, transport.c:fetch_refs_from_bundle() calls bundle.c:unbundle(). This function has three callers, i.e. "git bundle unbundle", normal fetching from a bundle, and more recently added bundle-uri codepaths.

I think one reasonable approach to take is to add an extra parameter that takes one of three values: (never, use-config, always), and conditionally add "--fsck-objects" to the command line of the index-pack. Teach "git bundle unbundle" the "--fsck-objects" option so that it can pass 'never' or 'always' from the command line, and pass 'use-config' from the code paths for normal fetching from a budnle and bundle-uri.

To implement use-config, you'd probably need to refactor a small part of fetch-pack.c:get_pack()

	if (fetch_fsck_objects >= 0
	    ? fetch_fsck_objects
	    : transfer_fsck_objects >= 0
	    ? transfer_fsck_objects
	    : 0)
		fsck_objects = 1;

into a public function (to support a caller like unbundle() that comes from sideways, the new function may also need to call fetch_pack_setup() to prime them).

A patch series may take a structure like so:

 * define enum { UNBUNDLE_FSCK_NEVER, UNBUNDLE_FSCK_ALWAYS } in
   bundle.h, have bundle.c:unbundle() accept a new parameter of that
   type, and conditionally add "--fsck-objects" to its call to
   "index-pack".  "git bundle unbundle" can pass 'never' to its
   invocation to unbundle() as an easy way to test it.  For the
   other two callers, we can start by passing 'always'.

 * (optional) teach "git bundle unbundle" a new "--fsck-objects"
   option to allow passing 'always' to its call to unbundle().  With
   that, add tests to feed it a bundle with questionable objects in
   it and make sure that unbundling notices.

 * refactor fetch-pack.c:get_pack() to make the fetch-then-transfer
   configuration logic available to external callers.

 * Add UNBUNDLE_FSCK_USE_CONFIG to the enum, enhance unbundle() to
   react to the value by calling the helper function you introduced
   in the previous step.

Thanks.

At 2024-05-21 01:19:02, "Junio C Hamano" <gitster@pobox.com> wrote:

I just submitted a new series mainly focusing on the unbundle handling during fetches. I would like to submit a new one for teaching "git bundle unbundle" a "--fsck-objects" option after this to make changes more targeted.

I tend to use `UNBUNDLE_FSCK_FOLLOW_FETCH` because this option is only used in fetches, though the current implementation is indeed reading configs.

The new patch series is constructed right as you suggested, thanks a lot for your help.

Xing Xin

At 2024-05-17 13:00:49, "Patrick Steinhardt" <ps@pks.im> wrote:

Glad to see you again here :)

I think so. For clones or fetches without using bundle-uri, we can hardly encounter the case that new packs are added during the negotiation process. So using `OBJECT_INFO_QUICK` can get some performance gain.

You are right! I tried dropping this `REF_SKIP_OID_VERIFICATION` flag and the negotiation works as expected.

After some further digging I find that without `REF_SKIP_OID_VERIFICATION`, both `write_ref_to_lockfile` for files backend and `reftable_be_transaction_prepare` for reftable backend would call `parse_object` to check the oid. `parse_object` can help refresh `packed_git` via `reprepare_packed_git`.

I favor the first approach because a validation on the object IDs we are writing is a safe guard . And the flag itself was designed to be used in testing scenarios.

/*
 * Blindly write an object_id. This is useful for testing data corruption
 * scenarios.
 */
#define REF_SKIP_OID_VERIFICATION (1 << 10)

"blanet via GitGitGadget" <gitgitgadget@gmail.com> writes:

s/a//

So the flow is:
1. `fetch_bundle_list` fetches all the bundles advertised via
`download_bundle_list` to local files.
2. It then calls `unbundle_all_bundles` to unbundle all the bundles.
3. Each bundle is then unbundled using `unbundle_from_file`.
4. Here, we first read the bundle header to get all the prerequisites
for the bundle, this is done in `read_bundle_header`.
5. Then we call `unbundle`, which calls `verify_bundle` to ensure that
the repository does indeed contain the prerequisites mentioned in the
bundle. Then it creates the index from the bundle file.

So because the objects are being checked, the `prepare_packed_git` function is eventually called, which means that the `raw_object_store->packed_git` data gets filled in and `packed_git_initialized` is set.

This means consecutive calls to `prepare_packed_git` doesn't re-initiate `raw_object_store->packed_git` since `packed_git_initialized` already is set.

So your explanation makes sense, as a _nit_ I would perhaps add the part about why consecutive calls to `prepare_packed_git` are ineffective.

s/unbundled/unbundling

The solution makes sense.

Would it make sense to move this to `bundle.c:unbundle()`, since that is also where the idx is created?

[snip]

At 2024-05-17 15:36:53, "Karthik Nayak" <karthik.188@gmail.com> wrote:

Thanks!

Thanks my friend, you have expressed this issue more clearly. I will post a new description based on your explanation with the creationToken case covered.

Copy that.

I wonder if we need a mental model that we should `reprepare_packed_git` that when a new pack and its corresponding idx is generated? Currently whether to call `reprepare_packed_git` is determined by the caller.

But within the scope of this bug, I tend to remove the `REF_SKIP_OID_VERIFICATION` flag when writing refs as Patrick suggested.

Xing Xin

From: Xing Xin <xingxin.xx@bytedance.com>

When using the bundle-uri mechanism with a bundle list containing multiple interrelated bundles, we encountered a bug where tips from downloaded bundles were not discovered, thus resulting in rather slow clones. This was particularly problematic when employing the heuristic `creationTokens`.

And this is easy to reproduce. Suppose we have a repository with a single branch `main` pointing to commit `A`, firstly we create a base bundle with

  git bundle create base.bundle main

Then let's add a new commit `B` on top of `A`, so that an incremental bundle for `main` can be created with

  git bundle create incr.bundle A..main

Now we can generate a bundle list with the following content:

  [bundle]
      version = 1
      mode = all
      heuristic = creationToken

  [bundle "base"]
      uri = base.bundle
      creationToken = 1

  [bundle "incr"]
      uri = incr.bundle
      creationToken = 2

A fresh clone with the bundle list above would give the expected `refs/bundles/main` pointing at `B` in new repository, in other words we already had everything locally from the bundles, but git would still download everything from server as if we got nothing.

So why the `refs/bundles/main` is not discovered? After some digging I found that:

1. Bundles in bundle list are downloaded to local files via
   `download_bundle_list` or via `fetch_bundles_by_token` for the
   creationToken heuristic case.
2. Then it tries to unbundle each bundle via `unbundle_from_file`, which
   is called by `unbundle_all_bundles` or called within
   `fetch_bundles_by_token` for the creationToken heuristic case.
3. Here, we first read the bundle header to get all the prerequisites
   for the bundle, this is done in `read_bundle_header`.
4. Then we call `unbundle`, which calls `verify_bundle` to ensure that
   the repository does indeed contain the prerequisites mentioned in the
   bundle.
5. The `verify_bundle` will call `parse_object`, within which the
   `prepare_packed_git` or `reprepare_packed_git` is eventually called,
   which means that the `raw_object_store->packed_git` data gets filled
   in and ``packed_git_initialized` is set. This also means consecutive
   calls to `prepare_packed_git` doesn't re-initiate
   `raw_object_store->packed_git` since `packed_git_initialized` already
   is set.
6. If `unbundle` succeeds, it writes some refs via `refs_update_ref`
   with `REF_SKIP_OID_VERIFICATION` set. So the bundle refs which can
   target arbitrary objects are written to the repository.
7. Finally in `do_fetch_pack_v2`, `mark_complete_and_common_ref` and
   `mark_tips` are called with `OBJECT_INFO_QUICK` set to find local
   tips. Here it won't call `reprepare_packed_git` anymore so it would
   fail to parse oids that only reside in the last bundle.

Back to the example above, when unbunding `incr.bundle`, `base.pack` is enlisted to `packed_git` bacause of the prerequisites to verify. While we can not find `B` for negotiation at a latter time because `B` exists in `incr.pack` which is not enlisted in `packed_git`.

This commit fixes this bug by dropping the `REF_SKIP_OID_VERIFICATION` flag when writing bundle refs, so we can:

1. Ensure that the bundle refs we are writing are pointing to valid
   objects.
2. Ensure all the tips from bundle refs can be correctly parsed.

And a set of negotiation related tests for bundle-uri are added.

Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
    bundle-uri: refresh packed_git if unbundle succeed
    
    cc: Patrick Steinhardt ps@pks.im cc: Karthik Nayak karthik.188@gmail.com

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1730%2Fblanet%2Fxx%2Fbundle-uri-bug-using-bundle-list-v2
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1730/blanet/xx/bundle-uri-bug-using-bundle-list-v2
Pull-Request: https://github.com/gitgitgadget/git/pull/1730

Range-diff vs v1:

 1:  d4841eea556 ! 1:  8bdeacf1360 bundle-uri: refresh packed_git if unbundle succeed
     @@ Metadata
      Author: Xing Xin <xingxin.xx@bytedance.com>
      
       ## Commit message ##
     -    bundle-uri: refresh packed_git if unbundle succeed
     +    bundle-uri: verify oid before writing refs
      
          When using the bundle-uri mechanism with a bundle list containing
          multiple interrelated bundles, we encountered a bug where tips from
     -    downloaded bundles were not being discovered, resulting in rather slow
     +    downloaded bundles were not discovered, thus resulting in rather slow
          clones. This was particularly problematic when employing the heuristic
          `creationTokens`.
      
     @@ Commit message
          So why the `refs/bundles/main` is not discovered? After some digging I
          found that:
      
     -    1. when unbundling a downloaded bundle, a `verify_bundle` is called to
     -       check its prerequisites if any. The verify procedure would find oids
     -       so `packed_git` is initialized.
     -
     -    2. after unbundled all bundles, we would enter `do_fetch_pack_v2`,
     -       during which `mark_complete_and_common_ref` and `mark_tips` would
     -       find oids with `OBJECT_INFO_QUICK` flag set, so no new packs would be
     -       enlisted if `packed_git` has already initialized in 1.
     +    1. Bundles in bundle list are downloaded to local files via
     +       `download_bundle_list` or via `fetch_bundles_by_token` for the
     +       creationToken heuristic case.
     +    2. Then it tries to unbundle each bundle via `unbundle_from_file`, which
     +       is called by `unbundle_all_bundles` or called within
     +       `fetch_bundles_by_token` for the creationToken heuristic case.
     +    3. Here, we first read the bundle header to get all the prerequisites
     +       for the bundle, this is done in `read_bundle_header`.
     +    4. Then we call `unbundle`, which calls `verify_bundle` to ensure that
     +       the repository does indeed contain the prerequisites mentioned in the
     +       bundle.
     +    5. The `verify_bundle` will call `parse_object`, within which the
     +       `prepare_packed_git` or `reprepare_packed_git` is eventually called,
     +       which means that the `raw_object_store->packed_git` data gets filled
     +       in and ``packed_git_initialized` is set. This also means consecutive
     +       calls to `prepare_packed_git` doesn't re-initiate
     +       `raw_object_store->packed_git` since `packed_git_initialized` already
     +       is set.
     +    6. If `unbundle` succeeds, it writes some refs via `refs_update_ref`
     +       with `REF_SKIP_OID_VERIFICATION` set. So the bundle refs which can
     +       target arbitrary objects are written to the repository.
     +    7. Finally in `do_fetch_pack_v2`, `mark_complete_and_common_ref` and
     +       `mark_tips` are called with `OBJECT_INFO_QUICK` set to find local
     +       tips. Here it won't call `reprepare_packed_git` anymore so it would
     +       fail to parse oids that only reside in the last bundle.
      
          Back to the example above, when unbunding `incr.bundle`, `base.pack` is
     -    enlisted to `packed_git` bacause of the prerequisites to verify. Then we
     -    can not find `B` for negotiation at a latter time bacause `B` exists in
     -    `incr.pack` which is not enlisted in `packed_git`.
     +    enlisted to `packed_git` bacause of the prerequisites to verify. While
     +    we can not find `B` for negotiation at a latter time because `B` exists
     +    in `incr.pack` which is not enlisted in `packed_git`.
     +
     +    This commit fixes this bug by dropping the `REF_SKIP_OID_VERIFICATION`
     +    flag when writing bundle refs, so we can:
      
     -    This commit fixes this by adding a `reprepare_packed_git` call for every
     -    successfully unbundled bundle, which ensures to enlist all generated
     -    packs from bundle uri. And a set of negotiation related tests are added.
     +    1. Ensure that the bundle refs we are writing are pointing to valid
     +       objects.
     +    2. Ensure all the tips from bundle refs can be correctly parsed.
     +
     +    And a set of negotiation related tests for bundle-uri are added.
      
          Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
      
       ## bundle-uri.c ##
     -@@
     - #include "refs.h"
     - #include "run-command.h"
     - #include "hashmap.h"
     -+#include "packfile.h"
     - #include "pkt-line.h"
     - #include "config.h"
     - #include "remote.h"
      @@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *file)
     - 			       VERIFY_BUNDLE_QUIET)))
     - 		return 1;
     + 		refs_update_ref(get_main_ref_store(the_repository),
     + 				"fetched bundle", bundle_ref.buf, oid,
     + 				has_old ? &old_oid : NULL,
     +-				REF_SKIP_OID_VERIFICATION,
     +-				UPDATE_REFS_MSG_ON_ERR);
     ++				0, UPDATE_REFS_MSG_ON_ERR);
     + 	}
       
     -+	reprepare_packed_git(r);
     -+
     - 	/*
     - 	 * Convert all refs/heads/ from the bundle into refs/bundles/
     - 	 * in the local repository.
     + 	bundle_header_release(&header);
      
       ## t/t5558-clone-bundle-uri.sh ##
      @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'fail to clone from non-bundle file' '

 bundle-uri.c                |   3 +-
 t/t5558-clone-bundle-uri.sh | 129 ++++++++++++++++++++++++++++++++++++
 2 files changed, 130 insertions(+), 2 deletions(-)

"blanet via GitGitGadget" <gitgitgadget@gmail.com> writes:

s/bacause/because

[snip]

This update looks good to me, Thanks.

While attempting to fix a reference negotiation bug in bundle-uri, we discovered that the fetch process are missing some helpful object validation logic when processing bundles. The main issues are:

 * In the bundle-uri scenario, we did not validate object IDs before writing
   bundle references. This is the root cause of the original negotiation bug
   in bundle-uri, and can cause potential repository corruption.
 * The existing fetch.fsckObjects and transfer.fsckObjects are not detected
   when directly fetching bundles. In fact there is no object validation
   support for unbundle.

The first patch fixes the bundle-uri negotiation issue by dropping the REF_SKIP_OID_VERIFICATION flag when writing bundle references.

Patch 2~4 extend bundle.c:unbundle with a unbundle_fsck_flags to control object fscking in different scenarios, the implementation mainly follows what Junio suggested on the mailing list.

Xing Xin (4):
  bundle-uri: verify oid before writing refs
  unbundle: introduce unbundle_fsck_flags for fsckobjects handling
  fetch-pack: expose fsckObjects configuration logic
  unbundle: introduce new option UNBUNDLE_FSCK_FOLLOW_FETCH

 builtin/bundle.c            |   2 +-
 bundle-uri.c                |   5 +-
 bundle.c                    |  20 ++++-
 bundle.h                    |   9 +-
 fetch-pack.c                |  18 ++--
 fetch-pack.h                |   2 +
 t/t5558-clone-bundle-uri.sh | 163 +++++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     |  23 +++++
 transport.c                 |   2 +-
 9 files changed, 227 insertions(+), 17 deletions(-)

base-commit: b9cfe4845cb2562584837bc0101c0ab76490a239
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1730%2Fblanet%2Fxx%2Fbundle-uri-bug-using-bundle-list-v3
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1730/blanet/xx/bundle-uri-bug-using-bundle-list-v3
Pull-Request: https://github.com/gitgitgadget/git/pull/1730

Range-diff vs v2:

 1:  8bdeacf1360 = 1:  8f488a5eeaa bundle-uri: verify oid before writing refs
 -:  ----------- > 2:  057c697970f unbundle: introduce unbundle_fsck_flags for fsckobjects handling
 -:  ----------- > 3:  67401d4fbcb fetch-pack: expose fsckObjects configuration logic
 -:  ----------- > 4:  c19b8f633cb unbundle: introduce new option UNBUNDLE_FSCK_FOLLOW_FETCH

From: Xing Xin <xingxin.xx@bytedance.com>

When using the bundle-uri mechanism with a bundle list containing multiple interrelated bundles, we encountered a bug where tips from downloaded bundles were not discovered, thus resulting in rather slow clones. This was particularly problematic when employing the heuristic `creationTokens`.

And this is easy to reproduce. Suppose we have a repository with a single branch `main` pointing to commit `A`, firstly we create a base bundle with

  git bundle create base.bundle main

Then let's add a new commit `B` on top of `A`, so that an incremental bundle for `main` can be created with

  git bundle create incr.bundle A..main

Now we can generate a bundle list with the following content:

  [bundle]
      version = 1
      mode = all
      heuristic = creationToken

  [bundle "base"]
      uri = base.bundle
      creationToken = 1

  [bundle "incr"]
      uri = incr.bundle
      creationToken = 2

A fresh clone with the bundle list above would give the expected `refs/bundles/main` pointing at `B` in new repository, in other words we already had everything locally from the bundles, but git would still download everything from server as if we got nothing.

So why the `refs/bundles/main` is not discovered? After some digging I found that:

1. Bundles in bundle list are downloaded to local files via
   `download_bundle_list` or via `fetch_bundles_by_token` for the
   creationToken heuristic case.
2. Then it tries to unbundle each bundle via `unbundle_from_file`, which
   is called by `unbundle_all_bundles` or called within
   `fetch_bundles_by_token` for the creationToken heuristic case.
3. Here, we first read the bundle header to get all the prerequisites
   for the bundle, this is done in `read_bundle_header`.
4. Then we call `unbundle`, which calls `verify_bundle` to ensure that
   the repository does indeed contain the prerequisites mentioned in the
   bundle.
5. The `verify_bundle` will call `parse_object`, within which the
   `prepare_packed_git` or `reprepare_packed_git` is eventually called,
   which means that the `raw_object_store->packed_git` data gets filled
   in and ``packed_git_initialized` is set. This also means consecutive
   calls to `prepare_packed_git` doesn't re-initiate
   `raw_object_store->packed_git` since `packed_git_initialized` already
   is set.
6. If `unbundle` succeeds, it writes some refs via `refs_update_ref`
   with `REF_SKIP_OID_VERIFICATION` set. So the bundle refs which can
   target arbitrary objects are written to the repository.
7. Finally in `do_fetch_pack_v2`, `mark_complete_and_common_ref` and
   `mark_tips` are called with `OBJECT_INFO_QUICK` set to find local
   tips. Here it won't call `reprepare_packed_git` anymore so it would
   fail to parse oids that only reside in the last bundle.

Back to the example above, when unbunding `incr.bundle`, `base.pack` is enlisted to `packed_git` bacause of the prerequisites to verify. While we can not find `B` for negotiation at a latter time because `B` exists in `incr.pack` which is not enlisted in `packed_git`.

This commit fixes this bug by dropping the `REF_SKIP_OID_VERIFICATION` flag when writing bundle refs, so we can:

1. Ensure that the bundle refs we are writing are pointing to valid
   objects.
2. Ensure all the tips from bundle refs can be correctly parsed.

And a set of negotiation related tests for bundle-uri are added.

Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |   3 +-
 t/t5558-clone-bundle-uri.sh | 129 ++++++++++++++++++++++++++++++++++++
 2 files changed, 130 insertions(+), 2 deletions(-)

On Mon, May 27, 2024 at 03:41:54PM +0000, Xing Xin via GitGitGadget wrote:

[snip]

s/``/`/

[snip]

I think one angle that your explanation doesn't cover is why exactly dropping the flag fixes the observed issue.

s/And/Add/

[snip]

As far as I can see there is no test that verifies the case where the bundle contains refs, but misses the objects to satisfy the refs. Can we craft such a bundle and exercise this new failure mode?

Patrick

At 2024-05-28 19:55:15, "Patrick Steinhardt" <ps@pks.im> wrote: [snip]

Copy that.

[snip]

The "And" is correct here, though I changed the commit message in the new version. :)

[snip]

A new test is added in [PATCH v4 1/4], which shows it correctly refuses to target a bundle reference to a non-existent object.

Xing Xin

From: Xing Xin <xingxin.xx@bytedance.com>

This commit adds a new enum `unbundle_fsck_flags` which is designed to control the fsck behavior when unbundling. `unbundle` can use this newly passed in enum to further decide whether to enable `--fsck-objects` for "git-index-pack".

Currently only `UNBUNDLE_FSCK_NEVER` and `UNBUNDLE_FSCK_ALWAYS` are supported as the very basic options. Another interesting option would be added in later commits.

Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 builtin/bundle.c |  2 +-
 bundle-uri.c     |  2 +-
 bundle.c         | 12 +++++++++++-
 bundle.h         |  8 +++++++-
 transport.c      |  2 +-
 5 files changed, 21 insertions(+), 5 deletions(-)

On Mon, May 27, 2024 at 03:41:55PM +0000, Xing Xin via GitGitGadget wrote: [snip]

Wouldn't this have been a natural fit for the new flag, e.g. via something like `VERIFY_BUNDLE_FSCK`?

Patrick

At 2024-05-28 20:03:25, "Patrick Steinhardt" <ps@pks.im> wrote:

It makes sense to me. Currently, verify_bundle_flags controls the amount of information displayed when checking a bundle's prerequisites. The newly added unbundle_fsck_flags is designed to check for broken objects during the unbundle process, which is essentially a form of bundle verification. I believe we should extend some object verification capabilities to the git bundle verify command as well, perhaps by adding a --fsck-objects option.

With this in mind, I support adding new options to verify_bundle_flags. Since bundle.c:unbundle needs to combine multiple options, we must define new options using bitwise shifting:

	enum verify_bundle_flags {
		VERIFY_BUNDLE_VERBOSE = (1 << 0),
		VERIFY_BUNDLE_QUIET = (1 << 1),
		VERIFY_BUNDLE_FSCK_OBJECTS_ALWAYS = (1 << 2),
		VERIFY_BUNDLE_FSCK_OBJECTS_FOLLOW_FETCH = (1 << 3),
	};

How about the naming? I'm not very good at naming :)

Xing Xin

On Thu, May 30, 2024 at 02:12:47AM +0800, Xing Xin wrote:

I later noticed that you extend the `unbundle_fsck_flags` in a later patch. With that in mind I don't think it's all that important anymore to merge those into the `verify_bundle_flags` as you would otherwise allow for weirdness. What happens for example when both `ALWAYS` and `FOLLOW_FETCH` are set?

So feel free to ignore this advice. If you still think it's a good idea then the above naming looks okay to me.

Patrick

At 2024-05-30 12:38:49, "Patrick Steinhardt" <ps@pks.im> wrote: [snip]

With the idea of extending "--fsck-objects" support for "git bundle verify" and "git bundle unbundle", I prefer to grouping these options together. Especially in the "git bundle verify" scenario, adding a new parameter like `unbundle_fsck_flags` for `bundle.c:verify_bundle` is confusing.

Xing Xin

From: Xing Xin <xingxin.xx@bytedance.com>

Currently we can use "transfer.fsckObjects" or "fetch.fsckObjects" to control whether to enable checks for broken objects during fetching. But these configs are only acknowledged by `fetch-pack.c:get_pack` and do not make sense when fetching from bundles or using bundle-uris.

This commit exposed the fetch-then-transfer configuration logic by adding a new function `fetch_pack_fsck_objects` in fetch-pack.h. In next commit, this new function will be used by `unbundle` in fetching scenarios.

Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 fetch-pack.c | 18 ++++++++++++------
 fetch-pack.h |  2 ++
 2 files changed, 14 insertions(+), 6 deletions(-)

On Mon, May 27, 2024 at 03:41:56PM +0000, Xing Xin via GitGitGadget wrote:

Do they not make sense, or are they not effective? I assume you mean the latter, right?

s/exposed/exposes/

This statement is really weird to read, but that is certainly not the fault of this patch, but...

... can we maybe rewrite it to something more customary here? The following is way easier to read, at least for me.

	int fetch_pack_fsck_objects(void)
	{
		fetch_pack_setup();
		if (fetch_fsck_objects >= 0 ||
		    transfer_fsck_objects >= 0)
			return 1;
		return 0;
	}

Let's add a comment here saying what this function does.

Patrick

Patrick Steinhardt <ps@pks.im> writes:

But do they mean the same thing?  In a repository where

    [fetch] fsckobjects = no

is set, no matter what transfer.fsckobjects says (or left unspecified), we want to return "no, we are not doing fsck".

Junio C Hamano <gitster@pobox.com> writes:

The original before it was made into a helper function was written as a cascade of ?: operators, because it had to be a single expression. As the body of a helper function, we now can sprinkle multiple return statements in it. I think the way that is easiest to understand is

	/* the most specific, if specified */
	if (fetch_fsck_objects >= 0)
		return fetch_fsck_objects;
	/* the less specific, catch-all for both directions */
	if (transfer_fsck_objects >= 0)
        	return transfer_fsck_objects;
	/* the fallback hardcoded default */
	return 0;

without the /* comments */.

On Tue, May 28, 2024 at 10:24:35AM -0700, Junio C Hamano wrote:

Ah, right, didn't see this mail. My revised version looks the same as yours, except for the added comments.

Patrick

At 2024-05-29 01:24:35, "Junio C Hamano" <gitster@pobox.com> wrote: [snip]

Applied, thanks!

Xing Xin

On Tue, May 28, 2024 at 10:10:46AM -0700, Junio C Hamano wrote:

Oh, of course they don't. This here would be a faithful conversion:

	int fetch_pack_fsck_objects(void)
	{
		fetch_pack_setup();
		if (fetch_fsck_objects >= 0)
			return fetch_fsck_objects;
		if (transfer_fsck_objects >= 0)
			return transfer_fsck_objects;
		return 0;
	}

Still easier to read in my opinion.

Patrick

From: Xing Xin <xingxin.xx@bytedance.com>

This commit adds a new option `UNBUNDLE_FSCK_FOLLOW_FETCH` to `unbundle_fsck_flags`, this new flag is currently used in the _fetch_ process by

- `transport.c:fetch_refs_from_bundle` for fetching directly from a
  bundle.
- `bundle-uri.c:unbundle_from_file` for unbundling bundles downloaded
  from bundle-uri.

So we now have a relatively consistent logic for checking objects during fetching. Add tests for the above two situations are added.

Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |  2 +-
 bundle.c                    | 10 +++++++++-
 bundle.h                    |  1 +
 t/t5558-clone-bundle-uri.sh | 36 +++++++++++++++++++++++++++++++-----
 t/t5607-clone-bundle.sh     | 23 +++++++++++++++++++++++
 transport.c                 |  2 +-
 6 files changed, 66 insertions(+), 8 deletions(-)

On Mon, May 27, 2024 at 03:41:57PM +0000, Xing Xin via GitGitGadget wrote:

Do we also want to have a test for `transfer.fsckObjects`?

Patrick

At 2024-05-28 20:05:48, "Patrick Steinhardt" <ps@pks.im> wrote: [snip]

Sure, some more combinations of "fetch.fsckObjects" and "transfer.fsckObjects" are added in tests.

Xing Xin

While attempting to fix a reference negotiation bug in bundle-uri, we identified that the fetch process lacks some crucial object validation checks when processing bundles. The primary issues are:

 1. In the bundle-uri scenario, object IDs were not validated before writing
    bundle references. This was the root cause of the original negotiation
    bug in bundle-uri and could lead to potential repository corruption.
 2. The existing "fetch.fsckObjects" and "transfer.fsckObjects"
    configurations were not applied when directly fetching bundles or
    fetching with bundle-uri enabled. In fact, there were no object
    validation supports for unbundle.

The first patch addresses the bundle-uri negotiation issue by removing the REF_SKIP_OID_VERIFICATION flag when writing bundle references.

Patches 2 through 4 extend verify_bundle_flags for bundle.c:unbundle to add support for object validation (fsck) in different scenarios, mainly following the suggestions from Junio on the mailing list.

Xing Xin (4):
  bundle-uri: verify oid before writing refs
  unbundle: extend verify_bundle_flags to support fsck-objects
  fetch-pack: expose fsckObjects configuration logic
  unbundle: introduce option VERIFY_BUNDLE_FSCK_FOLLOW_FETCH

 bundle-uri.c                |   5 +-
 bundle.c                    |  10 ++
 bundle.h                    |   2 +
 fetch-pack.c                |  17 ++--
 fetch-pack.h                |   5 +
 t/t5558-clone-bundle-uri.sh | 186 +++++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     |  33 +++++++
 transport.c                 |   2 +-
 8 files changed, 246 insertions(+), 14 deletions(-)

base-commit: b9cfe4845cb2562584837bc0101c0ab76490a239
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1730%2Fblanet%2Fxx%2Fbundle-uri-bug-using-bundle-list-v4
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1730/blanet/xx/bundle-uri-bug-using-bundle-list-v4
Pull-Request: https://github.com/gitgitgadget/git/pull/1730

Range-diff vs v3:

 1:  8f488a5eeaa ! 1:  e958a3ab20c bundle-uri: verify oid before writing refs
     @@ Commit message
          When using the bundle-uri mechanism with a bundle list containing
          multiple interrelated bundles, we encountered a bug where tips from
          downloaded bundles were not discovered, thus resulting in rather slow
     -    clones. This was particularly problematic when employing the heuristic
     -    `creationTokens`.
     +    clones. This was particularly problematic when employing the
     +    "creationTokens" heuristic.
      
     -    And this is easy to reproduce. Suppose we have a repository with a
     -    single branch `main` pointing to commit `A`, firstly we create a base
     -    bundle with
     +    To reproduce this issue, consider a repository with a single branch
     +    "main" pointing to commit "A". Firstly, create a base bundle with:
      
            git bundle create base.bundle main
      
     -    Then let's add a new commit `B` on top of `A`, so that an incremental
     -    bundle for `main` can be created with
     +    Then, add a new commit "B" on top of "A", and create an incremental
     +    bundle for "main":
      
            git bundle create incr.bundle A..main
      
     -    Now we can generate a bundle list with the following content:
     +    Now, generate a bundle list with the following content:
      
            [bundle]
                version = 1
     @@ Commit message
                uri = incr.bundle
                creationToken = 2
      
     -    A fresh clone with the bundle list above would give the expected
     -    `refs/bundles/main` pointing at `B` in new repository, in other words we
     -    already had everything locally from the bundles, but git would still
     -    download everything from server as if we got nothing.
     +    A fresh clone with the bundle list above should result in a reference
     +    "refs/bundles/main" pointing to "B" in the new repository. However, git
     +    would still download everything from the server, as if it had fetched
     +    nothing locally.
      
     -    So why the `refs/bundles/main` is not discovered? After some digging I
     +    So why the "refs/bundles/main" is not discovered? After some digging I
          found that:
      
          1. Bundles in bundle list are downloaded to local files via
     -       `download_bundle_list` or via `fetch_bundles_by_token` for the
     -       creationToken heuristic case.
     -    2. Then it tries to unbundle each bundle via `unbundle_from_file`, which
     -       is called by `unbundle_all_bundles` or called within
     -       `fetch_bundles_by_token` for the creationToken heuristic case.
     -    3. Here, we first read the bundle header to get all the prerequisites
     -       for the bundle, this is done in `read_bundle_header`.
     -    4. Then we call `unbundle`, which calls `verify_bundle` to ensure that
     -       the repository does indeed contain the prerequisites mentioned in the
     +       `bundle-uri.c:download_bundle_list` or via
     +       `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
     +       heuristic.
     +    2. Each bundle is unbundled via `bundle-uri.c:unbundle_from_file`, which
     +       is called by `bundle-uri.c:unbundle_all_bundles` or called within
     +       `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
     +       heuristic.
     +    3. To get all prerequisites of the bundle, the bundle header is read
     +       inside `bundle-uri.c:unbundle_from_file` to by calling
     +       `bundle.c:read_bundle_header`.
     +    4. Then it calls `bundle.c:unbundle`, which calls
     +       `bundle.c:verify_bundle` to ensure the repository contains all the
     +       prerequisites.
     +    5. `bundle.c:verify_bundle` calls `parse_object`, which eventually
     +       invokes `packfile.c:prepare_packed_git` or
     +       `packfile.c:reprepare_packed_git`, filling
     +       `raw_object_store->packed_git` and setting `packed_git_initialized`.
     +    6. If `bundle.c:unbundle` succeeds, it writes refs via
     +       `refs.c:refs_update_ref` with `REF_SKIP_OID_VERIFICATION` set. Here
     +       bundle refs which can target arbitrary objects are written to the
     +       repository.
     +    7. Finally, in `fetch-pack.c:do_fetch_pack_v2`, the functions
     +       `fetch-pack.c:mark_complete_and_common_ref` and
     +       `fetch-pack.c:mark_tips` are called with `OBJECT_INFO_QUICK` set to
     +       find local tips for negotiation. The `OBJECT_INFO_QUICK` flag
     +       prevents `packfile.c:reprepare_packed_git` from being called,
     +       resulting in failures to parse OIDs that reside only in the latest
             bundle.
     -    5. The `verify_bundle` will call `parse_object`, within which the
     -       `prepare_packed_git` or `reprepare_packed_git` is eventually called,
     -       which means that the `raw_object_store->packed_git` data gets filled
     -       in and ``packed_git_initialized` is set. This also means consecutive
     -       calls to `prepare_packed_git` doesn't re-initiate
     -       `raw_object_store->packed_git` since `packed_git_initialized` already
     -       is set.
     -    6. If `unbundle` succeeds, it writes some refs via `refs_update_ref`
     -       with `REF_SKIP_OID_VERIFICATION` set. So the bundle refs which can
     -       target arbitrary objects are written to the repository.
     -    7. Finally in `do_fetch_pack_v2`, `mark_complete_and_common_ref` and
     -       `mark_tips` are called with `OBJECT_INFO_QUICK` set to find local
     -       tips. Here it won't call `reprepare_packed_git` anymore so it would
     -       fail to parse oids that only reside in the last bundle.
     -
     -    Back to the example above, when unbunding `incr.bundle`, `base.pack` is
     -    enlisted to `packed_git` bacause of the prerequisites to verify. While
     -    we can not find `B` for negotiation at a latter time because `B` exists
     -    in `incr.pack` which is not enlisted in `packed_git`.
     -
     -    This commit fixes this bug by dropping the `REF_SKIP_OID_VERIFICATION`
     -    flag when writing bundle refs, so we can:
     -
     -    1. Ensure that the bundle refs we are writing are pointing to valid
     -       objects.
     -    2. Ensure all the tips from bundle refs can be correctly parsed.
     -
     -    And a set of negotiation related tests for bundle-uri are added.
      
     +    In the example above, when unbunding "incr.bundle", "base.pack" is added
     +    to `packed_git` due to prerequisites verification. However, "B" cannot
     +    be found for negotiation because it exists in "incr.pack", which is not
     +    included in `packed_git`.
     +
     +    This commit fixes the bug by removing `REF_SKIP_OID_VERIFICATION` flag
     +    when writing bundle refs. When `refs.c:refs_update_ref` is called to to
     +    write the corresponding bundle refs, it triggers
     +    `refs.c:ref_transaction_commit`.  This, in turn, invokes
     +    `refs.c:ref_transaction_prepare`, which calls `transaction_prepare` of
     +    the refs storage backend. For files backend, this function is
     +    `files-backend.c:files_transaction_prepare`, and for reftable backend,
     +    it is `reftable-backend.c:reftable_be_transaction_prepare`. Both
     +    functions eventually call `object.c:parse_object`, which can invoke
     +    `packfile.c:reprepare_packed_git` to refresh `packed_git`. This ensures
     +    that bundle refs point to valid objects and that all tips from bundle
     +    refs are correctly parsed during subsequent negotiations.
     +
     +    A test has been added to demonstrate that bundles with incorrect
     +    headers, where refs point to non-existent objects, do not result in any
     +    bundle refs being created in the repository. Additionally, a set of
     +    negotiation-related tests for fetching with bundle-uri has been
     +    included.
     +
     +    Reviewed-by: Karthik Nayak <karthik.188@gmail.com>
     +    Reviewed-by: Patrick Steinhardt <ps@pks.im>
          Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
      
       ## bundle-uri.c ##
     @@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *fi
      
       ## t/t5558-clone-bundle-uri.sh ##
      @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'fail to clone from non-bundle file' '
     + 
       test_expect_success 'create bundle' '
       	git init clone-from &&
     - 	git -C clone-from checkout -b topic &&
     +-	git -C clone-from checkout -b topic &&
     +-	test_commit -C clone-from A &&
     +-	test_commit -C clone-from B &&
     +-	git -C clone-from bundle create B.bundle topic
     ++	(
     ++		cd clone-from &&
     ++		git checkout -b topic &&
     ++
     ++		test_commit A &&
     ++		git bundle create A.bundle topic &&
      +
     - 	test_commit -C clone-from A &&
     -+	git -C clone-from bundle create A.bundle topic &&
     ++		test_commit B &&
     ++		git bundle create B.bundle topic &&
      +
     - 	test_commit -C clone-from B &&
     - 	git -C clone-from bundle create B.bundle topic
     ++		# Create a bundle with reference pointing to non-existent object.
     ++		sed "s/$(git rev-parse A)/$(git rev-parse B)/" <A.bundle >bad-header.bundle
     ++	)
       '
     + 
     + test_expect_success 'clone with path bundle' '
     +@@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone with path bundle' '
     + 	test_cmp expect actual
     + '
     + 
     ++test_expect_success 'clone with bundle that has bad header' '
     ++	git clone --bundle-uri="clone-from/bad-header.bundle" \
     ++		clone-from clone-bad-header 2>err &&
     ++	# Write bundle ref fails, but clone can still proceed.
     ++	commit_b=$(git -C clone-from rev-parse B) &&
     ++	test_grep "trying to write ref '\''refs/bundles/topic'\'' with nonexistent object $commit_b" err &&
     ++	git -C clone-bad-header for-each-ref --format="%(refname)" >refs &&
     ++	! grep "refs/bundles/" refs
     ++'
     ++
     + test_expect_success 'clone with path bundle and non-default hash' '
     + 	test_when_finished "rm -rf clone-path-non-default-hash" &&
     + 	GIT_DEFAULT_HASH=sha256 git clone --bundle-uri="clone-from/B.bundle" \
      @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone bundle list (file, any mode, all failures)' '
       	! grep "refs/bundles/" refs
       '
 2:  057c697970f ! 2:  beb70735811 unbundle: introduce unbundle_fsck_flags for fsckobjects handling
     @@ Metadata
      Author: Xing Xin <xingxin.xx@bytedance.com>
      
       ## Commit message ##
     -    unbundle: introduce unbundle_fsck_flags for fsckobjects handling
     +    unbundle: extend verify_bundle_flags to support fsck-objects
      
     -    This commit adds a new enum `unbundle_fsck_flags` which is designed to
     -    control the fsck behavior when unbundling. `unbundle` can use this newly
     -    passed in enum to further decide whether to enable `--fsck-objects` for
     -    "git-index-pack".
     +    This commit extends `verify_bundle_flags` by adding a new option
     +    `VERIFY_BUNDLE_FSCK_ALWAYS`, which enables checks for broken objects in
     +    `bundle.c:unbundle`. This option is now used as the default for fetches
     +    involving bundles, specifically by `transport.c:fetch_refs_from_bundle`
     +    for direct bundle fetches and by `bundle-uri.c:unbundle_from_file` for
     +    _bundle-uri_ enabled fetches.
      
     -    Currently only `UNBUNDLE_FSCK_NEVER` and `UNBUNDLE_FSCK_ALWAYS` are
     -    supported as the very basic options. Another interesting option would be
     -    added in later commits.
     +    Upcoming commits will introduce another option as a replacement that
     +    fits better with fetch operations. `VERIFY_BUNDLE_FSCK_ALWAYS` will be
     +    further used to add "--fsck-objects" support for "git bundle unbundle"
     +    and "git bundle verify".
      
     +    Reviewed-by: Patrick Steinhardt <ps@pks.im>
          Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
      
     - ## builtin/bundle.c ##
     -@@ builtin/bundle.c: static int cmd_bundle_unbundle(int argc, const char **argv, const char *prefix)
     - 		strvec_pushl(&extra_index_pack_args, "-v", "--progress-title",
     - 			     _("Unbundling objects"), NULL);
     - 	ret = !!unbundle(the_repository, &header, bundle_fd,
     --			 &extra_index_pack_args, 0) ||
     -+			 &extra_index_pack_args, 0, UNBUNDLE_FSCK_NEVER) ||
     - 		list_bundle_refs(&header, argc, argv);
     - 	bundle_header_release(&header);
     - cleanup:
     -
       ## bundle-uri.c ##
      @@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *file)
       	 * the prerequisite commits.
       	 */
       	if ((result = unbundle(r, &header, bundle_fd, NULL,
      -			       VERIFY_BUNDLE_QUIET)))
     -+			       VERIFY_BUNDLE_QUIET, UNBUNDLE_FSCK_ALWAYS)))
     ++			       VERIFY_BUNDLE_QUIET | VERIFY_BUNDLE_FSCK_ALWAYS)))
       		return 1;
       
       	/*
      
       ## bundle.c ##
     -@@ bundle.c: int create_bundle(struct repository *r, const char *path,
     - 
     - int unbundle(struct repository *r, struct bundle_header *header,
     - 	     int bundle_fd, struct strvec *extra_index_pack_args,
     --	     enum verify_bundle_flags flags)
     -+	     enum verify_bundle_flags flags,
     -+	     enum unbundle_fsck_flags fsck_flags)
     - {
     - 	struct child_process ip = CHILD_PROCESS_INIT;
     - 
      @@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
       	if (header->filter.choice)
       		strvec_push(&ip.args, "--promisor=from-bundle");
       
     -+	switch (fsck_flags) {
     -+	case UNBUNDLE_FSCK_ALWAYS:
     ++	if (flags & VERIFY_BUNDLE_FSCK_ALWAYS)
      +		strvec_push(&ip.args, "--fsck-objects");
     -+		break;
     -+	case UNBUNDLE_FSCK_NEVER:
     -+	default:
     -+		break;
     -+	}
      +
       	if (extra_index_pack_args) {
       		strvec_pushv(&ip.args, extra_index_pack_args->v);
     @@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
      
       ## bundle.h ##
      @@ bundle.h: int create_bundle(struct repository *r, const char *path,
     - 		  int argc, const char **argv, struct strvec *pack_options,
     - 		  int version);
     - 
     -+enum unbundle_fsck_flags {
     -+	UNBUNDLE_FSCK_NEVER = 0,
     -+	UNBUNDLE_FSCK_ALWAYS,
     -+};
     -+
       enum verify_bundle_flags {
       	VERIFY_BUNDLE_VERBOSE = (1 << 0),
       	VERIFY_BUNDLE_QUIET = (1 << 1),
     -@@ bundle.h: int verify_bundle(struct repository *r, struct bundle_header *header,
     -  */
     - int unbundle(struct repository *r, struct bundle_header *header,
     - 	     int bundle_fd, struct strvec *extra_index_pack_args,
     --	     enum verify_bundle_flags flags);
     -+	     enum verify_bundle_flags flags,
     -+	     enum unbundle_fsck_flags fsck_flags);
     - int list_bundle_refs(struct bundle_header *header,
     - 		int argc, const char **argv);
     ++	VERIFY_BUNDLE_FSCK_ALWAYS = (1 << 2),
     + };
       
     + int verify_bundle(struct repository *r, struct bundle_header *header,
      
       ## transport.c ##
      @@ transport.c: static int fetch_refs_from_bundle(struct transport *transport,
     @@ transport.c: static int fetch_refs_from_bundle(struct transport *transport,
       		get_refs_from_bundle_inner(transport);
       	ret = unbundle(the_repository, &data->header, data->fd,
      -		       &extra_index_pack_args, 0);
     -+		       &extra_index_pack_args, 0, UNBUNDLE_FSCK_ALWAYS);
     ++		       &extra_index_pack_args, VERIFY_BUNDLE_FSCK_ALWAYS);
       	transport->hash_algo = data->header.hash_algo;
       	return ret;
       }
 3:  67401d4fbcb ! 3:  5ddc894c2c1 fetch-pack: expose fsckObjects configuration logic
     @@ Metadata
       ## Commit message ##
          fetch-pack: expose fsckObjects configuration logic
      
     -    Currently we can use "transfer.fsckObjects" or "fetch.fsckObjects" to
     -    control whether to enable checks for broken objects during fetching. But
     -    these configs are only acknowledged by `fetch-pack.c:get_pack` and do
     -    not make sense when fetching from bundles or using bundle-uris.
     +    Currently, we can use "transfer.fsckObjects" and the more specific
     +    "fetch.fsckObjects" to control checks for broken objects in received
     +    packs during fetches. However, these configurations were only
     +    acknowledged by `fetch-pack.c:get_pack` and did not take effect in
     +    direct bundle fetches and fetches with _bundle-uri_ enabled.
      
     -    This commit exposed the fetch-then-transfer configuration logic by
     -    adding a new function `fetch_pack_fsck_objects` in fetch-pack.h. In next
     -    commit, this new function will be used by `unbundle` in fetching
     -    scenarios.
     +    This commit exposes the fetch-then-transfer configuration logic by
     +    adding a new function `fetch_pack_fsck_objects` in fetch-pack.h. This
     +    new function is used to replace the assignment for `fsck_objects` in
     +    `fetch-pack.c:get_pack`. In the next commit, it will also be used by
     +    `bundle.c:unbundle` to better fit fetching scenarios.
      
     +    Helped-by: Junio C Hamano <gitster@pobox.com>
     +    Helped-by: Patrick Steinhardt <ps@pks.im>
          Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
      
       ## fetch-pack.c ##
     @@ fetch-pack.c: static const struct object_id *iterate_ref_map(void *cb_data)
      +int fetch_pack_fsck_objects(void)
      +{
      +	fetch_pack_setup();
     -+
     -+	return fetch_fsck_objects >= 0
     -+	       ? fetch_fsck_objects
     -+	       : transfer_fsck_objects >= 0
     -+	       ? transfer_fsck_objects
     -+	       : 0;
     ++	if (fetch_fsck_objects >= 0)
     ++		return fetch_fsck_objects;
     ++	if (transfer_fsck_objects >= 0)
     ++		return transfer_fsck_objects;
     ++	return 0;
      +}
      +
       struct ref *fetch_pack(struct fetch_pack_args *args,
     @@ fetch-pack.h: void negotiate_using_fetch(const struct oid_array *negotiation_tip
        */
       int report_unmatched_refs(struct ref **sought, int nr_sought);
       
     ++/*
     ++ * Return true if checks for broken objects in received pack are required.
     ++ */
      +int fetch_pack_fsck_objects(void);
      +
       #endif
 4:  c19b8f633cb ! 4:  68b9bca9f8b unbundle: introduce new option UNBUNDLE_FSCK_FOLLOW_FETCH
     @@ Metadata
      Author: Xing Xin <xingxin.xx@bytedance.com>
      
       ## Commit message ##
     -    unbundle: introduce new option UNBUNDLE_FSCK_FOLLOW_FETCH
     +    unbundle: introduce option VERIFY_BUNDLE_FSCK_FOLLOW_FETCH
      
     -    This commit adds a new option `UNBUNDLE_FSCK_FOLLOW_FETCH` to
     -    `unbundle_fsck_flags`, this new flag is currently used in the _fetch_
     -    process by
     +    This commit introduces a new option `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` to
     +    `verify_bundle_flags`. In `bundle.c:unbundle`, this new option controls
     +    whether broken object checks should be enabled by invoking
     +    `fetch-pack.c:fetch_pack_fsck_objects`. Note that the option
     +    `VERIFY_BUNDLE_FSCK_ALWAYS` takes precedence over
     +    `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH`.
      
     -    - `transport.c:fetch_refs_from_bundle` for fetching directly from a
     -      bundle.
     -    - `bundle-uri.c:unbundle_from_file` for unbundling bundles downloaded
     -      from bundle-uri.
     +    This flag is now used in the fetching process by:
      
     -    So we now have a relatively consistent logic for checking objects during
     -    fetching. Add tests for the above two situations are added.
     +    - `transport.c:fetch_refs_from_bundle` for direct bundle fetches.
     +    - `bundle-uri.c:unbundle_from_file` for bundle-uri enabled fetches.
     +
     +    This addition ensures a consistent logic for object verification during
     +    fetch operations. Tests have been added to confirm functionality in the
     +    scenarios mentioned above.
      
          Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
      
     @@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *fi
       	 * the prerequisite commits.
       	 */
       	if ((result = unbundle(r, &header, bundle_fd, NULL,
     --			       VERIFY_BUNDLE_QUIET, UNBUNDLE_FSCK_ALWAYS)))
     -+			       VERIFY_BUNDLE_QUIET, UNBUNDLE_FSCK_FOLLOW_FETCH)))
     +-			       VERIFY_BUNDLE_QUIET | VERIFY_BUNDLE_FSCK_ALWAYS)))
     ++			       VERIFY_BUNDLE_QUIET | VERIFY_BUNDLE_FSCK_FOLLOW_FETCH)))
       		return 1;
       
       	/*
     @@ bundle.c
       static const char v2_bundle_signature[] = "# v2 git bundle\n";
       static const char v3_bundle_signature[] = "# v3 git bundle\n";
      @@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
     - 	     enum unbundle_fsck_flags fsck_flags)
     + 	     enum verify_bundle_flags flags)
       {
       	struct child_process ip = CHILD_PROCESS_INIT;
      +	int fsck_objects = 0;
     @@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
       	if (verify_bundle(r, header, flags))
       		return -1;
      @@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
     + 		strvec_push(&ip.args, "--promisor=from-bundle");
       
     - 	switch (fsck_flags) {
     - 	case UNBUNDLE_FSCK_ALWAYS:
     --		strvec_push(&ip.args, "--fsck-objects");
     + 	if (flags & VERIFY_BUNDLE_FSCK_ALWAYS)
      +		fsck_objects = 1;
     -+		break;
     -+	case UNBUNDLE_FSCK_FOLLOW_FETCH:
     ++	else if (flags & VERIFY_BUNDLE_FSCK_FOLLOW_FETCH)
      +		fsck_objects = fetch_pack_fsck_objects();
     - 		break;
     - 	case UNBUNDLE_FSCK_NEVER:
     - 	default:
     - 		break;
     - 	}
     - 
     -+	if (fsck_objects)
     -+		strvec_push(&ip.args, "--fsck-objects");
      +
     ++	if (fsck_objects)
     + 		strvec_push(&ip.args, "--fsck-objects");
     + 
       	if (extra_index_pack_args) {
     - 		strvec_pushv(&ip.args, extra_index_pack_args->v);
     - 		strvec_clear(extra_index_pack_args);
      
       ## bundle.h ##
     -@@ bundle.h: int create_bundle(struct repository *r, const char *path,
     - enum unbundle_fsck_flags {
     - 	UNBUNDLE_FSCK_NEVER = 0,
     - 	UNBUNDLE_FSCK_ALWAYS,
     -+	UNBUNDLE_FSCK_FOLLOW_FETCH,
     +@@ bundle.h: enum verify_bundle_flags {
     + 	VERIFY_BUNDLE_VERBOSE = (1 << 0),
     + 	VERIFY_BUNDLE_QUIET = (1 << 1),
     + 	VERIFY_BUNDLE_FSCK_ALWAYS = (1 << 2),
     ++	VERIFY_BUNDLE_FSCK_FOLLOW_FETCH = (1 << 3),
       };
       
     - enum verify_bundle_flags {
     + int verify_bundle(struct repository *r, struct bundle_header *header,
      
       ## t/t5558-clone-bundle-uri.sh ##
     -@@ t/t5558-clone-bundle-uri.sh: test_expect_success 'fail to clone from non-bundle file' '
     +@@ t/t5558-clone-bundle-uri.sh: test_expect_success 'create bundle' '
     + 		git bundle create B.bundle topic &&
       
     - test_expect_success 'create bundle' '
     - 	git init clone-from &&
     --	git -C clone-from checkout -b topic &&
     -+	(
     -+		cd clone-from &&
     -+		git checkout -b topic &&
     -+
     -+		test_commit A &&
     -+		git bundle create A.bundle topic &&
     -+
     -+		test_commit B &&
     -+		git bundle create B.bundle topic &&
     + 		# Create a bundle with reference pointing to non-existent object.
     +-		sed "s/$(git rev-parse A)/$(git rev-parse B)/" <A.bundle >bad-header.bundle
     ++		sed "s/$(git rev-parse A)/$(git rev-parse B)/" <A.bundle >bad-header.bundle &&
      +
      +		cat >data <<-EOF &&
      +		tree $(git rev-parse HEAD^{tree})
      +		parent $(git rev-parse HEAD)
      +		author A U Thor
      +		committer A U Thor
     - 
     --	test_commit -C clone-from A &&
     --	git -C clone-from bundle create A.bundle topic &&
     ++
      +		commit: this is a commit with bad emails
     - 
     --	test_commit -C clone-from B &&
     --	git -C clone-from bundle create B.bundle topic
     ++
      +		EOF
      +		git hash-object --literally -t commit -w --stdin <data >commit &&
      +		git branch bad $(cat commit) &&
     -+		git bundle create bad.bundle bad &&
     ++		git bundle create bad-object.bundle bad &&
      +		git update-ref -d refs/heads/bad
     -+	)
     + 	)
       '
       
     - test_expect_success 'clone with path bundle' '
     -@@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone with path bundle' '
     - 	test_cmp expect actual
     +@@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone with bundle that has bad header' '
     + 	! grep "refs/bundles/" refs
       '
       
     -+test_expect_success 'clone with bad bundle' '
     -+	git -c fetch.fsckObjects=true clone --bundle-uri="clone-from/bad.bundle" \
     -+		clone-from clone-bad 2>err &&
     -+	# Unbundle fails, but clone can still proceed.
     ++test_expect_success 'clone with bundle that has bad object' '
     ++	# Unbundle succeeds if no fsckObjects confugured.
     ++	git clone --bundle-uri="clone-from/bad-object.bundle" \
     ++		clone-from clone-bad-object-no-fsck &&
     ++	git -C clone-bad-object-no-fsck for-each-ref --format="%(refname)" >refs &&
     ++	grep "refs/bundles/" refs >actual &&
     ++	cat >expect <<-\EOF &&
     ++	refs/bundles/bad
     ++	EOF
     ++	test_cmp expect actual &&
     ++
     ++	# Unbundle fails with fsckObjects set true, but clone can still proceed.
     ++	git -c fetch.fsckObjects=true clone --bundle-uri="clone-from/bad-object.bundle" \
     ++		clone-from clone-bad-object-fsck 2>err &&
      +	test_grep "missingEmail" err &&
     -+	git -C clone-bad for-each-ref --format="%(refname)" >refs &&
     ++	git -C clone-bad-object-fsck for-each-ref --format="%(refname)" >refs &&
      +	! grep "refs/bundles/" refs
      +'
      +
     @@ t/t5607-clone-bundle.sh: test_expect_success 'fetch SHA-1 from bundle' '
       	git fetch --no-tags foo/tip.bundle "$(cat hash)"
       '
       
     -+test_expect_success 'clone bundle with fetch.fsckObjects' '
     ++test_expect_success 'clone bundle with different fsckObjects configurations' '
      +	test_create_repo bundle-fsck &&
      +	(
      +		cd bundle-fsck &&
     @@ t/t5607-clone-bundle.sh: test_expect_success 'fetch SHA-1 from bundle' '
      +		git branch bad $(cat commit) &&
      +		git bundle create bad.bundle bad
      +	) &&
     ++
     ++	git clone bundle-fsck/bad.bundle bundle-no-fsck &&
     ++
     ++	git -c fetch.fsckObjects=false -c transfer.fsckObjects=true \
     ++		clone bundle-fsck/bad.bundle bundle-fetch-no-fsck &&
     ++
      +	test_must_fail git -c fetch.fsckObjects=true \
     -+		clone bundle-fsck/bad.bundle bundle-fsck-clone 2>err &&
     ++		clone bundle-fsck/bad.bundle bundle-fetch-fsck 2>err &&
     ++	test_grep "missingEmail" err &&
     ++
     ++	test_must_fail git -c transfer.fsckObjects=true \
     ++		clone bundle-fsck/bad.bundle bundle-transfer-fsck 2>err &&
      +	test_grep "missingEmail" err
      +'
      +
     @@ transport.c: static int fetch_refs_from_bundle(struct transport *transport,
       	if (!data->get_refs_from_bundle_called)
       		get_refs_from_bundle_inner(transport);
       	ret = unbundle(the_repository, &data->header, data->fd,
     --		       &extra_index_pack_args, 0, UNBUNDLE_FSCK_ALWAYS);
     -+		       &extra_index_pack_args, 0, UNBUNDLE_FSCK_FOLLOW_FETCH);
     +-		       &extra_index_pack_args, VERIFY_BUNDLE_FSCK_ALWAYS);
     ++		       &extra_index_pack_args, VERIFY_BUNDLE_FSCK_FOLLOW_FETCH);
       	transport->hash_algo = data->header.hash_algo;
       	return ret;
       }

From: Xing Xin <xingxin.xx@bytedance.com>

When using the bundle-uri mechanism with a bundle list containing multiple interrelated bundles, we encountered a bug where tips from downloaded bundles were not discovered, thus resulting in rather slow clones. This was particularly problematic when employing the "creationTokens" heuristic.

To reproduce this issue, consider a repository with a single branch "main" pointing to commit "A". Firstly, create a base bundle with:

  git bundle create base.bundle main

Then, add a new commit "B" on top of "A", and create an incremental bundle for "main":

  git bundle create incr.bundle A..main

Now, generate a bundle list with the following content:

  [bundle]
      version = 1
      mode = all
      heuristic = creationToken

  [bundle "base"]
      uri = base.bundle
      creationToken = 1

  [bundle "incr"]
      uri = incr.bundle
      creationToken = 2

A fresh clone with the bundle list above should result in a reference "refs/bundles/main" pointing to "B" in the new repository. However, git would still download everything from the server, as if it had fetched nothing locally.

So why the "refs/bundles/main" is not discovered? After some digging I found that:

1. Bundles in bundle list are downloaded to local files via
   `bundle-uri.c:download_bundle_list` or via
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
2. Each bundle is unbundled via `bundle-uri.c:unbundle_from_file`, which
   is called by `bundle-uri.c:unbundle_all_bundles` or called within
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
3. To get all prerequisites of the bundle, the bundle header is read
   inside `bundle-uri.c:unbundle_from_file` to by calling
   `bundle.c:read_bundle_header`.
4. Then it calls `bundle.c:unbundle`, which calls
   `bundle.c:verify_bundle` to ensure the repository contains all the
   prerequisites.
5. `bundle.c:verify_bundle` calls `parse_object`, which eventually
   invokes `packfile.c:prepare_packed_git` or
   `packfile.c:reprepare_packed_git`, filling
   `raw_object_store->packed_git` and setting `packed_git_initialized`.
6. If `bundle.c:unbundle` succeeds, it writes refs via
   `refs.c:refs_update_ref` with `REF_SKIP_OID_VERIFICATION` set. Here
   bundle refs which can target arbitrary objects are written to the
   repository.
7. Finally, in `fetch-pack.c:do_fetch_pack_v2`, the functions
   `fetch-pack.c:mark_complete_and_common_ref` and
   `fetch-pack.c:mark_tips` are called with `OBJECT_INFO_QUICK` set to
   find local tips for negotiation. The `OBJECT_INFO_QUICK` flag
   prevents `packfile.c:reprepare_packed_git` from being called,
   resulting in failures to parse OIDs that reside only in the latest
   bundle.

In the example above, when unbunding "incr.bundle", "base.pack" is added to `packed_git` due to prerequisites verification. However, "B" cannot be found for negotiation because it exists in "incr.pack", which is not included in `packed_git`.

This commit fixes the bug by removing `REF_SKIP_OID_VERIFICATION` flag when writing bundle refs. When `refs.c:refs_update_ref` is called to to write the corresponding bundle refs, it triggers `refs.c:ref_transaction_commit`. This, in turn, invokes `refs.c:ref_transaction_prepare`, which calls `transaction_prepare` of the refs storage backend. For files backend, this function is `files-backend.c:files_transaction_prepare`, and for reftable backend, it is `reftable-backend.c:reftable_be_transaction_prepare`. Both functions eventually call `object.c:parse_object`, which can invoke `packfile.c:reprepare_packed_git` to refresh `packed_git`. This ensures that bundle refs point to valid objects and that all tips from bundle refs are correctly parsed during subsequent negotiations.

A test has been added to demonstrate that bundles with incorrect headers, where refs point to non-existent objects, do not result in any bundle refs being created in the repository. Additionally, a set of negotiation-related tests for fetching with bundle-uri has been included.

Reviewed-by: Karthik Nayak <karthik.188@gmail.com>
Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |   3 +-
 t/t5558-clone-bundle-uri.sh | 153 +++++++++++++++++++++++++++++++++++-
 2 files changed, 150 insertions(+), 6 deletions(-)

From: Xing Xin <xingxin.xx@bytedance.com>

This commit extends `verify_bundle_flags` by adding a new option `VERIFY_BUNDLE_FSCK_ALWAYS`, which enables checks for broken objects in `bundle.c:unbundle`. This option is now used as the default for fetches involving bundles, specifically by `transport.c:fetch_refs_from_bundle` for direct bundle fetches and by `bundle-uri.c:unbundle_from_file` for _bundle-uri_ enabled fetches.

Upcoming commits will introduce another option as a replacement that fits better with fetch operations. `VERIFY_BUNDLE_FSCK_ALWAYS` will be further used to add "--fsck-objects" support for "git bundle unbundle" and "git bundle verify".

Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c | 2 +-
 bundle.c     | 3 +++
 bundle.h     | 1 +
 transport.c  | 2 +-
 4 files changed, 6 insertions(+), 2 deletions(-)

On Thu, May 30, 2024 at 08:21:28AM +0000, Xing Xin via GitGitGadget wrote:

Tiny nit: the important change in this commit is not that you wire up the flag, but rather that we start to execute git-fsck(1) now. I'd thus propose to adapt the commit title accordingly.

Patrick

At 2024-06-06 20:06:41, "Patrick Steinhardt" <ps@pks.im> wrote:

To be precise, it is adding a "--fsck-objects" flag to "git-index-pack".

This commit is mapped to [PATCH v5 3/4] due to some adjustments to the commit order and implementation details. Hope the new title can better describe the new patch.

Xing Xin

From: Xing Xin <xingxin.xx@bytedance.com>

Currently, we can use "transfer.fsckObjects" and the more specific "fetch.fsckObjects" to control checks for broken objects in received packs during fetches. However, these configurations were only acknowledged by `fetch-pack.c:get_pack` and did not take effect in direct bundle fetches and fetches with _bundle-uri_ enabled.

This commit exposes the fetch-then-transfer configuration logic by adding a new function `fetch_pack_fsck_objects` in fetch-pack.h. This new function is used to replace the assignment for `fsck_objects` in `fetch-pack.c:get_pack`. In the next commit, it will also be used by `bundle.c:unbundle` to better fit fetching scenarios.

Helped-by: Junio C Hamano <gitster@pobox.com>
Helped-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 fetch-pack.c | 17 +++++++++++------
 fetch-pack.h |  5 +++++
 2 files changed, 16 insertions(+), 6 deletions(-)

From: Xing Xin <xingxin.xx@bytedance.com>

This commit introduces a new option `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` to `verify_bundle_flags`. In `bundle.c:unbundle`, this new option controls whether broken object checks should be enabled by invoking `fetch-pack.c:fetch_pack_fsck_objects`. Note that the option `VERIFY_BUNDLE_FSCK_ALWAYS` takes precedence over `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH`.

This flag is now used in the fetching process by:

- `transport.c:fetch_refs_from_bundle` for direct bundle fetches.
- `bundle-uri.c:unbundle_from_file` for bundle-uri enabled fetches.

This addition ensures a consistent logic for object verification during fetch operations. Tests have been added to confirm functionality in the scenarios mentioned above.

Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |  2 +-
 bundle.c                    |  7 +++++++
 bundle.h                    |  1 +
 t/t5558-clone-bundle-uri.sh | 35 ++++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     | 33 +++++++++++++++++++++++++++++++++
 transport.c                 |  2 +-
 6 files changed, 77 insertions(+), 3 deletions(-)

On Thu, May 30, 2024 at 08:21:30AM +0000, Xing Xin via GitGitGadget wrote:

Same here, the important part is not that we introduce the flag, but that we start using it in `unbundle_from_file()`.

One thing that is a bit weird is that we first change `unbundle()` to use `FSCK_ALWAYS` in a preceding patch, and then convert it to use `FSCK_FOLLOW_FETCH` in the same series. It could be restructured a bit to first introduce the flags, only, while not modifying any of the callsites yet. Passing the respective flags would then be done in a separate commit.

Patrick

At 2024-06-06 20:06:47, "Patrick Steinhardt" <ps@pks.im> wrote:

This makes sense to me, thanks!

Xing Xin

While attempting to fix a reference negotiation bug in bundle-uri, we identified that the fetch process lacks some crucial object validation checks when processing bundles. The primary issues are:

 1. In the bundle-uri scenario, object IDs were not validated before writing
    bundle references. This was the root cause of the original negotiation
    bug in bundle-uri and could lead to potential repository corruption.
 2. The existing "fetch.fsckObjects" and "transfer.fsckObjects"
    configurations were not applied when directly fetching bundles or
    fetching with bundle-uri enabled. In fact, there were no object
    validation supports for unbundle.

The first patch addresses the bundle-uri negotiation issue by removing the REF_SKIP_OID_VERIFICATION flag when writing bundle references.

Patches 2 through 4 extend verify_bundle_flags for bundle.c:unbundle to add support for object validation (fsck) in different scenarios, mainly following the suggestions from Junio on the mailing list.

Xing Xin (4):
  bundle-uri: verify oid before writing refs
  fetch-pack: expose fsckObjects configuration logic
  unbundle: extend options to support object verification
  unbundle: use VERIFY_BUNDLE_FSCK_FOLLOW_FETCH for fetches

 bundle-uri.c                |   5 +-
 bundle.c                    |  10 ++
 bundle.h                    |   2 +
 fetch-pack.c                |  17 ++--
 fetch-pack.h                |   5 +
 t/t5558-clone-bundle-uri.sh | 186 +++++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     |  33 +++++++
 transport.c                 |   2 +-
 8 files changed, 246 insertions(+), 14 deletions(-)

base-commit: b9cfe4845cb2562584837bc0101c0ab76490a239
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1730%2Fblanet%2Fxx%2Fbundle-uri-bug-using-bundle-list-v5
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1730/blanet/xx/bundle-uri-bug-using-bundle-list-v5
Pull-Request: https://github.com/gitgitgadget/git/pull/1730

Range-diff vs v4:

 1:  e958a3ab20c = 1:  e958a3ab20c bundle-uri: verify oid before writing refs
 3:  5ddc894c2c1 = 2:  d21c236b8de fetch-pack: expose fsckObjects configuration logic
 2:  beb70735811 ! 3:  0a18d7839be unbundle: extend verify_bundle_flags to support fsck-objects
     @@ Metadata
      Author: Xing Xin <xingxin.xx@bytedance.com>
      
       ## Commit message ##
     -    unbundle: extend verify_bundle_flags to support fsck-objects
     +    unbundle: extend options to support object verification
      
     -    This commit extends `verify_bundle_flags` by adding a new option
     -    `VERIFY_BUNDLE_FSCK_ALWAYS`, which enables checks for broken objects in
     -    `bundle.c:unbundle`. This option is now used as the default for fetches
     -    involving bundles, specifically by `transport.c:fetch_refs_from_bundle`
     -    for direct bundle fetches and by `bundle-uri.c:unbundle_from_file` for
     -    _bundle-uri_ enabled fetches.
     +    This commit extends object verification support in `bundle.c:unbundle`
     +    by adding two new options to `verify_bundle_flags`:
      
     -    Upcoming commits will introduce another option as a replacement that
     -    fits better with fetch operations. `VERIFY_BUNDLE_FSCK_ALWAYS` will be
     -    further used to add "--fsck-objects" support for "git bundle unbundle"
     -    and "git bundle verify".
     +    - `VERIFY_BUNDLE_FSCK_ALWAYS` explicitly enables checks for broken
     +      objects. It will be used to add "--fsck-objects" support for "git
     +      bundle unbundle" in a separate series.
     +    - `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` is designed to be used during fetch
     +      operations, specifically for direct bundle fetches and _bundle-uri_
     +      enabled fetches. When enabled, `bundle.c:unbundle` invokes
     +      `fetch-pack.c:fetch_pack_fsck_objects` to determine whether to enable
     +      checks for broken objects. Passing this flag during fetching will be
     +      implemented in a subsequent commit.
     +
     +    Note that the option `VERIFY_BUNDLE_FSCK_ALWAYS` takes precedence over
     +    `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH`.
      
          Reviewed-by: Patrick Steinhardt <ps@pks.im>
          Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
      
     - ## bundle-uri.c ##
     -@@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *file)
     - 	 * the prerequisite commits.
     - 	 */
     - 	if ((result = unbundle(r, &header, bundle_fd, NULL,
     --			       VERIFY_BUNDLE_QUIET)))
     -+			       VERIFY_BUNDLE_QUIET | VERIFY_BUNDLE_FSCK_ALWAYS)))
     - 		return 1;
     - 
     - 	/*
     -
       ## bundle.c ##
     +@@
     + #include "list-objects-filter-options.h"
     + #include "connected.h"
     + #include "write-or-die.h"
     ++#include "fetch-pack.h"
     + 
     + static const char v2_bundle_signature[] = "# v2 git bundle\n";
     + static const char v3_bundle_signature[] = "# v3 git bundle\n";
     +@@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
     + 	     enum verify_bundle_flags flags)
     + {
     + 	struct child_process ip = CHILD_PROCESS_INIT;
     ++	int fsck_objects = 0;
     + 
     + 	if (verify_bundle(r, header, flags))
     + 		return -1;
      @@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
       	if (header->filter.choice)
       		strvec_push(&ip.args, "--promisor=from-bundle");
       
      +	if (flags & VERIFY_BUNDLE_FSCK_ALWAYS)
     ++		fsck_objects = 1;
     ++	else if (flags & VERIFY_BUNDLE_FSCK_FOLLOW_FETCH)
     ++		fsck_objects = fetch_pack_fsck_objects();
     ++
     ++	if (fsck_objects)
      +		strvec_push(&ip.args, "--fsck-objects");
      +
       	if (extra_index_pack_args) {
     @@ bundle.h: int create_bundle(struct repository *r, const char *path,
       	VERIFY_BUNDLE_VERBOSE = (1 << 0),
       	VERIFY_BUNDLE_QUIET = (1 << 1),
      +	VERIFY_BUNDLE_FSCK_ALWAYS = (1 << 2),
     ++	VERIFY_BUNDLE_FSCK_FOLLOW_FETCH = (1 << 3),
       };
       
       int verify_bundle(struct repository *r, struct bundle_header *header,
     -
     - ## transport.c ##
     -@@ transport.c: static int fetch_refs_from_bundle(struct transport *transport,
     - 	if (!data->get_refs_from_bundle_called)
     - 		get_refs_from_bundle_inner(transport);
     - 	ret = unbundle(the_repository, &data->header, data->fd,
     --		       &extra_index_pack_args, 0);
     -+		       &extra_index_pack_args, VERIFY_BUNDLE_FSCK_ALWAYS);
     - 	transport->hash_algo = data->header.hash_algo;
     - 	return ret;
     - }
 4:  68b9bca9f8b ! 4:  eb9f21f16b5 unbundle: introduce option VERIFY_BUNDLE_FSCK_FOLLOW_FETCH
     @@ Metadata
      Author: Xing Xin <xingxin.xx@bytedance.com>
      
       ## Commit message ##
     -    unbundle: introduce option VERIFY_BUNDLE_FSCK_FOLLOW_FETCH
     +    unbundle: use VERIFY_BUNDLE_FSCK_FOLLOW_FETCH for fetches
      
     -    This commit introduces a new option `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` to
     -    `verify_bundle_flags`. In `bundle.c:unbundle`, this new option controls
     -    whether broken object checks should be enabled by invoking
     -    `fetch-pack.c:fetch_pack_fsck_objects`. Note that the option
     -    `VERIFY_BUNDLE_FSCK_ALWAYS` takes precedence over
     -    `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH`.
     -
     -    This flag is now used in the fetching process by:
     +    This commit passes `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` to `unbundle` in
     +    the fetching process, including:
      
          - `transport.c:fetch_refs_from_bundle` for direct bundle fetches.
          - `bundle-uri.c:unbundle_from_file` for bundle-uri enabled fetches.
     @@ Commit message
          fetch operations. Tests have been added to confirm functionality in the
          scenarios mentioned above.
      
     +    Reviewed-by: Patrick Steinhardt <ps@pks.im>
          Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
      
       ## bundle-uri.c ##
     @@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *fi
       	 * the prerequisite commits.
       	 */
       	if ((result = unbundle(r, &header, bundle_fd, NULL,
     --			       VERIFY_BUNDLE_QUIET | VERIFY_BUNDLE_FSCK_ALWAYS)))
     +-			       VERIFY_BUNDLE_QUIET)))
      +			       VERIFY_BUNDLE_QUIET | VERIFY_BUNDLE_FSCK_FOLLOW_FETCH)))
       		return 1;
       
       	/*
      
     - ## bundle.c ##
     -@@
     - #include "list-objects-filter-options.h"
     - #include "connected.h"
     - #include "write-or-die.h"
     -+#include "fetch-pack.h"
     - 
     - static const char v2_bundle_signature[] = "# v2 git bundle\n";
     - static const char v3_bundle_signature[] = "# v3 git bundle\n";
     -@@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
     - 	     enum verify_bundle_flags flags)
     - {
     - 	struct child_process ip = CHILD_PROCESS_INIT;
     -+	int fsck_objects = 0;
     - 
     - 	if (verify_bundle(r, header, flags))
     - 		return -1;
     -@@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
     - 		strvec_push(&ip.args, "--promisor=from-bundle");
     - 
     - 	if (flags & VERIFY_BUNDLE_FSCK_ALWAYS)
     -+		fsck_objects = 1;
     -+	else if (flags & VERIFY_BUNDLE_FSCK_FOLLOW_FETCH)
     -+		fsck_objects = fetch_pack_fsck_objects();
     -+
     -+	if (fsck_objects)
     - 		strvec_push(&ip.args, "--fsck-objects");
     - 
     - 	if (extra_index_pack_args) {
     -
     - ## bundle.h ##
     -@@ bundle.h: enum verify_bundle_flags {
     - 	VERIFY_BUNDLE_VERBOSE = (1 << 0),
     - 	VERIFY_BUNDLE_QUIET = (1 << 1),
     - 	VERIFY_BUNDLE_FSCK_ALWAYS = (1 << 2),
     -+	VERIFY_BUNDLE_FSCK_FOLLOW_FETCH = (1 << 3),
     - };
     - 
     - int verify_bundle(struct repository *r, struct bundle_header *header,
     -
       ## t/t5558-clone-bundle-uri.sh ##
      @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'create bundle' '
       		git bundle create B.bundle topic &&
     @@ transport.c: static int fetch_refs_from_bundle(struct transport *transport,
       	if (!data->get_refs_from_bundle_called)
       		get_refs_from_bundle_inner(transport);
       	ret = unbundle(the_repository, &data->header, data->fd,
     --		       &extra_index_pack_args, VERIFY_BUNDLE_FSCK_ALWAYS);
     +-		       &extra_index_pack_args, 0);
      +		       &extra_index_pack_args, VERIFY_BUNDLE_FSCK_FOLLOW_FETCH);
       	transport->hash_algo = data->header.hash_algo;
       	return ret;

From: Xing Xin <xingxin.xx@bytedance.com>

Currently, we can use "transfer.fsckObjects" and the more specific "fetch.fsckObjects" to control checks for broken objects in received packs during fetches. However, these configurations were only acknowledged by `fetch-pack.c:get_pack` and did not take effect in direct bundle fetches and fetches with _bundle-uri_ enabled.

This commit exposes the fetch-then-transfer configuration logic by adding a new function `fetch_pack_fsck_objects` in fetch-pack.h. This new function is used to replace the assignment for `fsck_objects` in `fetch-pack.c:get_pack`. In the next commit, it will also be used by `bundle.c:unbundle` to better fit fetching scenarios.

Helped-by: Junio C Hamano <gitster@pobox.com>
Helped-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 fetch-pack.c | 17 +++++++++++------
 fetch-pack.h |  5 +++++
 2 files changed, 16 insertions(+), 6 deletions(-)

From: Xing Xin <xingxin.xx@bytedance.com>

When using the bundle-uri mechanism with a bundle list containing multiple interrelated bundles, we encountered a bug where tips from downloaded bundles were not discovered, thus resulting in rather slow clones. This was particularly problematic when employing the "creationTokens" heuristic.

To reproduce this issue, consider a repository with a single branch "main" pointing to commit "A". Firstly, create a base bundle with:

  git bundle create base.bundle main

Then, add a new commit "B" on top of "A", and create an incremental bundle for "main":

  git bundle create incr.bundle A..main

Now, generate a bundle list with the following content:

  [bundle]
      version = 1
      mode = all
      heuristic = creationToken

  [bundle "base"]
      uri = base.bundle
      creationToken = 1

  [bundle "incr"]
      uri = incr.bundle
      creationToken = 2

A fresh clone with the bundle list above should result in a reference "refs/bundles/main" pointing to "B" in the new repository. However, git would still download everything from the server, as if it had fetched nothing locally.

So why the "refs/bundles/main" is not discovered? After some digging I found that:

1. Bundles in bundle list are downloaded to local files via
   `bundle-uri.c:download_bundle_list` or via
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
2. Each bundle is unbundled via `bundle-uri.c:unbundle_from_file`, which
   is called by `bundle-uri.c:unbundle_all_bundles` or called within
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
3. To get all prerequisites of the bundle, the bundle header is read
   inside `bundle-uri.c:unbundle_from_file` to by calling
   `bundle.c:read_bundle_header`.
4. Then it calls `bundle.c:unbundle`, which calls
   `bundle.c:verify_bundle` to ensure the repository contains all the
   prerequisites.
5. `bundle.c:verify_bundle` calls `parse_object`, which eventually
   invokes `packfile.c:prepare_packed_git` or
   `packfile.c:reprepare_packed_git`, filling
   `raw_object_store->packed_git` and setting `packed_git_initialized`.
6. If `bundle.c:unbundle` succeeds, it writes refs via
   `refs.c:refs_update_ref` with `REF_SKIP_OID_VERIFICATION` set. Here
   bundle refs which can target arbitrary objects are written to the
   repository.
7. Finally, in `fetch-pack.c:do_fetch_pack_v2`, the functions
   `fetch-pack.c:mark_complete_and_common_ref` and
   `fetch-pack.c:mark_tips` are called with `OBJECT_INFO_QUICK` set to
   find local tips for negotiation. The `OBJECT_INFO_QUICK` flag
   prevents `packfile.c:reprepare_packed_git` from being called,
   resulting in failures to parse OIDs that reside only in the latest
   bundle.

In the example above, when unbunding "incr.bundle", "base.pack" is added to `packed_git` due to prerequisites verification. However, "B" cannot be found for negotiation because it exists in "incr.pack", which is not included in `packed_git`.

This commit fixes the bug by removing `REF_SKIP_OID_VERIFICATION` flag when writing bundle refs. When `refs.c:refs_update_ref` is called to to write the corresponding bundle refs, it triggers `refs.c:ref_transaction_commit`. This, in turn, invokes `refs.c:ref_transaction_prepare`, which calls `transaction_prepare` of the refs storage backend. For files backend, this function is `files-backend.c:files_transaction_prepare`, and for reftable backend, it is `reftable-backend.c:reftable_be_transaction_prepare`. Both functions eventually call `object.c:parse_object`, which can invoke `packfile.c:reprepare_packed_git` to refresh `packed_git`. This ensures that bundle refs point to valid objects and that all tips from bundle refs are correctly parsed during subsequent negotiations.

A test has been added to demonstrate that bundles with incorrect headers, where refs point to non-existent objects, do not result in any bundle refs being created in the repository. Additionally, a set of negotiation-related tests for fetching with bundle-uri has been included.

Reviewed-by: Karthik Nayak <karthik.188@gmail.com>
Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |   3 +-
 t/t5558-clone-bundle-uri.sh | 153 +++++++++++++++++++++++++++++++++++-
 2 files changed, 150 insertions(+), 6 deletions(-)

From: Xing Xin <xingxin.xx@bytedance.com>

This commit extends object verification support in `bundle.c:unbundle` by adding two new options to `verify_bundle_flags`:

- `VERIFY_BUNDLE_FSCK_ALWAYS` explicitly enables checks for broken
  objects. It will be used to add "--fsck-objects" support for "git
  bundle unbundle" in a separate series.
- `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` is designed to be used during fetch
  operations, specifically for direct bundle fetches and _bundle-uri_
  enabled fetches. When enabled, `bundle.c:unbundle` invokes
  `fetch-pack.c:fetch_pack_fsck_objects` to determine whether to enable
  checks for broken objects. Passing this flag during fetching will be
  implemented in a subsequent commit.

Note that the option `VERIFY_BUNDLE_FSCK_ALWAYS` takes precedence over `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH`.

Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle.c | 10 ++++++++++
 bundle.h |  2 ++
 2 files changed, 12 insertions(+)

On Tue, Jun 11, 2024 at 06:42:05AM +0000, Xing Xin via GitGitGadget wrote:

Thanks, the new sequence of commits is much easier to follow. It also shows that there is no user of `VERIFY_BUNDLE_FSCK_ALWAYS` at the end of this series. So maybe we should drop that flag?

If you do that, then I'd also propose to merge patches 2 and 3 into one given that both are quite trivial and related to each other.

Other than that this series looks good to me.

Patrick

At 2024-06-11 17:11:09, "Patrick Steinhardt" <ps@pks.im> wrote:

OK, let's focus on the fetching scenarios in this patch series.

I merged patches 3 and 4 instead to combine the new option definition and usage logic, as the two are more closely related and more trivial. :)

Xing Xin

From: Xing Xin <xingxin.xx@bytedance.com>

This commit passes `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` to `unbundle` in the fetching process, including:

- `transport.c:fetch_refs_from_bundle` for direct bundle fetches.
- `bundle-uri.c:unbundle_from_file` for bundle-uri enabled fetches.

This addition ensures a consistent logic for object verification during fetch operations. Tests have been added to confirm functionality in the scenarios mentioned above.

Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |  2 +-
 t/t5558-clone-bundle-uri.sh | 35 ++++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     | 33 +++++++++++++++++++++++++++++++++
 transport.c                 |  2 +-
 4 files changed, 69 insertions(+), 3 deletions(-)

While attempting to fix a reference negotiation bug in bundle-uri, we identified that the fetch process lacks some crucial object validation checks when processing bundles. The primary issues are:

 1. In the bundle-uri scenario, object IDs were not validated before writing
    bundle references. This was the root cause of the original negotiation
    bug in bundle-uri and could lead to potential repository corruption.
 2. The existing "fetch.fsckObjects" and "transfer.fsckObjects"
    configurations were not applied when directly fetching bundles or
    fetching with bundle-uri enabled. In fact, there were no object
    validation supports for unbundle.

The first patch addresses the bundle-uri negotiation issue by removing the REF_SKIP_OID_VERIFICATION flag when writing bundle references.

Patches 2 through 3 extend verify_bundle_flags for bundle.c:unbundle to add support for object validation (fsck) in fetch scenarios, mainly following the suggestions from Junio and Patrick on the mailing list.

Xing Xin (3):
  bundle-uri: verify oid before writing refs
  fetch-pack: expose fsckObjects configuration logic
  unbundle: support object verification for fetches

 bundle-uri.c                |   5 +-
 bundle.c                    |   5 +
 bundle.h                    |   1 +
 fetch-pack.c                |  17 ++--
 fetch-pack.h                |   5 +
 t/t5558-clone-bundle-uri.sh | 186 +++++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     |  33 +++++++
 transport.c                 |   2 +-
 8 files changed, 240 insertions(+), 14 deletions(-)

base-commit: b9cfe4845cb2562584837bc0101c0ab76490a239
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1730%2Fblanet%2Fxx%2Fbundle-uri-bug-using-bundle-list-v6
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1730/blanet/xx/bundle-uri-bug-using-bundle-list-v6
Pull-Request: https://github.com/gitgitgadget/git/pull/1730

Range-diff vs v5:

 1:  e958a3ab20c = 1:  e958a3ab20c bundle-uri: verify oid before writing refs
 2:  d21c236b8de = 2:  d21c236b8de fetch-pack: expose fsckObjects configuration logic
 3:  0a18d7839be < -:  ----------- unbundle: extend options to support object verification
 4:  eb9f21f16b5 ! 3:  53395e8c08a unbundle: use VERIFY_BUNDLE_FSCK_FOLLOW_FETCH for fetches
     @@ Metadata
      Author: Xing Xin <xingxin.xx@bytedance.com>
      
       ## Commit message ##
     -    unbundle: use VERIFY_BUNDLE_FSCK_FOLLOW_FETCH for fetches
     +    unbundle: support object verification for fetches
      
     -    This commit passes `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` to `unbundle` in
     -    the fetching process, including:
     +    This commit extends object verification support for fetches in
     +    `bundle.c:unbundle` by adding the `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH`
     +    option to `verify_bundle_flags`. When this option is enabled,
     +    `bundle.c:unbundle` invokes `fetch-pack.c:fetch_pack_fsck_objects` to
     +    determine whether to append the "--fsck-objects" flag to
     +    "git-index-pack".
     +
     +    `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` is now passed to `unbundle` in the
     +    fetching process, including:
      
          - `transport.c:fetch_refs_from_bundle` for direct bundle fetches.
          - `bundle-uri.c:unbundle_from_file` for bundle-uri enabled fetches.
     @@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *fi
       
       	/*
      
     + ## bundle.c ##
     +@@
     + #include "list-objects-filter-options.h"
     + #include "connected.h"
     + #include "write-or-die.h"
     ++#include "fetch-pack.h"
     + 
     + static const char v2_bundle_signature[] = "# v2 git bundle\n";
     + static const char v3_bundle_signature[] = "# v3 git bundle\n";
     +@@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
     + 	if (header->filter.choice)
     + 		strvec_push(&ip.args, "--promisor=from-bundle");
     + 
     ++	if (flags & VERIFY_BUNDLE_FSCK_FOLLOW_FETCH)
     ++		if (fetch_pack_fsck_objects())
     ++			strvec_push(&ip.args, "--fsck-objects");
     ++
     + 	if (extra_index_pack_args) {
     + 		strvec_pushv(&ip.args, extra_index_pack_args->v);
     + 		strvec_clear(extra_index_pack_args);
     +
     + ## bundle.h ##
     +@@ bundle.h: int create_bundle(struct repository *r, const char *path,
     + enum verify_bundle_flags {
     + 	VERIFY_BUNDLE_VERBOSE = (1 << 0),
     + 	VERIFY_BUNDLE_QUIET = (1 << 1),
     ++	VERIFY_BUNDLE_FSCK_FOLLOW_FETCH = (1 << 2),
     + };
     + 
     + int verify_bundle(struct repository *r, struct bundle_header *header,
     +
       ## t/t5558-clone-bundle-uri.sh ##
      @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'create bundle' '
       		git bundle create B.bundle topic &&

From: Xing Xin <xingxin.xx@bytedance.com>

When using the bundle-uri mechanism with a bundle list containing multiple interrelated bundles, we encountered a bug where tips from downloaded bundles were not discovered, thus resulting in rather slow clones. This was particularly problematic when employing the "creationTokens" heuristic.

To reproduce this issue, consider a repository with a single branch "main" pointing to commit "A". Firstly, create a base bundle with:

  git bundle create base.bundle main

Then, add a new commit "B" on top of "A", and create an incremental bundle for "main":

  git bundle create incr.bundle A..main

Now, generate a bundle list with the following content:

  [bundle]
      version = 1
      mode = all
      heuristic = creationToken

  [bundle "base"]
      uri = base.bundle
      creationToken = 1

  [bundle "incr"]
      uri = incr.bundle
      creationToken = 2

A fresh clone with the bundle list above should result in a reference "refs/bundles/main" pointing to "B" in the new repository. However, git would still download everything from the server, as if it had fetched nothing locally.

So why the "refs/bundles/main" is not discovered? After some digging I found that:

1. Bundles in bundle list are downloaded to local files via
   `bundle-uri.c:download_bundle_list` or via
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
2. Each bundle is unbundled via `bundle-uri.c:unbundle_from_file`, which
   is called by `bundle-uri.c:unbundle_all_bundles` or called within
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
3. To get all prerequisites of the bundle, the bundle header is read
   inside `bundle-uri.c:unbundle_from_file` to by calling
   `bundle.c:read_bundle_header`.
4. Then it calls `bundle.c:unbundle`, which calls
   `bundle.c:verify_bundle` to ensure the repository contains all the
   prerequisites.
5. `bundle.c:verify_bundle` calls `parse_object`, which eventually
   invokes `packfile.c:prepare_packed_git` or
   `packfile.c:reprepare_packed_git`, filling
   `raw_object_store->packed_git` and setting `packed_git_initialized`.
6. If `bundle.c:unbundle` succeeds, it writes refs via
   `refs.c:refs_update_ref` with `REF_SKIP_OID_VERIFICATION` set. Here
   bundle refs which can target arbitrary objects are written to the
   repository.
7. Finally, in `fetch-pack.c:do_fetch_pack_v2`, the functions
   `fetch-pack.c:mark_complete_and_common_ref` and
   `fetch-pack.c:mark_tips` are called with `OBJECT_INFO_QUICK` set to
   find local tips for negotiation. The `OBJECT_INFO_QUICK` flag
   prevents `packfile.c:reprepare_packed_git` from being called,
   resulting in failures to parse OIDs that reside only in the latest
   bundle.

In the example above, when unbunding "incr.bundle", "base.pack" is added to `packed_git` due to prerequisites verification. However, "B" cannot be found for negotiation because it exists in "incr.pack", which is not included in `packed_git`.

This commit fixes the bug by removing `REF_SKIP_OID_VERIFICATION` flag when writing bundle refs. When `refs.c:refs_update_ref` is called to to write the corresponding bundle refs, it triggers `refs.c:ref_transaction_commit`. This, in turn, invokes `refs.c:ref_transaction_prepare`, which calls `transaction_prepare` of the refs storage backend. For files backend, this function is `files-backend.c:files_transaction_prepare`, and for reftable backend, it is `reftable-backend.c:reftable_be_transaction_prepare`. Both functions eventually call `object.c:parse_object`, which can invoke `packfile.c:reprepare_packed_git` to refresh `packed_git`. This ensures that bundle refs point to valid objects and that all tips from bundle refs are correctly parsed during subsequent negotiations.

A test has been added to demonstrate that bundles with incorrect headers, where refs point to non-existent objects, do not result in any bundle refs being created in the repository. Additionally, a set of negotiation-related tests for fetching with bundle-uri has been included.

Reviewed-by: Karthik Nayak <karthik.188@gmail.com>
Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |   3 +-
 t/t5558-clone-bundle-uri.sh | 153 +++++++++++++++++++++++++++++++++++-
 2 files changed, 150 insertions(+), 6 deletions(-)

"Xing Xin via GitGitGadget" <gitgitgadget@gmail.com> writes:

"Fix the bug by ...".

"to to"

Nice. That does sound like the right fix. The one who did something to _require_ us to reprepare causes the packed_git list refreshed.

I suspect that this would be terribly unportable. The early part of a bundle file may be text and sed may be able to grok but are you sure everybody's implementation of sed would not barf (or even worse, corrupt) the pack stream data that follows?

The code used in t/lib-bundle.sh:convert_bundle_to_pack() has been in use since 8315588b (bundle: fix wrong check of read_header()'s return value & add tests, 2007-03-06), so munging the bundle with a code similar to it may have a better portability story.

Add something like:

        corrupt_bundle_header () {
                sed -e '/^$/q' "$@"
                cat
        }

to t/lib-bundle.sh, which can take an arbitrary sequence of command line parameters to drive "sed", and can be used like so:

	corrupt_bundle_header \
		-e "s/^$(git rev-parse A) /$(git rev-parse B) /" \
		<A.bndl >B.bndl

perhaps?

So this is the test the proposed log message discussed. The description gave a false impression that the "broken header" test that has not much to do with the bug being fixed was the only added test---we probably want to correct that impression.

Drop this divider and comment. The HTTP tests you see below has a much better reason to be separated like that in order to warn test writers (they shouldn't add their random new tests after that point, because everything after that one is skipped when HTTPD tests are disabled---see the beginning of t/lib-httpd.sh which is included after that divider line), but everything here you added is not special. Everybody should run these tests.

Do not overly depend on glob not matching at this point when you establish the when-finished handler (or glob matching the files you want to catch and later test not adding anything you would want to clean). Quote "rm -f trace*.txt" and drop "r" if you do not absolutely need it (and I would imagine you don't, given the .txt suffix).

Hmph, if the expected pattern is only a few lines without any magic,

	test_write_lines >expect refs/bundles/topic &&

may be easier to follow.

Using "test_grep" would make it easier to diagnose when test breaks. A failing "grep" will be silent (i.e., "I didn't find anything you told me to look for"). A failing "test_grep" will tell you "I was told to find this, but didn't find any in that".

OK. This is a good use of here-doc (as opposed to test_write_lines I sugested earlier for different purposes). I wondered if these $(pwd) and file://$(pwd) are safe (I always get confused by the need to sometimes use $PWD to help Windows), but I see them used in what Derrick wrote in this file, so they must be fine.

But there may be characters in the leading part of $(pwd) that we do not control that needs quoting (like a double quote '"'). The value of bundle.*.uri may need to be quoted a bit carefully. This is not a new problem this patch introduces, so you do not have to rewrite this part of the patch; I'll mark it with #leftoverbits here---the idea being somebody else who is too bored can come back, see if it is truly broken, and fix them after all dust settles.

Abusing "git config -f bundle-list" might be safer, e.g.

	$ git config -f bundle.list bundle.bundle-1.uri \
		'file:///home/abc"def/t/trash dir/clone-from/b1.bndl'
	$ cat bundle.list
        [bundle "bundle-1"]
                uri = file:///home/abc\"def/t/trash dir/clone-from/b1.bndl

as you do not know what other garbage character is in $(pwd) part.

Just FYI, to negate test_grep, use

	test_grep ! "clone > want " trace-packet.txt

not	

	! test_grep "clone > want " trace-packet.txt ;# WRONG

At 2024-06-12 03:08:04, "Junio C Hamano" <gitster@pobox.com> wrote:

Ah, always typos. :-)

[snip]

Thanks, I never knew sed could be used this way! It's so concise!

However, after applying these code, the added test "t5558.5 clone with bundle that has bad header" fails in some CI jobs showing that we can not get expected error. For example CI "linux-musl (alpine)" gives:

	+ git clone '--bundle-uri=clone-from/bad-header.bundle' clone-from clone-bad-header
	+ git -C clone-from rev-parse B
	+ commit_b=d9df4505cb3522088b9e29d6051ac16f1564154a
	+ test_grep 'trying to write ref '"'"'refs/bundles/topic'"'"' with nonexistent object d9df4505cb3522088b9e29d6051ac16f1564154a' err
	+ eval 'last_arg=${2}'
	+ last_arg=err
	+ test -f err
	+ test 2 -lt 2
	+ test 'x!' '=' 'xtrying to write ref '"'"'refs/bundles/topic'"'"' with nonexistent object d9df4505cb3522088b9e29d6051ac16f1564154a'
	+ test 'x!' '=' 'xtrying to write ref '"'"'refs/bundles/topic'"'"' with nonexistent object d9df4505cb3522088b9e29d6051ac16f1564154a'
	+ grep 'trying to write ref '"'"'refs/bundles/topic'"'"' with nonexistent object d9df4505cb3522088b9e29d6051ac16f1564154a' err
	+ echo 'error: '"'"'grep trying to write ref '"'"'refs/bundles/topic'"'"' with nonexistent object d9df4505cb3522088b9e29d6051ac16f1564154a' 'err'"'"' didn'"'"'t find a match in:'
	error: 'grep trying to write ref 'refs/bundles/topic' with nonexistent object d9df4505cb3522088b9e29d6051ac16f1564154a err' didn't find a match in:
	+ test -s err
	+ cat err
	Cloning into 'clone-bad-header'...
	fatal: pack signature mismatch
	error: index-pack died
	done.
	+ return 1
	error: last command exited with $?=1
	not ok 5 - clone with bundle that has bad header

	+# Create a bundle with reference pointing to non-existent object.
	+sed -e "/^$/q" -e "s/$(git rev-parse A) /$(git rev-parse B) /" \
	+	<A.bundle >bad-header.bundle &&
	+convert_bundle_to_pack \
	+	<A.bundle >>bad-header.bundle

More details can be found at https://github.com/blanet/git/actions/runs/9541478191/job/26294731254.

After some digging, I discovered that inside the Alpine container, `corrupt_bundle_header` is missing the leading "PACK\x00" in the pack section of the converted bundle. This is likely caused by the incompatibility of "sed" across different operating systems. I stopped investigating further due to my unfamiliarity with containers and "sed" itself. Instead, I found another approach that utilizes the suggested usage of "sed" and `lib-bundle.sh:convert_bundle_to_pack`.

 	+# Create a bundle with reference pointing to non-existent object.
	+sed -e "/^$/q" -e "s/$(git rev-parse A) /$(git rev-parse B) /" \
	+	<A.bundle >bad-header.bundle &&
	+convert_bundle_to_pack \
	+	<A.bundle >>bad-header.bundle

Adjusted the commit message.

Thanks for the explanation, dropped the divider in new series.

Yes, the "-r" is not safe and I just need to clean the "*.txt", not involving directories.

Applied!

Thanks, it just accelerated the digging process of the test failure mentioned above. :)

Thanks your through review!

Xing Xin

From: Xing Xin <xingxin.xx@bytedance.com>

Currently, we can use "transfer.fsckObjects" and the more specific "fetch.fsckObjects" to control checks for broken objects in received packs during fetches. However, these configurations were only acknowledged by `fetch-pack.c:get_pack` and did not take effect in direct bundle fetches and fetches with _bundle-uri_ enabled.

This commit exposes the fetch-then-transfer configuration logic by adding a new function `fetch_pack_fsck_objects` in fetch-pack.h. This new function is used to replace the assignment for `fsck_objects` in `fetch-pack.c:get_pack`. In the next commit, it will also be used by `bundle.c:unbundle` to better fit fetching scenarios.

Helped-by: Junio C Hamano <gitster@pobox.com>
Helped-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 fetch-pack.c | 17 +++++++++++------
 fetch-pack.h |  5 +++++
 2 files changed, 16 insertions(+), 6 deletions(-)

"Xing Xin via GitGitGadget" <gitgitgadget@gmail.com> writes:

OK, this one is designed to be as lightweight as possible when it has already been called, so a potentially duplicated call would not cause too much worries here.

Much easier to follow.  Nicely done.

From: Xing Xin <xingxin.xx@bytedance.com>

This commit extends object verification support for fetches in `bundle.c:unbundle` by adding the `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` option to `verify_bundle_flags`. When this option is enabled, `bundle.c:unbundle` invokes `fetch-pack.c:fetch_pack_fsck_objects` to determine whether to append the "--fsck-objects" flag to "git-index-pack".

`VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` is now passed to `unbundle` in the fetching process, including:

- `transport.c:fetch_refs_from_bundle` for direct bundle fetches.
- `bundle-uri.c:unbundle_from_file` for bundle-uri enabled fetches.

This addition ensures a consistent logic for object verification during fetch operations. Tests have been added to confirm functionality in the scenarios mentioned above.

Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |  2 +-
 bundle.c                    |  5 +++++
 bundle.h                    |  1 +
 t/t5558-clone-bundle-uri.sh | 35 ++++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     | 33 +++++++++++++++++++++++++++++++++
 transport.c                 |  2 +-
 6 files changed, 75 insertions(+), 3 deletions(-)

"Xing Xin via GitGitGadget" <gitgitgadget@gmail.com> writes:

Please start your proposed log message by stating what the perceived problem without this patch in the current world is. Without it, you cannot easily answer the most basic question: "why are we doing this?"

The name VERIFY_BUNDLE_FSCK_FOLLOW_FETCH does not read very well. VERIFY_BUNDLE part is common across various flags, so what is specific to the feature is "FSCK_FOLLOW_FETCH", and it is good that we convey the fact that we do a bit more than the normal VERIFY_BUNDLE (which is, to read the prerequisite headers and make sure we have them in the sense that they are reachable from our refs) with the word FSCK.

But is it necessary (or even a good idea) to limit its usability with "FOLLOW_FETCH" (which does not look even grammatical)? Aren't we closing the door to other folks who may want to do a more thorough fsck-like checks in other codepaths by saying "if you are not doing this immediately after you fetch, you are unwelcome to use this flag"?

OK (modulo the flag name).

OK, that's quite straight-forward. We are running "index-pack --fix-thin --stdin" and feeding the bundle data to it. We just tell it to also work in the "--fsck-objects" mode.

OK.

I wonder if something like this is a potential follow-up topic somebody may be interested in after the dust settles. That is exactly why the name of this bit matters.

 builtin/bundle.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

At 2024-06-12 04:05:35, "Junio C Hamano" <gitster@pobox.com> wrote:

Got it.

I initially considered adding another option VERIFY_BUNDLE_FSCK_ALWAYS for other scenarios, which would take a higher priority than VERIFY_BUNDLE_FSCK_FOLLOW_FETCH. However, that approach is also confusing, as we would have two flags both controlling the fsck behavior.

How about extending VERIFY_BUNDLE_FSCK and letting the callers decide whether to add the flag for fscking?

In bundle.c, we can make a small change like:

On Tue, Jun 11, 2024 at 12:45:40PM +0000, blanet via GitGitGadget wrote:

Thanks, this version looks good to me.

Patrick

While attempting to fix a reference negotiation bug in bundle-uri, we identified that the fetch process lacks some crucial object validation checks when processing bundles. The primary issues are:

 1. In the bundle-uri scenario, object IDs were not validated before writing
    bundle references. This was the root cause of the original negotiation
    bug in bundle-uri and could lead to potential repository corruption.
 2. The existing "fetch.fsckObjects" and "transfer.fsckObjects"
    configurations were not applied when directly fetching bundles or
    fetching with bundle-uri enabled. In fact, there were no object
    validation supports for unbundle.

The first patch addresses the bundle-uri negotiation issue by removing the REF_SKIP_OID_VERIFICATION flag when writing bundle references.

Patches 2 through 3 extend verify_bundle_flags for bundle.c:unbundle to add support for object validation (fsck) in fetch scenarios, mainly following the suggestions from Junio and Patrick on the mailing list.

Xing Xin (3):
  bundle-uri: verify oid before writing refs
  fetch-pack: expose fsckObjects configuration logic
  unbundle: extend object verification for fetches

 bundle-uri.c                |   6 +-
 bundle.c                    |   3 +
 bundle.h                    |   1 +
 fetch-pack.c                |  17 ++--
 fetch-pack.h                |   5 +
 t/t5558-clone-bundle-uri.sh | 181 +++++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     |  33 +++++++
 transport.c                 |   3 +-
 8 files changed, 235 insertions(+), 14 deletions(-)

base-commit: b9cfe4845cb2562584837bc0101c0ab76490a239
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1730%2Fblanet%2Fxx%2Fbundle-uri-bug-using-bundle-list-v7
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1730/blanet/xx/bundle-uri-bug-using-bundle-list-v7
Pull-Request: https://github.com/gitgitgadget/git/pull/1730

Range-diff vs v6:

 1:  e958a3ab20c ! 1:  fc9f44fda00 bundle-uri: verify oid before writing refs
     @@ Commit message
          be found for negotiation because it exists in "incr.pack", which is not
          included in `packed_git`.
      
     -    This commit fixes the bug by removing `REF_SKIP_OID_VERIFICATION` flag
     -    when writing bundle refs. When `refs.c:refs_update_ref` is called to to
     -    write the corresponding bundle refs, it triggers
     -    `refs.c:ref_transaction_commit`.  This, in turn, invokes
     -    `refs.c:ref_transaction_prepare`, which calls `transaction_prepare` of
     -    the refs storage backend. For files backend, this function is
     -    `files-backend.c:files_transaction_prepare`, and for reftable backend,
     -    it is `reftable-backend.c:reftable_be_transaction_prepare`. Both
     -    functions eventually call `object.c:parse_object`, which can invoke
     +    Fix the bug by removing `REF_SKIP_OID_VERIFICATION` flag when writing
     +    bundle refs. When `refs.c:refs_update_ref` is called to write the
     +    corresponding bundle refs, it triggers `refs.c:ref_transaction_commit`.
     +    This, in turn, invokes `refs.c:ref_transaction_prepare`, which calls
     +    `transaction_prepare` of the refs storage backend. For files backend, it
     +    is `files-backend.c:files_transaction_prepare`, and for reftable
     +    backend, it is `reftable-backend.c:reftable_be_transaction_prepare`.
     +    Both functions eventually call `object.c:parse_object`, which can invoke
          `packfile.c:reprepare_packed_git` to refresh `packed_git`. This ensures
          that bundle refs point to valid objects and that all tips from bundle
          refs are correctly parsed during subsequent negotiations.
      
     -    A test has been added to demonstrate that bundles with incorrect
     -    headers, where refs point to non-existent objects, do not result in any
     -    bundle refs being created in the repository. Additionally, a set of
     -    negotiation-related tests for fetching with bundle-uri has been
     -    included.
     +    A set of negotiation-related tests for cloning with bundle-uri has been
     +    included to demonstrate that downloaded bundles are utilized to
     +    accelerate fetching.
     +
     +    Additionally, another test has been added to show that bundles with
     +    incorrect headers, where refs point to non-existent objects, do not
     +    result in any bundle refs being created in the repository.
      
          Reviewed-by: Karthik Nayak <karthik.188@gmail.com>
          Reviewed-by: Patrick Steinhardt <ps@pks.im>
     @@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *fi
       	bundle_header_release(&header);
      
       ## t/t5558-clone-bundle-uri.sh ##
     +@@
     + test_description='test fetching bundles with --bundle-uri'
     + 
     + . ./test-lib.sh
     ++. "$TEST_DIRECTORY"/lib-bundle.sh
     + 
     + test_expect_success 'fail to clone from non-existent file' '
     + 	test_when_finished rm -rf test &&
      @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'fail to clone from non-bundle file' '
       
       test_expect_success 'create bundle' '
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'fail to clone from non-bundle
      +		git bundle create B.bundle topic &&
      +
      +		# Create a bundle with reference pointing to non-existent object.
     -+		sed "s/$(git rev-parse A)/$(git rev-parse B)/" <A.bundle >bad-header.bundle
     ++		sed -e "/^$/q" -e "s/$(git rev-parse A) /$(git rev-parse B) /" \
     ++			<A.bundle >bad-header.bundle &&
     ++		convert_bundle_to_pack \
     ++			<A.bundle >>bad-header.bundle
      +	)
       '
       
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone with path bundle' '
       '
       
      +test_expect_success 'clone with bundle that has bad header' '
     ++	# Write bundle ref fails, but clone can still proceed.
      +	git clone --bundle-uri="clone-from/bad-header.bundle" \
      +		clone-from clone-bad-header 2>err &&
     -+	# Write bundle ref fails, but clone can still proceed.
      +	commit_b=$(git -C clone-from rev-parse B) &&
      +	test_grep "trying to write ref '\''refs/bundles/topic'\'' with nonexistent object $commit_b" err &&
      +	git -C clone-bad-header for-each-ref --format="%(refname)" >refs &&
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone bundle list (file, any m
       	! grep "refs/bundles/" refs
       '
       
     -+#########################################################################
     -+# Clone negotiation related tests begin here
     -+
      +test_expect_success 'negotiation: bundle with part of wanted commits' '
     -+	test_when_finished rm -rf trace*.txt &&
     ++	test_when_finished "rm -f trace*.txt" &&
      +	GIT_TRACE_PACKET="$(pwd)/trace-packet.txt" \
      +	git clone --no-local --bundle-uri="clone-from/A.bundle" \
      +		clone-from nego-bundle-part &&
      +	git -C nego-bundle-part for-each-ref --format="%(refname)" >refs &&
      +	grep "refs/bundles/" refs >actual &&
     -+	cat >expect <<-\EOF &&
     -+	refs/bundles/topic
     -+	EOF
     ++	test_write_lines refs/bundles/topic >expect &&
      +	test_cmp expect actual &&
      +	# Ensure that refs/bundles/topic are sent as "have".
     -+	grep "clone> have $(git -C clone-from rev-parse A)" trace-packet.txt
     ++	test_grep "clone> have $(git -C clone-from rev-parse A)" trace-packet.txt
      +'
      +
      +test_expect_success 'negotiation: bundle with all wanted commits' '
     -+	test_when_finished rm -rf trace*.txt &&
     ++	test_when_finished "rm -f trace*.txt" &&
      +	GIT_TRACE_PACKET="$(pwd)/trace-packet.txt" \
      +	git clone --no-local --single-branch --branch=topic --no-tags \
      +		--bundle-uri="clone-from/B.bundle" \
      +		clone-from nego-bundle-all &&
      +	git -C nego-bundle-all for-each-ref --format="%(refname)" >refs &&
      +	grep "refs/bundles/" refs >actual &&
     -+	cat >expect <<-\EOF &&
     -+	refs/bundles/topic
     -+	EOF
     ++	test_write_lines refs/bundles/topic >expect &&
      +	test_cmp expect actual &&
      +	# We already have all needed commits so no "want" needed.
      +	! grep "clone> want " trace-packet.txt
      +'
      +
      +test_expect_success 'negotiation: bundle list (no heuristic)' '
     -+	test_when_finished rm -f trace*.txt &&
     ++	test_when_finished "rm -f trace*.txt" &&
      +	cat >bundle-list <<-EOF &&
      +	[bundle]
      +		version = 1
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone bundle list (file, any m
      +	refs/bundles/left
      +	EOF
      +	test_cmp expect actual &&
     -+	grep "clone> have $(git -C nego-bundle-list-no-heuristic rev-parse refs/bundles/left)" trace-packet.txt
     ++	test_grep "clone> have $(git -C nego-bundle-list-no-heuristic rev-parse refs/bundles/left)" trace-packet.txt
      +'
      +
      +test_expect_success 'negotiation: bundle list (creationToken)' '
     -+	test_when_finished rm -f trace*.txt &&
     ++	test_when_finished "rm -f trace*.txt" &&
      +	cat >bundle-list <<-EOF &&
      +	[bundle]
      +		version = 1
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone bundle list (file, any m
      +	refs/bundles/left
      +	EOF
      +	test_cmp expect actual &&
     -+	grep "clone> have $(git -C nego-bundle-list-heuristic rev-parse refs/bundles/left)" trace-packet.txt
     ++	test_grep "clone> have $(git -C nego-bundle-list-heuristic rev-parse refs/bundles/left)" trace-packet.txt
      +'
      +
      +test_expect_success 'negotiation: bundle list with all wanted commits' '
     -+	test_when_finished rm -f trace*.txt &&
     ++	test_when_finished "rm -f trace*.txt" &&
      +	cat >bundle-list <<-EOF &&
      +	[bundle]
      +		version = 1
 2:  d21c236b8de = 2:  3dc0d9dd22f fetch-pack: expose fsckObjects configuration logic
 3:  53395e8c08a ! 3:  2f15099bbb9 unbundle: support object verification for fetches
     @@ Metadata
      Author: Xing Xin <xingxin.xx@bytedance.com>
      
       ## Commit message ##
     -    unbundle: support object verification for fetches
     +    unbundle: extend object verification for fetches
      
     -    This commit extends object verification support for fetches in
     -    `bundle.c:unbundle` by adding the `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH`
     -    option to `verify_bundle_flags`. When this option is enabled,
     -    `bundle.c:unbundle` invokes `fetch-pack.c:fetch_pack_fsck_objects` to
     -    determine whether to append the "--fsck-objects" flag to
     -    "git-index-pack".
     +    The existing fetch.fsckObjects and transfer.fsckObjects configurations
     +    were not fully applied to bundle-involved fetches, including direct
     +    bundle fetches and bundle-uri enabled fetches. Furthermore, there was no
     +    object verification support for unbundle.
      
     -    `VERIFY_BUNDLE_FSCK_FOLLOW_FETCH` is now passed to `unbundle` in the
     -    fetching process, including:
     +    This commit extends object verification support in `bundle.c:unbundle`
     +    by adding the `VERIFY_BUNDLE_FSCK` option to `verify_bundle_flags`. When
     +    this option is enabled, we append the `--fsck-objects` flag to
     +    `git-index-pack`.
     +
     +    The `VERIFY_BUNDLE_FSCK` option is now used by bundle-involved fetches,
     +    where we use `fetch-pack.c:fetch_pack_fsck_objects` to determine whether
     +    to enable this option for `bundle.c:unbundle`, specifically in:
      
          - `transport.c:fetch_refs_from_bundle` for direct bundle fetches.
          - `bundle-uri.c:unbundle_from_file` for bundle-uri enabled fetches.
      
          This addition ensures a consistent logic for object verification during
     -    fetch operations. Tests have been added to confirm functionality in the
     -    scenarios mentioned above.
     +    fetches. Tests have been added to confirm functionality in the scenarios
     +    mentioned above.
      
          Reviewed-by: Patrick Steinhardt <ps@pks.im>
          Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
      
       ## bundle-uri.c ##
     +@@
     + #include "hashmap.h"
     + #include "pkt-line.h"
     + #include "config.h"
     ++#include "fetch-pack.h"
     + #include "remote.h"
     + 
     + static struct {
      @@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *file)
       	 * the prerequisite commits.
       	 */
       	if ((result = unbundle(r, &header, bundle_fd, NULL,
      -			       VERIFY_BUNDLE_QUIET)))
     -+			       VERIFY_BUNDLE_QUIET | VERIFY_BUNDLE_FSCK_FOLLOW_FETCH)))
     ++			       VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0))))
       		return 1;
       
       	/*
      
       ## bundle.c ##
     -@@
     - #include "list-objects-filter-options.h"
     - #include "connected.h"
     - #include "write-or-die.h"
     -+#include "fetch-pack.h"
     - 
     - static const char v2_bundle_signature[] = "# v2 git bundle\n";
     - static const char v3_bundle_signature[] = "# v3 git bundle\n";
      @@ bundle.c: int unbundle(struct repository *r, struct bundle_header *header,
       	if (header->filter.choice)
       		strvec_push(&ip.args, "--promisor=from-bundle");
       
     -+	if (flags & VERIFY_BUNDLE_FSCK_FOLLOW_FETCH)
     -+		if (fetch_pack_fsck_objects())
     -+			strvec_push(&ip.args, "--fsck-objects");
     ++	if (flags & VERIFY_BUNDLE_FSCK)
     ++		strvec_push(&ip.args, "--fsck-objects");
      +
       	if (extra_index_pack_args) {
       		strvec_pushv(&ip.args, extra_index_pack_args->v);
     @@ bundle.h: int create_bundle(struct repository *r, const char *path,
       enum verify_bundle_flags {
       	VERIFY_BUNDLE_VERBOSE = (1 << 0),
       	VERIFY_BUNDLE_QUIET = (1 << 1),
     -+	VERIFY_BUNDLE_FSCK_FOLLOW_FETCH = (1 << 2),
     ++	VERIFY_BUNDLE_FSCK = (1 << 2),
       };
       
       int verify_bundle(struct repository *r, struct bundle_header *header,
      
       ## t/t5558-clone-bundle-uri.sh ##
      @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'create bundle' '
     - 		git bundle create B.bundle topic &&
     - 
     - 		# Create a bundle with reference pointing to non-existent object.
     --		sed "s/$(git rev-parse A)/$(git rev-parse B)/" <A.bundle >bad-header.bundle
     -+		sed "s/$(git rev-parse A)/$(git rev-parse B)/" <A.bundle >bad-header.bundle &&
     + 		sed -e "/^$/q" -e "s/$(git rev-parse A) /$(git rev-parse B) /" \
     + 			<A.bundle >bad-header.bundle &&
     + 		convert_bundle_to_pack \
     +-			<A.bundle >>bad-header.bundle
     ++			<A.bundle >>bad-header.bundle &&
      +
      +		cat >data <<-EOF &&
      +		tree $(git rev-parse HEAD^{tree})
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone with bundle that has bad
      +		clone-from clone-bad-object-no-fsck &&
      +	git -C clone-bad-object-no-fsck for-each-ref --format="%(refname)" >refs &&
      +	grep "refs/bundles/" refs >actual &&
     -+	cat >expect <<-\EOF &&
     -+	refs/bundles/bad
     -+	EOF
     ++	test_write_lines refs/bundles/bad >expect &&
      +	test_cmp expect actual &&
      +
      +	# Unbundle fails with fsckObjects set true, but clone can still proceed.
     @@ transport.c: static int fetch_refs_from_bundle(struct transport *transport,
       		get_refs_from_bundle_inner(transport);
       	ret = unbundle(the_repository, &data->header, data->fd,
      -		       &extra_index_pack_args, 0);
     -+		       &extra_index_pack_args, VERIFY_BUNDLE_FSCK_FOLLOW_FETCH);
     ++		       &extra_index_pack_args,
     ++		       fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0);
       	transport->hash_algo = data->header.hash_algo;
       	return ret;
       }

From: Xing Xin <xingxin.xx@bytedance.com>

When using the bundle-uri mechanism with a bundle list containing multiple interrelated bundles, we encountered a bug where tips from downloaded bundles were not discovered, thus resulting in rather slow clones. This was particularly problematic when employing the "creationTokens" heuristic.

To reproduce this issue, consider a repository with a single branch "main" pointing to commit "A". Firstly, create a base bundle with:

  git bundle create base.bundle main

Then, add a new commit "B" on top of "A", and create an incremental bundle for "main":

  git bundle create incr.bundle A..main

Now, generate a bundle list with the following content:

  [bundle]
      version = 1
      mode = all
      heuristic = creationToken

  [bundle "base"]
      uri = base.bundle
      creationToken = 1

  [bundle "incr"]
      uri = incr.bundle
      creationToken = 2

A fresh clone with the bundle list above should result in a reference "refs/bundles/main" pointing to "B" in the new repository. However, git would still download everything from the server, as if it had fetched nothing locally.

So why the "refs/bundles/main" is not discovered? After some digging I found that:

1. Bundles in bundle list are downloaded to local files via
   `bundle-uri.c:download_bundle_list` or via
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
2. Each bundle is unbundled via `bundle-uri.c:unbundle_from_file`, which
   is called by `bundle-uri.c:unbundle_all_bundles` or called within
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
3. To get all prerequisites of the bundle, the bundle header is read
   inside `bundle-uri.c:unbundle_from_file` to by calling
   `bundle.c:read_bundle_header`.
4. Then it calls `bundle.c:unbundle`, which calls
   `bundle.c:verify_bundle` to ensure the repository contains all the
   prerequisites.
5. `bundle.c:verify_bundle` calls `parse_object`, which eventually
   invokes `packfile.c:prepare_packed_git` or
   `packfile.c:reprepare_packed_git`, filling
   `raw_object_store->packed_git` and setting `packed_git_initialized`.
6. If `bundle.c:unbundle` succeeds, it writes refs via
   `refs.c:refs_update_ref` with `REF_SKIP_OID_VERIFICATION` set. Here
   bundle refs which can target arbitrary objects are written to the
   repository.
7. Finally, in `fetch-pack.c:do_fetch_pack_v2`, the functions
   `fetch-pack.c:mark_complete_and_common_ref` and
   `fetch-pack.c:mark_tips` are called with `OBJECT_INFO_QUICK` set to
   find local tips for negotiation. The `OBJECT_INFO_QUICK` flag
   prevents `packfile.c:reprepare_packed_git` from being called,
   resulting in failures to parse OIDs that reside only in the latest
   bundle.

In the example above, when unbunding "incr.bundle", "base.pack" is added to `packed_git` due to prerequisites verification. However, "B" cannot be found for negotiation because it exists in "incr.pack", which is not included in `packed_git`.

Fix the bug by removing `REF_SKIP_OID_VERIFICATION` flag when writing bundle refs. When `refs.c:refs_update_ref` is called to write the corresponding bundle refs, it triggers `refs.c:ref_transaction_commit`. This, in turn, invokes `refs.c:ref_transaction_prepare`, which calls `transaction_prepare` of the refs storage backend. For files backend, it is `files-backend.c:files_transaction_prepare`, and for reftable backend, it is `reftable-backend.c:reftable_be_transaction_prepare`. Both functions eventually call `object.c:parse_object`, which can invoke `packfile.c:reprepare_packed_git` to refresh `packed_git`. This ensures that bundle refs point to valid objects and that all tips from bundle refs are correctly parsed during subsequent negotiations.

A set of negotiation-related tests for cloning with bundle-uri has been included to demonstrate that downloaded bundles are utilized to accelerate fetching.

Additionally, another test has been added to show that bundles with incorrect headers, where refs point to non-existent objects, do not result in any bundle refs being created in the repository.

Reviewed-by: Karthik Nayak <karthik.188@gmail.com>
Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |   3 +-
 t/t5558-clone-bundle-uri.sh | 150 +++++++++++++++++++++++++++++++++++-
 2 files changed, 147 insertions(+), 6 deletions(-)

"Xing Xin via GitGitGadget" <gitgitgadget@gmail.com> writes:

Why not "test_grep !" here? There are other uses of bare grep in the newly added lines, but I won't repeat them here; the same comment applies to them.

Can this rev-parse fail (the failure would be hidden from the test)? If so,

	...
	test_cmp expect actual &&
	# Ensure that refs/bundles/topic is sent as "have"
	tip=$(git -C clone-from rev-parse A) &&
	test_grep "clone> have $tip" trace-packet.txt

would catch such a failure. You are doing so in the previous test in the hunk starting at 33/46 above with commit_b variable already.

There are other uses of git command in $(command substitution) whose exit status are ignored in the newly added lines, but I won't repeat them here; the same comment applies to them.

Thanks.

At 2024-06-19 01:37:58, "Junio C Hamano" <gitster@pobox.com> wrote:

Both `test_grep !` and `! grep` are widely used in tests. Sorry for not realizing that the former is preferred.

Fixed in the new series. Thanks.

Xing Xin

From: Xing Xin <xingxin.xx@bytedance.com>

Currently, we can use "transfer.fsckObjects" and the more specific "fetch.fsckObjects" to control checks for broken objects in received packs during fetches. However, these configurations were only acknowledged by `fetch-pack.c:get_pack` and did not take effect in direct bundle fetches and fetches with _bundle-uri_ enabled.

This commit exposes the fetch-then-transfer configuration logic by adding a new function `fetch_pack_fsck_objects` in fetch-pack.h. This new function is used to replace the assignment for `fsck_objects` in `fetch-pack.c:get_pack`. In the next commit, it will also be used by `bundle.c:unbundle` to better fit fetching scenarios.

Helped-by: Junio C Hamano <gitster@pobox.com>
Helped-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 fetch-pack.c | 17 +++++++++++------
 fetch-pack.h |  5 +++++
 2 files changed, 16 insertions(+), 6 deletions(-)

From: Xing Xin <xingxin.xx@bytedance.com>

The existing fetch.fsckObjects and transfer.fsckObjects configurations were not fully applied to bundle-involved fetches, including direct bundle fetches and bundle-uri enabled fetches. Furthermore, there was no object verification support for unbundle.

This commit extends object verification support in `bundle.c:unbundle` by adding the `VERIFY_BUNDLE_FSCK` option to `verify_bundle_flags`. When this option is enabled, we append the `--fsck-objects` flag to `git-index-pack`.

The `VERIFY_BUNDLE_FSCK` option is now used by bundle-involved fetches, where we use `fetch-pack.c:fetch_pack_fsck_objects` to determine whether to enable this option for `bundle.c:unbundle`, specifically in:

- `transport.c:fetch_refs_from_bundle` for direct bundle fetches.
- `bundle-uri.c:unbundle_from_file` for bundle-uri enabled fetches.

This addition ensures a consistent logic for object verification during fetches. Tests have been added to confirm functionality in the scenarios mentioned above.

Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |  3 ++-
 bundle.c                    |  3 +++
 bundle.h                    |  1 +
 t/t5558-clone-bundle-uri.sh | 33 ++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     | 33 +++++++++++++++++++++++++++++++++
 transport.c                 |  3 ++-
 6 files changed, 73 insertions(+), 3 deletions(-)

While attempting to fix a reference negotiation bug in bundle-uri, we identified that the fetch process lacks some crucial object validation checks when processing bundles. The primary issues are:

 1. In the bundle-uri scenario, object IDs were not validated before writing
    bundle references. This was the root cause of the original negotiation
    bug in bundle-uri and could lead to potential repository corruption.
 2. The existing "fetch.fsckObjects" and "transfer.fsckObjects"
    configurations were not applied when directly fetching bundles or
    fetching with bundle-uri enabled. In fact, there were no object
    validation supports for unbundle.

The first patch addresses the bundle-uri negotiation issue by removing the REF_SKIP_OID_VERIFICATION flag when writing bundle references.

Patches 2 through 3 extend verify_bundle_flags for bundle.c:unbundle to add support for object validation (fsck) in fetch scenarios, mainly following the suggestions from Junio and Patrick on the mailing list.

Xing Xin (3):
  bundle-uri: verify oid before writing refs
  fetch-pack: expose fsckObjects configuration logic
  unbundle: extend object verification for fetches

 bundle-uri.c                |   6 +-
 bundle.c                    |   3 +
 bundle.h                    |   1 +
 fetch-pack.c                |  17 ++--
 fetch-pack.h                |   5 +
 t/t5558-clone-bundle-uri.sh | 187 +++++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     |  35 +++++++
 transport.c                 |   3 +-
 8 files changed, 243 insertions(+), 14 deletions(-)

base-commit: b9cfe4845cb2562584837bc0101c0ab76490a239
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1730%2Fblanet%2Fxx%2Fbundle-uri-bug-using-bundle-list-v8
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1730/blanet/xx/bundle-uri-bug-using-bundle-list-v8
Pull-Request: https://github.com/gitgitgadget/git/pull/1730

Range-diff vs v7:

 1:  fc9f44fda00 ! 1:  d8fbde2dcd4 bundle-uri: verify oid before writing refs
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'fail to clone from non-bundle
      +		git bundle create B.bundle topic &&
      +
      +		# Create a bundle with reference pointing to non-existent object.
     -+		sed -e "/^$/q" -e "s/$(git rev-parse A) /$(git rev-parse B) /" \
     ++		commit_a=$(git rev-parse A) &&
     ++		commit_b=$(git rev-parse B) &&
     ++		sed -e "/^$/q" -e "s/$commit_a /$commit_b /" \
      +			<A.bundle >bad-header.bundle &&
      +		convert_bundle_to_pack \
      +			<A.bundle >>bad-header.bundle
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone with path bundle' '
      +	commit_b=$(git -C clone-from rev-parse B) &&
      +	test_grep "trying to write ref '\''refs/bundles/topic'\'' with nonexistent object $commit_b" err &&
      +	git -C clone-bad-header for-each-ref --format="%(refname)" >refs &&
     -+	! grep "refs/bundles/" refs
     ++	test_grep ! "refs/bundles/" refs
      +'
      +
       test_expect_success 'clone with path bundle and non-default hash' '
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone bundle list (file, any m
      +	test_write_lines refs/bundles/topic >expect &&
      +	test_cmp expect actual &&
      +	# Ensure that refs/bundles/topic are sent as "have".
     -+	test_grep "clone> have $(git -C clone-from rev-parse A)" trace-packet.txt
     ++	tip=$(git -C clone-from rev-parse A) &&
     ++	test_grep "clone> have $tip" trace-packet.txt
      +'
      +
      +test_expect_success 'negotiation: bundle with all wanted commits' '
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone bundle list (file, any m
      +	test_write_lines refs/bundles/topic >expect &&
      +	test_cmp expect actual &&
      +	# We already have all needed commits so no "want" needed.
     -+	! grep "clone> want " trace-packet.txt
     ++	test_grep ! "clone> want " trace-packet.txt
      +'
      +
      +test_expect_success 'negotiation: bundle list (no heuristic)' '
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone bundle list (file, any m
      +	refs/bundles/left
      +	EOF
      +	test_cmp expect actual &&
     -+	test_grep "clone> have $(git -C nego-bundle-list-no-heuristic rev-parse refs/bundles/left)" trace-packet.txt
     ++	tip=$(git -C nego-bundle-list-no-heuristic rev-parse refs/bundles/left) &&
     ++	test_grep "clone> have $tip" trace-packet.txt
      +'
      +
      +test_expect_success 'negotiation: bundle list (creationToken)' '
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone bundle list (file, any m
      +	refs/bundles/left
      +	EOF
      +	test_cmp expect actual &&
     -+	test_grep "clone> have $(git -C nego-bundle-list-heuristic rev-parse refs/bundles/left)" trace-packet.txt
     ++	tip=$(git -C nego-bundle-list-heuristic rev-parse refs/bundles/left) &&
     ++	test_grep "clone> have $tip" trace-packet.txt
      +'
      +
      +test_expect_success 'negotiation: bundle list with all wanted commits' '
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone bundle list (file, any m
      +	EOF
      +	test_cmp expect actual &&
      +	# We already have all needed commits so no "want" needed.
     -+	! grep "clone> want " trace-packet.txt
     ++	test_grep ! "clone> want " trace-packet.txt
      +'
      +
       #########################################################################
 2:  3dc0d9dd22f ! 2:  518584c8698 fetch-pack: expose fsckObjects configuration logic
     @@ Commit message
          "fetch.fsckObjects" to control checks for broken objects in received
          packs during fetches. However, these configurations were only
          acknowledged by `fetch-pack.c:get_pack` and did not take effect in
     -    direct bundle fetches and fetches with _bundle-uri_ enabled.
     +    direct bundle fetches or fetches with _bundle-uri_ enabled.
      
          This commit exposes the fetch-then-transfer configuration logic by
          adding a new function `fetch_pack_fsck_objects` in fetch-pack.h. This
          new function is used to replace the assignment for `fsck_objects` in
     -    `fetch-pack.c:get_pack`. In the next commit, it will also be used by
     -    `bundle.c:unbundle` to better fit fetching scenarios.
     +    `fetch-pack.c:get_pack`. In the next commit, this function will also be
     +    used to extend fsck support for bundle-involved fetches.
      
          Helped-by: Junio C Hamano <gitster@pobox.com>
          Helped-by: Patrick Steinhardt <ps@pks.im>
 3:  2f15099bbb9 ! 3:  698dd6e49b7 unbundle: extend object verification for fetches
     @@ bundle.h: int create_bundle(struct repository *r, const char *path,
      
       ## t/t5558-clone-bundle-uri.sh ##
      @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'create bundle' '
     - 		sed -e "/^$/q" -e "s/$(git rev-parse A) /$(git rev-parse B) /" \
     + 		sed -e "/^$/q" -e "s/$commit_a /$commit_b /" \
       			<A.bundle >bad-header.bundle &&
       		convert_bundle_to_pack \
      -			<A.bundle >>bad-header.bundle
      +			<A.bundle >>bad-header.bundle &&
      +
     ++		tree_b=$(git rev-parse B^{tree}) &&
      +		cat >data <<-EOF &&
     -+		tree $(git rev-parse HEAD^{tree})
     -+		parent $(git rev-parse HEAD)
     ++		tree $tree_b
     ++		parent $commit_b
      +		author A U Thor
      +		committer A U Thor
      +
      +		commit: this is a commit with bad emails
      +
      +		EOF
     -+		git hash-object --literally -t commit -w --stdin <data >commit &&
     -+		git branch bad $(cat commit) &&
     ++		bad_commit=$(git hash-object --literally -t commit -w --stdin <data) &&
     ++		git branch bad $bad_commit &&
      +		git bundle create bad-object.bundle bad &&
      +		git update-ref -d refs/heads/bad
       	)
       '
       
      @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone with bundle that has bad header' '
     - 	! grep "refs/bundles/" refs
     + 	test_grep ! "refs/bundles/" refs
       '
       
      +test_expect_success 'clone with bundle that has bad object' '
     -+	# Unbundle succeeds if no fsckObjects confugured.
     ++	# Unbundle succeeds if no fsckObjects configured.
      +	git clone --bundle-uri="clone-from/bad-object.bundle" \
      +		clone-from clone-bad-object-no-fsck &&
      +	git -C clone-bad-object-no-fsck for-each-ref --format="%(refname)" >refs &&
     @@ t/t5558-clone-bundle-uri.sh: test_expect_success 'clone with bundle that has bad
      +		clone-from clone-bad-object-fsck 2>err &&
      +	test_grep "missingEmail" err &&
      +	git -C clone-bad-object-fsck for-each-ref --format="%(refname)" >refs &&
     -+	! grep "refs/bundles/" refs
     ++	test_grep ! "refs/bundles/" refs
      +'
      +
       test_expect_success 'clone with path bundle and non-default hash' '
     @@ t/t5607-clone-bundle.sh: test_expect_success 'fetch SHA-1 from bundle' '
      +	test_create_repo bundle-fsck &&
      +	(
      +		cd bundle-fsck &&
     -+		test_commit first &&
     ++		test_commit A &&
     ++		commit_a=$(git rev-parse A) &&
     ++		tree_a=$(git rev-parse A^{tree}) &&
      +		cat >data <<-EOF &&
     -+		tree $(git rev-parse HEAD^{tree})
     -+		parent $(git rev-parse HEAD)
     ++		tree $tree_a
     ++		parent $commit_a
      +		author A U Thor
      +		committer A U Thor
      +
      +		commit: this is a commit with bad emails
      +
      +		EOF
     -+		git hash-object --literally -t commit -w --stdin <data >commit &&
     -+		git branch bad $(cat commit) &&
     ++		bad_commit=$(git hash-object --literally -t commit -w --stdin <data) &&
     ++		git branch bad $bad_commit &&
      +		git bundle create bad.bundle bad
      +	) &&
      +

From: Xing Xin <xingxin.xx@bytedance.com>

Currently, we can use "transfer.fsckObjects" and the more specific "fetch.fsckObjects" to control checks for broken objects in received packs during fetches. However, these configurations were only acknowledged by `fetch-pack.c:get_pack` and did not take effect in direct bundle fetches or fetches with _bundle-uri_ enabled.

This commit exposes the fetch-then-transfer configuration logic by adding a new function `fetch_pack_fsck_objects` in fetch-pack.h. This new function is used to replace the assignment for `fsck_objects` in `fetch-pack.c:get_pack`. In the next commit, this function will also be used to extend fsck support for bundle-involved fetches.

Helped-by: Junio C Hamano <gitster@pobox.com>
Helped-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 fetch-pack.c | 17 +++++++++++------
 fetch-pack.h |  5 +++++
 2 files changed, 16 insertions(+), 6 deletions(-)

From: Xing Xin <xingxin.xx@bytedance.com>

When using the bundle-uri mechanism with a bundle list containing multiple interrelated bundles, we encountered a bug where tips from downloaded bundles were not discovered, thus resulting in rather slow clones. This was particularly problematic when employing the "creationTokens" heuristic.

To reproduce this issue, consider a repository with a single branch "main" pointing to commit "A". Firstly, create a base bundle with:

  git bundle create base.bundle main

Then, add a new commit "B" on top of "A", and create an incremental bundle for "main":

  git bundle create incr.bundle A..main

Now, generate a bundle list with the following content:

  [bundle]
      version = 1
      mode = all
      heuristic = creationToken

  [bundle "base"]
      uri = base.bundle
      creationToken = 1

  [bundle "incr"]
      uri = incr.bundle
      creationToken = 2

A fresh clone with the bundle list above should result in a reference "refs/bundles/main" pointing to "B" in the new repository. However, git would still download everything from the server, as if it had fetched nothing locally.

So why the "refs/bundles/main" is not discovered? After some digging I found that:

1. Bundles in bundle list are downloaded to local files via
   `bundle-uri.c:download_bundle_list` or via
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
2. Each bundle is unbundled via `bundle-uri.c:unbundle_from_file`, which
   is called by `bundle-uri.c:unbundle_all_bundles` or called within
   `bundle-uri.c:fetch_bundles_by_token` for the "creationToken"
   heuristic.
3. To get all prerequisites of the bundle, the bundle header is read
   inside `bundle-uri.c:unbundle_from_file` to by calling
   `bundle.c:read_bundle_header`.
4. Then it calls `bundle.c:unbundle`, which calls
   `bundle.c:verify_bundle` to ensure the repository contains all the
   prerequisites.
5. `bundle.c:verify_bundle` calls `parse_object`, which eventually
   invokes `packfile.c:prepare_packed_git` or
   `packfile.c:reprepare_packed_git`, filling
   `raw_object_store->packed_git` and setting `packed_git_initialized`.
6. If `bundle.c:unbundle` succeeds, it writes refs via
   `refs.c:refs_update_ref` with `REF_SKIP_OID_VERIFICATION` set. Here
   bundle refs which can target arbitrary objects are written to the
   repository.
7. Finally, in `fetch-pack.c:do_fetch_pack_v2`, the functions
   `fetch-pack.c:mark_complete_and_common_ref` and
   `fetch-pack.c:mark_tips` are called with `OBJECT_INFO_QUICK` set to
   find local tips for negotiation. The `OBJECT_INFO_QUICK` flag
   prevents `packfile.c:reprepare_packed_git` from being called,
   resulting in failures to parse OIDs that reside only in the latest
   bundle.

In the example above, when unbunding "incr.bundle", "base.pack" is added to `packed_git` due to prerequisites verification. However, "B" cannot be found for negotiation because it exists in "incr.pack", which is not included in `packed_git`.

Fix the bug by removing `REF_SKIP_OID_VERIFICATION` flag when writing bundle refs. When `refs.c:refs_update_ref` is called to write the corresponding bundle refs, it triggers `refs.c:ref_transaction_commit`. This, in turn, invokes `refs.c:ref_transaction_prepare`, which calls `transaction_prepare` of the refs storage backend. For files backend, it is `files-backend.c:files_transaction_prepare`, and for reftable backend, it is `reftable-backend.c:reftable_be_transaction_prepare`. Both functions eventually call `object.c:parse_object`, which can invoke `packfile.c:reprepare_packed_git` to refresh `packed_git`. This ensures that bundle refs point to valid objects and that all tips from bundle refs are correctly parsed during subsequent negotiations.

A set of negotiation-related tests for cloning with bundle-uri has been included to demonstrate that downloaded bundles are utilized to accelerate fetching.

Additionally, another test has been added to show that bundles with incorrect headers, where refs point to non-existent objects, do not result in any bundle refs being created in the repository.

Reviewed-by: Karthik Nayak <karthik.188@gmail.com>
Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |   3 +-
 t/t5558-clone-bundle-uri.sh | 155 +++++++++++++++++++++++++++++++++++-
 2 files changed, 152 insertions(+), 6 deletions(-)

From: Xing Xin <xingxin.xx@bytedance.com>

The existing fetch.fsckObjects and transfer.fsckObjects configurations were not fully applied to bundle-involved fetches, including direct bundle fetches and bundle-uri enabled fetches. Furthermore, there was no object verification support for unbundle.

This commit extends object verification support in `bundle.c:unbundle` by adding the `VERIFY_BUNDLE_FSCK` option to `verify_bundle_flags`. When this option is enabled, we append the `--fsck-objects` flag to `git-index-pack`.

The `VERIFY_BUNDLE_FSCK` option is now used by bundle-involved fetches, where we use `fetch-pack.c:fetch_pack_fsck_objects` to determine whether to enable this option for `bundle.c:unbundle`, specifically in:

- `transport.c:fetch_refs_from_bundle` for direct bundle fetches.
- `bundle-uri.c:unbundle_from_file` for bundle-uri enabled fetches.

This addition ensures a consistent logic for object verification during fetches. Tests have been added to confirm functionality in the scenarios mentioned above.

Reviewed-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Xing Xin <xingxin.xx@bytedance.com>
---
 bundle-uri.c                |  3 ++-
 bundle.c                    |  3 +++
 bundle.h                    |  1 +
 t/t5558-clone-bundle-uri.sh | 34 +++++++++++++++++++++++++++++++++-
 t/t5607-clone-bundle.sh     | 35 +++++++++++++++++++++++++++++++++++
 transport.c                 |  3 ++-
 6 files changed, 76 insertions(+), 3 deletions(-)