git@vger.kernel.org list mirror (unofficial, one of many)
 help / color / mirror / code / Atom feed
From: Jiang Xin <worldhello.net@gmail.com>
To: Junio C Hamano <gitster@pobox.com>
Cc: Jiang Xin <worldhello.net@gmail.com>,
	Han Xin <chiyutianyi@gmail.com>,
	Han Xin <hanxin.hx@alibaba-inc.com>,
	Git List <git@vger.kernel.org>,
	Jiang Xin <zhiyou.jx@alibaba-inc.com>
Subject: Re: [PATCH 2/2] send-pack: check atomic push before running GPG
Date: Wed, 16 Sep 2020 07:49:58 -0400	[thread overview]
Message-ID: <20200916114958.1123-1-worldhello.net@gmail.com> (raw)
In-Reply-To: <xmqqeen2xrok.fsf@gitster.c.googlers.com>

Junio C Hamano <gitster@pobox.com> writes:\r
\r
> > The next block ("Finally, tell the other end!") is what we send\r
> > commands to "receive-pack", right after some of the status are reset\r
> > to REF_STATUS_OK or REF_STATUS_EXPECTING_REPORT by this chunk of code.\r
> > So moving the generate_push_cert() part right before the "Finally,\r
> > tell the other end!" part LGTM.\r
> \r
> Sorry, I do not follow.  The loop in question is the one before\r
> "Finally tell the other end".  The loop ends like so:\r
> \r
> 	for (ref = remote_refs; ref; ref = ref->next) {\r
> 		...\r
> 		if (args->dry_run || !status_report)\r
> 			ref->status = REF_STATUS_OK;\r
> 		else\r
> 			ref->status = REF_STATUS_EXPECTING_REPORT;\r
> 	}\r
> \r
> and the patch moves a call to generate_push_cert() that looks at\r
> remote_refs _after_ this loop, but generate_push_cert() function\r
> uses a loop over remote_refs that uses check_to_send_update(), which\r
> looks at ref->status's value to decide what to do.  Its correct\r
> operation relies on ref->status NOT updated by the above loop.\r
> \r
\r
To make it clear, I refactor the Han Xin's patch, quote and add comments\r
as follows (changes on whitespace are ignored):\r
\r
\r
>>         /*\r
>>          * NEEDSWORK: why does delete-refs have to be so specific to\r
>>          * send-pack machinery that set_ref_status_for_push() cannot\r
>>          * set this bit for us???\r
>>          */\r
>>         for (ref = remote_refs; ref; ref = ref->next)\r
>>             if (ref->deletion && !allow_deleting_refs)\r
>>                 ref->status = REF_STATUS_REJECT_NODELETE;\r
>>     \r
>>    -    if (!args->dry_run)\r
>>    -        advertise_shallow_grafts_buf(&req_buf);\r
>>    -\r
>>    -    if (!args->dry_run && push_cert_nonce)\r
>>    -        cmds_sent = generate_push_cert(&req_buf, remote_refs, args,\r
>>    -                                       cap_buf.buf, push_cert_nonce);\r
>>    -\r
>>         /*\r
>>          * Clear the status for each ref and see if we need to send\r
>>          * the pack data.\r
>>          */\r
>>         for (ref = remote_refs; ref; ref = ref->next) {\r
>>             switch (check_to_send_update(ref, args)) {\r
>>             case 0: /* no error */\r
>>                 break;\r
>>             case CHECK_REF_STATUS_REJECTED:\r
>>                 /*\r
>>                  * When we know the server would reject a ref update if\r
>>                  * we were to send it and we're trying to send the refs\r
>>                  * atomically, abort the whole operation.\r
>>                  */\r
>>                 if (use_atomic) {\r
>>                     strbuf_release(&req_buf);\r
>>                     strbuf_release(&cap_buf);\r
>>                     reject_atomic_push(remote_refs, args->send_mirror);\r
>>                     error("atomic push failed for ref %s. status: %d\n",\r
>>                           ref->name, ref->status);\r
>>                     return args->porcelain ? 0 : -1;\r
>>                 }\r
>>                 /* else fallthrough */\r
>>             default:\r
>>                 continue;\r
>>             }\r
>>             if (!ref->deletion)\r
>>                 need_pack_data = 1;\r
>>     \r
>>             if (args->dry_run || !status_report)\r
>>                 ref->status = REF_STATUS_OK;\r
>>             else\r
>>                 ref->status = REF_STATUS_EXPECTING_REPORT;\r
>>         }\r
>>     \r
>>    +    if (!args->dry_run)\r
>>    +        advertise_shallow_grafts_buf(&req_buf);\r
>>    +\r
>>    +\r
>>         /*\r
>>          * Finally, tell the other end!\r
>>          */\r
>>    +    if (!args->dry_run && push_cert_nonce)\r
>>    +        cmds_sent = generate_push_cert(&req_buf, remote_refs, args,\r
>>    +                           cap_buf.buf, push_cert_nonce);\r
\r
Moving `generate_push_cert()` here, will: \r
1. Increase the perforcemance a little bit for failed atomic push.\r
2. Make it clear that the commands will be sent only once.\r
   For GPG-signed push, commands will be sent via `generate_push_cert()`,\r
   and for non-GPG-signed push, commands will be sent using the following code.\r
3. For GPG-signed push, won't run the following loop.\r
\r
>>    +    else if (!args->dry_run)\r
>>             for (ref = remote_refs; ref; ref = ref->next) {\r
>>                 char *old_hex, *new_hex;\r
>>     \r
>>    -            if (args->dry_run || push_cert_nonce)\r
>>    -                continue;\r
>>    -\r
>>                 if (check_to_send_update(ref, args) < 0)\r
>>                     continue;\r
\r
In the original "Finally, tell the other end" block, the function\r
`check_to_send_update()` is also called for non-PGP-signed push.\r
The 'ref->status' changed by the "Clear the status" block won't \r
make any difference for the return value of the function\r
`check_to_send_update()`. Refs even with status REF_STATUS_OK and\r
REF_STATUS_EXPECTING_REPORT will be sent to the server side.\r
\r
>>     \r
>>                 old_hex = oid_to_hex(&ref->old_oid);\r
>>                 new_hex = oid_to_hex(&ref->new_oid);\r
>>                 if (!cmds_sent) {\r
>>                     packet_buf_write(&req_buf,\r
>>                              "%s %s %s%c%s",\r
>>                              old_hex, new_hex, ref->name, 0,\r
>>                              cap_buf.buf);\r
>>                     cmds_sent = 1;\r
>>                 } else {\r
>>                     packet_buf_write(&req_buf, "%s %s %s",\r
>>                              old_hex, new_hex, ref->name);\r
>>                 }\r
>>             }\r
\r
\r
> The loop prepares the status field so that we can then read and\r
> record the response against each ref updates from the other side.\r
> \r
> The ref->status field is set to EXPECTING_REPORT, later to be\r
> updated to REF_STATUS_OK or REF_STATUS_REMOTE_REJECT.  We can\r
> clobber the original value of ref->status at this point only because\r
> the loop depends on the fact that no check_to_send_update() call\r
> will happen after the loop (because the ref->status field the\r
> helper's operation depends on is already reset for the next phase of\r
> operation).  The patch that moves generate_push_cert() call below\r
> the loop, whether it is before or after the "Finally tell the other\r
> end" loop, is therefore fundamentally broken, isn't it?\r
> \r
> I do not think it would introduce such breakage if we teach\r
> generate_push_cert() to pay attention to the atomicity, and that can\r
> be done without reordering the calls in send_pack() to break the\r
> control flow.\r
\r

  reply	other threads:[~2020-09-16 20:23 UTC|newest]

Thread overview: 17+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-09-15  9:58 [PATCH 1/2] t5534: new test case for atomic signed push Han Xin
2020-09-15  9:58 ` [PATCH 2/2] send-pack: check atomic push before running GPG Han Xin
2020-09-15 21:02   ` Junio C Hamano
2020-09-15 21:40     ` Junio C Hamano
2020-09-16  1:53     ` Jiang Xin
2020-09-16  4:37       ` Junio C Hamano
2020-09-16 11:49         ` Jiang Xin [this message]
2020-09-16 23:44           ` Junio C Hamano
2020-09-18  4:50             ` [PATCH v2] send-pack: run GPG after atomic push checking Han Xin
2020-09-19  0:02               ` Junio C Hamano
2020-09-19 14:47                 ` [PATCH v3] " Han Xin
2020-09-19 23:02                   ` Junio C Hamano
2020-09-20  6:20                     ` [PATCH v4] " Han Xin
2020-09-16 17:35         ` [PATCH 2/2] send-pack: check atomic push before running GPG 韩欣(炽天)
2020-09-15 20:31 ` [PATCH 1/2] t5534: new test case for atomic signed push Junio C Hamano
2020-09-16  0:34   ` brian m. carlson
2020-09-15 20:34 ` Junio C Hamano

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

  List information: http://vger.kernel.org/majordomo-info.html

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200916114958.1123-1-worldhello.net@gmail.com \
    --to=worldhello.net@gmail.com \
    --cc=chiyutianyi@gmail.com \
    --cc=git@vger.kernel.org \
    --cc=gitster@pobox.com \
    --cc=hanxin.hx@alibaba-inc.com \
    --cc=zhiyou.jx@alibaba-inc.com \
    --subject='Re: [PATCH 2/2] send-pack: check atomic push before running GPG' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

Code repositories for project(s) associated with this inbox:

	https://80x24.org/mirrors/git.git

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).