public inbox for git@vger.kernel.org 
 help / color / mirror / Atom feed
From: Andy Whitcroft <apw@shadowen•org>
To: Junio C Hamano <junkio@cox•net>
Cc: Andy Whitcroft <apw@shadowen•org>, git@vger•kernel.org
Subject: Re: [PATCH] use xread where we are not checking for EAGAIN/EINTR
Date: Mon, 08 Jan 2007 15:56:15 +0000	[thread overview]
Message-ID: <45A2699F.5060100@shadowen.org> (raw)
In-Reply-To: <459E4270.9090307@shadowen.org>

[-- Attachment #1: Type: text/plain, Size: 1869 bytes --]

Andy Whitcroft wrote:
> Junio C Hamano wrote:
>> Andy Whitcroft <apw@shadowen•org> writes:
>>
>>>     We have an xread() wrapper to help us with those nasty
>>>     interrupt returns and yet we fail to use it consistently.
>>>     This patch updates those plain read()'s which do not
>>>     have any handling for errors, or which treat those errors
>>>     as user visible fatal errors.
>>>
>>>     This feels right to me, but perhaps there is some good
>>>     reason that things are done this way ... if so could
>>>     someone elighten me.
>> Thanks.
>>
>> I do not think any of the changes you did introduced new bugs,
>> but I think some of them are still wrong.  xread() protects us
>> from EINTR happening before any byte is read, but it can still
>> give a short read.  Many callers have a loop like this:
>>
>> 	do {
>>         	size = xread(...);
>>                 yet_to_go -= size;
>> 	} while (yet_to_go);
>>
>> but some are not (e.g. add_excludes_from_file_1() in dir.c
>> expects xread() does not return before reading full buffer).
> 
> Yes, that is true.  I was going to fix that in the next step with the
> writes.  But yes thats likely to involve them becoming 'read_in_full'
> style thing and in fact churn us more.
> 
> Ignore this one and I'll look to do it 'right'.

Ok, after much hacking about I think I've got something sensible sorted
out for this.  Following this email are four patches which convert the
various read/write/xread/xwrite users over to xread/read_in_full and
xwrite/write_in_full as appropriate.  Its pretty invasive as obviously
I/O is pretty common in an SCM.

next with this stack passes the builtin test suite.  I additionally used
the attached patch to induce severe short read/write semantics.  Before
this patch series git was _very_ unhappy, unable to even create a
repository.

This series is against next.

-apw


[-- Attachment #2: PATCH --]
[-- Type: text/plain, Size: 688 bytes --]

diff --git a/git-compat-util.h b/git-compat-util.h
index 55456da..931cbed 100644
--- a/git-compat-util.h
+++ b/git-compat-util.h
@@ -210,6 +210,7 @@ static inline void *xmmap(void *start, size_t length,
 static inline ssize_t xread(int fd, void *buf, size_t len)
 {
 	ssize_t nr;
+	if (len > 5) len = 5;
 	while (1) {
 		nr = read(fd, buf, len);
 		if ((nr < 0) && (errno == EAGAIN || errno == EINTR))
@@ -221,6 +222,7 @@ static inline ssize_t xread(int fd, void *buf, size_t len)
 static inline ssize_t xwrite(int fd, const void *buf, size_t len)
 {
 	ssize_t nr;
+	if (len > 5) len = 5;
 	while (1) {
 		nr = write(fd, buf, len);
 		if ((nr < 0) && (errno == EAGAIN || errno == EINTR))

  reply	other threads:[~2007-01-08 15:56 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2007-01-05 10:54 [PATCH] use xread where we are not checking for EAGAIN/EINTR Andy Whitcroft
2007-01-05 11:19 ` Junio C Hamano
2007-01-05 12:20   ` Andy Whitcroft
2007-01-08 15:56     ` Andy Whitcroft [this message]
2007-01-08 15:57       ` [PATCH 1/4] short i/o: clean up the naming for the write_{in,or}_xxx family Andy Whitcroft
2007-01-08 15:58       ` [PATCH 2/4] short i/o: fix calls to read to use xread or read_in_full Andy Whitcroft
2007-01-08 15:58       ` [PATCH 3/4] short i/o: fix calls to write to use xwrite or write_in_full Andy Whitcroft
2007-01-08 15:58       ` [PATCH 4/4] short i/o: fix config updates to use write_in_full Andy Whitcroft
2007-01-08 20:13       ` [PATCH] use xread where we are not checking for EAGAIN/EINTR Junio C Hamano
2007-01-11 21:43       ` [PATCH] Avoid errors and warnings when attempting to do I/O on zero bytes Eric Wong

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

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

  git send-email \
    --in-reply-to=45A2699F.5060100@shadowen.org \
    --to=apw@shadowen$(echo .)org \
    --cc=git@vger$(echo .)kernel.org \
    --cc=junkio@cox$(echo .)net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox