code review 1032041: syscall: force O_LARGEFILE in Linux open system call

syscall: force O_LARGEFILE in Linux open system call

Fixes issue 717.

[rsc, 2010-04-28 22:23:29]

Hello adg (cc: golang-dev@googlegroups.com),

I'd like you to review this change.

[peterGo, 2010-04-28 23:34:22]

On 2010/04/28 22:23:29, rsc wrote:
> Hello adg (cc: mailto:golang-dev@googlegroups.com),
> 
> I'd like you to review this change.

Russ,
I'll take a look too, and run some tests.
Peter

[adg, 2010-04-29 00:54:13]

LGTM
but I'd like to hear if it resolves Peter's issue.

Are there side-effects to using O_LARGEFILE ? If not, why wouldn't it
be the default?

On 29 April 2010 08:23,  <rsc@golang.org> wrote:
> Reviewers: adg,
>
> Message:
> Hello adg (cc: golang-dev@googlegroups.com),
>
> I'd like you to review this change.
>
>
> Description:
> syscall: force O_LARGEFILE in Linux open system call
>
> Fixes issue 717.
>
> Please review this at http://codereview.appspot.com/1032041/show
>
> Affected files:
>  M src/pkg/syscall/mkall.sh
>  M src/pkg/syscall/mkerrors.sh
>  M src/pkg/syscall/syscall_linux.go
>  M src/pkg/syscall/zerrors_linux_386.go
>  M src/pkg/syscall/zerrors_linux_amd64.go
>  M src/pkg/syscall/zsyscall_linux_386.go
>  M src/pkg/syscall/zsyscall_linux_amd64.go
>
>
>

[rsc, 2010-04-29 02:35:47]

> Are there side-effects to using O_LARGEFILE ? If not, why wouldn't it
> be the default?

it's a clumsy hack to signal to the kernel whether
the library is aware of 64-bit file offsets or not.

[rsc, 2010-04-30 06:34:13]

*** Submitted as http://code.google.com/p/go/source/detail?r=907f7b4512eb ***

syscall: force O_LARGEFILE in Linux open system call

Fixes issue 717.

R=adg, PeterGo
CC=golang-dev
http://codereview.appspot.com/1032041

[cw, 2010-04-30 07:36:02]

questions

http://codereview.appspot.com/1032041/diff/6002/9003
File src/pkg/syscall/syscall_linux.go (left):

http://codereview.appspot.com/1032041/diff/6002/9003#oldcode580
src/pkg/syscall/syscall_linux.go:580: //sys	Openat(dirfd int, path string, flags
int, mode int) (fd int, errno int)
Open at should get the same treatment surely? (& Creat)

http://codereview.appspot.com/1032041/diff/6002/9004
File src/pkg/syscall/zerrors_linux_386.go (left):

http://codereview.appspot.com/1032041/diff/6002/9004#oldcode2
src/pkg/syscall/zerrors_linux_386.go:2: // MACHINE GENERATED BY THE COMMAND
ABOVE; DO NOT EDIT
fine but...

http://codereview.appspot.com/1032041/diff/6002/9004
File src/pkg/syscall/zerrors_linux_386.go (right):

http://codereview.appspot.com/1032041/diff/6002/9004#newcode52
src/pkg/syscall/zerrors_linux_386.go:52: EAX                              = 0x6
seems like this should be here

http://codereview.appspot.com/1032041/diff/6002/9004#newcode61
src/pkg/syscall/zerrors_linux_386.go:61: EBP                              = 0x5
or this...

http://codereview.appspot.com/1032041/diff/6002/9004#newcode225
src/pkg/syscall/zerrors_linux_386.go:225: F_SETLKW64                       = 0xe
and these changed

http://codereview.appspot.com/1032041/diff/6002/9004#newcode554
src/pkg/syscall/zerrors_linux_386.go:554: WORDSIZE                         =
0x20
????

did we use the wrong defs (64) earlier not 32?

[rsc1, 2010-04-30 07:40:19]

http://codereview.appspot.com/1032041/diff/6002/9004
File src/pkg/syscall/zerrors_linux_386.go (right):

http://codereview.appspot.com/1032041/diff/6002/9004#newcode185
src/pkg/syscall/zerrors_linux_386.go:185: ESI                              = 0x3
haha

http://codereview.appspot.com/1032041/diff/6002/9004#newcode554
src/pkg/syscall/zerrors_linux_386.go:554: WORDSIZE                         =
0x20
On 2010/04/30 07:36:02, cw wrote:
> ????
> 
> did we use the wrong defs (64) earlier not 32?

Yes, actually.  See the changes to mkall.sh in this CL.
It turned up because O_LARGEFILE is 0 in 64-bit mode
so part of making this work was getting the right (non-zero)
value for O_LARGEFILE.

[rsc, 2010-04-30 07:42:34]

I'll do Openat tomorrow; Creat is helpless because it
doesn't take flags at all.

Russ