xfs
[Top] [All Lists]

Re: [PATCH 15/30] db: separate out straight buffer IO from map based IO.

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 15/30] db: separate out straight buffer IO from map based IO.
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Thu, 31 Oct 2013 07:56:09 -0700
Cc: xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <1383107481-28937-16-git-send-email-david@xxxxxxxxxxxxx>
References: <1383107481-28937-1-git-send-email-david@xxxxxxxxxxxxx> <1383107481-28937-16-git-send-email-david@xxxxxxxxxxxxx>
User-agent: Mutt/1.5.21 (2010-09-15)
On Wed, Oct 30, 2013 at 03:31:06PM +1100, Dave Chinner wrote:
> From: Dave Chinner <dchinner@xxxxxxxxxx>
> 
> First step in converting to libxfs based IO.
> 
> Signed-off-by: Dave Chinner <dchinner@xxxxxxxxxx>

The patch description is a little too short, there's not real
explanation of what it actually does.

> -     if (read_bbs(XFS_SB_DADDR, 1, &bufp, NULL)) {
> +     if (read_buf(XFS_SB_DADDR, 1, bufp)) {

E.g. why isn't this already using the normal libxfs routines?

(there probably is an explanation but I don't quite see it yet..)

>  int
> +read_buf(
> +     xfs_daddr_t     bbno,
> +     int             count,
> +     void            *bufp)
> +{

Is read_buf really a good name for something that is a trivial pread
wrapper and doesn't deal with buffers?  Should the function have some
comments explaining when to use it?  The same also applies to the write
side.

> +     int             err;
> +
> +     err = pwrite64(x.dfd, bufp, BBTOB(count), BBTOB(bbno));
> +     if (err < 0)
> +             err = errno;
> +     else if (err < count)
> +             err = -1;
> +     return err;
> +}

> +static void
> +write_cur_buf(void)
> +{
> +     int ret;
> +
> +     ret = write_buf(iocur_top->bb, iocur_top->blen, iocur_top->buf);
> +
> +     if (ret == -1)
> +             dbprintf(_("incomplete write, block: %lld\n"),
> +                      (iocur_base + iocur_sp)->bb);
> +     else if (ret != 0)
> +             dbprintf(_("write error: %s\n"), strerror(ret));
> +
> +     /* re-read buffer from disk */
> +     ret = read_buf(iocur_top->bb, iocur_top->blen, iocur_top->buf);
> +     if (ret == -1)
> +             dbprintf(_("incomplete read, block: %lld\n"),
> +                      (iocur_base + iocur_sp)->bb);
> +     else if (ret != 0)
> +             dbprintf(_("read error: %s\n"), strerror(ret));
> +}

What is the point of the write and re-read cycle?

> +     for (j = 0; j < count; j++) {
> +             bbno = bbmap->b[j];
>               if (lseek64(x.dfd, bbno << BBSHIFT, SEEK_SET) < 0) {
>                       rval = errno;
>                       dbprintf(_("can't seek in filesystem at bb %lld\n"), 
> bbno);
>                       return rval;
>               }
> -             c = BBTOB(bbmap ? 1 : count);
> +             c = BBTOB(1);
>               i = (int)write(x.dfd, (char *)bufp + BBTOB(j), c);

Shoiuldn't this use the write_buf helper above?

> +     for (j = 0; j < count; j++) {
> +             bbno = bbmap->b[j];
>               if (lseek64(x.dfd, bbno << BBSHIFT, SEEK_SET) < 0) {
>                       rval = errno;
>                       dbprintf(_("can't seek in filesystem at bb %lld\n"), 
> bbno);
> @@ -483,7 +534,7 @@ read_bbs(
>                               xfree(buf);
>                       buf = NULL;
>               } else {
> -                     c = BBTOB(bbmap ? 1 : count);
> +                     c = BBTOB(1);
>                       i = (int)read(x.dfd, (char *)buf + BBTOB(j), c);

And read_buf here?

<Prev in Thread] Current Thread [Next in Thread>