[stgt] [PATCH] fix check of sent len in mtask_recv_send_handler(), report exceptions in concat_write()

Brad.Goodman at emc.com Brad.Goodman at emc.com
Wed Apr 18 14:43:42 CEST 2012


Trying the patch now - all indications show it has resolved the issue!

Thanks!

-BKG

> -----Original Message-----
> From: stgt-owner at vger.kernel.org [mailto:stgt-owner at vger.kernel.org] On
> Behalf Of Brad.Goodman at emc.com
> Sent: Tuesday, April 17, 2012 8:38 PM
> To: fujita.tomonori at lab.ntt.co.jp
> Cc: nezhinsky at gmail.com; stgt at vger.kernel.org; alexandern at mellanox.com
> Subject: Re: [PATCH] fix check of sent len in
> mtask_recv_send_handler(), report exceptions in concat_write()
> 
> Absolutely. We will try it first thing in the morning.
> 
> Thanks again,
> 
> -BKG
> 
> Sent from my iPhone (Please excuse any typos)
> 
> On Apr 17, 2012, at 8:24 PM, "FUJITA Tomonori"
> <fujita.tomonori at lab.ntt.co.jp> wrote:
> 
> > CC'ed Brad,
> >
> > Can you try the patch?
> >
> > Thanks,
> >
> > On Wed, 18 Apr 2012 01:17:35 +0300
> > nezhinsky at gmail.com wrote:
> >
> >> From: Alexander Nezhinsky <alexandern at mellanox.com>
> >>
> >> This patch hopefully fixes the problem reported by Brad. Please
> verify.
> >>
> >> 1) mtask->done is reset to 0 when transitioning to
> MTASK_STATE_PDU_SEND because it is used
> >> as the offset into concat_buf, while rsp->len accounts for the total
> response length, incl. the header;
> >> thus the check if the entire mtask resp is sent should subtract
> header len.
> >> 2) when writing concat_buf to a socket file, if the buffer offset is
> past the buffer end,
> >> return -1 and set errno=EINVAL instead of just returning 0.
> >> The problem is that concat_write() regularly returns the retcode
> from write(), which is either
> >> the number of written bytes or -1 (and then errno describes the
> error). Thus when we avoid
> >> calling write() the same convention should be simulated.
> >> Similar fix in case of a previous error, then concat_buf->err stores
> the relevant errno,
> >> so -1 is returned and errno is reset to the error code.
> >>
> >> Signed-off-by: Alexander Nezhinsky <alexandern at mellanox.com>
> >> ---
> >> usr/concat_buf.c |   12 ++++++++----
> >> usr/mgmt.c       |    4 ++--
> >> 2 files changed, 10 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/usr/concat_buf.c b/usr/concat_buf.c
> >> index 9395f4b..00661d4 100644
> >> --- a/usr/concat_buf.c
> >> +++ b/usr/concat_buf.c
> >> @@ -74,13 +74,17 @@ int concat_write(struct concat_buf *b, int fd,
> int offset)
> >> {
> >>    concat_buf_finish(b);
> >>
> >> -    if (b->err)
> >> -        return b->err;
> >> +    if (b->err) {
> >> +        errno = b->err;
> >> +        return -1;
> >> +    }
> >>
> >>    if (b->size - offset > 0)
> >>        return write(fd, b->buf + offset, b->size - offset);
> >> -    else
> >> -        return 0;
> >> +    else {
> >> +        errno = EINVAL;
> >> +        return -1;
> >> +    }
> >> }
> >>
> >> void concat_buf_release(struct concat_buf *b)
> >> diff --git a/usr/mgmt.c b/usr/mgmt.c
> >> index 8b62d13..8915dea 100644
> >> --- a/usr/mgmt.c
> >> +++ b/usr/mgmt.c
> >> @@ -590,9 +590,9 @@ static void mtask_recv_send_handler(int fd, int
> events, void *data)
> >>        break;
> >>    case MTASK_STATE_PDU_SEND:
> >>        err = concat_write(&mtask->rsp_concat, fd, mtask->done);
> >> -        if (err > 0) {
> >> +        if (err >= 0) {
> >>            mtask->done += err;
> >> -            if (mtask->done == rsp->len) {
> >> +            if (mtask->done == (rsp->len - sizeof(*rsp))) {
> >>                if (req->mode == MODE_SYSTEM &&
> >>                    req->op == OP_DELETE &&
> >>                    !rsp->err)
> >> --
> >> 1.7.9.6
> >>
> >> --
> >> To unsubscribe from this list: send the line "unsubscribe stgt" in
> >> the body of a message to majordomo at vger.kernel.org
> >> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> >>
> >>
> >
> --
> To unsubscribe from this list: send the line "unsubscribe stgt" in
> the body of a message to majordomo at vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe stgt" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html



More information about the stgt mailing list