| Message ID | 20230623225513.2732256-5-dhowells@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [net-next,v5,01/16] tcp_bpf, smc, tls, espintcp, siw: Reduce MSG_SENDPAGE_NOTLAST usage | expand |
On Sat, Jun 24, 2023 at 12:55 AM David Howells <dhowells@redhat.com> wrote: > > Use sendmsg() and MSG_SPLICE_PAGES rather than sendpage in ceph when > transmitting data. For the moment, this can only transmit one page at a > time because of the architecture of net/ceph/, but if > write_partial_message_data() can be given a bvec[] at a time by the Hi David, write_partial_message_data() is net/ceph/messenger_v1.c specific, so it doesn't apply here. I would suggest squashing the two net/ceph patches into one since even the titles are the same. Also, we tend to use "libceph: " prefix for net/ceph changes. > iteration code, this would allow pages to be sent in a batch. > > Signed-off-by: David Howells <dhowells@redhat.com> > cc: Ilya Dryomov <idryomov@gmail.com> > cc: Xiubo Li <xiubli@redhat.com> > cc: Jeff Layton <jlayton@kernel.org> > cc: "David S. Miller" <davem@davemloft.net> > cc: Eric Dumazet <edumazet@google.com> > cc: Jakub Kicinski <kuba@kernel.org> > cc: Paolo Abeni <pabeni@redhat.com> > cc: Jens Axboe <axboe@kernel.dk> > cc: Matthew Wilcox <willy@infradead.org> > cc: ceph-devel@vger.kernel.org > cc: netdev@vger.kernel.org > --- > net/ceph/messenger_v2.c | 91 +++++++++-------------------------------- > 1 file changed, 19 insertions(+), 72 deletions(-) > > diff --git a/net/ceph/messenger_v2.c b/net/ceph/messenger_v2.c > index 301a991dc6a6..87ac97073e75 100644 > --- a/net/ceph/messenger_v2.c > +++ b/net/ceph/messenger_v2.c > @@ -117,91 +117,38 @@ static int ceph_tcp_recv(struct ceph_connection *con) > return ret; > } > > -static int do_sendmsg(struct socket *sock, struct iov_iter *it) > -{ > - struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS }; > - int ret; > - > - msg.msg_iter = *it; > - while (iov_iter_count(it)) { > - ret = sock_sendmsg(sock, &msg); > - if (ret <= 0) { > - if (ret == -EAGAIN) > - ret = 0; > - return ret; > - } > - > - iov_iter_advance(it, ret); > - } > - > - WARN_ON(msg_data_left(&msg)); > - return 1; > -} > - > -static int do_try_sendpage(struct socket *sock, struct iov_iter *it) > -{ > - struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS }; > - struct bio_vec bv; > - int ret; > - > - if (WARN_ON(!iov_iter_is_bvec(it))) > - return -EINVAL; > - > - while (iov_iter_count(it)) { > - /* iov_iter_iovec() for ITER_BVEC */ > - bvec_set_page(&bv, it->bvec->bv_page, > - min(iov_iter_count(it), > - it->bvec->bv_len - it->iov_offset), > - it->bvec->bv_offset + it->iov_offset); > - > - /* > - * sendpage cannot properly handle pages with > - * page_count == 0, we need to fall back to sendmsg if > - * that's the case. > - * > - * Same goes for slab pages: skb_can_coalesce() allows > - * coalescing neighboring slab objects into a single frag > - * which triggers one of hardened usercopy checks. > - */ > - if (sendpage_ok(bv.bv_page)) { > - ret = sock->ops->sendpage(sock, bv.bv_page, > - bv.bv_offset, bv.bv_len, > - CEPH_MSG_FLAGS); > - } else { > - iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len); > - ret = sock_sendmsg(sock, &msg); > - } > - if (ret <= 0) { > - if (ret == -EAGAIN) > - ret = 0; > - return ret; > - } > - > - iov_iter_advance(it, ret); > - } > - > - return 1; > -} > - > /* > * Write as much as possible. The socket is expected to be corked, > - * so we don't bother with MSG_MORE/MSG_SENDPAGE_NOTLAST here. > + * so we don't bother with MSG_MORE here. > * > * Return: > - * 1 - done, nothing (else) to write > + * >0 - done, nothing (else) to write It would be nice to avoid making tweaks like this to the outer interface as part of switching to a new internal API. > * 0 - socket is full, need to wait > * <0 - error > */ > static int ceph_tcp_send(struct ceph_connection *con) > { > + struct msghdr msg = { > + .msg_iter = con->v2.out_iter, > + .msg_flags = CEPH_MSG_FLAGS, > + }; > int ret; > > + if (WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter))) > + return -EINVAL; Previously, this WARN_ON + error applied only to the "try sendpage" path. There is a ton of kvec usage in net/ceph/messenger_v2.c, so I'm pretty sure that placing it here breaks everything. > + > + if (con->v2.out_iter_sendpage) > + msg.msg_flags |= MSG_SPLICE_PAGES; > + > dout("%s con %p have %zu try_sendpage %d\n", __func__, con, > iov_iter_count(&con->v2.out_iter), con->v2.out_iter_sendpage); > - if (con->v2.out_iter_sendpage) > - ret = do_try_sendpage(con->sock, &con->v2.out_iter); > - else > - ret = do_sendmsg(con->sock, &con->v2.out_iter); > + > + ret = sock_sendmsg(con->sock, &msg); > + if (ret > 0) > + iov_iter_advance(&con->v2.out_iter, ret); > + else if (ret == -EAGAIN) > + ret = 0; Hrm, is sock_sendmsg() now guaranteed to exhaust the iterator (i.e. a "short write" is no longer possible)? Unless that is the case, this is not an equivalent transformation. This is actually the reason for > * Return: > * 1 - done, nothing (else) to write specification which you also tweaked. It doesn't make sense for ceph_tcp_send() to return the number of bytes sent because the caller expects everything to be sent when a positive number is returned. Thanks, Ilya
Ilya Dryomov <idryomov@gmail.com> wrote: > write_partial_message_data() is net/ceph/messenger_v1.c specific, so it > doesn't apply here. I would suggest squashing the two net/ceph patches > into one since even the titles are the same. I would, but they're now applied to net-next, so we need to patch that. > > * Write as much as possible. The socket is expected to be corked, > > - * so we don't bother with MSG_MORE/MSG_SENDPAGE_NOTLAST here. > > + * so we don't bother with MSG_MORE here. > > * > > * Return: > > - * 1 - done, nothing (else) to write > > + * >0 - done, nothing (else) to write > > It would be nice to avoid making tweaks like this to the outer > interface as part of switching to a new internal API. Ok. I'll change that and wrap the sendmsg in a loop. Though, as I asked in an earlier reply, why is MSG_DONTWAIT used here? > > + if (WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter))) > > + return -EINVAL; > > Previously, this WARN_ON + error applied only to the "try sendpage" > path. There is a ton of kvec usage in net/ceph/messenger_v2.c, so I'm > pretty sure that placing it here breaks everything. This should have been removed as MSG_SPLICE_PAGES now accepts KVEC and XARRAY iterators also. Btw, is it feasible to use con->v2.out_iter_sendpage to apply MSG_SPLICE_PAGES to the iterator to be transmitted as a whole? It seems to be set depending on iterator type. David
On Mon, Jun 26, 2023 at 5:30 PM David Howells <dhowells@redhat.com> wrote: > > Ilya Dryomov <idryomov@gmail.com> wrote: > > > write_partial_message_data() is net/ceph/messenger_v1.c specific, so it > > doesn't apply here. I would suggest squashing the two net/ceph patches > > into one since even the titles are the same. > > I would, but they're now applied to net-next, so we need to patch that. I don't see a problem with that given that the patches themselves have major issues (i.e. it's not just a commit message/title nit). > > > > * Write as much as possible. The socket is expected to be corked, > > > - * so we don't bother with MSG_MORE/MSG_SENDPAGE_NOTLAST here. > > > + * so we don't bother with MSG_MORE here. > > > * > > > * Return: > > > - * 1 - done, nothing (else) to write > > > + * >0 - done, nothing (else) to write > > > > It would be nice to avoid making tweaks like this to the outer > > interface as part of switching to a new internal API. > > Ok. I'll change that and wrap the sendmsg in a loop. Though, as I asked in > an earlier reply, why is MSG_DONTWAIT used here? See my reply there. > > > > + if (WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter))) > > > + return -EINVAL; > > > > Previously, this WARN_ON + error applied only to the "try sendpage" > > path. There is a ton of kvec usage in net/ceph/messenger_v2.c, so I'm > > pretty sure that placing it here breaks everything. > > This should have been removed as MSG_SPLICE_PAGES now accepts KVEC and XARRAY > iterators also. > > Btw, is it feasible to use con->v2.out_iter_sendpage to apply MSG_SPLICE_PAGES > to the iterator to be transmitted as a whole? It seems to be set depending on > iterator type. I'm not sure I understand what you mean by "transmitted as a whole". con->v2.out_iter_sendpage is set only when zerocopy is desired. If the underlying data is not guaranteed to remain stable, zerocopy behavior is not safe. Thanks, Ilya
Ilya Dryomov <idryomov@gmail.com> wrote: > > Btw, is it feasible to use con->v2.out_iter_sendpage to apply > > MSG_SPLICE_PAGES to the iterator to be transmitted as a whole? It seems > > to be set depending on iterator type. > > I'm not sure I understand what you mean by "transmitted as a whole". > con->v2.out_iter_sendpage is set only when zerocopy is desired. If the > underlying data is not guaranteed to remain stable, zerocopy behavior > is not safe. I think I need to reinstate the per-page sendpage_ok() check here also - though Al pointed out it isn't sufficiently exhaustive. There are pages that sendpage_ok() will return true on that you shouldn't be passing to sendpage(). I'll whip up a patch to partially revert this also. David
diff --git a/net/ceph/messenger_v2.c b/net/ceph/messenger_v2.c index 301a991dc6a6..87ac97073e75 100644 --- a/net/ceph/messenger_v2.c +++ b/net/ceph/messenger_v2.c @@ -117,91 +117,38 @@ static int ceph_tcp_recv(struct ceph_connection *con) return ret; } -static int do_sendmsg(struct socket *sock, struct iov_iter *it) -{ - struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS }; - int ret; - - msg.msg_iter = *it; - while (iov_iter_count(it)) { - ret = sock_sendmsg(sock, &msg); - if (ret <= 0) { - if (ret == -EAGAIN) - ret = 0; - return ret; - } - - iov_iter_advance(it, ret); - } - - WARN_ON(msg_data_left(&msg)); - return 1; -} - -static int do_try_sendpage(struct socket *sock, struct iov_iter *it) -{ - struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS }; - struct bio_vec bv; - int ret; - - if (WARN_ON(!iov_iter_is_bvec(it))) - return -EINVAL; - - while (iov_iter_count(it)) { - /* iov_iter_iovec() for ITER_BVEC */ - bvec_set_page(&bv, it->bvec->bv_page, - min(iov_iter_count(it), - it->bvec->bv_len - it->iov_offset), - it->bvec->bv_offset + it->iov_offset); - - /* - * sendpage cannot properly handle pages with - * page_count == 0, we need to fall back to sendmsg if - * that's the case. - * - * Same goes for slab pages: skb_can_coalesce() allows - * coalescing neighboring slab objects into a single frag - * which triggers one of hardened usercopy checks. - */ - if (sendpage_ok(bv.bv_page)) { - ret = sock->ops->sendpage(sock, bv.bv_page, - bv.bv_offset, bv.bv_len, - CEPH_MSG_FLAGS); - } else { - iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len); - ret = sock_sendmsg(sock, &msg); - } - if (ret <= 0) { - if (ret == -EAGAIN) - ret = 0; - return ret; - } - - iov_iter_advance(it, ret); - } - - return 1; -} - /* * Write as much as possible. The socket is expected to be corked, - * so we don't bother with MSG_MORE/MSG_SENDPAGE_NOTLAST here. + * so we don't bother with MSG_MORE here. * * Return: - * 1 - done, nothing (else) to write + * >0 - done, nothing (else) to write * 0 - socket is full, need to wait * <0 - error */ static int ceph_tcp_send(struct ceph_connection *con) { + struct msghdr msg = { + .msg_iter = con->v2.out_iter, + .msg_flags = CEPH_MSG_FLAGS, + }; int ret; + if (WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter))) + return -EINVAL; + + if (con->v2.out_iter_sendpage) + msg.msg_flags |= MSG_SPLICE_PAGES; + dout("%s con %p have %zu try_sendpage %d\n", __func__, con, iov_iter_count(&con->v2.out_iter), con->v2.out_iter_sendpage); - if (con->v2.out_iter_sendpage) - ret = do_try_sendpage(con->sock, &con->v2.out_iter); - else - ret = do_sendmsg(con->sock, &con->v2.out_iter); + + ret = sock_sendmsg(con->sock, &msg); + if (ret > 0) + iov_iter_advance(&con->v2.out_iter, ret); + else if (ret == -EAGAIN) + ret = 0; + dout("%s con %p ret %d left %zu\n", __func__, con, ret, iov_iter_count(&con->v2.out_iter)); return ret;
Use sendmsg() and MSG_SPLICE_PAGES rather than sendpage in ceph when transmitting data. For the moment, this can only transmit one page at a time because of the architecture of net/ceph/, but if write_partial_message_data() can be given a bvec[] at a time by the iteration code, this would allow pages to be sent in a batch. Signed-off-by: David Howells <dhowells@redhat.com> cc: Ilya Dryomov <idryomov@gmail.com> cc: Xiubo Li <xiubli@redhat.com> cc: Jeff Layton <jlayton@kernel.org> cc: "David S. Miller" <davem@davemloft.net> cc: Eric Dumazet <edumazet@google.com> cc: Jakub Kicinski <kuba@kernel.org> cc: Paolo Abeni <pabeni@redhat.com> cc: Jens Axboe <axboe@kernel.dk> cc: Matthew Wilcox <willy@infradead.org> cc: ceph-devel@vger.kernel.org cc: netdev@vger.kernel.org --- net/ceph/messenger_v2.c | 91 +++++++++-------------------------------- 1 file changed, 19 insertions(+), 72 deletions(-)