From: "Marc-André Lureau" <mlureau@redhat.com>
To: "Michael S. Tsirkin" <mst@redhat.com>
Cc: marcandre lureau <marcandre.lureau@redhat.com>,
qemu-devel@nongnu.org, mukawa@igel.co.jp,
yuanhan liu <yuanhan.liu@linux.intel.com>,
victork@redhat.com, jonshin@cisco.com
Subject: Re: [Qemu-devel] [PATCH v3 10/28] vhost: change some assert() for error_report() or silent fail
Date: Thu, 21 Jul 2016 03:57:53 -0400 (EDT) [thread overview]
Message-ID: <1468020466.6954884.1469087873857.JavaMail.zimbra@redhat.com> (raw)
In-Reply-To: <20160720164529-mutt-send-email-mst@kernel.org>
Hi
----- Original Message -----
> On Wed, Jul 20, 2016 at 09:41:26AM -0400, Marc-André Lureau wrote:
> >
> >
> > ----- Original Message -----
> > > On Wed, Jul 06, 2016 at 08:47:03PM +0200, marcandre.lureau@redhat.com
> > > wrote:
> > > > From: Marc-André Lureau <marcandre.lureau@redhat.com>
> > > >
> > > > Calling a vhost operation may fail, especially with disconnectable
> > > > backends. Treat some that look harmless as recoverable errors (print
> > > > error, or ignore on error code path).
> > > >
> > > > Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
> > >
> > > These might be recoverable for vhost-user not for vhost_net.
> >
> > I don't think we can hide all the error handling in vhost-user very
> > long, soon enough we will need to reset the guest device state.
>
> Interesting. This will need some thought.
>
>
> > If
> > vhost-net doesn't support error, it should rather assert() there, but
> > having the error handling done at higher level, at the vhost interface
> > level at least, not at the backend level.
>
> Interesting. That might be reasonable too but would increase the scope
> of this already large patchset even further.
>
>
> > > IMO the backend should return 0 if error is benign,
> > > report errors to vhost only if they are fatal.
> >
> > Imho whether it's fatal and how to recover as not much to do with the
> > backend (which each kind of just a proxy), it should be handled at higher
> > level, possibly up to the guest.
>
> Consider example below. EBADF means fd not writeable - remote exited
> so that's benign. EFAULT means code bug. vhost has no idea there's
> an fd though.
>
It will probably be EPIPE, right? unless the fd was closed. Hhmpf, tcp_chr_write() actually returns the number of bytes to write if the peer is disconnected... and io_channel_send_full() on failure EINVAL..
>
> > > For example, consider set mem table. Write failing is one thing,
> > > and it's benign, but e.g. table too big is another thing and isn't.
It depends, if the backend disconnects during that call, it isn't "fatal". A later reconnection will restart and reset vhost-user tables.
> > > Also, we might want to distinguish between EBADF (fd closed)
> > > and other types of errors. All this knowledge belomgs in vhost user.
> > >
It's hard to hide disconnected state away, beside the need to report errors higher up, the handling of the disconnected state is not just in the vhost-user backend, but also in net/vhost-user.
I also notice that qemu_chr_fe_write*() will not trigger disconnect events, while read qemu_chr_fe_read() will: vhost_dev struct will be 0'ed during the call, by net_vhost_user_event() handler.
In most cases (there is a minor exception in set_vring_endian_legacy), vhost actually doesn't care about errno. However, it reports up errors using errno values. It seems it is only reported by vhost_dev_init() and vhost_dev_start(), and the value is used for strerror/error_report() (I notice also that failing vhost_dev_start() in vhost_scsi_start() is fatal)
> > > > ---
> > > > hw/virtio/vhost.c | 32 +++++++++++++++++++++-----------
> > > > 1 file changed, 21 insertions(+), 11 deletions(-)
> > > >
> > > > diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
> > > > index 75bc51e..e03a031 100644
> > > > --- a/hw/virtio/vhost.c
> > > > +++ b/hw/virtio/vhost.c
> > > > @@ -400,7 +400,10 @@ static inline void vhost_dev_log_resize(struct
> > > > vhost_dev *dev, uint64_t size)
> > > > /* inform backend of log switching, this must be done before
> > > > releasing the current log, to ensure no logging is lost */
> > > > r = dev->vhost_ops->vhost_set_log_base(dev, log_base, log);
> > > > - assert(r >= 0);
> > > > + if (r < 0) {
> > > > + error_report("Failed to change backend log");
> > > > + }
> > > > +
> > > > vhost_log_put(dev, true);
> > > > dev->log = log;
> > > > dev->log_size = size;
> > > > @@ -567,7 +570,9 @@ static void vhost_commit(MemoryListener *listener)
> > > >
> > > > if (!dev->log_enabled) {
> > > > r = dev->vhost_ops->vhost_set_mem_table(dev, dev->mem);
> > > > - assert(r >= 0);
> > > > + if (r < 0) {
> > > > + error_report("Failed to set mem table");
> > > > + }
> > > > dev->memory_changed = false;
> > > > return;
> > > > }
> > > > @@ -580,7 +585,9 @@ static void vhost_commit(MemoryListener *listener)
> > > > vhost_dev_log_resize(dev, log_size + VHOST_LOG_BUFFER);
> > > > }
> > > > r = dev->vhost_ops->vhost_set_mem_table(dev, dev->mem);
> > > > - assert(r >= 0);
> > > > + if (r < 0) {
> > > > + error_report("Failed to set mem table");
> > > > + }
> > > > /* To log less, can only decrease log size after table update. */
> > > > if (dev->log_size > log_size + VHOST_LOG_BUFFER) {
> > > > vhost_dev_log_resize(dev, log_size);
> > > > @@ -649,6 +656,7 @@ static int vhost_virtqueue_set_addr(struct
> > > > vhost_dev
> > > > *dev,
> > > > };
> > > > int r = dev->vhost_ops->vhost_set_vring_addr(dev, &addr);
> > > > if (r < 0) {
> > > > + error_report("Failed to set vring addr");
> > > > return -errno;
> > > > }
> > > > return 0;
> > > > @@ -662,12 +670,15 @@ static int vhost_dev_set_features(struct
> > > > vhost_dev
> > > > *dev, bool enable_log)
> > > > features |= 0x1ULL << VHOST_F_LOG_ALL;
> > > > }
> > > > r = dev->vhost_ops->vhost_set_features(dev, features);
> > > > + if (r < 0) {
> > > > + error_report("Failed to set features");
> > > > + }
> > > > return r < 0 ? -errno : 0;
> > > > }
> > > >
> > > > static int vhost_dev_set_log(struct vhost_dev *dev, bool enable_log)
> > > > {
> > > > - int r, t, i, idx;
> > > > + int r, i, idx;
> > > > r = vhost_dev_set_features(dev, enable_log);
> > > > if (r < 0) {
> > > > goto err_features;
> > > > @@ -684,12 +695,10 @@ static int vhost_dev_set_log(struct vhost_dev
> > > > *dev,
> > > > bool enable_log)
> > > > err_vq:
> > > > for (; i >= 0; --i) {
> > > > idx = dev->vhost_ops->vhost_get_vq_index(dev, dev->vq_index +
> > > > i);
> > > > - t = vhost_virtqueue_set_addr(dev, dev->vqs + i, idx,
> > > > - dev->log_enabled);
> > > > - assert(t >= 0);
> > > > + vhost_virtqueue_set_addr(dev, dev->vqs + i, idx,
> > > > + dev->log_enabled);
> > > > }
> > > > - t = vhost_dev_set_features(dev, dev->log_enabled);
> > > > - assert(t >= 0);
> > > > + vhost_dev_set_features(dev, dev->log_enabled);
> > > > err_features:
> > > > return r;
> > > > }
> > > > @@ -937,7 +946,6 @@ static void vhost_virtqueue_stop(struct vhost_dev
> > > > *dev,
> > > > }
> > > > }
> > > >
> > > > - assert (r >= 0);
> > > > cpu_physical_memory_unmap(vq->ring,
> > > > virtio_queue_get_ring_size(vdev,
> > > > idx),
> > > > 0, virtio_queue_get_ring_size(vdev,
> > > > idx));
> > > > cpu_physical_memory_unmap(vq->used,
> > > > virtio_queue_get_used_size(vdev,
> > > > idx),
> > > > @@ -1191,7 +1199,9 @@ void vhost_virtqueue_mask(struct vhost_dev *hdev,
> > > > VirtIODevice *vdev, int n,
> > > >
> > > > file.index = hdev->vhost_ops->vhost_get_vq_index(hdev, n);
> > > > r = hdev->vhost_ops->vhost_set_vring_call(hdev, &file);
> > > > - assert(r >= 0);
> > > > + if (r < 0) {
> > > > + error_report("Failed to set vring call");
> > > > + }
> > > > }
> > > >
> > > > uint64_t vhost_get_features(struct vhost_dev *hdev, const int
> > > > *feature_bits,
> > > > --
> > > > 2.9.0
> > >
>
next prev parent reply other threads:[~2016-07-21 7:58 UTC|newest]
Thread overview: 38+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-07-06 18:46 [Qemu-devel] [PATCH v3 00/28] vhost-user reconnect fixes marcandre.lureau
2016-07-06 18:46 ` [Qemu-devel] [PATCH v3 01/28] misc: indentation marcandre.lureau
2016-07-06 18:46 ` [Qemu-devel] [PATCH v3 02/28] vhost-user: minor simplification marcandre.lureau
2016-07-06 18:46 ` [Qemu-devel] [PATCH v3 03/28] vhost: don't assume opaque is a fd, use backend cleanup marcandre.lureau
2016-07-06 18:46 ` [Qemu-devel] [PATCH v3 04/28] vhost: make vhost_log_put() idempotent marcandre.lureau
2016-07-06 18:46 ` [Qemu-devel] [PATCH v3 05/28] vhost: call vhost_log_put() on cleanup marcandre.lureau
2016-07-06 18:46 ` [Qemu-devel] [PATCH v3 06/28] vhost: add vhost device only after all success marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 07/28] vhost: make vhost_dev_cleanup() idempotent marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 08/28] vhost-net: always call vhost_dev_cleanup() on failure marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 09/28] vhost: fix calling vhost_dev_cleanup() after vhost_dev_init() marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 10/28] vhost: change some assert() for error_report() or silent fail marcandre.lureau
2016-07-20 13:24 ` Michael S. Tsirkin
2016-07-20 13:33 ` Michael S. Tsirkin
2016-07-20 13:41 ` Marc-André Lureau
2016-07-20 13:55 ` Michael S. Tsirkin
2016-07-21 7:57 ` Marc-André Lureau [this message]
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 11/28] vhost: use error_report() instead of fprintf(stderr, ...) marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 12/28] vhost-user: check qemu_chr_fe_set_msgfds() return value marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 13/28] vhost-user: check vhost_user_{read, write}() " marcandre.lureau
2016-07-20 13:28 ` Michael S. Tsirkin
2016-07-21 7:55 ` Marc-André Lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 14/28] qemu-char: check socket is actually connected marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 15/28] vhost-user: keep vhost_net after a disconnection marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 16/28] Revert "vhost-net: do not crash if backend is not present" marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 17/28] get_vhost_net() should be != null after vhost_user_init marcandre.lureau
2016-07-20 13:36 ` Michael S. Tsirkin
2016-07-21 7:55 ` Marc-André Lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 18/28] vhost-net: success if backend has no ops->vhost_migration_done marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 19/28] vhost: add assert() to check runtime behaviour marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 20/28] char: add chr_wait_connected callback marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 21/28] char: add and use tcp_chr_wait_connected marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 22/28] vhost-user: wait until backend init is completed marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 23/28] tests: plug some leaks in virtio-net-test marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 24/28] tests: fix vhost-user-test leak marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 25/28] tests: add /vhost-user/connect-fail test marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 26/28] tests: add a simple /vhost-user/multiqueue test marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 27/28] vhost-user: add error report in vhost_user_write() marcandre.lureau
2016-07-06 18:47 ` [Qemu-devel] [PATCH v3 28/28] vhost: add vhost_net_set_backend() marcandre.lureau
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=1468020466.6954884.1469087873857.JavaMail.zimbra@redhat.com \
--to=mlureau@redhat.com \
--cc=jonshin@cisco.com \
--cc=marcandre.lureau@redhat.com \
--cc=mst@redhat.com \
--cc=mukawa@igel.co.jp \
--cc=qemu-devel@nongnu.org \
--cc=victork@redhat.com \
--cc=yuanhan.liu@linux.intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
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;
as well as URLs for NNTP newsgroup(s).