Skip to contentSkip to conversation

Limit websockets memory usage & other bug fixes

PATCH Superseded Daniel P. Berrange 2 participants 19 messages Oct 10, 2017 – Oct 11, 2017

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover Limit websockets memory usage & other bug fixes

The core motivation for this patch series is to fix a security issue publically reported, where websockets code can consume arbitrary amounts of RAM with slow clients:

https://bugs.launchpad.net/qemu/+bug/1718964

I've asked for a CVE but its not assigned yet. Since the bug is public we might as well get on with code review while waiting for the CVE number.

The first patch is the minimum required to fix the actual CVE in git master, taking advantage of how we know the VNC server will call us. The 5th patch lets us tighten up buffer limiting of writes further, so we're not making assumptions about VNC server code.

The websockets code is broken right back to the day it was merged in QEMU 1.2.1

The fix in patch 1 can apply to stable branches from 2.6 -> 2.10 inclusive, provided another fix from master is cherry-picked first

commit eefa3d8ef649f9055611361e2201cca49f8c3433
Author: Brandon Carpenter <[email protected]>
Date: Tue Sep 12 08:21:48 2017 -0700

io: Small updates in preparation for websocket changes

since that refactors code duplication in the GSource impl. Once we merge for master, I'll send a patch to qemu-stable.

Versions prior to 2.6 would require a fix to be done in the ui/vnc.c file vnc_update_client method instead. It would need to check vs->ws_output buffer size. I'm not intending to write any such patch - this is just info in case anyone is stuck on such ancient versions and needs to figure out a fix.

Daniel P. Berrange (7):
io: monitor encoutput buffer size from websocket GSource
io: simplify websocket ping reply handling
io: get rid of qio_channel_websock_encode helper method
io: pass a struct iovec into qio_channel_websock_encode
io: get rid of bounce buffering in websock write path
io: cope with websock 'Connection' header having multiple values
io: add trace points for websocket HTTP protocol headers

 include/io/channel-websock.h |   3 +-
 io/channel-websock.c         | 163 ++++++++++++++++++++++++-------------------
 io/trace-events              |   2 +
 3 files changed, 93 insertions(+), 75 deletions(-)
Signature · 1 line
2.13.5
#1 lore raw

PATCH 1/7 io: monitor encoutput buffer size from websocket GSource

The websocket GSource is monitoring the size of the rawoutput buffer to determine if the channel can accepts more writes. The rawoutput buffer, however, is merely a temporary staging buffer before data is copied into the encoutput buffer. This its size will always be zero when the GSource runs.

This flaw causes the encoutput buffer to grow without bound if the other end of the underlying data channel doesn't read data being sent. This can be seen with VNC if a client is on a slow WAN link and the guest OS is sending many screen updates. A malicious VNC client can act like it is on a slow link by playing a video in the guest and then reading data very slowly, causing QEMU host memory to expand arbitrarily.

This issue is assigned CVE-2017-????, publically reported in

https://bugs.launchpad.net/qemu/+bug/1718964

---
 io/channel-websock.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
io/channel-websock.c +2 -2
index d1d471f86e..04bcc059cd 100644
@@ -28,7 +28,7 @@
 #include <time.h>
 
 
-/* Max amount to allow in rawinput/rawoutput buffers */
+/* Max amount to allow in rawinput/encoutput buffers */
 #define QIO_CHANNEL_WEBSOCK_MAX_BUFFER 8192
 
 #define QIO_CHANNEL_WEBSOCK_CLIENT_KEY_LEN 24
@@ -1208,7 +1208,7 @@ qio_channel_websock_source_check(GSource *source)
     if (wsource->wioc->rawinput.offset || wsource->wioc->io_eof) {
         cond |= G_IO_IN;
     }
-    if (wsource->wioc->rawoutput.offset < QIO_CHANNEL_WEBSOCK_MAX_BUFFER) {
+    if (wsource->wioc->encoutput.offset < QIO_CHANNEL_WEBSOCK_MAX_BUFFER) {
         cond |= G_IO_OUT;
     }
 
Signature · 1 line
2.13.5
Reviewed-by on v1 1/7
#2 lore raw

Re: [Qemu-devel] [PATCH v1 1/7] io: monitor encoutput buffer size from websocket GSource

Quoted 4 lines · Daniel P. Berrange

The websocket GSource is monitoring the size of the rawoutput buffer to determine if the channel can accepts more writes. The rawoutput buffer, however, is merely a temporary staging buffer before data is copied into the encoutput buffer. This

s/This/Thus/

Quoted 11 lines

its size will always be zero when the GSource runs.

This flaw causes the encoutput buffer to grow without bound if the other end of the underlying data channel doesn't read data being sent. This can be seen with VNC if a client is on a slow WAN link and the guest OS is sending many screen updates. A malicious VNC client can act like it is on a slow link by playing a video in the guest and then reading data very slowly, causing QEMU host memory to expand arbitrarily.

This issue is assigned CVE-2017-????, publically reported in

If we get the assignment in time, I'm sure you'll update this before the PULL request.

Quoted 7 lines

https://bugs.launchpad.net/qemu/+bug/1718964

---
 io/channel-websock.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 24 lines
io/channel-websock.c +2 -2
index d1d471f86e..04bcc059cd 100644
@@ -28,7 +28,7 @@
 #include <time.h>
 
 
-/* Max amount to allow in rawinput/rawoutput buffers */
+/* Max amount to allow in rawinput/encoutput buffers */
 #define QIO_CHANNEL_WEBSOCK_MAX_BUFFER 8192
 
 #define QIO_CHANNEL_WEBSOCK_CLIENT_KEY_LEN 24
@@ -1208,7 +1208,7 @@ qio_channel_websock_source_check(GSource *source)
     if (wsource->wioc->rawinput.offset || wsource->wioc->io_eof) {
         cond |= G_IO_IN;
     }
-    if (wsource->wioc->rawoutput.offset < QIO_CHANNEL_WEBSOCK_MAX_BUFFER) {
+    if (wsource->wioc->encoutput.offset < QIO_CHANNEL_WEBSOCK_MAX_BUFFER) {
         cond |= G_IO_OUT;
     }
 
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#3 lore raw

Re: [Qemu-devel] [PATCH v1 1/7] io: monitor encoutput buffer size from websocket GSource

Quoted 18 lines · Eric Blake
Quoted 4 lines · Daniel P. Berrange

The websocket GSource is monitoring the size of the rawoutput buffer to determine if the channel can accepts more writes. The rawoutput buffer, however, is merely a temporary staging buffer before data is copied into the encoutput buffer. This

s/This/Thus/

Quoted 10 lines

its size will always be zero when the GSource runs.

This flaw causes the encoutput buffer to grow without bound if the other end of the underlying data channel doesn't read data being sent. This can be seen with VNC if a client is on a slow WAN link and the guest OS is sending many screen updates. A malicious VNC client can act like it is on a slow link by playing a video in the guest and then reading data very slowly, causing QEMU host memory to expand arbitrarily.

Quoted 4 lines
Quoted 1 line

This issue is assigned CVE-2017-????, publically reported in

If we get the assignment in time, I'm sure you'll update this before the PULL request.

Yes, exactly the plan...

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#4 lore raw

PATCH 2/7 io: simplify websocket ping reply handling

In reply to Daniel P. Berrange

We must ensure we don't get flooded with ping replies if the outbound channel is slow. Currently we do this by keeping the ping reply in a separate temporary buffer and only writing it if the encoutput buffer is completely empty. This is overly pessimistic, as it is reasonable to add a ping reply to the encoutput buffer even if it has previous data in it, as long as that previous data doesn't include a ping reply.

To track this better, put the ping reply directly into the encoutput buffer, and then record the size of encoutput at this time in ping_remain. As we write encoutput to the underlying channel, we can decrement the ping_remain counter. Once it hits zero, we can accept further ping replies for transmission.

---
 include/io/channel-websock.h |  2 +-
 io/channel-websock.c         | 28 +++++++++++++++-------------
 2 files changed, 16 insertions(+), 14 deletions(-)
include/io/channel-websock.h +1 -1
index ff32d8651b..3f92535cae 100644
@@ -60,8 +60,8 @@ struct QIOChannelWebsock {
     Buffer encoutput;
     Buffer rawinput;
     Buffer rawoutput;
-    Buffer ping_reply;
     size_t payload_remain;
+    size_t ping_remain;
     QIOChannelWebsockMask mask;
     guint io_tag;
     Error *io_err;
io/channel-websock.c +15 -13
index 04bcc059cd..1c34f68012 100644
@@ -825,11 +825,14 @@ static int qio_channel_websock_decode_payload(QIOChannelWebsock *ioc,
         }
         return -1;
     } else if (ioc->opcode == QIO_CHANNEL_WEBSOCK_OPCODE_PING) {
-        /* ping frames produce an immediate reply */
-        buffer_reset(&ioc->ping_reply);
-        qio_channel_websock_encode_buffer(
-            ioc, &ioc->ping_reply, QIO_CHANNEL_WEBSOCK_OPCODE_PONG,
-            &ioc->encinput);
+        /* ping frames produce an immediate reply, as long as we've not still
+         * got a previous ping queued, in which case we drop the new pong */
+        if (ioc->ping_remain == 0) {
+            qio_channel_websock_encode_buffer(
+                ioc, &ioc->encoutput, QIO_CHANNEL_WEBSOCK_OPCODE_PONG,
+                &ioc->encinput);
+            ioc->ping_remain = ioc->encoutput.offset;
+        }
     }   /* pong frames are ignored */
 
     if (payload_len) {
@@ -888,7 +891,6 @@ static void qio_channel_websock_finalize(Object *obj)
     buffer_free(&ioc->encoutput);
     buffer_free(&ioc->rawinput);
     buffer_free(&ioc->rawoutput);
-    buffer_free(&ioc->ping_reply);
     object_unref(OBJECT(ioc->master));
     if (ioc->io_tag) {
         g_source_remove(ioc->io_tag);
@@ -946,12 +948,7 @@ static ssize_t qio_channel_websock_write_wire(QIOChannelWebsock *ioc,
     ssize_t ret;
     ssize_t done = 0;
 
-    /* ping replies take priority over binary data */
-    if (!ioc->ping_reply.offset) {
-        qio_channel_websock_encode(ioc);
-    } else if (!ioc->encoutput.offset) {
-        buffer_move_empty(&ioc->encoutput, &ioc->ping_reply);
-    }
+    qio_channel_websock_encode(ioc);
 
     while (ioc->encoutput.offset > 0) {
         ret = qio_channel_write(ioc->master,
@@ -968,6 +965,11 @@ static ssize_t qio_channel_websock_write_wire(QIOChannelWebsock *ioc,
         }
         buffer_advance(&ioc->encoutput, ret);
         done += ret;
+        if (ioc->ping_remain < ret) {
+            ioc->ping_remain = 0;
+        } else {
+            ioc->ping_remain -= ret;
+        }
     }
     return done;
 }
@@ -1026,7 +1028,7 @@ static void qio_channel_websock_set_watch(QIOChannelWebsock *ioc)
         return;
     }
 
-    if (ioc->encoutput.offset || ioc->ping_reply.offset) {
+    if (ioc->encoutput.offset) {
         cond |= G_IO_OUT;
     }
     if (ioc->encinput.offset < QIO_CHANNEL_WEBSOCK_MAX_BUFFER &&
Signature · 1 line
2.13.5
Reviewed-by on v1 2/7
#5 lore raw

Re: [Qemu-devel] [PATCH v1 2/7] io: simplify websocket ping reply handling

Quoted 23 lines · Daniel P. Berrange

We must ensure we don't get flooded with ping replies if the outbound channel is slow. Currently we do this by keeping the ping reply in a separate temporary buffer and only writing it if the encoutput buffer is completely empty. This is overly pessimistic, as it is reasonable to add a ping reply to the encoutput buffer even if it has previous data in it, as long as that previous data doesn't include a ping reply.

To track this better, put the ping reply directly into the encoutput buffer, and then record the size of encoutput at this time in ping_remain. As we write encoutput to the underlying channel, we can decrement the ping_remain counter. Once it hits zero, we can accept further ping replies for transmission.

---
 include/io/channel-websock.h |  2 +-
 io/channel-websock.c         | 28 +++++++++++++++-------------
 2 files changed, 16 insertions(+), 14 deletions(-)

+++ b/io/channel-websock.c

io/channel-websock.c static int qio_channel_websock_decode_payload(QIOChannelWebsock *ioc, · line 829

Quoted 10 lines
io/channel-websock.c +2 -5
@@ -825,11 +825,14 @@ static int qio_channel_websock_decode_payload(QIOChannelWebsock *ioc,
         }
         return -1;
     } else if (ioc->opcode == QIO_CHANNEL_WEBSOCK_OPCODE_PING) {
-        /* ping frames produce an immediate reply */
-        buffer_reset(&ioc->ping_reply);
-        qio_channel_websock_encode_buffer(
-            ioc, &ioc->ping_reply, QIO_CHANNEL_WEBSOCK_OPCODE_PONG,
-            &ioc->encinput);
+        /* ping frames produce an immediate reply, as long as we've not still
+         * got a previous ping queued, in which case we drop the new pong */

Wouldn't that be a 'previous pong queued'?

io/channel-websock.c static int qio_channel_websock_decode_payload(QIOChannelWebsock *ioc, · line 835

Quoted 6 lines
io/channel-websock.c +6 -0
+        if (ioc->ping_remain == 0) {
+            qio_channel_websock_encode_buffer(
+                ioc, &ioc->encoutput, QIO_CHANNEL_WEBSOCK_OPCODE_PONG,
+                &ioc->encinput);
+            ioc->ping_remain = ioc->encoutput.offset;
+        }

But if you change the comment, then naming the variable pong_remain may make more sense.

But naming is a bikeshed issue, so either way,

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#6 lore raw

Re: [Qemu-devel] [PATCH v1 2/7] io: simplify websocket ping reply handling

Quoted 22 lines · Eric Blake
Quoted 20 lines · Daniel P. Berrange

We must ensure we don't get flooded with ping replies if the outbound channel is slow. Currently we do this by keeping the ping reply in a separate temporary buffer and only writing it if the encoutput buffer is completely empty. This is overly pessimistic, as it is reasonable to add a ping reply to the encoutput buffer even if it has previous data in it, as long as that previous data doesn't include a ping reply.

To track this better, put the ping reply directly into the encoutput buffer, and then record the size of encoutput at this time in ping_remain. As we write encoutput to the underlying channel, we can decrement the ping_remain counter. Once it hits zero, we can accept further ping replies for transmission.

---
 include/io/channel-websock.h |  2 +-
 io/channel-websock.c         | 28 +++++++++++++++-------------
 2 files changed, 16 insertions(+), 14 deletions(-)
Quoted 14 lines

io/channel-websock.c static int qio_channel_websock_decode_payload(QIOChannelWebsock *ioc, · line 829

Quoted 12 lines

+++ b/io/channel-websock.c

io/channel-websock.c +2 -5
@@ -825,11 +825,14 @@ static int qio_channel_websock_decode_payload(QIOChannelWebsock *ioc,
         }
         return -1;
     } else if (ioc->opcode == QIO_CHANNEL_WEBSOCK_OPCODE_PING) {
-        /* ping frames produce an immediate reply */
-        buffer_reset(&ioc->ping_reply);
-        qio_channel_websock_encode_buffer(
-            ioc, &ioc->ping_reply, QIO_CHANNEL_WEBSOCK_OPCODE_PONG,
-            &ioc->encinput);
+        /* ping frames produce an immediate reply, as long as we've not still
+         * got a previous ping queued, in which case we drop the new pong */

Wouldn't that be a 'previous pong queued'?

Indeed

Quoted 10 lines

io/channel-websock.c static int qio_channel_websock_decode_payload(QIOChannelWebsock *ioc, · line 835

Quoted 7 lines
io/channel-websock.c +6 -0
+        if (ioc->ping_remain == 0) {
+            qio_channel_websock_encode_buffer(
+                ioc, &ioc->encoutput, QIO_CHANNEL_WEBSOCK_OPCODE_PONG,
+                &ioc->encinput);
+            ioc->ping_remain = ioc->encoutput.offset;
+        }

But if you change the comment, then naming the variable pong_remain may make more sense.

Yeah, that's sensible.

Quoted 4 lines

But naming is a bikeshed issue, so either way,

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#7 lore raw

PATCH 3/7 io: get rid of qio_channel_websock_encode helper method

In reply to Daniel P. Berrange

The qio_channel_websock_encode method is only used in one place, everything else calls qio_channel_websock_encode_buffer directly. It can also be pushed up a level into the qio_channel_websock_writev method, since every other caller of qio_channel_websock_write_wire has already filled encoutput.

---
 io/channel-websock.c | 20 ++++++--------------
 1 file changed, 6 insertions(+), 14 deletions(-)
io/channel-websock.c +6 -14
index 1c34f68012..17f50f5bb1 100644
@@ -616,18 +616,6 @@ static void qio_channel_websock_encode_buffer(QIOChannelWebsock *ioc,
 }
 
 
-static void qio_channel_websock_encode(QIOChannelWebsock *ioc)
-{
-    if (!ioc->rawoutput.offset) {
-        return;
-    }
-    qio_channel_websock_encode_buffer(
-        ioc, &ioc->encoutput, QIO_CHANNEL_WEBSOCK_OPCODE_BINARY_FRAME,
-        &ioc->rawoutput);
-    buffer_reset(&ioc->rawoutput);
-}
-
-
 static ssize_t qio_channel_websock_write_wire(QIOChannelWebsock *, Error **);
 
 
@@ -948,8 +936,6 @@ static ssize_t qio_channel_websock_write_wire(QIOChannelWebsock *ioc,
     ssize_t ret;
     ssize_t done = 0;
 
-    qio_channel_websock_encode(ioc);
-
     while (ioc->encoutput.offset > 0) {
         ret = qio_channel_write(ioc->master,
                                 (char *)ioc->encoutput.buffer,
@@ -1134,6 +1120,12 @@ static ssize_t qio_channel_websock_writev(QIOChannel *ioc,
     }
 
  done:
+    if (ioc->rawoutput.offset) {
+        qio_channel_websock_encode_buffer(
+            ioc, &ioc->encoutput, QIO_CHANNEL_WEBSOCK_OPCODE_BINARY_FRAME,
+            &ioc->rawoutput);
+        buffer_reset(&ioc->rawoutput);
+    }
     ret = qio_channel_websock_write_wire(wioc, errp);
     if (ret < 0 &&
         ret != QIO_CHANNEL_ERR_BLOCK) {
Signature · 1 line
2.13.5
Reviewed-by on v1 3/7
#8 lore raw

Re: [Qemu-devel] [PATCH v1 3/7] io: get rid of qio_channel_websock_encode helper method

Quoted 11 lines · Daniel P. Berrange

The qio_channel_websock_encode method is only used in one place, everything else calls qio_channel_websock_encode_buffer directly. It can also be pushed up a level into the qio_channel_websock_writev method, since every other caller of qio_channel_websock_write_wire has already filled encoutput.

---
 io/channel-websock.c | 20 ++++++--------------
 1 file changed, 6 insertions(+), 14 deletions(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#9 lore raw

PATCH 4/7 io: pass a struct iovec into qio_channel_websock_encode

In reply to Daniel P. Berrange

Instead of requiring use of another Buffer, pass a struct iovec into qio_channel_websock_encode, which gives callers more flexibility in how they process data.

---
 io/channel-websock.c | 69 ++++++++++++++++++++++++++++++++--------------------
 1 file changed, 42 insertions(+), 27 deletions(-)
io/channel-websock.c +42 -27
index 17f50f5bb1..ad62dda479 100644
@@ -582,11 +582,14 @@ static gboolean qio_channel_websock_handshake_io(QIOChannel *ioc,
 }
 
 
-static void qio_channel_websock_encode_buffer(QIOChannelWebsock *ioc,
-                                              Buffer *output,
-                                              uint8_t opcode, Buffer *buffer)
+static void qio_channel_websock_encode(QIOChannelWebsock *ioc,
+                                       uint8_t opcode,
+                                       const struct iovec *iov,
+                                       size_t niov,
+                                       size_t size)
 {
     size_t header_size;
+    size_t i;
     union {
         char buf[QIO_CHANNEL_WEBSOCK_HEADER_LEN_64_BIT];
         QIOChannelWebsockHeader ws;
@@ -594,25 +597,31 @@ static void qio_channel_websock_encode_buffer(QIOChannelWebsock *ioc,
 
     header.ws.b0 = QIO_CHANNEL_WEBSOCK_HEADER_FIELD_FIN |
         (opcode & QIO_CHANNEL_WEBSOCK_HEADER_FIELD_OPCODE);
-    if (buffer->offset < QIO_CHANNEL_WEBSOCK_PAYLOAD_LEN_THRESHOLD_7_BIT) {
-        header.ws.b1 = (uint8_t)buffer->offset;
+    if (size < QIO_CHANNEL_WEBSOCK_PAYLOAD_LEN_THRESHOLD_7_BIT) {
+        header.ws.b1 = (uint8_t)size;
         header_size = QIO_CHANNEL_WEBSOCK_HEADER_LEN_7_BIT;
-    } else if (buffer->offset <
-               QIO_CHANNEL_WEBSOCK_PAYLOAD_LEN_THRESHOLD_16_BIT) {
+    } else if (size < QIO_CHANNEL_WEBSOCK_PAYLOAD_LEN_THRESHOLD_16_BIT) {
         header.ws.b1 = QIO_CHANNEL_WEBSOCK_PAYLOAD_LEN_MAGIC_16_BIT;
-        header.ws.u.s16.l16 = cpu_to_be16((uint16_t)buffer->offset);
+        header.ws.u.s16.l16 = cpu_to_be16((uint16_t)size);
         header_size = QIO_CHANNEL_WEBSOCK_HEADER_LEN_16_BIT;
     } else {
         header.ws.b1 = QIO_CHANNEL_WEBSOCK_PAYLOAD_LEN_MAGIC_64_BIT;
-        header.ws.u.s64.l64 = cpu_to_be64(buffer->offset);
+        header.ws.u.s64.l64 = cpu_to_be64(size);
         header_size = QIO_CHANNEL_WEBSOCK_HEADER_LEN_64_BIT;
     }
     header_size -= QIO_CHANNEL_WEBSOCK_HEADER_LEN_MASK;
 
-    trace_qio_channel_websock_encode(ioc, opcode, header_size, buffer->offset);
-    buffer_reserve(output, header_size + buffer->offset);
-    buffer_append(output, header.buf, header_size);
-    buffer_append(output, buffer->buffer, buffer->offset);
+    trace_qio_channel_websock_encode(ioc, opcode, header_size, size);
+    buffer_reserve(&ioc->encoutput, header_size + size);
+    buffer_append(&ioc->encoutput, header.buf, header_size);
+    for (i = 0; i < niov && size != 0; i++) {
+        size_t want = iov->iov_len;
+        if (want > size) {
+            want = size;
+        }
+        buffer_append(&ioc->encoutput, iov->iov_base, want);
+        size -= want;
+    }
 }
 
 
@@ -622,6 +631,7 @@ static ssize_t qio_channel_websock_write_wire(QIOChannelWebsock *, Error **);
 static void qio_channel_websock_write_close(QIOChannelWebsock *ioc,
                                             uint16_t code, const char *reason)
 {
+    struct iovec iov;
     buffer_reserve(&ioc->rawoutput, 2 + (reason ? strlen(reason) : 0));
     *(uint16_t *)(ioc->rawoutput.buffer + ioc->rawoutput.offset) =
         cpu_to_be16(code);
@@ -629,9 +639,10 @@ static void qio_channel_websock_write_close(QIOChannelWebsock *ioc,
     if (reason) {
         buffer_append(&ioc->rawoutput, reason, strlen(reason));
     }
-    qio_channel_websock_encode_buffer(
-        ioc, &ioc->encoutput, QIO_CHANNEL_WEBSOCK_OPCODE_CLOSE,
-        &ioc->rawoutput);
+    iov.iov_base = ioc->rawoutput.buffer;
+    iov.iov_len = ioc->rawoutput.offset;
+    qio_channel_websock_encode(ioc, QIO_CHANNEL_WEBSOCK_OPCODE_CLOSE,
+                               &iov, 1, iov.iov_len);
     buffer_reset(&ioc->rawoutput);
     qio_channel_websock_write_wire(ioc, NULL);
     qio_channel_shutdown(ioc->master, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
@@ -801,9 +812,10 @@ static int qio_channel_websock_decode_payload(QIOChannelWebsock *ioc,
         error_setg(errp, "websocket closed by peer");
         if (payload_len) {
             /* echo client status */
-            qio_channel_websock_encode_buffer(
-                ioc, &ioc->encoutput, QIO_CHANNEL_WEBSOCK_OPCODE_CLOSE,
-                &ioc->encinput);
+            struct iovec iov = { .iov_base = ioc->encinput.buffer,
+                                 .iov_len = ioc->encinput.offset };
+            qio_channel_websock_encode(ioc, QIO_CHANNEL_WEBSOCK_OPCODE_CLOSE,
+                                       &iov, 1, iov.iov_len);
             qio_channel_websock_write_wire(ioc, NULL);
             qio_channel_shutdown(ioc->master, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
         } else {
@@ -816,9 +828,10 @@ static int qio_channel_websock_decode_payload(QIOChannelWebsock *ioc,
         /* ping frames produce an immediate reply, as long as we've not still
          * got a previous ping queued, in which case we drop the new pong */
         if (ioc->ping_remain == 0) {
-            qio_channel_websock_encode_buffer(
-                ioc, &ioc->encoutput, QIO_CHANNEL_WEBSOCK_OPCODE_PONG,
-                &ioc->encinput);
+            struct iovec iov = { .iov_base = ioc->encinput.buffer,
+                                 .iov_len = ioc->encinput.offset };
+            qio_channel_websock_encode(ioc, QIO_CHANNEL_WEBSOCK_OPCODE_PONG,
+                                       &iov, 1, iov.iov_len);
             ioc->ping_remain = ioc->encoutput.offset;
         }
     }   /* pong frames are ignored */
@@ -1120,11 +1133,13 @@ static ssize_t qio_channel_websock_writev(QIOChannel *ioc,
     }
 
  done:
-    if (ioc->rawoutput.offset) {
-        qio_channel_websock_encode_buffer(
-            ioc, &ioc->encoutput, QIO_CHANNEL_WEBSOCK_OPCODE_BINARY_FRAME,
-            &ioc->rawoutput);
-        buffer_reset(&ioc->rawoutput);
+    if (wioc->rawoutput.offset) {
+        struct iovec iov = { .iov_base = wioc->rawoutput.buffer,
+                             .iov_len = wioc->rawoutput.offset };
+        qio_channel_websock_encode(wioc,
+                                   QIO_CHANNEL_WEBSOCK_OPCODE_BINARY_FRAME,
+                                   &iov, 1, iov.iov_len);
+        buffer_reset(&wioc->rawoutput);
     }
     ret = qio_channel_websock_write_wire(wioc, errp);
     if (ret < 0 &&
Signature · 1 line
2.13.5
#10 lore raw

Re: [Qemu-devel] [PATCH v1 4/7] io: pass a struct iovec into qio_channel_websock_encode

Quoted 10 lines · Daniel P. Berrange

Instead of requiring use of another Buffer, pass a struct iovec into qio_channel_websock_encode, which gives callers more flexibility in how they process data.

---
 io/channel-websock.c | 69 ++++++++++++++++++++++++++++++++--------------------
 1 file changed, 42 insertions(+), 27 deletions(-)

io/channel-websock.c static gboolean qio_channel_websock_handshake_io(QIOChannel *ioc, · line 590

Quoted 6 lines
io/channel-websock.c +5 -0
+static void qio_channel_websock_encode(QIOChannelWebsock *ioc,
+                                       uint8_t opcode,
+                                       const struct iovec *iov,
+                                       size_t niov,
+                                       size_t size)
 {

Is size redundant with iov_size(iov, niov)? Or is a caller allowed to pass a smaller size (tail end of iov is not encoded) or larger size (encoding stops at end of iov, even if size is not exhausted)? I'd lean towards the former (one less parameter), especially since all callers in this patch could have passed iov_size(&iov, 1) for the same effect.

Quoted 5 lines
io/channel-websock.c +1 -4
-    trace_qio_channel_websock_encode(ioc, opcode, header_size, buffer->offset);
-    buffer_reserve(output, header_size + buffer->offset);
-    buffer_append(output, header.buf, header_size);
-    buffer_append(output, buffer->buffer, buffer->offset);
+    trace_qio_channel_websock_encode(ioc, opcode, header_size, size);

io/channel-websock.c static void qio_channel_websock_encode_buffer(QIOChannelWebsock *ioc, · line 624

Quoted 10 lines
io/channel-websock.c +10 -0
+    buffer_reserve(&ioc->encoutput, header_size + size);
+    buffer_append(&ioc->encoutput, header.buf, header_size);
+    for (i = 0; i < niov && size != 0; i++) {
+        size_t want = iov->iov_len;
+        if (want > size) {
+            want = size;
+        }
+        buffer_append(&ioc->encoutput, iov->iov_base, want);
+        size -= want;
+    }

Umm, where are you incrementing iov? It appears you only tested with niov == 1.

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#11 lore raw

Re: [Qemu-devel] [PATCH v1 4/7] io: pass a struct iovec into qio_channel_websock_encode

Quoted 18 lines · Eric Blake
Quoted 10 lines · Daniel P. Berrange

Instead of requiring use of another Buffer, pass a struct iovec into qio_channel_websock_encode, which gives callers more flexibility in how they process data.

---
 io/channel-websock.c | 69 ++++++++++++++++++++++++++++++++--------------------
 1 file changed, 42 insertions(+), 27 deletions(-)

io/channel-websock.c static gboolean qio_channel_websock_handshake_io(QIOChannel *ioc, · line 590

Quoted 6 lines
io/channel-websock.c +5 -0
+static void qio_channel_websock_encode(QIOChannelWebsock *ioc,
+                                       uint8_t opcode,
+                                       const struct iovec *iov,
+                                       size_t niov,
+                                       size_t size)
 {
Quoted 5 lines

Is size redundant with iov_size(iov, niov)? Or is a caller allowed to pass a smaller size (tail end of iov is not encoded) or larger size (encoding stops at end of iov, even if size is not exhausted)? I'd lean towards the former (one less parameter), especially since all callers in this patch could have passed iov_size(&iov, 1) for the same effect.

You've already noticed the code in later patch which passes a smaller size. This lets me avoid having to duplicate the iovec array and set smaller iov_len

Quoted 17 lines
Quoted 6 lines
io/channel-websock.c +1 -4
-    trace_qio_channel_websock_encode(ioc, opcode, header_size, buffer->offset);
-    buffer_reserve(output, header_size + buffer->offset);
-    buffer_append(output, header.buf, header_size);
-    buffer_append(output, buffer->buffer, buffer->offset);
+    trace_qio_channel_websock_encode(ioc, opcode, header_size, size);

io/channel-websock.c static void qio_channel_websock_encode_buffer(QIOChannelWebsock *ioc, · line 624

Quoted 10 lines
io/channel-websock.c +10 -0
+    buffer_reserve(&ioc->encoutput, header_size + size);
+    buffer_append(&ioc->encoutput, header.buf, header_size);
+    for (i = 0; i < niov && size != 0; i++) {
+        size_t want = iov->iov_len;
+        if (want > size) {
+            want = size;
+        }
+        buffer_append(&ioc->encoutput, iov->iov_base, want);
+        size -= want;
+    }
Quoted 2 lines

Umm, where are you incrementing iov? It appears you only tested with niov == 1.

Opps. Yeah, we need to cope with niov > 0 to satisfy qio_channel_writev API contract, but the VNC server only ever sends a single iov element so I didn't hit the bug.

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#12 lore raw

PATCH 5/7 io: get rid of bounce buffering in websock write path

In reply to Daniel P. Berrange

Currently most outbound I/O on the websock channel gets copied into the rawoutput buffer, and then immediately copied again into the encoutput buffer, with a header prepended. Now that qio_channel_websock_encode accepts a struct iovec, we can trivially remove this bounce buffering and write directly to encoutput.

In doing so, we also now correctly validate the encoutput size against the QIO_CHANNEL_WEBSOCK_MAX_BUFFER limit.

---
 include/io/channel-websock.h |  1 -
 io/channel-websock.c         | 64 +++++++++++++++++++-------------------------
 2 files changed, 28 insertions(+), 37 deletions(-)
include/io/channel-websock.h +0 -1
index 3f92535cae..f2ac0fdad1 100644
@@ -59,7 +59,6 @@ struct QIOChannelWebsock {
     Buffer encinput;
     Buffer encoutput;
     Buffer rawinput;
-    Buffer rawoutput;
     size_t payload_remain;
     size_t ping_remain;
     QIOChannelWebsockMask mask;
io/channel-websock.c +28 -36

111 lines not on this page · raw message

Signature · 1 line
2.13.5
#13 lore raw

Re: [Qemu-devel] [PATCH v1 5/7] io: get rid of bounce buffering in websock write path

Quoted 16 lines · Daniel P. Berrange

Currently most outbound I/O on the websock channel gets copied into the rawoutput buffer, and then immediately copied again into the encoutput buffer, with a header prepended. Now that qio_channel_websock_encode accepts a struct iovec, we can trivially remove this bounce buffering and write directly to encoutput.

In doing so, we also now correctly validate the encoutput size against the QIO_CHANNEL_WEBSOCK_MAX_BUFFER limit.

---
 include/io/channel-websock.h |  1 -
 io/channel-websock.c         | 64 +++++++++++++++++++-------------------------
 2 files changed, 28 insertions(+), 37 deletions(-)

io/channel-websock.c static ssize_t qio_channel_websock_write_wire(QIOChannelWebsock *, Error **); · line 650

Quoted 6 lines
io/channel-websock.c +1 -4
-    iov.iov_base = ioc->rawoutput.buffer;
-    iov.iov_len = ioc->rawoutput.offset;
     qio_channel_websock_encode(ioc, QIO_CHANNEL_WEBSOCK_OPCODE_CLOSE,
-                               &iov, 1, iov.iov_len);
-    buffer_reset(&ioc->rawoutput);
+                               iov, niov, size);

If 4/7 changes this to compute size from the iov/niov, then this has knock-on impact.

Quoted 9 lines
io/channel-websock.c +4 -0
@@ -1115,32 +1118,21 @@ static ssize_t qio_channel_websock_writev(QIOChannel *ioc,
         return -1;
 
+    avail = wioc->encoutput.offset >= QIO_CHANNEL_WEBSOCK_MAX_BUFFER ?
+        0 : (QIO_CHANNEL_WEBSOCK_MAX_BUFFER - wioc->encoutput.offset);
+    if (want > avail) {
+        want = avail;
     }
 

io/channel-websock.c static ssize_t qio_channel_websock_writev(QIOChannel *ioc, · line 1129

Quoted 10 lines
io/channel-websock.c +2 -6
- done:
-    if (wioc->rawoutput.offset) {
-        struct iovec iov = { .iov_base = wioc->rawoutput.buffer,
-                             .iov_len = wioc->rawoutput.offset };
+    if (want) {
         qio_channel_websock_encode(wioc,
                                    QIO_CHANNEL_WEBSOCK_OPCODE_BINARY_FRAME,
-                                   &iov, 1, iov.iov_len);
-        buffer_reset(&wioc->rawoutput);
+                                   iov, niov, want);

then again, I see that you DO support want < iov_size(iov, niov).

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#14 lore raw

PATCH 6/7 io: cope with websock 'Connection' header having multiple values

In reply to Daniel P. Berrange

The noVNC server sends a header "Connection: keep-alive, Upgrade" which fails our simple equality test. Split the header on ',', trim whitespace and then check for 'upgrade' token.

---
 io/channel-websock.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
io/channel-websock.c +13 -1
index 455f5e322c..a21915881d 100644
@@ -374,6 +374,9 @@ static void qio_channel_websock_handshake_process(QIOChannelWebsock *ioc,
     size_t nhdrs = G_N_ELEMENTS(hdrs);
     const char *protocols = NULL, *version = NULL, *key = NULL,
         *host = NULL, *connection = NULL, *upgrade = NULL;
+    char **connectionv;
+    bool upgraded = false;
+    size_t i;
 
     nhdrs = qio_channel_websock_extract_headers(ioc, buffer, hdrs, nhdrs, errp);
     if (!nhdrs) {
@@ -440,7 +443,16 @@ static void qio_channel_websock_handshake_process(QIOChannelWebsock *ioc,
         goto bad_request;
     }
 
-    if (strcasecmp(connection, QIO_CHANNEL_WEBSOCK_CONNECTION_UPGRADE) != 0) {
+    connectionv = g_strsplit(connection, ",", 0);
+    for (i = 0; connectionv != NULL && connectionv[i] != NULL; i++) {
+        g_strstrip(connectionv[i]);
+        if (strcasecmp(connectionv[i],
+                       QIO_CHANNEL_WEBSOCK_CONNECTION_UPGRADE) == 0) {
+            upgraded = true;
+        }
+    }
+    g_strfreev(connectionv);
+    if (!upgraded) {
         error_setg(errp, "No connection upgrade requested '%s'", connection);
         goto bad_request;
     }
Signature · 1 line
2.13.5
Reviewed-by on v1 6/7
#15 lore raw

Re: [Qemu-devel] [PATCH v1 6/7] io: cope with websock 'Connection' header having multiple values

Quoted 11 lines · Daniel P. Berrange

The noVNC server sends a header "Connection: keep-alive, Upgrade" which fails our simple equality test. Split the header on ',', trim whitespace and then check for 'upgrade' token.

---
 io/channel-websock.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)

io/channel-websock.c static void qio_channel_websock_handshake_process(QIOChannelWebsock *ioc, · line 443

Quoted 4 lines
io/channel-websock.c +0 -1
@@ -440,7 +443,16 @@ static void qio_channel_websock_handshake_process(QIOChannelWebsock *ioc,
         goto bad_request;
     }
 
-    if (strcasecmp(connection, QIO_CHANNEL_WEBSOCK_CONNECTION_UPGRADE) != 0) {

My first thought was whether you could just use strcasestr() instead of strcasecmp(), rather than the malloc overhead of g_strsplit(). But that would treat "noUpgradeGarbage" as success, making your approach a bit stricter.

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org
#16 lore raw

Re: [Qemu-devel] [PATCH v1 6/7] io: cope with websock 'Connection' header having multiple values

Quoted 15 lines · Eric Blake
Quoted 14 lines · Daniel P. Berrange

The noVNC server sends a header "Connection: keep-alive, Upgrade" which fails our simple equality test. Split the header on ',', trim whitespace and then check for 'upgrade' token.

---
 io/channel-websock.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
io/channel-websock.c +0 -0
@@ -440,7 +443,16 @@ static void qio_channel_websock_handshake_process(QIOChannelWebsock *ioc,
         goto bad_request;
     }
 
Quoted 6 lines

io/channel-websock.c static void qio_channel_websock_handshake_process(QIOChannelWebsock *ioc, · line 443

Quoted 1 line
io/channel-websock.c +0 -1
-    if (strcasecmp(connection, QIO_CHANNEL_WEBSOCK_CONNECTION_UPGRADE) != 0) {

My first thought was whether you could just use strcasestr() instead of strcasecmp(), rather than the malloc overhead of g_strsplit(). But that would treat "noUpgradeGarbage" as success, making your approach a bit stricter.

Also note that when reading HTTP headers we've already limited max data size to 4k for the entire HTTP header set. So we're doing g_strsplit over a pretty short piece of data, so negligible perf implications of that.

Quoted 2 lines

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#17 lore raw

PATCH 7/7 io: add trace points for websocket HTTP protocol headers

In reply to Daniel P. Berrange
---
 io/channel-websock.c | 4 ++++
 io/trace-events      | 2 ++
 2 files changed, 6 insertions(+)
io/channel-websock.c +4 -0
index a21915881d..003d467715 100644
@@ -224,6 +224,7 @@ qio_channel_websock_extract_headers(QIOChannelWebsock *ioc,
         goto bad_request;
     }
     *nl = '\0';
+    trace_qio_channel_websock_http_greeting(ioc, buffer);
 
     tmp = strchr(buffer, ' ');
     if (!tmp) {
@@ -425,6 +426,9 @@ static void qio_channel_websock_handshake_process(QIOChannelWebsock *ioc,
         goto bad_request;
     }
 
+    trace_qio_channel_websock_http_request(ioc, protocols, version,
+                                           host, connection, upgrade, key);
+
     if (!g_strrstr(protocols, QIO_CHANNEL_WEBSOCK_PROTOCOL_BINARY)) {
         error_setg(errp, "No '%s' protocol is supported by client '%s'",
                    QIO_CHANNEL_WEBSOCK_PROTOCOL_BINARY, protocols);
io/trace-events +2 -0
index 801b5dcb61..f70bad7cbe 100644
@@ -48,6 +48,8 @@ qio_channel_websock_handshake_pending(void *ioc, int status) "Websock handshake
 qio_channel_websock_handshake_reply(void *ioc) "Websock handshake reply ioc=%p"
 qio_channel_websock_handshake_fail(void *ioc, const char *msg) "Websock handshake fail ioc=%p err=%s"
 qio_channel_websock_handshake_complete(void *ioc) "Websock handshake complete ioc=%p"
+qio_channel_websock_http_greeting(void *ioc, const char *greeting) "Websocket HTTP request ioc=%p greeting='%s'"
+qio_channel_websock_http_request(void *ioc, const char *protocols, const char *version, const char *host, const char *connection, const char *upgrade, const char *key) "Websocket HTTP request ioc=%p protocols='%s' version='%s' host='%s' connection='%s' upgrade='%s' key='%s'"
 qio_channel_websock_header_partial_decode(void *ioc, size_t payloadlen, unsigned char fin, unsigned char opcode, unsigned char has_mask) "Websocket header decoded ioc=%p payload-len=%zu fin=0x%x opcode=0x%x has_mask=0x%x"
 qio_channel_websock_header_full_decode(void *ioc, size_t headerlen, size_t payloadlen, uint32_t mask) "Websocket header decoded ioc=%p header-len=%zu payload-len=%zu mask=0x%x"
 qio_channel_websock_payload_decode(void *ioc, uint8_t opcode, size_t payload_remain) "Websocket header decoded ioc=%p opcode=0x%x payload-remain=%zu"
Signature · 1 line
2.13.5
Reviewed-by on v1 7/7
#18 lore raw

Re: [Qemu-devel] [PATCH v1 7/7] io: add trace points for websocket HTTP protocol headers

Quoted 6 lines · Daniel P. Berrange
---
 io/channel-websock.c | 4 ++++
 io/trace-events      | 2 ++
 2 files changed, 6 insertions(+)
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org