Skip to contentSkip to conversation

refine vhost-vdpa initialization

PATCH In review Jason Wang 4 participants 20 messages Aug 31, 2020 – Sep 16, 2020 Networking

lore
Share & tools
lore

Jason Wang

author
#0 lore raw

cover refine vhost-vdpa initialization

Hi all:

This series refine vhost-vdpa initialization by:

- fixing SIGSEV when vhostdev is not specificed
- tweak the codes for vhost-vdpa initialization
- allow to accept preopen vhost-vdpa file descriptor

Please review.

Thanks

Jason Wang (9):
vhost-vdpa: remove the default devname
vhost-vdpa: mandate vhostdev
vhost-vdpa: remove the unnecessary assert(s->vhost_net)
vhost-vdpa: refine info string
vhost-vdpa: remove the unnecessary initialization
vhost-vdpa: remove the unnecessary queue index assignment
vhost-vdpa: squash net_vhost_vdpa_init() into net_init_vhost_vdpa()
vhost-vdpa: add accurate error string when fail to open vhost vDPA
device
vhost-vdpa: allow pre-opend file descriptor

 net/vhost-vdpa.c | 63 ++++++++++++++++++++++++++++++------------------
 qapi/net.json    |  6 +++--
 2 files changed, 43 insertions(+), 26 deletions(-)
Signature · 1 line
2.20.1
author
#1 lore raw

PATCH 1/9 vhost-vdpa: remove the default devname

The code doesn't have a default vhostdev, so remove the default description in net.json.

---
 qapi/net.json | 1 -
 1 file changed, 1 deletion(-)
qapi/net.json +0 -1
index ddb113e5e5..a2a94fad3e 100644
@@ -438,7 +438,6 @@
 # specifications with a vendor specific control path.
 #
 # @vhostdev: path of vhost-vdpa device
-#            (default:'/dev/vhost-vdpa-0')
 #
 # @queues: number of queues to be created for multiqueue vhost-vdpa
 #          (default: 1)
Signature · 1 line
2.20.1
#2 lore raw

Re: [PATCH 1/9] vhost-vdpa: remove the default devname

Quoted 14 lines · Jason Wang

The code doesn't have a default vhostdev, so remove the default description in net.json.

---
 qapi/net.json | 1 -
 1 file changed, 1 deletion(-)
qapi/net.json +0 -0
index ddb113e5e5..a2a94fad3e 100644

qapi/net.json line 443

Quoted 7 lines
qapi/net.json +0 -1
@@ -438,7 +438,6 @@
 # specifications with a vendor specific control path.
 #
 # @vhostdev: path of vhost-vdpa device
-#            (default:'/dev/vhost-vdpa-0')
 #
 # @queues: number of queues to be created for multiqueue vhost-vdpa
 #          (default: 1)

If you remove the default, you must also set the parameter as not optional (remove the star in front of its declaration):

qapi/net.json +1 -1
index ddb113e5e5a8..012830ca1a27 100644
@@ -447,7 +447,7 @@
 ##
 { 'struct': 'NetdevVhostVDPAOptions',
   'data': {
-    '*vhostdev':     'str',
+    'vhostdev':     'str',
     '*queues':       'int' } }
 
 ##

And then you'll have:

$ ./qemu-system-x86_64  -netdev vhost-vdpa,id=hostnet1
qemu-system-x86_64: Parameter 'vhostdev' is missing

And PATCH 2 becomes useless.

Thanks,
Laurent

author
#3 lore raw

PATCH 2/9 vhost-vdpa: mandate vhostdev

In reply to Jason Wang

vhost-dev is mandatory for vhost-vdpa to be initialized otherwise net_vhost_vdpa_init will pass an uninitialized pointer to qemu_open() which will lead a SIGSEV:

#0  0x0000555555c3a640 in strstart (str=str@entry=0x0, val=val@entry=0x555555dbcb65 "/dev/fdset/", ptr=ptr@entry=0x7fffffffdfb8) at ../util/cutils.c:77
#1  0x0000555555c572a7 in qemu_open (name=name@entry=0x0, flags=flags@entry=2) at ../util/osdep.c:294
#2  0x000055555584314a in net_vhost_vdpa_init (device=0x555555c81baa "vhost-vdpa", vhostdev=0x0, name=0x555556396600 "hn0", peer=0x0) at ../net/vhost-vdpa.c:187
#3  0x000055555584314a in net_init_vhost_vdpa (netdev=<optimized out>, name=0x555556396600 "hn0", peer=0x0, errp=<optimized out>) at ../net/vhost-vdpa.c:227
#4  0x000055555587e8c9 in net_client_init1 (netdev=0x555556398790, is_netdev=is_netdev@entry=true, errp=errp@entry=0x7fffffffe290) at ../net/net.c:1008
#5  0x000055555587ecc7 in net_client_init (opts=0x555556192ff0, is_netdev=<optimized out>, errp=0x7fffffffe290) at ../net/net.c:1113
#6  0x0000555555c33082 in qemu_opts_foreach
    (list=<optimized out>, func=func@entry=0x55555587ed50 <net_init_netdev>, opaque=opaque@entry=0x0, errp=errp@entry=0x7fffffffe290) at ../util/qemu-option.c:1172
#7  0x0000555555880057 in net_init_clients (errp=errp@entry=0x7fffffffe290) at ../net/net.c:1494
#8  0x0000555555a7f8e9 in qemu_init (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) at ../softmmu/vl.c:4250
#9  0x00005555557f26cd in main (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) at ../softmmu/main.c:49
---
 net/vhost-vdpa.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
net/vhost-vdpa.c +5 -1
index bc0e0d2d35..b7221beaa1 100644
@@ -206,7 +206,7 @@ static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
     }
     if (strcmp(netdev, name) == 0 &&
         !g_str_has_prefix(driver, "virtio-net-")) {
-        error_setg(errp, "vhost-vdpa requires frontend driver virtio-net-*");
+        error_setg(errp, "Vhost-vdpa requires frontend driver virtio-net-*");
         return -1;
     }
     return 0;
@@ -224,5 +224,9 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
                           (char *)name, errp)) {
         return -1;
     }
+    if (!opts->has_vhostdev) {
+        error_setg(errp, "vhost-vdpa requires vhostdev to be set");
+        return -1;
+    }
     return net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, opts->vhostdev);
 }
Signature · 1 line
2.20.1
#4 lore raw

Re: [PATCH 2/9] vhost-vdpa: mandate vhostdev

Quoted 36 lines · Jason Wang

vhost-dev is mandatory for vhost-vdpa to be initialized otherwise net_vhost_vdpa_init will pass an uninitialized pointer to qemu_open() which will lead a SIGSEV:

#0  0x0000555555c3a640 in strstart (str=str@entry=0x0, val=val@entry=0x555555dbcb65 "/dev/fdset/", ptr=ptr@entry=0x7fffffffdfb8) at ../util/cutils.c:77
#1  0x0000555555c572a7 in qemu_open (name=name@entry=0x0, flags=flags@entry=2) at ../util/osdep.c:294
#2  0x000055555584314a in net_vhost_vdpa_init (device=0x555555c81baa "vhost-vdpa", vhostdev=0x0, name=0x555556396600 "hn0", peer=0x0) at ../net/vhost-vdpa.c:187
#3  0x000055555584314a in net_init_vhost_vdpa (netdev=<optimized out>, name=0x555556396600 "hn0", peer=0x0, errp=<optimized out>) at ../net/vhost-vdpa.c:227
#4  0x000055555587e8c9 in net_client_init1 (netdev=0x555556398790, is_netdev=is_netdev@entry=true, errp=errp@entry=0x7fffffffe290) at ../net/net.c:1008
#5  0x000055555587ecc7 in net_client_init (opts=0x555556192ff0, is_netdev=<optimized out>, errp=0x7fffffffe290) at ../net/net.c:1113
#6  0x0000555555c33082 in qemu_opts_foreach
    (list=<optimized out>, func=func@entry=0x55555587ed50 <net_init_netdev>, opaque=opaque@entry=0x0, errp=errp@entry=0x7fffffffe290) at ../util/qemu-option.c:1172
#7  0x0000555555880057 in net_init_clients (errp=errp@entry=0x7fffffffe290) at ../net/net.c:1494
#8  0x0000555555a7f8e9 in qemu_init (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) at ../softmmu/vl.c:4250
#9  0x00005555557f26cd in main (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) at ../softmmu/main.c:49
---
 net/vhost-vdpa.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
net/vhost-vdpa.c +1 -1
index bc0e0d2d35..b7221beaa1 100644
@@ -206,7 +206,7 @@ static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
     }
     if (strcmp(netdev, name) == 0 &&
         !g_str_has_prefix(driver, "virtio-net-")) {
-        error_setg(errp, "vhost-vdpa requires frontend driver virtio-net-*");
+        error_setg(errp, "Vhost-vdpa requires frontend driver virtio-net-*");
         return -1;
     }
     return 0;

net/vhost-vdpa.c int net_init_vhost_vdpa(const Netdev *netdev, const char *name, · line 230

Quoted 7 lines
net/vhost-vdpa.c +4 -0
@@ -224,5 +224,9 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
                           (char *)name, errp)) {
         return -1;
     }
+    if (!opts->has_vhostdev) {
+        error_setg(errp, "vhost-vdpa requires vhostdev to be set");
+        return -1;
+    }

Useless if you remove the '*' in qapi/net.json.

Thanks,
Laurent

author
#5 lore raw

PATCH 3/9 vhost-vdpa: remove the unnecessary assert(s->vhost_net)

In reply to Jason Wang

vhost_vdpa_add() can fail before s->vhost_net is set and net_vhost_vdpa_init() can report error. So there's no need for this assert. Let's remove it.

---
 net/vhost-vdpa.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/vhost-vdpa.c +1 -1
index b7221beaa1..c4568b885e 100644
@@ -190,7 +190,7 @@ static int net_vhost_vdpa_init(NetClientState *peer, const char *device,
     }
     s->vhost_vdpa.device_fd = vdpa_device_fd;
     ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa);
-    assert(s->vhost_net);
+
     return ret;
 }
 
Signature · 1 line
2.20.1
author
#6 lore raw

PATCH 4/9 vhost-vdpa: refine info string

In reply to Jason Wang

We use "vhost-vdpa" as nic info string which is not useful, let's add the vhostdev path.

before:
info network
v0: index=0,type=vhost-vdpa,vhost-vdpa

after:
info network
v0: index=0,type=vhost-vdpa,vhostdev=/dev/vhost-vdpa-0

---
 net/vhost-vdpa.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/vhost-vdpa.c +1 -1
index c4568b885e..397d4d3082 100644
@@ -181,7 +181,7 @@ static int net_vhost_vdpa_init(NetClientState *peer, const char *device,
     int ret = 0;
     assert(name);
     nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, name);
-    snprintf(nc->info_str, sizeof(nc->info_str), TYPE_VHOST_VDPA);
+    snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", vhostdev);
     nc->queue_index = 0;
     s = DO_UPCAST(VhostVDPAState, nc, nc);
     vdpa_device_fd = qemu_open(vhostdev, O_RDWR);
Signature · 1 line
2.20.1
Reviewed-by on v1 4/9
#7 lore raw

Re: [PATCH 4/9] vhost-vdpa: refine info string

Quoted 22 lines · Jason Wang

We use "vhost-vdpa" as nic info string which is not useful, let's add the vhostdev path.

before:
info network
v0: index=0,type=vhost-vdpa,vhost-vdpa

after:
info network
v0: index=0,type=vhost-vdpa,vhostdev=/dev/vhost-vdpa-0

---
 net/vhost-vdpa.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/vhost-vdpa.c +0 -0
index c4568b885e..397d4d3082 100644

net/vhost-vdpa.c static int net_vhost_vdpa_init(NetClientState *peer, const char *device, · line 187

Quoted 8 lines
net/vhost-vdpa.c +1 -1
@@ -181,7 +181,7 @@ static int net_vhost_vdpa_init(NetClientState *peer, const char *device,
     int ret = 0;
     assert(name);
     nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, name);
-    snprintf(nc->info_str, sizeof(nc->info_str), TYPE_VHOST_VDPA);
+    snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", vhostdev);
     nc->queue_index = 0;
     s = DO_UPCAST(VhostVDPAState, nc, nc);
     vdpa_device_fd = qemu_open(vhostdev, O_RDWR);
author
#8 lore raw

PATCH 5/9 vhost-vdpa: remove the unnecessary initialization

In reply to Jason Wang
---
 net/vhost-vdpa.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
net/vhost-vdpa.c +3 -3
index 397d4d3082..bcbf49d55b 100644
@@ -175,10 +175,10 @@ static NetClientInfo net_vhost_vdpa_info = {
 static int net_vhost_vdpa_init(NetClientState *peer, const char *device,
                                const char *name, const char *vhostdev)
 {
-    NetClientState *nc = NULL;
+    NetClientState *nc;
     VhostVDPAState *s;
-    int vdpa_device_fd = -1;
-    int ret = 0;
+    int vdpa_device_fd;
+    int ret;
     assert(name);
     nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, name);
     snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", vhostdev);
Signature · 1 line
2.20.1
Reviewed-by on v1 5/9
#9 lore raw

Re: [PATCH 5/9] vhost-vdpa: remove the unnecessary initialization

Quoted 14 lines · Jason Wang
---
 net/vhost-vdpa.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
net/vhost-vdpa.c +0 -0
index 397d4d3082..bcbf49d55b 100644
@@ -175,10 +175,10 @@ static NetClientInfo net_vhost_vdpa_info = {
 static int net_vhost_vdpa_init(NetClientState *peer, const char *device,
                                const char *name, const char *vhostdev)
 {

net/vhost-vdpa.c static NetClientInfo net_vhost_vdpa_info = { · line 184

Quoted 10 lines
net/vhost-vdpa.c +3 -3
-    NetClientState *nc = NULL;
+    NetClientState *nc;
     VhostVDPAState *s;
-    int vdpa_device_fd = -1;
-    int ret = 0;
+    int vdpa_device_fd;
+    int ret;
     assert(name);
     nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, name);
     snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", vhostdev);
author
#10 lore raw

PATCH 6/9 vhost-vdpa: remove the unnecessary queue index assignment

In reply to Jason Wang

The nc is allocated through g_malloc0(), so no need to assign its queue_index to zero.

---
 net/vhost-vdpa.c | 1 -
 1 file changed, 1 deletion(-)
net/vhost-vdpa.c +0 -1
index bcbf49d55b..1d3ac89135 100644
@@ -182,7 +182,6 @@ static int net_vhost_vdpa_init(NetClientState *peer, const char *device,
     assert(name);
     nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, name);
     snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", vhostdev);
-    nc->queue_index = 0;
     s = DO_UPCAST(VhostVDPAState, nc, nc);
     vdpa_device_fd = qemu_open(vhostdev, O_RDWR);
     if (vdpa_device_fd == -1) {
Signature · 1 line
2.20.1
Reviewed-by on v1 6/9
#11 lore raw

Re: [PATCH 6/9] vhost-vdpa: remove the unnecessary queue index assignment

Quoted 14 lines · Jason Wang

The nc is allocated through g_malloc0(), so no need to assign its queue_index to zero.

---
 net/vhost-vdpa.c | 1 -
 1 file changed, 1 deletion(-)
net/vhost-vdpa.c +0 -0
index bcbf49d55b..1d3ac89135 100644

net/vhost-vdpa.c static int net_vhost_vdpa_init(NetClientState *peer, const char *device, · line 187

Quoted 7 lines
net/vhost-vdpa.c +0 -1
@@ -182,7 +182,6 @@ static int net_vhost_vdpa_init(NetClientState *peer, const char *device,
     assert(name);
     nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, name);
     snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", vhostdev);
-    nc->queue_index = 0;
     s = DO_UPCAST(VhostVDPAState, nc, nc);
     vdpa_device_fd = qemu_open(vhostdev, O_RDWR);
     if (vdpa_device_fd == -1) {
author
#12 lore raw

PATCH 7/9 vhost-vdpa: squash net_vhost_vdpa_init() into net_init_vhost_vdpa()

In reply to Jason Wang

This patch squashes net_vhost_vdpa_init() into net_init_vhost_vdpa(). This will simplify adding pre open file descriptor support.

---
 net/vhost-vdpa.c | 41 +++++++++++++++++++----------------------
 1 file changed, 19 insertions(+), 22 deletions(-)
net/vhost-vdpa.c +19 -22
index 1d3ac89135..f5cc4e8326 100644
@@ -172,27 +172,6 @@ static NetClientInfo net_vhost_vdpa_info = {
         .has_ufo = vhost_vdpa_has_ufo,
 };
 
-static int net_vhost_vdpa_init(NetClientState *peer, const char *device,
-                               const char *name, const char *vhostdev)
-{
-    NetClientState *nc;
-    VhostVDPAState *s;
-    int vdpa_device_fd;
-    int ret;
-    assert(name);
-    nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, name);
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", vhostdev);
-    s = DO_UPCAST(VhostVDPAState, nc, nc);
-    vdpa_device_fd = qemu_open(vhostdev, O_RDWR);
-    if (vdpa_device_fd == -1) {
-        return -errno;
-    }
-    s->vhost_vdpa.device_fd = vdpa_device_fd;
-    ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa);
-
-    return ret;
-}
-
 static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
 {
     const char *name = opaque;
@@ -215,6 +194,10 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
                         NetClientState *peer, Error **errp)
 {
     const NetdevVhostVDPAOptions *opts;
+    NetClientState *nc;
+    VhostVDPAState *s;
+    int vdpa_device_fd;
+    int ret;
 
     assert(netdev->type == NET_CLIENT_DRIVER_VHOST_VDPA);
     opts = &netdev->u.vhost_vdpa;
@@ -227,5 +210,19 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
         error_setg(errp, "vhost-vdpa requires vhostdev to be set");
         return -1;
     }
-    return net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, opts->vhostdev);
+
+    assert(name);
+
+    nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, TYPE_VHOST_VDPA, name);
+    snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", opts->vhostdev);
+
+    s = DO_UPCAST(VhostVDPAState, nc, nc);
+    vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
+    if (vdpa_device_fd == -1) {
+        return -errno;
+    }
+    s->vhost_vdpa.device_fd = vdpa_device_fd;
+    ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa);
+
+    return ret;
 }
Signature · 1 line
2.20.1
Reviewed-by on v1 7/9
#13 lore raw

Re: [PATCH 7/9] vhost-vdpa: squash net_vhost_vdpa_init() into net_init_vhost_vdpa()

Quoted 62 lines · Jason Wang

This patch squashes net_vhost_vdpa_init() into net_init_vhost_vdpa(). This will simplify adding pre open file descriptor support.

---
 net/vhost-vdpa.c | 41 +++++++++++++++++++----------------------
 1 file changed, 19 insertions(+), 22 deletions(-)
net/vhost-vdpa.c +9 -22
index 1d3ac89135..f5cc4e8326 100644
@@ -172,27 +172,6 @@ static NetClientInfo net_vhost_vdpa_info = {
         .has_ufo = vhost_vdpa_has_ufo,
 };
 
-static int net_vhost_vdpa_init(NetClientState *peer, const char *device,
-                               const char *name, const char *vhostdev)
-{
-    NetClientState *nc;
-    VhostVDPAState *s;
-    int vdpa_device_fd;
-    int ret;
-    assert(name);
-    nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, name);
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", vhostdev);
-    s = DO_UPCAST(VhostVDPAState, nc, nc);
-    vdpa_device_fd = qemu_open(vhostdev, O_RDWR);
-    if (vdpa_device_fd == -1) {
-        return -errno;
-    }
-    s->vhost_vdpa.device_fd = vdpa_device_fd;
-    ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa);
-
-    return ret;
-}
-
 static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
 {
     const char *name = opaque;
@@ -215,6 +194,10 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
                         NetClientState *peer, Error **errp)
 {
     const NetdevVhostVDPAOptions *opts;
+    NetClientState *nc;
+    VhostVDPAState *s;
+    int vdpa_device_fd;
+    int ret;
 
     assert(netdev->type == NET_CLIENT_DRIVER_VHOST_VDPA);
     opts = &netdev->u.vhost_vdpa;
@@ -227,5 +210,19 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
         error_setg(errp, "vhost-vdpa requires vhostdev to be set");
         return -1;
     }
-    return net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, opts->vhostdev);
+
+    assert(name);
+
+    nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, TYPE_VHOST_VDPA, name);
+    snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", opts->vhostdev);

net/vhost-vdpa.c int net_init_vhost_vdpa(const Netdev *netdev, const char *name, · line 227

Quoted 10 lines
net/vhost-vdpa.c +10 -0
+
+    s = DO_UPCAST(VhostVDPAState, nc, nc);
+    vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
+    if (vdpa_device_fd == -1) {
+        return -errno;
+    }
+    s->vhost_vdpa.device_fd = vdpa_device_fd;
+    ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa);
+
+    return ret;

You can avoid to declare and use "ret" by doing directly

"return vhost_vdpa_add()"

Anyway, the code move is correct:

author
#14 lore raw

PATCH 8/9 vhost-vdpa: add accurate error string when fail to open vhost vDPA device

In reply to Jason Wang

This patch adds more accurate error string when fail to open vhost vDPA device.

---
 net/vhost-vdpa.c | 1 +
 1 file changed, 1 insertion(+)
net/vhost-vdpa.c +1 -0
index f5cc4e8326..9a6f0b63d3 100644
@@ -219,6 +219,7 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
     s = DO_UPCAST(VhostVDPAState, nc, nc);
     vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
     if (vdpa_device_fd == -1) {
+        error_setg(errp, "Fail to open vhost-vdpa device %s", opts->vhostdev);
         return -errno;
     }
     s->vhost_vdpa.device_fd = vdpa_device_fd;
Signature · 1 line
2.20.1
#15 lore raw

Re: [PATCH 8/9] vhost-vdpa: add accurate error string when fail to open vhost vDPA device

net/vhost-vdpa.c int net_init_vhost_vdpa(const Netdev *netdev, const char *name,

Quoted 13 lines · Jason Wang

This patch adds more accurate error string when fail to open vhost vDPA device.

---
 net/vhost-vdpa.c | 1 +
 1 file changed, 1 insertion(+)
net/vhost-vdpa.c +0 -0
index f5cc4e8326..9a6f0b63d3 100644
@@ -219,6 +219,7 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,

I think you can also remove the "assert(name)", qemu_net_client_setup() in qemu_new_net_client() uses a default value if NULL.

net/vhost-vdpa.c int net_init_vhost_vdpa(const Netdev *netdev, const char *name, · line 223

Quoted 5 lines
net/vhost-vdpa.c +1 -0
     s = DO_UPCAST(VhostVDPAState, nc, nc);
     vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
     if (vdpa_device_fd == -1) {
+        error_setg(errp, "Fail to open vhost-vdpa device %s", opts->vhostdev);
         return -errno;

you can "return -1" now: no one checks for the exact returned value, all the other init functions from net_client_init_fun[] return -1 or 0, and the errno is now in the error string.

net/vhost-vdpa.c int net_init_vhost_vdpa(const Netdev *netdev, const char *name, · line 225

Quoted 3 lines
net/vhost-vdpa.c +0 -0
     }
     s->vhost_vdpa.device_fd = vdpa_device_fd;

Thanks,
Laurent

author
#16 lore raw

PATCH 9/9 vhost-vdpa: allow pre-opend file descriptor

In reply to Jason Wang

This patch allows to initialize vhost-vdpa network backend with pre opened vhost-vdpa file descriptor. This is useful for running unprivileged qemu through libvirt.

Cc Eric Blake <[email protected]>
Cc Markus Armbruster <[email protected]>
---
 net/vhost-vdpa.c | 24 +++++++++++++++++++-----
 qapi/net.json    |  5 ++++-
 2 files changed, 23 insertions(+), 6 deletions(-)
net/vhost-vdpa.c +19 -5
index 9a6f0b63d3..f6385cd264 100644
@@ -206,20 +206,34 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
                           (char *)name, errp)) {
         return -1;
     }
-    if (!opts->has_vhostdev) {
-        error_setg(errp, "vhost-vdpa requires vhostdev to be set");
+    if (!(opts->has_vhostdev ^ opts->has_fd)) {
+        error_setg(errp, "Vhost-vdpa requires either vhostdev or fd to be set");
         return -1;
     }
 
     assert(name);
 
     nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, TYPE_VHOST_VDPA, name);
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", opts->vhostdev);
+    if (opts->has_vhostdev) {
+        snprintf(nc->info_str, sizeof(nc->info_str),
+                 "vhostdev=%s", opts->vhostdev);
+        vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
+        if (vdpa_device_fd == -1) {
+            error_setg(errp, "Fail to open vhost-vdpa device %s",
+                       opts->vhostdev);
+            return -errno;
+        }
+    } else {
+        snprintf(nc->info_str, sizeof(nc->info_str), "fd=%s", opts->fd);
+        vdpa_device_fd = monitor_fd_param(cur_mon, opts->fd, errp);
+        if (vdpa_device_fd == -1) {
+            return -1;
+        }
+    }
 
     s = DO_UPCAST(VhostVDPAState, nc, nc);
-    vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
     if (vdpa_device_fd == -1) {
-        error_setg(errp, "Fail to open vhost-vdpa device %s", opts->vhostdev);
+
         return -errno;
     }
     s->vhost_vdpa.device_fd = vdpa_device_fd;
qapi/net.json +4 -1
index a2a94fad3e..5ad60c3045 100644
@@ -442,12 +442,15 @@
 # @queues: number of queues to be created for multiqueue vhost-vdpa
 #          (default: 1)
 #
+# @fd: file descriptor of an already opened vhost-vdpa (since 5.2)
+#
 # Since: 5.1
 ##
 { 'struct': 'NetdevVhostVDPAOptions',
   'data': {
     '*vhostdev':     'str',
-    '*queues':       'int' } }
+    '*queues':       'int',
+    '*fd':           'str' } }
 
 ##
 # @NetClientDriver:
Signature · 1 line
2.20.1
#17 lore raw

Re: [PATCH 9/9] vhost-vdpa: allow pre-opend file descriptor

Quoted 79 lines · Jason Wang

This patch allows to initialize vhost-vdpa network backend with pre opened vhost-vdpa file descriptor. This is useful for running unprivileged qemu through libvirt.

Cc Eric Blake <[email protected]>
Cc Markus Armbruster <[email protected]>
---
 net/vhost-vdpa.c | 24 +++++++++++++++++++-----
 qapi/net.json    |  5 ++++-
 2 files changed, 23 insertions(+), 6 deletions(-)
net/vhost-vdpa.c +19 -5
index 9a6f0b63d3..f6385cd264 100644
@@ -206,20 +206,34 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
                           (char *)name, errp)) {
         return -1;
     }
-    if (!opts->has_vhostdev) {
-        error_setg(errp, "vhost-vdpa requires vhostdev to be set");
+    if (!(opts->has_vhostdev ^ opts->has_fd)) {
+        error_setg(errp, "Vhost-vdpa requires either vhostdev or fd to be set");
         return -1;
     }
 
     assert(name);
 
     nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, TYPE_VHOST_VDPA, name);
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", opts->vhostdev);
+    if (opts->has_vhostdev) {
+        snprintf(nc->info_str, sizeof(nc->info_str),
+                 "vhostdev=%s", opts->vhostdev);
+        vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
+        if (vdpa_device_fd == -1) {
+            error_setg(errp, "Fail to open vhost-vdpa device %s",
+                       opts->vhostdev);
+            return -errno;
+        }
+    } else {
+        snprintf(nc->info_str, sizeof(nc->info_str), "fd=%s", opts->fd);
+        vdpa_device_fd = monitor_fd_param(cur_mon, opts->fd, errp);
+        if (vdpa_device_fd == -1) {
+            return -1;
+        }
+    }
 
     s = DO_UPCAST(VhostVDPAState, nc, nc);
-    vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
     if (vdpa_device_fd == -1) {
-        error_setg(errp, "Fail to open vhost-vdpa device %s", opts->vhostdev);
+
         return -errno;
     }
     s->vhost_vdpa.device_fd = vdpa_device_fd;
qapi/net.json +4 -2
index a2a94fad3e..5ad60c3045 100644
@@ -442,12 +442,15 @@
 # @queues: number of queues to be created for multiqueue vhost-vdpa
 #          (default: 1)
 #
+# @fd: file descriptor of an already opened vhost-vdpa (since 5.2)
+#
 # Since: 5.1
 ##
 { 'struct': 'NetdevVhostVDPAOptions',
   'data': {
     '*vhostdev':     'str',
-    '*queues':       'int' } }
+    '*queues':       'int',
+    '*fd':           'str' } }
 
 ##
 # @NetClientDriver:
--
Quoted 2 lines

2.20.1

I think the latest code supported this part.
you can pass a pre open file descriptor to it via the add-fd QMP command to /dev/fdset/NNN, and then pass the string "/dev/fdset/NNN" as vhostdev. so we don't need a special fd parameter here.

Thanks
Cindy

author
#18 lore raw

Re: [PATCH 9/9] vhost-vdpa: allow pre-opend file descriptor

Quoted 76 lines · Cindy Lu
Quoted 75 lines · Jason Wang

This patch allows to initialize vhost-vdpa network backend with pre opened vhost-vdpa file descriptor. This is useful for running unprivileged qemu through libvirt.

Cc Eric Blake <[email protected]>
Cc Markus Armbruster <[email protected]>
---
  net/vhost-vdpa.c | 24 +++++++++++++++++++-----
  qapi/net.json    |  5 ++++-
  2 files changed, 23 insertions(+), 6 deletions(-)
net/vhost-vdpa.c +19 -5
index 9a6f0b63d3..f6385cd264 100644
@@ -206,20 +206,34 @@ int net_init_vhost_vdpa(const Netdev *netdev, const char *name,
                            (char *)name, errp)) {
          return -1;
      }
-    if (!opts->has_vhostdev) {
-        error_setg(errp, "vhost-vdpa requires vhostdev to be set");
+    if (!(opts->has_vhostdev ^ opts->has_fd)) {
+        error_setg(errp, "Vhost-vdpa requires either vhostdev or fd to be set");
          return -1;
      }
 
      assert(name);
 
      nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, TYPE_VHOST_VDPA, name);
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhostdev=%s", opts->vhostdev);
+    if (opts->has_vhostdev) {
+        snprintf(nc->info_str, sizeof(nc->info_str),
+                 "vhostdev=%s", opts->vhostdev);
+        vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
+        if (vdpa_device_fd == -1) {
+            error_setg(errp, "Fail to open vhost-vdpa device %s",
+                       opts->vhostdev);
+            return -errno;
+        }
+    } else {
+        snprintf(nc->info_str, sizeof(nc->info_str), "fd=%s", opts->fd);
+        vdpa_device_fd = monitor_fd_param(cur_mon, opts->fd, errp);
+        if (vdpa_device_fd == -1) {
+            return -1;
+        }
+    }
 
      s = DO_UPCAST(VhostVDPAState, nc, nc);
-    vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR);
      if (vdpa_device_fd == -1) {
-        error_setg(errp, "Fail to open vhost-vdpa device %s", opts->vhostdev);
+
          return -errno;
      }
      s->vhost_vdpa.device_fd = vdpa_device_fd;
qapi/net.json +4 -1
index a2a94fad3e..5ad60c3045 100644
@@ -442,12 +442,15 @@
  # @queues: number of queues to be created for multiqueue vhost-vdpa
  #          (default: 1)
  #
+# @fd: file descriptor of an already opened vhost-vdpa (since 5.2)
+#
  # Since: 5.1
  ##
  { 'struct': 'NetdevVhostVDPAOptions',
    'data': {
      '*vhostdev':     'str',
-    '*queues':       'int' } }
+    '*queues':       'int',
+    '*fd':           'str' } }
 
Quoted 9 lines
Quoted 5 lines
qapi/net.json +0 -1
  ##
  # @NetClientDriver:
--

2.20.1

I think the latest code supported this part.
you can pass a pre open file descriptor to it via the add-fd QMP command to /dev/fdset/NNN, and then pass the string "/dev/fdset/NNN" as vhostdev. so we don't need a special fd parameter here.

Right, I forgot the fdset tricks.

Thanks

Quoted 4 lines

Thanks
Cindy

#19 lore raw

Re: [PATCH 9/9] vhost-vdpa: allow pre-opend file descriptor

In reply to Cindy Lu
Quoted 27 lines · Cindy Lu
Quoted 26 lines · Jason Wang

This patch allows to initialize vhost-vdpa network backend with pre opened vhost-vdpa file descriptor. This is useful for running unprivileged qemu through libvirt.

Cc Eric Blake <[email protected]>
Cc Markus Armbruster <[email protected]>

---

+++ b/qapi/net.json

qapi/net.json +4 -1
@@ -442,12 +442,15 @@
  # @queues: number of queues to be created for multiqueue vhost-vdpa
  #          (default: 1)
  #
+# @fd: file descriptor of an already opened vhost-vdpa (since 5.2)
+#
  # Since: 5.1
  ##
  { 'struct': 'NetdevVhostVDPAOptions',
    'data': {
      '*vhostdev':     'str',
-    '*queues':       'int' } }
+    '*queues':       'int',
+    '*fd':           'str' } }
 
Quoted 9 lines
Quoted 5 lines
qapi/net.json +0 -1
  ##
  # @NetClientDriver:
--

2.20.1

I think the latest code supported this part.
you can pass a pre open file descriptor to it via the add-fd QMP command to /dev/fdset/NNN, and then pass the string "/dev/fdset/NNN" as vhostdev. so we don't need a special fd parameter here.

Correct - the 'vhostdev' parameter + magic filename is all the more that is needed to access a pre-opened fd, so no 'fd' parameter should be added.

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org