summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--include/linux/socket.h4
-rw-r--r--include/uapi/linux/vm_sockets.h171
-rw-r--r--net/Kconfig1
-rw-r--r--net/Makefile1
-rw-r--r--net/vmw_vsock/Kconfig28
-rw-r--r--net/vmw_vsock/Makefile7
-rw-r--r--net/vmw_vsock/af_vsock.c2015
-rw-r--r--net/vmw_vsock/af_vsock.h175
-rw-r--r--net/vmw_vsock/vmci_transport.c2157
-rw-r--r--net/vmw_vsock/vmci_transport.h139
-rw-r--r--net/vmw_vsock/vmci_transport_notify.c680
-rw-r--r--net/vmw_vsock/vmci_transport_notify.h83
-rw-r--r--net/vmw_vsock/vmci_transport_notify_qstate.c438
-rw-r--r--net/vmw_vsock/vsock_addr.c86
-rw-r--r--net/vmw_vsock/vsock_addr.h32
-rw-r--r--net/vmw_vsock/vsock_version.h22
16 files changed, 6038 insertions, 1 deletions
diff --git a/include/linux/socket.h b/include/linux/socket.h
index 9a546ff853dc..2b9f74b0ffea 100644
--- a/include/linux/socket.h
+++ b/include/linux/socket.h
@@ -178,7 +178,8 @@ struct ucred {
178#define AF_CAIF 37 /* CAIF sockets */ 178#define AF_CAIF 37 /* CAIF sockets */
179#define AF_ALG 38 /* Algorithm sockets */ 179#define AF_ALG 38 /* Algorithm sockets */
180#define AF_NFC 39 /* NFC sockets */ 180#define AF_NFC 39 /* NFC sockets */
181#define AF_MAX 40 /* For now.. */ 181#define AF_VSOCK 40 /* vSockets */
182#define AF_MAX 41 /* For now.. */
182 183
183/* Protocol families, same as address families. */ 184/* Protocol families, same as address families. */
184#define PF_UNSPEC AF_UNSPEC 185#define PF_UNSPEC AF_UNSPEC
@@ -221,6 +222,7 @@ struct ucred {
221#define PF_CAIF AF_CAIF 222#define PF_CAIF AF_CAIF
222#define PF_ALG AF_ALG 223#define PF_ALG AF_ALG
223#define PF_NFC AF_NFC 224#define PF_NFC AF_NFC
225#define PF_VSOCK AF_VSOCK
224#define PF_MAX AF_MAX 226#define PF_MAX AF_MAX
225 227
226/* Maximum queue length specifiable by listen. */ 228/* Maximum queue length specifiable by listen. */
diff --git a/include/uapi/linux/vm_sockets.h b/include/uapi/linux/vm_sockets.h
new file mode 100644
index 000000000000..f7f2e99dec84
--- /dev/null
+++ b/include/uapi/linux/vm_sockets.h
@@ -0,0 +1,171 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#ifndef _VM_SOCKETS_H_
17#define _VM_SOCKETS_H_
18
19#if !defined(__KERNEL__)
20#include <sys/socket.h>
21#endif
22
23/* Option name for STREAM socket buffer size. Use as the option name in
24 * setsockopt(3) or getsockopt(3) to set or get an unsigned long long that
25 * specifies the size of the buffer underlying a vSockets STREAM socket.
26 * Value is clamped to the MIN and MAX.
27 */
28
29#define SO_VM_SOCKETS_BUFFER_SIZE 0
30
31/* Option name for STREAM socket minimum buffer size. Use as the option name
32 * in setsockopt(3) or getsockopt(3) to set or get an unsigned long long that
33 * specifies the minimum size allowed for the buffer underlying a vSockets
34 * STREAM socket.
35 */
36
37#define SO_VM_SOCKETS_BUFFER_MIN_SIZE 1
38
39/* Option name for STREAM socket maximum buffer size. Use as the option name
40 * in setsockopt(3) or getsockopt(3) to set or get an unsigned long long
41 * that specifies the maximum size allowed for the buffer underlying a
42 * vSockets STREAM socket.
43 */
44
45#define SO_VM_SOCKETS_BUFFER_MAX_SIZE 2
46
47/* Option name for socket peer's host-specific VM ID. Use as the option name
48 * in getsockopt(3) to get a host-specific identifier for the peer endpoint's
49 * VM. The identifier is a signed integer.
50 * Only available for hypervisor endpoints.
51 */
52
53#define SO_VM_SOCKETS_PEER_HOST_VM_ID 3
54
55/* Option name for socket's service label. Use as the option name in
56 * setsockopt(3) or getsockopt(3) to set or get the service label for a socket.
57 * The service label is a C-style NUL-terminated string. Only available for
58 * hypervisor endpoints.
59 */
60
61#define SO_VM_SOCKETS_SERVICE_LABEL 4
62
63/* Option name for determining if a socket is trusted. Use as the option name
64 * in getsockopt(3) to determine if a socket is trusted. The value is a
65 * signed integer.
66 */
67
68#define SO_VM_SOCKETS_TRUSTED 5
69
70/* Option name for STREAM socket connection timeout. Use as the option name
71 * in setsockopt(3) or getsockopt(3) to set or get the connection
72 * timeout for a STREAM socket.
73 */
74
75#define SO_VM_SOCKETS_CONNECT_TIMEOUT 6
76
77/* Option name for using non-blocking send/receive. Use as the option name
78 * for setsockopt(3) or getsockopt(3) to set or get the non-blocking
79 * transmit/receive flag for a STREAM socket. This flag determines whether
80 * send() and recv() can be called in non-blocking contexts for the given
81 * socket. The value is a signed integer.
82 *
83 * This option is only relevant to kernel endpoints, where descheduling the
84 * thread of execution is not allowed, for example, while holding a spinlock.
85 * It is not to be confused with conventional non-blocking socket operations.
86 *
87 * Only available for hypervisor endpoints.
88 */
89
90#define SO_VM_SOCKETS_NONBLOCK_TXRX 7
91
92/* The vSocket equivalent of INADDR_ANY. This works for the svm_cid field of
93 * sockaddr_vm and indicates the context ID of the current endpoint.
94 */
95
96#define VMADDR_CID_ANY -1U
97
98/* Bind to any available port. Works for the svm_port field of
99 * sockaddr_vm.
100 */
101
102#define VMADDR_PORT_ANY -1U
103
104/* Use this as the destination CID in an address when referring to the
105 * hypervisor. VMCI relies on it being 0, but this would be useful for other
106 * transports too.
107 */
108
109#define VMADDR_CID_HYPERVISOR 0
110
111/* This CID is specific to VMCI and can be considered reserved (even VMCI
112 * doesn't use it anymore, it's a legacy value from an older release).
113 */
114
115#define VMADDR_CID_RESERVED 1
116
117/* Use this as the destination CID in an address when referring to the host
118 * (any process other than the hypervisor). VMCI relies on it being 2, but
119 * this would be useful for other transports too.
120 */
121
122#define VMADDR_CID_HOST 2
123
124/* Invalid vSockets version. */
125
126#define VM_SOCKETS_INVALID_VERSION -1U
127
128/* The epoch (first) component of the vSockets version. A single byte
129 * representing the epoch component of the vSockets version.
130 */
131
132#define VM_SOCKETS_VERSION_EPOCH(_v) (((_v) & 0xFF000000) >> 24)
133
134/* The major (second) component of the vSockets version. A single byte
135 * representing the major component of the vSockets version. Typically
136 * changes for every major release of a product.
137 */
138
139#define VM_SOCKETS_VERSION_MAJOR(_v) (((_v) & 0x00FF0000) >> 16)
140
141/* The minor (third) component of the vSockets version. Two bytes representing
142 * the minor component of the vSockets version.
143 */
144
145#define VM_SOCKETS_VERSION_MINOR(_v) (((_v) & 0x0000FFFF))
146
147/* Address structure for vSockets. The address family should be set to
148 * whatever vmci_sock_get_af_value_fd() returns. The structure members should
149 * all align on their natural boundaries without resorting to compiler packing
150 * directives. The total size of this structure should be exactly the same as
151 * that of struct sockaddr.
152 */
153
154struct sockaddr_vm {
155 sa_family_t svm_family;
156 unsigned short svm_reserved1;
157 unsigned int svm_port;
158 unsigned int svm_cid;
159 unsigned char svm_zero[sizeof(struct sockaddr) -
160 sizeof(sa_family_t) -
161 sizeof(unsigned short) -
162 sizeof(unsigned int) - sizeof(unsigned int)];
163};
164
165#define IOCTL_VM_SOCKETS_GET_LOCAL_CID _IO(7, 0xb9)
166
167#if defined(__KERNEL__)
168int vm_sockets_get_local_cid(void);
169#endif
170
171#endif
diff --git a/net/Kconfig b/net/Kconfig
index c31348e70aad..5a1888bb036d 100644
--- a/net/Kconfig
+++ b/net/Kconfig
@@ -217,6 +217,7 @@ source "net/dcb/Kconfig"
217source "net/dns_resolver/Kconfig" 217source "net/dns_resolver/Kconfig"
218source "net/batman-adv/Kconfig" 218source "net/batman-adv/Kconfig"
219source "net/openvswitch/Kconfig" 219source "net/openvswitch/Kconfig"
220source "net/vmw_vsock/Kconfig"
220 221
221config RPS 222config RPS
222 boolean 223 boolean
diff --git a/net/Makefile b/net/Makefile
index c5aa8b3b49dc..091e7b04f301 100644
--- a/net/Makefile
+++ b/net/Makefile
@@ -69,3 +69,4 @@ obj-$(CONFIG_CEPH_LIB) += ceph/
69obj-$(CONFIG_BATMAN_ADV) += batman-adv/ 69obj-$(CONFIG_BATMAN_ADV) += batman-adv/
70obj-$(CONFIG_NFC) += nfc/ 70obj-$(CONFIG_NFC) += nfc/
71obj-$(CONFIG_OPENVSWITCH) += openvswitch/ 71obj-$(CONFIG_OPENVSWITCH) += openvswitch/
72obj-$(CONFIG_VSOCKETS) += vmw_vsock/
diff --git a/net/vmw_vsock/Kconfig b/net/vmw_vsock/Kconfig
new file mode 100644
index 000000000000..b5fa7e40cdcb
--- /dev/null
+++ b/net/vmw_vsock/Kconfig
@@ -0,0 +1,28 @@
1#
2# Vsock protocol
3#
4
5config VSOCKETS
6 tristate "Virtual Socket protocol"
7 help
8 Virtual Socket Protocol is a socket protocol similar to TCP/IP
9 allowing comunication between Virtual Machines and hypervisor
10 or host.
11
12 You should also select one or more hypervisor-specific transports
13 below.
14
15 To compile this driver as a module, choose M here: the module
16 will be called vsock. If unsure, say N.
17
18config VMWARE_VMCI_VSOCKETS
19 tristate "VMware VMCI transport for Virtual Sockets"
20 depends on VSOCKETS && VMWARE_VMCI
21 help
22 This module implements a VMCI transport for Virtual Sockets.
23
24 Enable this transport if your Virtual Machine runs on a VMware
25 hypervisor.
26
27 To compile this driver as a module, choose M here: the module
28 will be called vmw_vsock_vmci_transport. If unsure, say N.
diff --git a/net/vmw_vsock/Makefile b/net/vmw_vsock/Makefile
new file mode 100644
index 000000000000..2ce52d70f224
--- /dev/null
+++ b/net/vmw_vsock/Makefile
@@ -0,0 +1,7 @@
1obj-$(CONFIG_VSOCKETS) += vsock.o
2obj-$(CONFIG_VMWARE_VMCI_VSOCKETS) += vmw_vsock_vmci_transport.o
3
4vsock-y += af_vsock.o vsock_addr.o
5
6vmw_vsock_vmci_transport-y += vmci_transport.o vmci_transport_notify.o \
7 vmci_transport_notify_qstate.o
diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
new file mode 100644
index 000000000000..54bb7bdf92d3
--- /dev/null
+++ b/net/vmw_vsock/af_vsock.c
@@ -0,0 +1,2015 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16/* Implementation notes:
17 *
18 * - There are two kinds of sockets: those created by user action (such as
19 * calling socket(2)) and those created by incoming connection request packets.
20 *
21 * - There are two "global" tables, one for bound sockets (sockets that have
22 * specified an address that they are responsible for) and one for connected
23 * sockets (sockets that have established a connection with another socket).
24 * These tables are "global" in that all sockets on the system are placed
25 * within them. - Note, though, that the bound table contains an extra entry
26 * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
27 * that list. The bound table is used solely for lookup of sockets when packets
28 * are received and that's not necessary for SOCK_DGRAM sockets since we create
29 * a datagram handle for each and need not perform a lookup. Keeping SOCK_DGRAM
30 * sockets out of the bound hash buckets will reduce the chance of collisions
31 * when looking for SOCK_STREAM sockets and prevents us from having to check the
32 * socket type in the hash table lookups.
33 *
34 * - Sockets created by user action will either be "client" sockets that
35 * initiate a connection or "server" sockets that listen for connections; we do
36 * not support simultaneous connects (two "client" sockets connecting).
37 *
38 * - "Server" sockets are referred to as listener sockets throughout this
39 * implementation because they are in the SS_LISTEN state. When a connection
40 * request is received (the second kind of socket mentioned above), we create a
41 * new socket and refer to it as a pending socket. These pending sockets are
42 * placed on the pending connection list of the listener socket. When future
43 * packets are received for the address the listener socket is bound to, we
44 * check if the source of the packet is from one that has an existing pending
45 * connection. If it does, we process the packet for the pending socket. When
46 * that socket reaches the connected state, it is removed from the listener
47 * socket's pending list and enqueued in the listener socket's accept queue.
48 * Callers of accept(2) will accept connected sockets from the listener socket's
49 * accept queue. If the socket cannot be accepted for some reason then it is
50 * marked rejected. Once the connection is accepted, it is owned by the user
51 * process and the responsibility for cleanup falls with that user process.
52 *
53 * - It is possible that these pending sockets will never reach the connected
54 * state; in fact, we may never receive another packet after the connection
55 * request. Because of this, we must schedule a cleanup function to run in the
56 * future, after some amount of time passes where a connection should have been
57 * established. This function ensures that the socket is off all lists so it
58 * cannot be retrieved, then drops all references to the socket so it is cleaned
59 * up (sock_put() -> sk_free() -> our sk_destruct implementation). Note this
60 * function will also cleanup rejected sockets, those that reach the connected
61 * state but leave it before they have been accepted.
62 *
63 * - Sockets created by user action will be cleaned up when the user process
64 * calls close(2), causing our release implementation to be called. Our release
65 * implementation will perform some cleanup then drop the last reference so our
66 * sk_destruct implementation is invoked. Our sk_destruct implementation will
67 * perform additional cleanup that's common for both types of sockets.
68 *
69 * - A socket's reference count is what ensures that the structure won't be
70 * freed. Each entry in a list (such as the "global" bound and connected tables
71 * and the listener socket's pending list and connected queue) ensures a
72 * reference. When we defer work until process context and pass a socket as our
73 * argument, we must ensure the reference count is increased to ensure the
74 * socket isn't freed before the function is run; the deferred function will
75 * then drop the reference.
76 */
77
78#include <linux/types.h>
79
80#define EXPORT_SYMTAB
81#include <linux/bitops.h>
82#include <linux/cred.h>
83#include <linux/init.h>
84#include <linux/io.h>
85#include <linux/kernel.h>
86#include <linux/kmod.h>
87#include <linux/list.h>
88#include <linux/miscdevice.h>
89#include <linux/module.h>
90#include <linux/mutex.h>
91#include <linux/net.h>
92#include <linux/poll.h>
93#include <linux/skbuff.h>
94#include <linux/smp.h>
95#include <linux/socket.h>
96#include <linux/stddef.h>
97#include <linux/unistd.h>
98#include <linux/wait.h>
99#include <linux/workqueue.h>
100#include <net/sock.h>
101
102#include "af_vsock.h"
103#include "vsock_version.h"
104
105static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
106static void vsock_sk_destruct(struct sock *sk);
107static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
108
109/* Protocol family. */
110static struct proto vsock_proto = {
111 .name = "AF_VSOCK",
112 .owner = THIS_MODULE,
113 .obj_size = sizeof(struct vsock_sock),
114};
115
116/* The default peer timeout indicates how long we will wait for a peer response
117 * to a control message.
118 */
119#define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
120
121#define SS_LISTEN 255
122
123static const struct vsock_transport *transport;
124static DEFINE_MUTEX(vsock_register_mutex);
125
126/**** EXPORTS ****/
127
128/* Get the ID of the local context. This is transport dependent. */
129
130int vm_sockets_get_local_cid(void)
131{
132 return transport->get_local_cid();
133}
134EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
135
136/**** UTILS ****/
137
138/* Each bound VSocket is stored in the bind hash table and each connected
139 * VSocket is stored in the connected hash table.
140 *
141 * Unbound sockets are all put on the same list attached to the end of the hash
142 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in
143 * the bucket that their local address hashes to (vsock_bound_sockets(addr)
144 * represents the list that addr hashes to).
145 *
146 * Specifically, we initialize the vsock_bind_table array to a size of
147 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
148 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
149 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function
150 * mods with VSOCK_HASH_SIZE - 1 to ensure this.
151 */
152#define VSOCK_HASH_SIZE 251
153#define MAX_PORT_RETRIES 24
154
155#define VSOCK_HASH(addr) ((addr)->svm_port % (VSOCK_HASH_SIZE - 1))
156#define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
157#define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE])
158
159/* XXX This can probably be implemented in a better way. */
160#define VSOCK_CONN_HASH(src, dst) \
161 (((src)->svm_cid ^ (dst)->svm_port) % (VSOCK_HASH_SIZE - 1))
162#define vsock_connected_sockets(src, dst) \
163 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
164#define vsock_connected_sockets_vsk(vsk) \
165 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
166
167static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
168static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
169static DEFINE_SPINLOCK(vsock_table_lock);
170
171static __init void vsock_init_tables(void)
172{
173 int i;
174
175 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
176 INIT_LIST_HEAD(&vsock_bind_table[i]);
177
178 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
179 INIT_LIST_HEAD(&vsock_connected_table[i]);
180}
181
182static void __vsock_insert_bound(struct list_head *list,
183 struct vsock_sock *vsk)
184{
185 sock_hold(&vsk->sk);
186 list_add(&vsk->bound_table, list);
187}
188
189static void __vsock_insert_connected(struct list_head *list,
190 struct vsock_sock *vsk)
191{
192 sock_hold(&vsk->sk);
193 list_add(&vsk->connected_table, list);
194}
195
196static void __vsock_remove_bound(struct vsock_sock *vsk)
197{
198 list_del_init(&vsk->bound_table);
199 sock_put(&vsk->sk);
200}
201
202static void __vsock_remove_connected(struct vsock_sock *vsk)
203{
204 list_del_init(&vsk->connected_table);
205 sock_put(&vsk->sk);
206}
207
208static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
209{
210 struct vsock_sock *vsk;
211
212 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
213 if (vsock_addr_equals_addr_any(addr, &vsk->local_addr))
214 return sk_vsock(vsk);
215
216 return NULL;
217}
218
219static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
220 struct sockaddr_vm *dst)
221{
222 struct vsock_sock *vsk;
223
224 list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
225 connected_table) {
226 if (vsock_addr_equals_addr(src, &vsk->remote_addr)
227 && vsock_addr_equals_addr(dst, &vsk->local_addr)) {
228 return sk_vsock(vsk);
229 }
230 }
231
232 return NULL;
233}
234
235static bool __vsock_in_bound_table(struct vsock_sock *vsk)
236{
237 return !list_empty(&vsk->bound_table);
238}
239
240static bool __vsock_in_connected_table(struct vsock_sock *vsk)
241{
242 return !list_empty(&vsk->connected_table);
243}
244
245static void vsock_insert_unbound(struct vsock_sock *vsk)
246{
247 spin_lock_bh(&vsock_table_lock);
248 __vsock_insert_bound(vsock_unbound_sockets, vsk);
249 spin_unlock_bh(&vsock_table_lock);
250}
251
252void vsock_insert_connected(struct vsock_sock *vsk)
253{
254 struct list_head *list = vsock_connected_sockets(
255 &vsk->remote_addr, &vsk->local_addr);
256
257 spin_lock_bh(&vsock_table_lock);
258 __vsock_insert_connected(list, vsk);
259 spin_unlock_bh(&vsock_table_lock);
260}
261EXPORT_SYMBOL_GPL(vsock_insert_connected);
262
263void vsock_remove_bound(struct vsock_sock *vsk)
264{
265 spin_lock_bh(&vsock_table_lock);
266 __vsock_remove_bound(vsk);
267 spin_unlock_bh(&vsock_table_lock);
268}
269EXPORT_SYMBOL_GPL(vsock_remove_bound);
270
271void vsock_remove_connected(struct vsock_sock *vsk)
272{
273 spin_lock_bh(&vsock_table_lock);
274 __vsock_remove_connected(vsk);
275 spin_unlock_bh(&vsock_table_lock);
276}
277EXPORT_SYMBOL_GPL(vsock_remove_connected);
278
279struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
280{
281 struct sock *sk;
282
283 spin_lock_bh(&vsock_table_lock);
284 sk = __vsock_find_bound_socket(addr);
285 if (sk)
286 sock_hold(sk);
287
288 spin_unlock_bh(&vsock_table_lock);
289
290 return sk;
291}
292EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
293
294struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
295 struct sockaddr_vm *dst)
296{
297 struct sock *sk;
298
299 spin_lock_bh(&vsock_table_lock);
300 sk = __vsock_find_connected_socket(src, dst);
301 if (sk)
302 sock_hold(sk);
303
304 spin_unlock_bh(&vsock_table_lock);
305
306 return sk;
307}
308EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
309
310static bool vsock_in_bound_table(struct vsock_sock *vsk)
311{
312 bool ret;
313
314 spin_lock_bh(&vsock_table_lock);
315 ret = __vsock_in_bound_table(vsk);
316 spin_unlock_bh(&vsock_table_lock);
317
318 return ret;
319}
320
321static bool vsock_in_connected_table(struct vsock_sock *vsk)
322{
323 bool ret;
324
325 spin_lock_bh(&vsock_table_lock);
326 ret = __vsock_in_connected_table(vsk);
327 spin_unlock_bh(&vsock_table_lock);
328
329 return ret;
330}
331
332void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
333{
334 int i;
335
336 spin_lock_bh(&vsock_table_lock);
337
338 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
339 struct vsock_sock *vsk;
340 list_for_each_entry(vsk, &vsock_connected_table[i],
341 connected_table);
342 fn(sk_vsock(vsk));
343 }
344
345 spin_unlock_bh(&vsock_table_lock);
346}
347EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
348
349void vsock_add_pending(struct sock *listener, struct sock *pending)
350{
351 struct vsock_sock *vlistener;
352 struct vsock_sock *vpending;
353
354 vlistener = vsock_sk(listener);
355 vpending = vsock_sk(pending);
356
357 sock_hold(pending);
358 sock_hold(listener);
359 list_add_tail(&vpending->pending_links, &vlistener->pending_links);
360}
361EXPORT_SYMBOL_GPL(vsock_add_pending);
362
363void vsock_remove_pending(struct sock *listener, struct sock *pending)
364{
365 struct vsock_sock *vpending = vsock_sk(pending);
366
367 list_del_init(&vpending->pending_links);
368 sock_put(listener);
369 sock_put(pending);
370}
371EXPORT_SYMBOL_GPL(vsock_remove_pending);
372
373void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
374{
375 struct vsock_sock *vlistener;
376 struct vsock_sock *vconnected;
377
378 vlistener = vsock_sk(listener);
379 vconnected = vsock_sk(connected);
380
381 sock_hold(connected);
382 sock_hold(listener);
383 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
384}
385EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
386
387static struct sock *vsock_dequeue_accept(struct sock *listener)
388{
389 struct vsock_sock *vlistener;
390 struct vsock_sock *vconnected;
391
392 vlistener = vsock_sk(listener);
393
394 if (list_empty(&vlistener->accept_queue))
395 return NULL;
396
397 vconnected = list_entry(vlistener->accept_queue.next,
398 struct vsock_sock, accept_queue);
399
400 list_del_init(&vconnected->accept_queue);
401 sock_put(listener);
402 /* The caller will need a reference on the connected socket so we let
403 * it call sock_put().
404 */
405
406 return sk_vsock(vconnected);
407}
408
409static bool vsock_is_accept_queue_empty(struct sock *sk)
410{
411 struct vsock_sock *vsk = vsock_sk(sk);
412 return list_empty(&vsk->accept_queue);
413}
414
415static bool vsock_is_pending(struct sock *sk)
416{
417 struct vsock_sock *vsk = vsock_sk(sk);
418 return !list_empty(&vsk->pending_links);
419}
420
421static int vsock_send_shutdown(struct sock *sk, int mode)
422{
423 return transport->shutdown(vsock_sk(sk), mode);
424}
425
426void vsock_pending_work(struct work_struct *work)
427{
428 struct sock *sk;
429 struct sock *listener;
430 struct vsock_sock *vsk;
431 bool cleanup;
432
433 vsk = container_of(work, struct vsock_sock, dwork.work);
434 sk = sk_vsock(vsk);
435 listener = vsk->listener;
436 cleanup = true;
437
438 lock_sock(listener);
439 lock_sock(sk);
440
441 if (vsock_is_pending(sk)) {
442 vsock_remove_pending(listener, sk);
443 } else if (!vsk->rejected) {
444 /* We are not on the pending list and accept() did not reject
445 * us, so we must have been accepted by our user process. We
446 * just need to drop our references to the sockets and be on
447 * our way.
448 */
449 cleanup = false;
450 goto out;
451 }
452
453 listener->sk_ack_backlog--;
454
455 /* We need to remove ourself from the global connected sockets list so
456 * incoming packets can't find this socket, and to reduce the reference
457 * count.
458 */
459 if (vsock_in_connected_table(vsk))
460 vsock_remove_connected(vsk);
461
462 sk->sk_state = SS_FREE;
463
464out:
465 release_sock(sk);
466 release_sock(listener);
467 if (cleanup)
468 sock_put(sk);
469
470 sock_put(sk);
471 sock_put(listener);
472}
473EXPORT_SYMBOL_GPL(vsock_pending_work);
474
475/**** SOCKET OPERATIONS ****/
476
477static int __vsock_bind_stream(struct vsock_sock *vsk,
478 struct sockaddr_vm *addr)
479{
480 static u32 port = LAST_RESERVED_PORT + 1;
481 struct sockaddr_vm new_addr;
482
483 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
484
485 if (addr->svm_port == VMADDR_PORT_ANY) {
486 bool found = false;
487 unsigned int i;
488
489 for (i = 0; i < MAX_PORT_RETRIES; i++) {
490 if (port <= LAST_RESERVED_PORT)
491 port = LAST_RESERVED_PORT + 1;
492
493 new_addr.svm_port = port++;
494
495 if (!__vsock_find_bound_socket(&new_addr)) {
496 found = true;
497 break;
498 }
499 }
500
501 if (!found)
502 return -EADDRNOTAVAIL;
503 } else {
504 /* If port is in reserved range, ensure caller
505 * has necessary privileges.
506 */
507 if (addr->svm_port <= LAST_RESERVED_PORT &&
508 !capable(CAP_NET_BIND_SERVICE)) {
509 return -EACCES;
510 }
511
512 if (__vsock_find_bound_socket(&new_addr))
513 return -EADDRINUSE;
514 }
515
516 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
517
518 /* Remove stream sockets from the unbound list and add them to the hash
519 * table for easy lookup by its address. The unbound list is simply an
520 * extra entry at the end of the hash table, a trick used by AF_UNIX.
521 */
522 __vsock_remove_bound(vsk);
523 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
524
525 return 0;
526}
527
528static int __vsock_bind_dgram(struct vsock_sock *vsk,
529 struct sockaddr_vm *addr)
530{
531 return transport->dgram_bind(vsk, addr);
532}
533
534static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
535{
536 struct vsock_sock *vsk = vsock_sk(sk);
537 u32 cid;
538 int retval;
539
540 /* First ensure this socket isn't already bound. */
541 if (vsock_addr_bound(&vsk->local_addr))
542 return -EINVAL;
543
544 /* Now bind to the provided address or select appropriate values if
545 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that
546 * like AF_INET prevents binding to a non-local IP address (in most
547 * cases), we only allow binding to the local CID.
548 */
549 cid = transport->get_local_cid();
550 if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
551 return -EADDRNOTAVAIL;
552
553 switch (sk->sk_socket->type) {
554 case SOCK_STREAM:
555 spin_lock_bh(&vsock_table_lock);
556 retval = __vsock_bind_stream(vsk, addr);
557 spin_unlock_bh(&vsock_table_lock);
558 break;
559
560 case SOCK_DGRAM:
561 retval = __vsock_bind_dgram(vsk, addr);
562 break;
563
564 default:
565 retval = -EINVAL;
566 break;
567 }
568
569 return retval;
570}
571
572struct sock *__vsock_create(struct net *net,
573 struct socket *sock,
574 struct sock *parent,
575 gfp_t priority,
576 unsigned short type)
577{
578 struct sock *sk;
579 struct vsock_sock *psk;
580 struct vsock_sock *vsk;
581
582 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto);
583 if (!sk)
584 return NULL;
585
586 sock_init_data(sock, sk);
587
588 /* sk->sk_type is normally set in sock_init_data, but only if sock is
589 * non-NULL. We make sure that our sockets always have a type by
590 * setting it here if needed.
591 */
592 if (!sock)
593 sk->sk_type = type;
594
595 vsk = vsock_sk(sk);
596 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
597 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
598
599 sk->sk_destruct = vsock_sk_destruct;
600 sk->sk_backlog_rcv = vsock_queue_rcv_skb;
601 sk->sk_state = 0;
602 sock_reset_flag(sk, SOCK_DONE);
603
604 INIT_LIST_HEAD(&vsk->bound_table);
605 INIT_LIST_HEAD(&vsk->connected_table);
606 vsk->listener = NULL;
607 INIT_LIST_HEAD(&vsk->pending_links);
608 INIT_LIST_HEAD(&vsk->accept_queue);
609 vsk->rejected = false;
610 vsk->sent_request = false;
611 vsk->ignore_connecting_rst = false;
612 vsk->peer_shutdown = 0;
613
614 psk = parent ? vsock_sk(parent) : NULL;
615 if (parent) {
616 vsk->trusted = psk->trusted;
617 vsk->owner = get_cred(psk->owner);
618 vsk->connect_timeout = psk->connect_timeout;
619 } else {
620 vsk->trusted = capable(CAP_NET_ADMIN);
621 vsk->owner = get_current_cred();
622 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
623 }
624
625 if (transport->init(vsk, psk) < 0) {
626 sk_free(sk);
627 return NULL;
628 }
629
630 if (sock)
631 vsock_insert_unbound(vsk);
632
633 return sk;
634}
635EXPORT_SYMBOL_GPL(__vsock_create);
636
637static void __vsock_release(struct sock *sk)
638{
639 if (sk) {
640 struct sk_buff *skb;
641 struct sock *pending;
642 struct vsock_sock *vsk;
643
644 vsk = vsock_sk(sk);
645 pending = NULL; /* Compiler warning. */
646
647 if (vsock_in_bound_table(vsk))
648 vsock_remove_bound(vsk);
649
650 if (vsock_in_connected_table(vsk))
651 vsock_remove_connected(vsk);
652
653 transport->release(vsk);
654
655 lock_sock(sk);
656 sock_orphan(sk);
657 sk->sk_shutdown = SHUTDOWN_MASK;
658
659 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
660 kfree_skb(skb);
661
662 /* Clean up any sockets that never were accepted. */
663 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
664 __vsock_release(pending);
665 sock_put(pending);
666 }
667
668 release_sock(sk);
669 sock_put(sk);
670 }
671}
672
673static void vsock_sk_destruct(struct sock *sk)
674{
675 struct vsock_sock *vsk = vsock_sk(sk);
676
677 transport->destruct(vsk);
678
679 /* When clearing these addresses, there's no need to set the family and
680 * possibly register the address family with the kernel.
681 */
682 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
683 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
684
685 put_cred(vsk->owner);
686}
687
688static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
689{
690 int err;
691
692 err = sock_queue_rcv_skb(sk, skb);
693 if (err)
694 kfree_skb(skb);
695
696 return err;
697}
698
699s64 vsock_stream_has_data(struct vsock_sock *vsk)
700{
701 return transport->stream_has_data(vsk);
702}
703EXPORT_SYMBOL_GPL(vsock_stream_has_data);
704
705s64 vsock_stream_has_space(struct vsock_sock *vsk)
706{
707 return transport->stream_has_space(vsk);
708}
709EXPORT_SYMBOL_GPL(vsock_stream_has_space);
710
711static int vsock_release(struct socket *sock)
712{
713 __vsock_release(sock->sk);
714 sock->sk = NULL;
715 sock->state = SS_FREE;
716
717 return 0;
718}
719
720static int
721vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
722{
723 int err;
724 struct sock *sk;
725 struct sockaddr_vm *vm_addr;
726
727 sk = sock->sk;
728
729 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
730 return -EINVAL;
731
732 lock_sock(sk);
733 err = __vsock_bind(sk, vm_addr);
734 release_sock(sk);
735
736 return err;
737}
738
739static int vsock_getname(struct socket *sock,
740 struct sockaddr *addr, int *addr_len, int peer)
741{
742 int err;
743 struct sock *sk;
744 struct vsock_sock *vsk;
745 struct sockaddr_vm *vm_addr;
746
747 sk = sock->sk;
748 vsk = vsock_sk(sk);
749 err = 0;
750
751 lock_sock(sk);
752
753 if (peer) {
754 if (sock->state != SS_CONNECTED) {
755 err = -ENOTCONN;
756 goto out;
757 }
758 vm_addr = &vsk->remote_addr;
759 } else {
760 vm_addr = &vsk->local_addr;
761 }
762
763 if (!vm_addr) {
764 err = -EINVAL;
765 goto out;
766 }
767
768 /* sys_getsockname() and sys_getpeername() pass us a
769 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately
770 * that macro is defined in socket.c instead of .h, so we hardcode its
771 * value here.
772 */
773 BUILD_BUG_ON(sizeof(*vm_addr) > 128);
774 memcpy(addr, vm_addr, sizeof(*vm_addr));
775 *addr_len = sizeof(*vm_addr);
776
777out:
778 release_sock(sk);
779 return err;
780}
781
782static int vsock_shutdown(struct socket *sock, int mode)
783{
784 int err;
785 struct sock *sk;
786
787 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
788 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
789 * here like the other address families do. Note also that the
790 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
791 * which is what we want.
792 */
793 mode++;
794
795 if ((mode & ~SHUTDOWN_MASK) || !mode)
796 return -EINVAL;
797
798 /* If this is a STREAM socket and it is not connected then bail out
799 * immediately. If it is a DGRAM socket then we must first kick the
800 * socket so that it wakes up from any sleeping calls, for example
801 * recv(), and then afterwards return the error.
802 */
803
804 sk = sock->sk;
805 if (sock->state == SS_UNCONNECTED) {
806 err = -ENOTCONN;
807 if (sk->sk_type == SOCK_STREAM)
808 return err;
809 } else {
810 sock->state = SS_DISCONNECTING;
811 err = 0;
812 }
813
814 /* Receive and send shutdowns are treated alike. */
815 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
816 if (mode) {
817 lock_sock(sk);
818 sk->sk_shutdown |= mode;
819 sk->sk_state_change(sk);
820 release_sock(sk);
821
822 if (sk->sk_type == SOCK_STREAM) {
823 sock_reset_flag(sk, SOCK_DONE);
824 vsock_send_shutdown(sk, mode);
825 }
826 }
827
828 return err;
829}
830
831static unsigned int vsock_poll(struct file *file, struct socket *sock,
832 poll_table *wait)
833{
834 struct sock *sk;
835 unsigned int mask;
836 struct vsock_sock *vsk;
837
838 sk = sock->sk;
839 vsk = vsock_sk(sk);
840
841 poll_wait(file, sk_sleep(sk), wait);
842 mask = 0;
843
844 if (sk->sk_err)
845 /* Signify that there has been an error on this socket. */
846 mask |= POLLERR;
847
848 /* INET sockets treat local write shutdown and peer write shutdown as a
849 * case of POLLHUP set.
850 */
851 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
852 ((sk->sk_shutdown & SEND_SHUTDOWN) &&
853 (vsk->peer_shutdown & SEND_SHUTDOWN))) {
854 mask |= POLLHUP;
855 }
856
857 if (sk->sk_shutdown & RCV_SHUTDOWN ||
858 vsk->peer_shutdown & SEND_SHUTDOWN) {
859 mask |= POLLRDHUP;
860 }
861
862 if (sock->type == SOCK_DGRAM) {
863 /* For datagram sockets we can read if there is something in
864 * the queue and write as long as the socket isn't shutdown for
865 * sending.
866 */
867 if (!skb_queue_empty(&sk->sk_receive_queue) ||
868 (sk->sk_shutdown & RCV_SHUTDOWN)) {
869 mask |= POLLIN | POLLRDNORM;
870 }
871
872 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
873 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
874
875 } else if (sock->type == SOCK_STREAM) {
876 lock_sock(sk);
877
878 /* Listening sockets that have connections in their accept
879 * queue can be read.
880 */
881 if (sk->sk_state == SS_LISTEN
882 && !vsock_is_accept_queue_empty(sk))
883 mask |= POLLIN | POLLRDNORM;
884
885 /* If there is something in the queue then we can read. */
886 if (transport->stream_is_active(vsk) &&
887 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
888 bool data_ready_now = false;
889 int ret = transport->notify_poll_in(
890 vsk, 1, &data_ready_now);
891 if (ret < 0) {
892 mask |= POLLERR;
893 } else {
894 if (data_ready_now)
895 mask |= POLLIN | POLLRDNORM;
896
897 }
898 }
899
900 /* Sockets whose connections have been closed, reset, or
901 * terminated should also be considered read, and we check the
902 * shutdown flag for that.
903 */
904 if (sk->sk_shutdown & RCV_SHUTDOWN ||
905 vsk->peer_shutdown & SEND_SHUTDOWN) {
906 mask |= POLLIN | POLLRDNORM;
907 }
908
909 /* Connected sockets that can produce data can be written. */
910 if (sk->sk_state == SS_CONNECTED) {
911 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
912 bool space_avail_now = false;
913 int ret = transport->notify_poll_out(
914 vsk, 1, &space_avail_now);
915 if (ret < 0) {
916 mask |= POLLERR;
917 } else {
918 if (space_avail_now)
919 /* Remove POLLWRBAND since INET
920 * sockets are not setting it.
921 */
922 mask |= POLLOUT | POLLWRNORM;
923
924 }
925 }
926 }
927
928 /* Simulate INET socket poll behaviors, which sets
929 * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
930 * but local send is not shutdown.
931 */
932 if (sk->sk_state == SS_UNCONNECTED) {
933 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
934 mask |= POLLOUT | POLLWRNORM;
935
936 }
937
938 release_sock(sk);
939 }
940
941 return mask;
942}
943
944static int vsock_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
945 struct msghdr *msg, size_t len)
946{
947 int err;
948 struct sock *sk;
949 struct vsock_sock *vsk;
950 struct sockaddr_vm *remote_addr;
951
952 if (msg->msg_flags & MSG_OOB)
953 return -EOPNOTSUPP;
954
955 /* For now, MSG_DONTWAIT is always assumed... */
956 err = 0;
957 sk = sock->sk;
958 vsk = vsock_sk(sk);
959
960 lock_sock(sk);
961
962 if (!vsock_addr_bound(&vsk->local_addr)) {
963 struct sockaddr_vm local_addr;
964
965 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
966 err = __vsock_bind(sk, &local_addr);
967 if (err != 0)
968 goto out;
969
970 }
971
972 /* If the provided message contains an address, use that. Otherwise
973 * fall back on the socket's remote handle (if it has been connected).
974 */
975 if (msg->msg_name &&
976 vsock_addr_cast(msg->msg_name, msg->msg_namelen,
977 &remote_addr) == 0) {
978 /* Ensure this address is of the right type and is a valid
979 * destination.
980 */
981
982 if (remote_addr->svm_cid == VMADDR_CID_ANY)
983 remote_addr->svm_cid = transport->get_local_cid();
984
985 if (!vsock_addr_bound(remote_addr)) {
986 err = -EINVAL;
987 goto out;
988 }
989 } else if (sock->state == SS_CONNECTED) {
990 remote_addr = &vsk->remote_addr;
991
992 if (remote_addr->svm_cid == VMADDR_CID_ANY)
993 remote_addr->svm_cid = transport->get_local_cid();
994
995 /* XXX Should connect() or this function ensure remote_addr is
996 * bound?
997 */
998 if (!vsock_addr_bound(&vsk->remote_addr)) {
999 err = -EINVAL;
1000 goto out;
1001 }
1002 } else {
1003 err = -EINVAL;
1004 goto out;
1005 }
1006
1007 if (!transport->dgram_allow(remote_addr->svm_cid,
1008 remote_addr->svm_port)) {
1009 err = -EINVAL;
1010 goto out;
1011 }
1012
1013 err = transport->dgram_enqueue(vsk, remote_addr, msg->msg_iov, len);
1014
1015out:
1016 release_sock(sk);
1017 return err;
1018}
1019
1020static int vsock_dgram_connect(struct socket *sock,
1021 struct sockaddr *addr, int addr_len, int flags)
1022{
1023 int err;
1024 struct sock *sk;
1025 struct vsock_sock *vsk;
1026 struct sockaddr_vm *remote_addr;
1027
1028 sk = sock->sk;
1029 vsk = vsock_sk(sk);
1030
1031 err = vsock_addr_cast(addr, addr_len, &remote_addr);
1032 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1033 lock_sock(sk);
1034 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1035 VMADDR_PORT_ANY);
1036 sock->state = SS_UNCONNECTED;
1037 release_sock(sk);
1038 return 0;
1039 } else if (err != 0)
1040 return -EINVAL;
1041
1042 lock_sock(sk);
1043
1044 if (!vsock_addr_bound(&vsk->local_addr)) {
1045 struct sockaddr_vm local_addr;
1046
1047 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
1048 err = __vsock_bind(sk, &local_addr);
1049 if (err != 0)
1050 goto out;
1051
1052 }
1053
1054 if (!transport->dgram_allow(remote_addr->svm_cid,
1055 remote_addr->svm_port)) {
1056 err = -EINVAL;
1057 goto out;
1058 }
1059
1060 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1061 sock->state = SS_CONNECTED;
1062
1063out:
1064 release_sock(sk);
1065 return err;
1066}
1067
1068static int vsock_dgram_recvmsg(struct kiocb *kiocb, struct socket *sock,
1069 struct msghdr *msg, size_t len, int flags)
1070{
1071 return transport->dgram_dequeue(kiocb, vsock_sk(sock->sk), msg, len,
1072 flags);
1073}
1074
1075static const struct proto_ops vsock_dgram_ops = {
1076 .family = PF_VSOCK,
1077 .owner = THIS_MODULE,
1078 .release = vsock_release,
1079 .bind = vsock_bind,
1080 .connect = vsock_dgram_connect,
1081 .socketpair = sock_no_socketpair,
1082 .accept = sock_no_accept,
1083 .getname = vsock_getname,
1084 .poll = vsock_poll,
1085 .ioctl = sock_no_ioctl,
1086 .listen = sock_no_listen,
1087 .shutdown = vsock_shutdown,
1088 .setsockopt = sock_no_setsockopt,
1089 .getsockopt = sock_no_getsockopt,
1090 .sendmsg = vsock_dgram_sendmsg,
1091 .recvmsg = vsock_dgram_recvmsg,
1092 .mmap = sock_no_mmap,
1093 .sendpage = sock_no_sendpage,
1094};
1095
1096static void vsock_connect_timeout(struct work_struct *work)
1097{
1098 struct sock *sk;
1099 struct vsock_sock *vsk;
1100
1101 vsk = container_of(work, struct vsock_sock, dwork.work);
1102 sk = sk_vsock(vsk);
1103
1104 lock_sock(sk);
1105 if (sk->sk_state == SS_CONNECTING &&
1106 (sk->sk_shutdown != SHUTDOWN_MASK)) {
1107 sk->sk_state = SS_UNCONNECTED;
1108 sk->sk_err = ETIMEDOUT;
1109 sk->sk_error_report(sk);
1110 }
1111 release_sock(sk);
1112
1113 sock_put(sk);
1114}
1115
1116static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1117 int addr_len, int flags)
1118{
1119 int err;
1120 struct sock *sk;
1121 struct vsock_sock *vsk;
1122 struct sockaddr_vm *remote_addr;
1123 long timeout;
1124 DEFINE_WAIT(wait);
1125
1126 err = 0;
1127 sk = sock->sk;
1128 vsk = vsock_sk(sk);
1129
1130 lock_sock(sk);
1131
1132 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1133 switch (sock->state) {
1134 case SS_CONNECTED:
1135 err = -EISCONN;
1136 goto out;
1137 case SS_DISCONNECTING:
1138 err = -EINVAL;
1139 goto out;
1140 case SS_CONNECTING:
1141 /* This continues on so we can move sock into the SS_CONNECTED
1142 * state once the connection has completed (at which point err
1143 * will be set to zero also). Otherwise, we will either wait
1144 * for the connection or return -EALREADY should this be a
1145 * non-blocking call.
1146 */
1147 err = -EALREADY;
1148 break;
1149 default:
1150 if ((sk->sk_state == SS_LISTEN) ||
1151 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1152 err = -EINVAL;
1153 goto out;
1154 }
1155
1156 /* The hypervisor and well-known contexts do not have socket
1157 * endpoints.
1158 */
1159 if (!transport->stream_allow(remote_addr->svm_cid,
1160 remote_addr->svm_port)) {
1161 err = -ENETUNREACH;
1162 goto out;
1163 }
1164
1165 /* Set the remote address that we are connecting to. */
1166 memcpy(&vsk->remote_addr, remote_addr,
1167 sizeof(vsk->remote_addr));
1168
1169 /* Autobind this socket to the local address if necessary. */
1170 if (!vsock_addr_bound(&vsk->local_addr)) {
1171 struct sockaddr_vm local_addr;
1172
1173 vsock_addr_init(&local_addr, VMADDR_CID_ANY,
1174 VMADDR_PORT_ANY);
1175 err = __vsock_bind(sk, &local_addr);
1176 if (err != 0)
1177 goto out;
1178
1179 }
1180
1181 sk->sk_state = SS_CONNECTING;
1182
1183 err = transport->connect(vsk);
1184 if (err < 0)
1185 goto out;
1186
1187 /* Mark sock as connecting and set the error code to in
1188 * progress in case this is a non-blocking connect.
1189 */
1190 sock->state = SS_CONNECTING;
1191 err = -EINPROGRESS;
1192 }
1193
1194 /* The receive path will handle all communication until we are able to
1195 * enter the connected state. Here we wait for the connection to be
1196 * completed or a notification of an error.
1197 */
1198 timeout = vsk->connect_timeout;
1199 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1200
1201 while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1202 if (flags & O_NONBLOCK) {
1203 /* If we're not going to block, we schedule a timeout
1204 * function to generate a timeout on the connection
1205 * attempt, in case the peer doesn't respond in a
1206 * timely manner. We hold on to the socket until the
1207 * timeout fires.
1208 */
1209 sock_hold(sk);
1210 INIT_DELAYED_WORK(&vsk->dwork,
1211 vsock_connect_timeout);
1212 schedule_delayed_work(&vsk->dwork, timeout);
1213
1214 /* Skip ahead to preserve error code set above. */
1215 goto out_wait;
1216 }
1217
1218 release_sock(sk);
1219 timeout = schedule_timeout(timeout);
1220 lock_sock(sk);
1221
1222 if (signal_pending(current)) {
1223 err = sock_intr_errno(timeout);
1224 goto out_wait_error;
1225 } else if (timeout == 0) {
1226 err = -ETIMEDOUT;
1227 goto out_wait_error;
1228 }
1229
1230 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1231 }
1232
1233 if (sk->sk_err) {
1234 err = -sk->sk_err;
1235 goto out_wait_error;
1236 } else
1237 err = 0;
1238
1239out_wait:
1240 finish_wait(sk_sleep(sk), &wait);
1241out:
1242 release_sock(sk);
1243 return err;
1244
1245out_wait_error:
1246 sk->sk_state = SS_UNCONNECTED;
1247 sock->state = SS_UNCONNECTED;
1248 goto out_wait;
1249}
1250
1251static int vsock_accept(struct socket *sock, struct socket *newsock, int flags)
1252{
1253 struct sock *listener;
1254 int err;
1255 struct sock *connected;
1256 struct vsock_sock *vconnected;
1257 long timeout;
1258 DEFINE_WAIT(wait);
1259
1260 err = 0;
1261 listener = sock->sk;
1262
1263 lock_sock(listener);
1264
1265 if (sock->type != SOCK_STREAM) {
1266 err = -EOPNOTSUPP;
1267 goto out;
1268 }
1269
1270 if (listener->sk_state != SS_LISTEN) {
1271 err = -EINVAL;
1272 goto out;
1273 }
1274
1275 /* Wait for children sockets to appear; these are the new sockets
1276 * created upon connection establishment.
1277 */
1278 timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1279 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1280
1281 while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1282 listener->sk_err == 0) {
1283 release_sock(listener);
1284 timeout = schedule_timeout(timeout);
1285 lock_sock(listener);
1286
1287 if (signal_pending(current)) {
1288 err = sock_intr_errno(timeout);
1289 goto out_wait;
1290 } else if (timeout == 0) {
1291 err = -EAGAIN;
1292 goto out_wait;
1293 }
1294
1295 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1296 }
1297
1298 if (listener->sk_err)
1299 err = -listener->sk_err;
1300
1301 if (connected) {
1302 listener->sk_ack_backlog--;
1303
1304 lock_sock(connected);
1305 vconnected = vsock_sk(connected);
1306
1307 /* If the listener socket has received an error, then we should
1308 * reject this socket and return. Note that we simply mark the
1309 * socket rejected, drop our reference, and let the cleanup
1310 * function handle the cleanup; the fact that we found it in
1311 * the listener's accept queue guarantees that the cleanup
1312 * function hasn't run yet.
1313 */
1314 if (err) {
1315 vconnected->rejected = true;
1316 release_sock(connected);
1317 sock_put(connected);
1318 goto out_wait;
1319 }
1320
1321 newsock->state = SS_CONNECTED;
1322 sock_graft(connected, newsock);
1323 release_sock(connected);
1324 sock_put(connected);
1325 }
1326
1327out_wait:
1328 finish_wait(sk_sleep(listener), &wait);
1329out:
1330 release_sock(listener);
1331 return err;
1332}
1333
1334static int vsock_listen(struct socket *sock, int backlog)
1335{
1336 int err;
1337 struct sock *sk;
1338 struct vsock_sock *vsk;
1339
1340 sk = sock->sk;
1341
1342 lock_sock(sk);
1343
1344 if (sock->type != SOCK_STREAM) {
1345 err = -EOPNOTSUPP;
1346 goto out;
1347 }
1348
1349 if (sock->state != SS_UNCONNECTED) {
1350 err = -EINVAL;
1351 goto out;
1352 }
1353
1354 vsk = vsock_sk(sk);
1355
1356 if (!vsock_addr_bound(&vsk->local_addr)) {
1357 err = -EINVAL;
1358 goto out;
1359 }
1360
1361 sk->sk_max_ack_backlog = backlog;
1362 sk->sk_state = SS_LISTEN;
1363
1364 err = 0;
1365
1366out:
1367 release_sock(sk);
1368 return err;
1369}
1370
1371static int vsock_stream_setsockopt(struct socket *sock,
1372 int level,
1373 int optname,
1374 char __user *optval,
1375 unsigned int optlen)
1376{
1377 int err;
1378 struct sock *sk;
1379 struct vsock_sock *vsk;
1380 u64 val;
1381
1382 if (level != AF_VSOCK)
1383 return -ENOPROTOOPT;
1384
1385#define COPY_IN(_v) \
1386 do { \
1387 if (optlen < sizeof(_v)) { \
1388 err = -EINVAL; \
1389 goto exit; \
1390 } \
1391 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) { \
1392 err = -EFAULT; \
1393 goto exit; \
1394 } \
1395 } while (0)
1396
1397 err = 0;
1398 sk = sock->sk;
1399 vsk = vsock_sk(sk);
1400
1401 lock_sock(sk);
1402
1403 switch (optname) {
1404 case SO_VM_SOCKETS_BUFFER_SIZE:
1405 COPY_IN(val);
1406 transport->set_buffer_size(vsk, val);
1407 break;
1408
1409 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1410 COPY_IN(val);
1411 transport->set_max_buffer_size(vsk, val);
1412 break;
1413
1414 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1415 COPY_IN(val);
1416 transport->set_min_buffer_size(vsk, val);
1417 break;
1418
1419 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1420 struct timeval tv;
1421 COPY_IN(tv);
1422 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1423 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1424 vsk->connect_timeout = tv.tv_sec * HZ +
1425 DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1426 if (vsk->connect_timeout == 0)
1427 vsk->connect_timeout =
1428 VSOCK_DEFAULT_CONNECT_TIMEOUT;
1429
1430 } else {
1431 err = -ERANGE;
1432 }
1433 break;
1434 }
1435
1436 default:
1437 err = -ENOPROTOOPT;
1438 break;
1439 }
1440
1441#undef COPY_IN
1442
1443exit:
1444 release_sock(sk);
1445 return err;
1446}
1447
1448static int vsock_stream_getsockopt(struct socket *sock,
1449 int level, int optname,
1450 char __user *optval,
1451 int __user *optlen)
1452{
1453 int err;
1454 int len;
1455 struct sock *sk;
1456 struct vsock_sock *vsk;
1457 u64 val;
1458
1459 if (level != AF_VSOCK)
1460 return -ENOPROTOOPT;
1461
1462 err = get_user(len, optlen);
1463 if (err != 0)
1464 return err;
1465
1466#define COPY_OUT(_v) \
1467 do { \
1468 if (len < sizeof(_v)) \
1469 return -EINVAL; \
1470 \
1471 len = sizeof(_v); \
1472 if (copy_to_user(optval, &_v, len) != 0) \
1473 return -EFAULT; \
1474 \
1475 } while (0)
1476
1477 err = 0;
1478 sk = sock->sk;
1479 vsk = vsock_sk(sk);
1480
1481 switch (optname) {
1482 case SO_VM_SOCKETS_BUFFER_SIZE:
1483 val = transport->get_buffer_size(vsk);
1484 COPY_OUT(val);
1485 break;
1486
1487 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1488 val = transport->get_max_buffer_size(vsk);
1489 COPY_OUT(val);
1490 break;
1491
1492 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1493 val = transport->get_min_buffer_size(vsk);
1494 COPY_OUT(val);
1495 break;
1496
1497 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1498 struct timeval tv;
1499 tv.tv_sec = vsk->connect_timeout / HZ;
1500 tv.tv_usec =
1501 (vsk->connect_timeout -
1502 tv.tv_sec * HZ) * (1000000 / HZ);
1503 COPY_OUT(tv);
1504 break;
1505 }
1506 default:
1507 return -ENOPROTOOPT;
1508 }
1509
1510 err = put_user(len, optlen);
1511 if (err != 0)
1512 return -EFAULT;
1513
1514#undef COPY_OUT
1515
1516 return 0;
1517}
1518
1519static int vsock_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1520 struct msghdr *msg, size_t len)
1521{
1522 struct sock *sk;
1523 struct vsock_sock *vsk;
1524 ssize_t total_written;
1525 long timeout;
1526 int err;
1527 struct vsock_transport_send_notify_data send_data;
1528
1529 DEFINE_WAIT(wait);
1530
1531 sk = sock->sk;
1532 vsk = vsock_sk(sk);
1533 total_written = 0;
1534 err = 0;
1535
1536 if (msg->msg_flags & MSG_OOB)
1537 return -EOPNOTSUPP;
1538
1539 lock_sock(sk);
1540
1541 /* Callers should not provide a destination with stream sockets. */
1542 if (msg->msg_namelen) {
1543 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1544 goto out;
1545 }
1546
1547 /* Send data only if both sides are not shutdown in the direction. */
1548 if (sk->sk_shutdown & SEND_SHUTDOWN ||
1549 vsk->peer_shutdown & RCV_SHUTDOWN) {
1550 err = -EPIPE;
1551 goto out;
1552 }
1553
1554 if (sk->sk_state != SS_CONNECTED ||
1555 !vsock_addr_bound(&vsk->local_addr)) {
1556 err = -ENOTCONN;
1557 goto out;
1558 }
1559
1560 if (!vsock_addr_bound(&vsk->remote_addr)) {
1561 err = -EDESTADDRREQ;
1562 goto out;
1563 }
1564
1565 /* Wait for room in the produce queue to enqueue our user's data. */
1566 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1567
1568 err = transport->notify_send_init(vsk, &send_data);
1569 if (err < 0)
1570 goto out;
1571
1572 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1573
1574 while (total_written < len) {
1575 ssize_t written;
1576
1577 while (vsock_stream_has_space(vsk) == 0 &&
1578 sk->sk_err == 0 &&
1579 !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1580 !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1581
1582 /* Don't wait for non-blocking sockets. */
1583 if (timeout == 0) {
1584 err = -EAGAIN;
1585 goto out_wait;
1586 }
1587
1588 err = transport->notify_send_pre_block(vsk, &send_data);
1589 if (err < 0)
1590 goto out_wait;
1591
1592 release_sock(sk);
1593 timeout = schedule_timeout(timeout);
1594 lock_sock(sk);
1595 if (signal_pending(current)) {
1596 err = sock_intr_errno(timeout);
1597 goto out_wait;
1598 } else if (timeout == 0) {
1599 err = -EAGAIN;
1600 goto out_wait;
1601 }
1602
1603 prepare_to_wait(sk_sleep(sk), &wait,
1604 TASK_INTERRUPTIBLE);
1605 }
1606
1607 /* These checks occur both as part of and after the loop
1608 * conditional since we need to check before and after
1609 * sleeping.
1610 */
1611 if (sk->sk_err) {
1612 err = -sk->sk_err;
1613 goto out_wait;
1614 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1615 (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1616 err = -EPIPE;
1617 goto out_wait;
1618 }
1619
1620 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1621 if (err < 0)
1622 goto out_wait;
1623
1624 /* Note that enqueue will only write as many bytes as are free
1625 * in the produce queue, so we don't need to ensure len is
1626 * smaller than the queue size. It is the caller's
1627 * responsibility to check how many bytes we were able to send.
1628 */
1629
1630 written = transport->stream_enqueue(
1631 vsk, msg->msg_iov,
1632 len - total_written);
1633 if (written < 0) {
1634 err = -ENOMEM;
1635 goto out_wait;
1636 }
1637
1638 total_written += written;
1639
1640 err = transport->notify_send_post_enqueue(
1641 vsk, written, &send_data);
1642 if (err < 0)
1643 goto out_wait;
1644
1645 }
1646
1647out_wait:
1648 if (total_written > 0)
1649 err = total_written;
1650 finish_wait(sk_sleep(sk), &wait);
1651out:
1652 release_sock(sk);
1653 return err;
1654}
1655
1656
1657static int
1658vsock_stream_recvmsg(struct kiocb *kiocb,
1659 struct socket *sock,
1660 struct msghdr *msg, size_t len, int flags)
1661{
1662 struct sock *sk;
1663 struct vsock_sock *vsk;
1664 int err;
1665 size_t target;
1666 ssize_t copied;
1667 long timeout;
1668 struct vsock_transport_recv_notify_data recv_data;
1669
1670 DEFINE_WAIT(wait);
1671
1672 sk = sock->sk;
1673 vsk = vsock_sk(sk);
1674 err = 0;
1675
1676 lock_sock(sk);
1677
1678 if (sk->sk_state != SS_CONNECTED) {
1679 /* Recvmsg is supposed to return 0 if a peer performs an
1680 * orderly shutdown. Differentiate between that case and when a
1681 * peer has not connected or a local shutdown occured with the
1682 * SOCK_DONE flag.
1683 */
1684 if (sock_flag(sk, SOCK_DONE))
1685 err = 0;
1686 else
1687 err = -ENOTCONN;
1688
1689 goto out;
1690 }
1691
1692 if (flags & MSG_OOB) {
1693 err = -EOPNOTSUPP;
1694 goto out;
1695 }
1696
1697 /* We don't check peer_shutdown flag here since peer may actually shut
1698 * down, but there can be data in the queue that a local socket can
1699 * receive.
1700 */
1701 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1702 err = 0;
1703 goto out;
1704 }
1705
1706 /* It is valid on Linux to pass in a zero-length receive buffer. This
1707 * is not an error. We may as well bail out now.
1708 */
1709 if (!len) {
1710 err = 0;
1711 goto out;
1712 }
1713
1714 /* We must not copy less than target bytes into the user's buffer
1715 * before returning successfully, so we wait for the consume queue to
1716 * have that much data to consume before dequeueing. Note that this
1717 * makes it impossible to handle cases where target is greater than the
1718 * queue size.
1719 */
1720 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1721 if (target >= transport->stream_rcvhiwat(vsk)) {
1722 err = -ENOMEM;
1723 goto out;
1724 }
1725 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1726 copied = 0;
1727
1728 err = transport->notify_recv_init(vsk, target, &recv_data);
1729 if (err < 0)
1730 goto out;
1731
1732 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1733
1734 while (1) {
1735 s64 ready = vsock_stream_has_data(vsk);
1736
1737 if (ready < 0) {
1738 /* Invalid queue pair content. XXX This should be
1739 * changed to a connection reset in a later change.
1740 */
1741
1742 err = -ENOMEM;
1743 goto out_wait;
1744 } else if (ready > 0) {
1745 ssize_t read;
1746
1747 err = transport->notify_recv_pre_dequeue(
1748 vsk, target, &recv_data);
1749 if (err < 0)
1750 break;
1751
1752 read = transport->stream_dequeue(
1753 vsk, msg->msg_iov,
1754 len - copied, flags);
1755 if (read < 0) {
1756 err = -ENOMEM;
1757 break;
1758 }
1759
1760 copied += read;
1761
1762 err = transport->notify_recv_post_dequeue(
1763 vsk, target, read,
1764 !(flags & MSG_PEEK), &recv_data);
1765 if (err < 0)
1766 goto out_wait;
1767
1768 if (read >= target || flags & MSG_PEEK)
1769 break;
1770
1771 target -= read;
1772 } else {
1773 if (sk->sk_err != 0 || (sk->sk_shutdown & RCV_SHUTDOWN)
1774 || (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1775 break;
1776 }
1777 /* Don't wait for non-blocking sockets. */
1778 if (timeout == 0) {
1779 err = -EAGAIN;
1780 break;
1781 }
1782
1783 err = transport->notify_recv_pre_block(
1784 vsk, target, &recv_data);
1785 if (err < 0)
1786 break;
1787
1788 release_sock(sk);
1789 timeout = schedule_timeout(timeout);
1790 lock_sock(sk);
1791
1792 if (signal_pending(current)) {
1793 err = sock_intr_errno(timeout);
1794 break;
1795 } else if (timeout == 0) {
1796 err = -EAGAIN;
1797 break;
1798 }
1799
1800 prepare_to_wait(sk_sleep(sk), &wait,
1801 TASK_INTERRUPTIBLE);
1802 }
1803 }
1804
1805 if (sk->sk_err)
1806 err = -sk->sk_err;
1807 else if (sk->sk_shutdown & RCV_SHUTDOWN)
1808 err = 0;
1809
1810 if (copied > 0) {
1811 /* We only do these additional bookkeeping/notification steps
1812 * if we actually copied something out of the queue pair
1813 * instead of just peeking ahead.
1814 */
1815
1816 if (!(flags & MSG_PEEK)) {
1817 /* If the other side has shutdown for sending and there
1818 * is nothing more to read, then modify the socket
1819 * state.
1820 */
1821 if (vsk->peer_shutdown & SEND_SHUTDOWN) {
1822 if (vsock_stream_has_data(vsk) <= 0) {
1823 sk->sk_state = SS_UNCONNECTED;
1824 sock_set_flag(sk, SOCK_DONE);
1825 sk->sk_state_change(sk);
1826 }
1827 }
1828 }
1829 err = copied;
1830 }
1831
1832out_wait:
1833 finish_wait(sk_sleep(sk), &wait);
1834out:
1835 release_sock(sk);
1836 return err;
1837}
1838
1839static const struct proto_ops vsock_stream_ops = {
1840 .family = PF_VSOCK,
1841 .owner = THIS_MODULE,
1842 .release = vsock_release,
1843 .bind = vsock_bind,
1844 .connect = vsock_stream_connect,
1845 .socketpair = sock_no_socketpair,
1846 .accept = vsock_accept,
1847 .getname = vsock_getname,
1848 .poll = vsock_poll,
1849 .ioctl = sock_no_ioctl,
1850 .listen = vsock_listen,
1851 .shutdown = vsock_shutdown,
1852 .setsockopt = vsock_stream_setsockopt,
1853 .getsockopt = vsock_stream_getsockopt,
1854 .sendmsg = vsock_stream_sendmsg,
1855 .recvmsg = vsock_stream_recvmsg,
1856 .mmap = sock_no_mmap,
1857 .sendpage = sock_no_sendpage,
1858};
1859
1860static int vsock_create(struct net *net, struct socket *sock,
1861 int protocol, int kern)
1862{
1863 if (!sock)
1864 return -EINVAL;
1865
1866 if (protocol)
1867 return -EPROTONOSUPPORT;
1868
1869 switch (sock->type) {
1870 case SOCK_DGRAM:
1871 sock->ops = &vsock_dgram_ops;
1872 break;
1873 case SOCK_STREAM:
1874 sock->ops = &vsock_stream_ops;
1875 break;
1876 default:
1877 return -ESOCKTNOSUPPORT;
1878 }
1879
1880 sock->state = SS_UNCONNECTED;
1881
1882 return __vsock_create(net, sock, NULL, GFP_KERNEL, 0) ? 0 : -ENOMEM;
1883}
1884
1885static const struct net_proto_family vsock_family_ops = {
1886 .family = AF_VSOCK,
1887 .create = vsock_create,
1888 .owner = THIS_MODULE,
1889};
1890
1891static long vsock_dev_do_ioctl(struct file *filp,
1892 unsigned int cmd, void __user *ptr)
1893{
1894 u32 __user *p = ptr;
1895 int retval = 0;
1896
1897 switch (cmd) {
1898 case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1899 if (put_user(transport->get_local_cid(), p) != 0)
1900 retval = -EFAULT;
1901 break;
1902
1903 default:
1904 pr_err("Unknown ioctl %d\n", cmd);
1905 retval = -EINVAL;
1906 }
1907
1908 return retval;
1909}
1910
1911static long vsock_dev_ioctl(struct file *filp,
1912 unsigned int cmd, unsigned long arg)
1913{
1914 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1915}
1916
1917#ifdef CONFIG_COMPAT
1918static long vsock_dev_compat_ioctl(struct file *filp,
1919 unsigned int cmd, unsigned long arg)
1920{
1921 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1922}
1923#endif
1924
1925static const struct file_operations vsock_device_ops = {
1926 .owner = THIS_MODULE,
1927 .unlocked_ioctl = vsock_dev_ioctl,
1928#ifdef CONFIG_COMPAT
1929 .compat_ioctl = vsock_dev_compat_ioctl,
1930#endif
1931 .open = nonseekable_open,
1932};
1933
1934static struct miscdevice vsock_device = {
1935 .name = "vsock",
1936 .minor = MISC_DYNAMIC_MINOR,
1937 .fops = &vsock_device_ops,
1938};
1939
1940static int __vsock_core_init(void)
1941{
1942 int err;
1943
1944 vsock_init_tables();
1945
1946 err = misc_register(&vsock_device);
1947 if (err) {
1948 pr_err("Failed to register misc device\n");
1949 return -ENOENT;
1950 }
1951
1952 err = proto_register(&vsock_proto, 1); /* we want our slab */
1953 if (err) {
1954 pr_err("Cannot register vsock protocol\n");
1955 goto err_misc_deregister;
1956 }
1957
1958 err = sock_register(&vsock_family_ops);
1959 if (err) {
1960 pr_err("could not register af_vsock (%d) address family: %d\n",
1961 AF_VSOCK, err);
1962 goto err_unregister_proto;
1963 }
1964
1965 return 0;
1966
1967err_unregister_proto:
1968 proto_unregister(&vsock_proto);
1969err_misc_deregister:
1970 misc_deregister(&vsock_device);
1971 return err;
1972}
1973
1974int vsock_core_init(const struct vsock_transport *t)
1975{
1976 int retval = mutex_lock_interruptible(&vsock_register_mutex);
1977 if (retval)
1978 return retval;
1979
1980 if (transport) {
1981 retval = -EBUSY;
1982 goto out;
1983 }
1984
1985 transport = t;
1986 retval = __vsock_core_init();
1987 if (retval)
1988 transport = NULL;
1989
1990out:
1991 mutex_unlock(&vsock_register_mutex);
1992 return retval;
1993}
1994EXPORT_SYMBOL_GPL(vsock_core_init);
1995
1996void vsock_core_exit(void)
1997{
1998 mutex_lock(&vsock_register_mutex);
1999
2000 misc_deregister(&vsock_device);
2001 sock_unregister(AF_VSOCK);
2002 proto_unregister(&vsock_proto);
2003
2004 /* We do not want the assignment below re-ordered. */
2005 mb();
2006 transport = NULL;
2007
2008 mutex_unlock(&vsock_register_mutex);
2009}
2010EXPORT_SYMBOL_GPL(vsock_core_exit);
2011
2012MODULE_AUTHOR("VMware, Inc.");
2013MODULE_DESCRIPTION("VMware Virtual Socket Family");
2014MODULE_VERSION(VSOCK_DRIVER_VERSION_STRING);
2015MODULE_LICENSE("GPL v2");
diff --git a/net/vmw_vsock/af_vsock.h b/net/vmw_vsock/af_vsock.h
new file mode 100644
index 000000000000..7d64d3609ec9
--- /dev/null
+++ b/net/vmw_vsock/af_vsock.h
@@ -0,0 +1,175 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#ifndef __AF_VSOCK_H__
17#define __AF_VSOCK_H__
18
19#include <linux/kernel.h>
20#include <linux/workqueue.h>
21#include <linux/vm_sockets.h>
22
23#include "vsock_addr.h"
24
25#define LAST_RESERVED_PORT 1023
26
27#define vsock_sk(__sk) ((struct vsock_sock *)__sk)
28#define sk_vsock(__vsk) (&(__vsk)->sk)
29
30struct vsock_sock {
31 /* sk must be the first member. */
32 struct sock sk;
33 struct sockaddr_vm local_addr;
34 struct sockaddr_vm remote_addr;
35 /* Links for the global tables of bound and connected sockets. */
36 struct list_head bound_table;
37 struct list_head connected_table;
38 /* Accessed without the socket lock held. This means it can never be
39 * modified outsided of socket create or destruct.
40 */
41 bool trusted;
42 bool cached_peer_allow_dgram; /* Dgram communication allowed to
43 * cached peer?
44 */
45 u32 cached_peer; /* Context ID of last dgram destination check. */
46 const struct cred *owner;
47 /* Rest are SOCK_STREAM only. */
48 long connect_timeout;
49 /* Listening socket that this came from. */
50 struct sock *listener;
51 /* Used for pending list and accept queue during connection handshake.
52 * The listening socket is the head for both lists. Sockets created
53 * for connection requests are placed in the pending list until they
54 * are connected, at which point they are put in the accept queue list
55 * so they can be accepted in accept(). If accept() cannot accept the
56 * connection, it is marked as rejected so the cleanup function knows
57 * to clean up the socket.
58 */
59 struct list_head pending_links;
60 struct list_head accept_queue;
61 bool rejected;
62 struct delayed_work dwork;
63 u32 peer_shutdown;
64 bool sent_request;
65 bool ignore_connecting_rst;
66
67 /* Private to transport. */
68 void *trans;
69};
70
71s64 vsock_stream_has_data(struct vsock_sock *vsk);
72s64 vsock_stream_has_space(struct vsock_sock *vsk);
73void vsock_pending_work(struct work_struct *work);
74struct sock *__vsock_create(struct net *net,
75 struct socket *sock,
76 struct sock *parent,
77 gfp_t priority, unsigned short type);
78
79/**** TRANSPORT ****/
80
81struct vsock_transport_recv_notify_data {
82 u64 data1; /* Transport-defined. */
83 u64 data2; /* Transport-defined. */
84 bool notify_on_block;
85};
86
87struct vsock_transport_send_notify_data {
88 u64 data1; /* Transport-defined. */
89 u64 data2; /* Transport-defined. */
90};
91
92struct vsock_transport {
93 /* Initialize/tear-down socket. */
94 int (*init)(struct vsock_sock *, struct vsock_sock *);
95 void (*destruct)(struct vsock_sock *);
96 void (*release)(struct vsock_sock *);
97
98 /* Connections. */
99 int (*connect)(struct vsock_sock *);
100
101 /* DGRAM. */
102 int (*dgram_bind)(struct vsock_sock *, struct sockaddr_vm *);
103 int (*dgram_dequeue)(struct kiocb *kiocb, struct vsock_sock *vsk,
104 struct msghdr *msg, size_t len, int flags);
105 int (*dgram_enqueue)(struct vsock_sock *, struct sockaddr_vm *,
106 struct iovec *, size_t len);
107 bool (*dgram_allow)(u32 cid, u32 port);
108
109 /* STREAM. */
110 /* TODO: stream_bind() */
111 ssize_t (*stream_dequeue)(struct vsock_sock *, struct iovec *,
112 size_t len, int flags);
113 ssize_t (*stream_enqueue)(struct vsock_sock *, struct iovec *,
114 size_t len);
115 s64 (*stream_has_data)(struct vsock_sock *);
116 s64 (*stream_has_space)(struct vsock_sock *);
117 u64 (*stream_rcvhiwat)(struct vsock_sock *);
118 bool (*stream_is_active)(struct vsock_sock *);
119 bool (*stream_allow)(u32 cid, u32 port);
120
121 /* Notification. */
122 int (*notify_poll_in)(struct vsock_sock *, size_t, bool *);
123 int (*notify_poll_out)(struct vsock_sock *, size_t, bool *);
124 int (*notify_recv_init)(struct vsock_sock *, size_t,
125 struct vsock_transport_recv_notify_data *);
126 int (*notify_recv_pre_block)(struct vsock_sock *, size_t,
127 struct vsock_transport_recv_notify_data *);
128 int (*notify_recv_pre_dequeue)(struct vsock_sock *, size_t,
129 struct vsock_transport_recv_notify_data *);
130 int (*notify_recv_post_dequeue)(struct vsock_sock *, size_t,
131 ssize_t, bool, struct vsock_transport_recv_notify_data *);
132 int (*notify_send_init)(struct vsock_sock *,
133 struct vsock_transport_send_notify_data *);
134 int (*notify_send_pre_block)(struct vsock_sock *,
135 struct vsock_transport_send_notify_data *);
136 int (*notify_send_pre_enqueue)(struct vsock_sock *,
137 struct vsock_transport_send_notify_data *);
138 int (*notify_send_post_enqueue)(struct vsock_sock *, ssize_t,
139 struct vsock_transport_send_notify_data *);
140
141 /* Shutdown. */
142 int (*shutdown)(struct vsock_sock *, int);
143
144 /* Buffer sizes. */
145 void (*set_buffer_size)(struct vsock_sock *, u64);
146 void (*set_min_buffer_size)(struct vsock_sock *, u64);
147 void (*set_max_buffer_size)(struct vsock_sock *, u64);
148 u64 (*get_buffer_size)(struct vsock_sock *);
149 u64 (*get_min_buffer_size)(struct vsock_sock *);
150 u64 (*get_max_buffer_size)(struct vsock_sock *);
151
152 /* Addressing. */
153 u32 (*get_local_cid)(void);
154};
155
156/**** CORE ****/
157
158int vsock_core_init(const struct vsock_transport *t);
159void vsock_core_exit(void);
160
161/**** UTILS ****/
162
163void vsock_release_pending(struct sock *pending);
164void vsock_add_pending(struct sock *listener, struct sock *pending);
165void vsock_remove_pending(struct sock *listener, struct sock *pending);
166void vsock_enqueue_accept(struct sock *listener, struct sock *connected);
167void vsock_insert_connected(struct vsock_sock *vsk);
168void vsock_remove_bound(struct vsock_sock *vsk);
169void vsock_remove_connected(struct vsock_sock *vsk);
170struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr);
171struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
172 struct sockaddr_vm *dst);
173void vsock_for_each_connected_socket(void (*fn)(struct sock *sk));
174
175#endif /* __AF_VSOCK_H__ */
diff --git a/net/vmw_vsock/vmci_transport.c b/net/vmw_vsock/vmci_transport.c
new file mode 100644
index 000000000000..e8a87cf37072
--- /dev/null
+++ b/net/vmw_vsock/vmci_transport.c
@@ -0,0 +1,2157 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#include <linux/types.h>
17
18#define EXPORT_SYMTAB
19#include <linux/bitops.h>
20#include <linux/cred.h>
21#include <linux/init.h>
22#include <linux/io.h>
23#include <linux/kernel.h>
24#include <linux/kmod.h>
25#include <linux/list.h>
26#include <linux/miscdevice.h>
27#include <linux/module.h>
28#include <linux/mutex.h>
29#include <linux/net.h>
30#include <linux/poll.h>
31#include <linux/skbuff.h>
32#include <linux/smp.h>
33#include <linux/socket.h>
34#include <linux/stddef.h>
35#include <linux/unistd.h>
36#include <linux/wait.h>
37#include <linux/workqueue.h>
38#include <net/sock.h>
39
40#include "af_vsock.h"
41#include "vmci_transport_notify.h"
42
43static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
44static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
45static void vmci_transport_peer_attach_cb(u32 sub_id,
46 const struct vmci_event_data *ed,
47 void *client_data);
48static void vmci_transport_peer_detach_cb(u32 sub_id,
49 const struct vmci_event_data *ed,
50 void *client_data);
51static void vmci_transport_recv_pkt_work(struct work_struct *work);
52static int vmci_transport_recv_listen(struct sock *sk,
53 struct vmci_transport_packet *pkt);
54static int vmci_transport_recv_connecting_server(
55 struct sock *sk,
56 struct sock *pending,
57 struct vmci_transport_packet *pkt);
58static int vmci_transport_recv_connecting_client(
59 struct sock *sk,
60 struct vmci_transport_packet *pkt);
61static int vmci_transport_recv_connecting_client_negotiate(
62 struct sock *sk,
63 struct vmci_transport_packet *pkt);
64static int vmci_transport_recv_connecting_client_invalid(
65 struct sock *sk,
66 struct vmci_transport_packet *pkt);
67static int vmci_transport_recv_connected(struct sock *sk,
68 struct vmci_transport_packet *pkt);
69static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
70static u16 vmci_transport_new_proto_supported_versions(void);
71static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
72 bool old_pkt_proto);
73
74struct vmci_transport_recv_pkt_info {
75 struct work_struct work;
76 struct sock *sk;
77 struct vmci_transport_packet pkt;
78};
79
80static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
81 VMCI_INVALID_ID };
82static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
83
84static int PROTOCOL_OVERRIDE = -1;
85
86#define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN 128
87#define VMCI_TRANSPORT_DEFAULT_QP_SIZE 262144
88#define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX 262144
89
90/* The default peer timeout indicates how long we will wait for a peer response
91 * to a control message.
92 */
93#define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
94
95#define SS_LISTEN 255
96
97/* Helper function to convert from a VMCI error code to a VSock error code. */
98
99static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
100{
101 int err;
102
103 switch (vmci_error) {
104 case VMCI_ERROR_NO_MEM:
105 err = ENOMEM;
106 break;
107 case VMCI_ERROR_DUPLICATE_ENTRY:
108 case VMCI_ERROR_ALREADY_EXISTS:
109 err = EADDRINUSE;
110 break;
111 case VMCI_ERROR_NO_ACCESS:
112 err = EPERM;
113 break;
114 case VMCI_ERROR_NO_RESOURCES:
115 err = ENOBUFS;
116 break;
117 case VMCI_ERROR_INVALID_RESOURCE:
118 err = EHOSTUNREACH;
119 break;
120 case VMCI_ERROR_INVALID_ARGS:
121 default:
122 err = EINVAL;
123 }
124
125 return err > 0 ? -err : err;
126}
127
128static inline void
129vmci_transport_packet_init(struct vmci_transport_packet *pkt,
130 struct sockaddr_vm *src,
131 struct sockaddr_vm *dst,
132 u8 type,
133 u64 size,
134 u64 mode,
135 struct vmci_transport_waiting_info *wait,
136 u16 proto,
137 struct vmci_handle handle)
138{
139 /* We register the stream control handler as an any cid handle so we
140 * must always send from a source address of VMADDR_CID_ANY
141 */
142 pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
143 VMCI_TRANSPORT_PACKET_RID);
144 pkt->dg.dst = vmci_make_handle(dst->svm_cid,
145 VMCI_TRANSPORT_PACKET_RID);
146 pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
147 pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
148 pkt->type = type;
149 pkt->src_port = src->svm_port;
150 pkt->dst_port = dst->svm_port;
151 memset(&pkt->proto, 0, sizeof(pkt->proto));
152 memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
153
154 switch (pkt->type) {
155 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
156 pkt->u.size = 0;
157 break;
158
159 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
160 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
161 pkt->u.size = size;
162 break;
163
164 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
165 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
166 pkt->u.handle = handle;
167 break;
168
169 case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
170 case VMCI_TRANSPORT_PACKET_TYPE_READ:
171 case VMCI_TRANSPORT_PACKET_TYPE_RST:
172 pkt->u.size = 0;
173 break;
174
175 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
176 pkt->u.mode = mode;
177 break;
178
179 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
180 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
181 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
182 break;
183
184 case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
185 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
186 pkt->u.size = size;
187 pkt->proto = proto;
188 break;
189 }
190}
191
192static inline void
193vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
194 struct sockaddr_vm *local,
195 struct sockaddr_vm *remote)
196{
197 vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
198 vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
199}
200
201static int
202__vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
203 struct sockaddr_vm *src,
204 struct sockaddr_vm *dst,
205 enum vmci_transport_packet_type type,
206 u64 size,
207 u64 mode,
208 struct vmci_transport_waiting_info *wait,
209 u16 proto,
210 struct vmci_handle handle,
211 bool convert_error)
212{
213 int err;
214
215 vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
216 proto, handle);
217 err = vmci_datagram_send(&pkt->dg);
218 if (convert_error && (err < 0))
219 return vmci_transport_error_to_vsock_error(err);
220
221 return err;
222}
223
224static int
225vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
226 enum vmci_transport_packet_type type,
227 u64 size,
228 u64 mode,
229 struct vmci_transport_waiting_info *wait,
230 struct vmci_handle handle)
231{
232 struct vmci_transport_packet reply;
233 struct sockaddr_vm src, dst;
234
235 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
236 return 0;
237 } else {
238 vmci_transport_packet_get_addresses(pkt, &src, &dst);
239 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
240 type,
241 size, mode, wait,
242 VSOCK_PROTO_INVALID,
243 handle, true);
244 }
245}
246
247static int
248vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
249 struct sockaddr_vm *dst,
250 enum vmci_transport_packet_type type,
251 u64 size,
252 u64 mode,
253 struct vmci_transport_waiting_info *wait,
254 struct vmci_handle handle)
255{
256 /* Note that it is safe to use a single packet across all CPUs since
257 * two tasklets of the same type are guaranteed to not ever run
258 * simultaneously. If that ever changes, or VMCI stops using tasklets,
259 * we can use per-cpu packets.
260 */
261 static struct vmci_transport_packet pkt;
262
263 return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
264 size, mode, wait,
265 VSOCK_PROTO_INVALID, handle,
266 false);
267}
268
269static int
270vmci_transport_send_control_pkt(struct sock *sk,
271 enum vmci_transport_packet_type type,
272 u64 size,
273 u64 mode,
274 struct vmci_transport_waiting_info *wait,
275 u16 proto,
276 struct vmci_handle handle)
277{
278 struct vmci_transport_packet *pkt;
279 struct vsock_sock *vsk;
280 int err;
281
282 vsk = vsock_sk(sk);
283
284 if (!vsock_addr_bound(&vsk->local_addr))
285 return -EINVAL;
286
287 if (!vsock_addr_bound(&vsk->remote_addr))
288 return -EINVAL;
289
290 pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
291 if (!pkt)
292 return -ENOMEM;
293
294 err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
295 &vsk->remote_addr, type, size,
296 mode, wait, proto, handle,
297 true);
298 kfree(pkt);
299
300 return err;
301}
302
303static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
304 struct sockaddr_vm *src,
305 struct vmci_transport_packet *pkt)
306{
307 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
308 return 0;
309 return vmci_transport_send_control_pkt_bh(
310 dst, src,
311 VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
312 0, NULL, VMCI_INVALID_HANDLE);
313}
314
315static int vmci_transport_send_reset(struct sock *sk,
316 struct vmci_transport_packet *pkt)
317{
318 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
319 return 0;
320 return vmci_transport_send_control_pkt(sk,
321 VMCI_TRANSPORT_PACKET_TYPE_RST,
322 0, 0, NULL, VSOCK_PROTO_INVALID,
323 VMCI_INVALID_HANDLE);
324}
325
326static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
327{
328 return vmci_transport_send_control_pkt(
329 sk,
330 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
331 size, 0, NULL,
332 VSOCK_PROTO_INVALID,
333 VMCI_INVALID_HANDLE);
334}
335
336static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
337 u16 version)
338{
339 return vmci_transport_send_control_pkt(
340 sk,
341 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
342 size, 0, NULL, version,
343 VMCI_INVALID_HANDLE);
344}
345
346static int vmci_transport_send_qp_offer(struct sock *sk,
347 struct vmci_handle handle)
348{
349 return vmci_transport_send_control_pkt(
350 sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
351 0, NULL,
352 VSOCK_PROTO_INVALID, handle);
353}
354
355static int vmci_transport_send_attach(struct sock *sk,
356 struct vmci_handle handle)
357{
358 return vmci_transport_send_control_pkt(
359 sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
360 0, 0, NULL, VSOCK_PROTO_INVALID,
361 handle);
362}
363
364static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
365{
366 return vmci_transport_reply_control_pkt_fast(
367 pkt,
368 VMCI_TRANSPORT_PACKET_TYPE_RST,
369 0, 0, NULL,
370 VMCI_INVALID_HANDLE);
371}
372
373static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
374 struct sockaddr_vm *src)
375{
376 return vmci_transport_send_control_pkt_bh(
377 dst, src,
378 VMCI_TRANSPORT_PACKET_TYPE_INVALID,
379 0, 0, NULL, VMCI_INVALID_HANDLE);
380}
381
382int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
383 struct sockaddr_vm *src)
384{
385 return vmci_transport_send_control_pkt_bh(
386 dst, src,
387 VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
388 0, NULL, VMCI_INVALID_HANDLE);
389}
390
391int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
392 struct sockaddr_vm *src)
393{
394 return vmci_transport_send_control_pkt_bh(
395 dst, src,
396 VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
397 0, NULL, VMCI_INVALID_HANDLE);
398}
399
400int vmci_transport_send_wrote(struct sock *sk)
401{
402 return vmci_transport_send_control_pkt(
403 sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
404 0, NULL, VSOCK_PROTO_INVALID,
405 VMCI_INVALID_HANDLE);
406}
407
408int vmci_transport_send_read(struct sock *sk)
409{
410 return vmci_transport_send_control_pkt(
411 sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
412 0, NULL, VSOCK_PROTO_INVALID,
413 VMCI_INVALID_HANDLE);
414}
415
416int vmci_transport_send_waiting_write(struct sock *sk,
417 struct vmci_transport_waiting_info *wait)
418{
419 return vmci_transport_send_control_pkt(
420 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
421 0, 0, wait, VSOCK_PROTO_INVALID,
422 VMCI_INVALID_HANDLE);
423}
424
425int vmci_transport_send_waiting_read(struct sock *sk,
426 struct vmci_transport_waiting_info *wait)
427{
428 return vmci_transport_send_control_pkt(
429 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
430 0, 0, wait, VSOCK_PROTO_INVALID,
431 VMCI_INVALID_HANDLE);
432}
433
434static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
435{
436 return vmci_transport_send_control_pkt(
437 &vsk->sk,
438 VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
439 0, mode, NULL,
440 VSOCK_PROTO_INVALID,
441 VMCI_INVALID_HANDLE);
442}
443
444static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
445{
446 return vmci_transport_send_control_pkt(sk,
447 VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
448 size, 0, NULL,
449 VSOCK_PROTO_INVALID,
450 VMCI_INVALID_HANDLE);
451}
452
453static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
454 u16 version)
455{
456 return vmci_transport_send_control_pkt(
457 sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
458 size, 0, NULL, version,
459 VMCI_INVALID_HANDLE);
460}
461
462static struct sock *vmci_transport_get_pending(
463 struct sock *listener,
464 struct vmci_transport_packet *pkt)
465{
466 struct vsock_sock *vlistener;
467 struct vsock_sock *vpending;
468 struct sock *pending;
469
470 vlistener = vsock_sk(listener);
471
472 list_for_each_entry(vpending, &vlistener->pending_links,
473 pending_links) {
474 struct sockaddr_vm src;
475 struct sockaddr_vm dst;
476
477 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
478 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
479
480 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
481 vsock_addr_equals_addr(&dst, &vpending->local_addr)) {
482 pending = sk_vsock(vpending);
483 sock_hold(pending);
484 goto found;
485 }
486 }
487
488 pending = NULL;
489found:
490 return pending;
491
492}
493
494static void vmci_transport_release_pending(struct sock *pending)
495{
496 sock_put(pending);
497}
498
499/* We allow two kinds of sockets to communicate with a restricted VM: 1)
500 * trusted sockets 2) sockets from applications running as the same user as the
501 * VM (this is only true for the host side and only when using hosted products)
502 */
503
504static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
505{
506 return vsock->trusted ||
507 vmci_is_context_owner(peer_cid, vsock->owner->uid);
508}
509
510/* We allow sending datagrams to and receiving datagrams from a restricted VM
511 * only if it is trusted as described in vmci_transport_is_trusted.
512 */
513
514static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
515{
516 if (vsock->cached_peer != peer_cid) {
517 vsock->cached_peer = peer_cid;
518 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
519 (vmci_context_get_priv_flags(peer_cid) &
520 VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
521 vsock->cached_peer_allow_dgram = false;
522 } else {
523 vsock->cached_peer_allow_dgram = true;
524 }
525 }
526
527 return vsock->cached_peer_allow_dgram;
528}
529
530static int
531vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
532 struct vmci_handle *handle,
533 u64 produce_size,
534 u64 consume_size,
535 u32 peer, u32 flags, bool trusted)
536{
537 int err = 0;
538
539 if (trusted) {
540 /* Try to allocate our queue pair as trusted. This will only
541 * work if vsock is running in the host.
542 */
543
544 err = vmci_qpair_alloc(qpair, handle, produce_size,
545 consume_size,
546 peer, flags,
547 VMCI_PRIVILEGE_FLAG_TRUSTED);
548 if (err != VMCI_ERROR_NO_ACCESS)
549 goto out;
550
551 }
552
553 err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
554 peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
555out:
556 if (err < 0) {
557 pr_err("Could not attach to queue pair with %d\n",
558 err);
559 err = vmci_transport_error_to_vsock_error(err);
560 }
561
562 return err;
563}
564
565static int
566vmci_transport_datagram_create_hnd(u32 resource_id,
567 u32 flags,
568 vmci_datagram_recv_cb recv_cb,
569 void *client_data,
570 struct vmci_handle *out_handle)
571{
572 int err = 0;
573
574 /* Try to allocate our datagram handler as trusted. This will only work
575 * if vsock is running in the host.
576 */
577
578 err = vmci_datagram_create_handle_priv(resource_id, flags,
579 VMCI_PRIVILEGE_FLAG_TRUSTED,
580 recv_cb,
581 client_data, out_handle);
582
583 if (err == VMCI_ERROR_NO_ACCESS)
584 err = vmci_datagram_create_handle(resource_id, flags,
585 recv_cb, client_data,
586 out_handle);
587
588 return err;
589}
590
591/* This is invoked as part of a tasklet that's scheduled when the VMCI
592 * interrupt fires. This is run in bottom-half context and if it ever needs to
593 * sleep it should defer that work to a work queue.
594 */
595
596static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
597{
598 struct sock *sk;
599 size_t size;
600 struct sk_buff *skb;
601 struct vsock_sock *vsk;
602
603 sk = (struct sock *)data;
604
605 /* This handler is privileged when this module is running on the host.
606 * We will get datagrams from all endpoints (even VMs that are in a
607 * restricted context). If we get one from a restricted context then
608 * the destination socket must be trusted.
609 *
610 * NOTE: We access the socket struct without holding the lock here.
611 * This is ok because the field we are interested is never modified
612 * outside of the create and destruct socket functions.
613 */
614 vsk = vsock_sk(sk);
615 if (!vmci_transport_allow_dgram(vsk, dg->src.context))
616 return VMCI_ERROR_NO_ACCESS;
617
618 size = VMCI_DG_SIZE(dg);
619
620 /* Attach the packet to the socket's receive queue as an sk_buff. */
621 skb = alloc_skb(size, GFP_ATOMIC);
622 if (skb) {
623 /* sk_receive_skb() will do a sock_put(), so hold here. */
624 sock_hold(sk);
625 skb_put(skb, size);
626 memcpy(skb->data, dg, size);
627 sk_receive_skb(sk, skb, 0);
628 }
629
630 return VMCI_SUCCESS;
631}
632
633static bool vmci_transport_stream_allow(u32 cid, u32 port)
634{
635 static const u32 non_socket_contexts[] = {
636 VMADDR_CID_HYPERVISOR,
637 VMADDR_CID_RESERVED,
638 };
639 int i;
640
641 BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
642
643 for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
644 if (cid == non_socket_contexts[i])
645 return false;
646 }
647
648 return true;
649}
650
651/* This is invoked as part of a tasklet that's scheduled when the VMCI
652 * interrupt fires. This is run in bottom-half context but it defers most of
653 * its work to the packet handling work queue.
654 */
655
656static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
657{
658 struct sock *sk;
659 struct sockaddr_vm dst;
660 struct sockaddr_vm src;
661 struct vmci_transport_packet *pkt;
662 struct vsock_sock *vsk;
663 bool bh_process_pkt;
664 int err;
665
666 sk = NULL;
667 err = VMCI_SUCCESS;
668 bh_process_pkt = false;
669
670 /* Ignore incoming packets from contexts without sockets, or resources
671 * that aren't vsock implementations.
672 */
673
674 if (!vmci_transport_stream_allow(dg->src.context, -1)
675 || VMCI_TRANSPORT_PACKET_RID != dg->src.resource)
676 return VMCI_ERROR_NO_ACCESS;
677
678 if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
679 /* Drop datagrams that do not contain full VSock packets. */
680 return VMCI_ERROR_INVALID_ARGS;
681
682 pkt = (struct vmci_transport_packet *)dg;
683
684 /* Find the socket that should handle this packet. First we look for a
685 * connected socket and if there is none we look for a socket bound to
686 * the destintation address.
687 */
688 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
689 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
690
691 sk = vsock_find_connected_socket(&src, &dst);
692 if (!sk) {
693 sk = vsock_find_bound_socket(&dst);
694 if (!sk) {
695 /* We could not find a socket for this specified
696 * address. If this packet is a RST, we just drop it.
697 * If it is another packet, we send a RST. Note that
698 * we do not send a RST reply to RSTs so that we do not
699 * continually send RSTs between two endpoints.
700 *
701 * Note that since this is a reply, dst is src and src
702 * is dst.
703 */
704 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
705 pr_err("unable to send reset\n");
706
707 err = VMCI_ERROR_NOT_FOUND;
708 goto out;
709 }
710 }
711
712 /* If the received packet type is beyond all types known to this
713 * implementation, reply with an invalid message. Hopefully this will
714 * help when implementing backwards compatibility in the future.
715 */
716 if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
717 vmci_transport_send_invalid_bh(&dst, &src);
718 err = VMCI_ERROR_INVALID_ARGS;
719 goto out;
720 }
721
722 /* This handler is privileged when this module is running on the host.
723 * We will get datagram connect requests from all endpoints (even VMs
724 * that are in a restricted context). If we get one from a restricted
725 * context then the destination socket must be trusted.
726 *
727 * NOTE: We access the socket struct without holding the lock here.
728 * This is ok because the field we are interested is never modified
729 * outside of the create and destruct socket functions.
730 */
731 vsk = vsock_sk(sk);
732 if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
733 err = VMCI_ERROR_NO_ACCESS;
734 goto out;
735 }
736
737 /* We do most everything in a work queue, but let's fast path the
738 * notification of reads and writes to help data transfer performance.
739 * We can only do this if there is no process context code executing
740 * for this socket since that may change the state.
741 */
742 bh_lock_sock(sk);
743
744 if (!sock_owned_by_user(sk) && sk->sk_state == SS_CONNECTED)
745 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
746 sk, pkt, true, &dst, &src,
747 &bh_process_pkt);
748
749 bh_unlock_sock(sk);
750
751 if (!bh_process_pkt) {
752 struct vmci_transport_recv_pkt_info *recv_pkt_info;
753
754 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
755 if (!recv_pkt_info) {
756 if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
757 pr_err("unable to send reset\n");
758
759 err = VMCI_ERROR_NO_MEM;
760 goto out;
761 }
762
763 recv_pkt_info->sk = sk;
764 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
765 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
766
767 schedule_work(&recv_pkt_info->work);
768 /* Clear sk so that the reference count incremented by one of
769 * the Find functions above is not decremented below. We need
770 * that reference count for the packet handler we've scheduled
771 * to run.
772 */
773 sk = NULL;
774 }
775
776out:
777 if (sk)
778 sock_put(sk);
779
780 return err;
781}
782
783static void vmci_transport_peer_attach_cb(u32 sub_id,
784 const struct vmci_event_data *e_data,
785 void *client_data)
786{
787 struct sock *sk = client_data;
788 const struct vmci_event_payload_qp *e_payload;
789 struct vsock_sock *vsk;
790
791 e_payload = vmci_event_data_const_payload(e_data);
792
793 vsk = vsock_sk(sk);
794
795 /* We don't ask for delayed CBs when we subscribe to this event (we
796 * pass 0 as flags to vmci_event_subscribe()). VMCI makes no
797 * guarantees in that case about what context we might be running in,
798 * so it could be BH or process, blockable or non-blockable. So we
799 * need to account for all possible contexts here.
800 */
801 local_bh_disable();
802 bh_lock_sock(sk);
803
804 /* XXX This is lame, we should provide a way to lookup sockets by
805 * qp_handle.
806 */
807 if (vmci_handle_is_equal(vmci_trans(vsk)->qp_handle,
808 e_payload->handle)) {
809 /* XXX This doesn't do anything, but in the future we may want
810 * to set a flag here to verify the attach really did occur and
811 * we weren't just sent a datagram claiming it was.
812 */
813 goto out;
814 }
815
816out:
817 bh_unlock_sock(sk);
818 local_bh_enable();
819}
820
821static void vmci_transport_handle_detach(struct sock *sk)
822{
823 struct vsock_sock *vsk;
824
825 vsk = vsock_sk(sk);
826 if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
827 sock_set_flag(sk, SOCK_DONE);
828
829 /* On a detach the peer will not be sending or receiving
830 * anymore.
831 */
832 vsk->peer_shutdown = SHUTDOWN_MASK;
833
834 /* We should not be sending anymore since the peer won't be
835 * there to receive, but we can still receive if there is data
836 * left in our consume queue.
837 */
838 if (vsock_stream_has_data(vsk) <= 0) {
839 if (sk->sk_state == SS_CONNECTING) {
840 /* The peer may detach from a queue pair while
841 * we are still in the connecting state, i.e.,
842 * if the peer VM is killed after attaching to
843 * a queue pair, but before we complete the
844 * handshake. In that case, we treat the detach
845 * event like a reset.
846 */
847
848 sk->sk_state = SS_UNCONNECTED;
849 sk->sk_err = ECONNRESET;
850 sk->sk_error_report(sk);
851 return;
852 }
853 sk->sk_state = SS_UNCONNECTED;
854 }
855 sk->sk_state_change(sk);
856 }
857}
858
859static void vmci_transport_peer_detach_cb(u32 sub_id,
860 const struct vmci_event_data *e_data,
861 void *client_data)
862{
863 struct sock *sk = client_data;
864 const struct vmci_event_payload_qp *e_payload;
865 struct vsock_sock *vsk;
866
867 e_payload = vmci_event_data_const_payload(e_data);
868 vsk = vsock_sk(sk);
869 if (vmci_handle_is_invalid(e_payload->handle))
870 return;
871
872 /* Same rules for locking as for peer_attach_cb(). */
873 local_bh_disable();
874 bh_lock_sock(sk);
875
876 /* XXX This is lame, we should provide a way to lookup sockets by
877 * qp_handle.
878 */
879 if (vmci_handle_is_equal(vmci_trans(vsk)->qp_handle,
880 e_payload->handle))
881 vmci_transport_handle_detach(sk);
882
883 bh_unlock_sock(sk);
884 local_bh_enable();
885}
886
887static void vmci_transport_qp_resumed_cb(u32 sub_id,
888 const struct vmci_event_data *e_data,
889 void *client_data)
890{
891 vsock_for_each_connected_socket(vmci_transport_handle_detach);
892}
893
894static void vmci_transport_recv_pkt_work(struct work_struct *work)
895{
896 struct vmci_transport_recv_pkt_info *recv_pkt_info;
897 struct vmci_transport_packet *pkt;
898 struct sock *sk;
899
900 recv_pkt_info =
901 container_of(work, struct vmci_transport_recv_pkt_info, work);
902 sk = recv_pkt_info->sk;
903 pkt = &recv_pkt_info->pkt;
904
905 lock_sock(sk);
906
907 switch (sk->sk_state) {
908 case SS_LISTEN:
909 vmci_transport_recv_listen(sk, pkt);
910 break;
911 case SS_CONNECTING:
912 /* Processing of pending connections for servers goes through
913 * the listening socket, so see vmci_transport_recv_listen()
914 * for that path.
915 */
916 vmci_transport_recv_connecting_client(sk, pkt);
917 break;
918 case SS_CONNECTED:
919 vmci_transport_recv_connected(sk, pkt);
920 break;
921 default:
922 /* Because this function does not run in the same context as
923 * vmci_transport_recv_stream_cb it is possible that the
924 * socket has closed. We need to let the other side know or it
925 * could be sitting in a connect and hang forever. Send a
926 * reset to prevent that.
927 */
928 vmci_transport_send_reset(sk, pkt);
929 goto out;
930 }
931
932out:
933 release_sock(sk);
934 kfree(recv_pkt_info);
935 /* Release reference obtained in the stream callback when we fetched
936 * this socket out of the bound or connected list.
937 */
938 sock_put(sk);
939}
940
941static int vmci_transport_recv_listen(struct sock *sk,
942 struct vmci_transport_packet *pkt)
943{
944 struct sock *pending;
945 struct vsock_sock *vpending;
946 int err;
947 u64 qp_size;
948 bool old_request = false;
949 bool old_pkt_proto = false;
950
951 err = 0;
952
953 /* Because we are in the listen state, we could be receiving a packet
954 * for ourself or any previous connection requests that we received.
955 * If it's the latter, we try to find a socket in our list of pending
956 * connections and, if we do, call the appropriate handler for the
957 * state that that socket is in. Otherwise we try to service the
958 * connection request.
959 */
960 pending = vmci_transport_get_pending(sk, pkt);
961 if (pending) {
962 lock_sock(pending);
963 switch (pending->sk_state) {
964 case SS_CONNECTING:
965 err = vmci_transport_recv_connecting_server(sk,
966 pending,
967 pkt);
968 break;
969 default:
970 vmci_transport_send_reset(pending, pkt);
971 err = -EINVAL;
972 }
973
974 if (err < 0)
975 vsock_remove_pending(sk, pending);
976
977 release_sock(pending);
978 vmci_transport_release_pending(pending);
979
980 return err;
981 }
982
983 /* The listen state only accepts connection requests. Reply with a
984 * reset unless we received a reset.
985 */
986
987 if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
988 pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
989 vmci_transport_reply_reset(pkt);
990 return -EINVAL;
991 }
992
993 if (pkt->u.size == 0) {
994 vmci_transport_reply_reset(pkt);
995 return -EINVAL;
996 }
997
998 /* If this socket can't accommodate this connection request, we send a
999 * reset. Otherwise we create and initialize a child socket and reply
1000 * with a connection negotiation.
1001 */
1002 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1003 vmci_transport_reply_reset(pkt);
1004 return -ECONNREFUSED;
1005 }
1006
1007 pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1008 sk->sk_type);
1009 if (!pending) {
1010 vmci_transport_send_reset(sk, pkt);
1011 return -ENOMEM;
1012 }
1013
1014 vpending = vsock_sk(pending);
1015
1016 vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1017 pkt->dst_port);
1018 vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1019 pkt->src_port);
1020
1021 /* If the proposed size fits within our min/max, accept it. Otherwise
1022 * propose our own size.
1023 */
1024 if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1025 pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1026 qp_size = pkt->u.size;
1027 } else {
1028 qp_size = vmci_trans(vpending)->queue_pair_size;
1029 }
1030
1031 /* Figure out if we are using old or new requests based on the
1032 * overrides pkt types sent by our peer.
1033 */
1034 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1035 old_request = old_pkt_proto;
1036 } else {
1037 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1038 old_request = true;
1039 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1040 old_request = false;
1041
1042 }
1043
1044 if (old_request) {
1045 /* Handle a REQUEST (or override) */
1046 u16 version = VSOCK_PROTO_INVALID;
1047 if (vmci_transport_proto_to_notify_struct(
1048 pending, &version, true))
1049 err = vmci_transport_send_negotiate(pending, qp_size);
1050 else
1051 err = -EINVAL;
1052
1053 } else {
1054 /* Handle a REQUEST2 (or override) */
1055 int proto_int = pkt->proto;
1056 int pos;
1057 u16 active_proto_version = 0;
1058
1059 /* The list of possible protocols is the intersection of all
1060 * protocols the client supports ... plus all the protocols we
1061 * support.
1062 */
1063 proto_int &= vmci_transport_new_proto_supported_versions();
1064
1065 /* We choose the highest possible protocol version and use that
1066 * one.
1067 */
1068 pos = fls(proto_int);
1069 if (pos) {
1070 active_proto_version = (1 << (pos - 1));
1071 if (vmci_transport_proto_to_notify_struct(
1072 pending, &active_proto_version, false))
1073 err = vmci_transport_send_negotiate2(pending,
1074 qp_size,
1075 active_proto_version);
1076 else
1077 err = -EINVAL;
1078
1079 } else {
1080 err = -EINVAL;
1081 }
1082 }
1083
1084 if (err < 0) {
1085 vmci_transport_send_reset(sk, pkt);
1086 sock_put(pending);
1087 err = vmci_transport_error_to_vsock_error(err);
1088 goto out;
1089 }
1090
1091 vsock_add_pending(sk, pending);
1092 sk->sk_ack_backlog++;
1093
1094 pending->sk_state = SS_CONNECTING;
1095 vmci_trans(vpending)->produce_size =
1096 vmci_trans(vpending)->consume_size = qp_size;
1097 vmci_trans(vpending)->queue_pair_size = qp_size;
1098
1099 vmci_trans(vpending)->notify_ops->process_request(pending);
1100
1101 /* We might never receive another message for this socket and it's not
1102 * connected to any process, so we have to ensure it gets cleaned up
1103 * ourself. Our delayed work function will take care of that. Note
1104 * that we do not ever cancel this function since we have few
1105 * guarantees about its state when calling cancel_delayed_work().
1106 * Instead we hold a reference on the socket for that function and make
1107 * it capable of handling cases where it needs to do nothing but
1108 * release that reference.
1109 */
1110 vpending->listener = sk;
1111 sock_hold(sk);
1112 sock_hold(pending);
1113 INIT_DELAYED_WORK(&vpending->dwork, vsock_pending_work);
1114 schedule_delayed_work(&vpending->dwork, HZ);
1115
1116out:
1117 return err;
1118}
1119
1120static int
1121vmci_transport_recv_connecting_server(struct sock *listener,
1122 struct sock *pending,
1123 struct vmci_transport_packet *pkt)
1124{
1125 struct vsock_sock *vpending;
1126 struct vmci_handle handle;
1127 struct vmci_qp *qpair;
1128 bool is_local;
1129 u32 flags;
1130 u32 detach_sub_id;
1131 int err;
1132 int skerr;
1133
1134 vpending = vsock_sk(pending);
1135 detach_sub_id = VMCI_INVALID_ID;
1136
1137 switch (pkt->type) {
1138 case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1139 if (vmci_handle_is_invalid(pkt->u.handle)) {
1140 vmci_transport_send_reset(pending, pkt);
1141 skerr = EPROTO;
1142 err = -EINVAL;
1143 goto destroy;
1144 }
1145 break;
1146 default:
1147 /* Close and cleanup the connection. */
1148 vmci_transport_send_reset(pending, pkt);
1149 skerr = EPROTO;
1150 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1151 goto destroy;
1152 }
1153
1154 /* In order to complete the connection we need to attach to the offered
1155 * queue pair and send an attach notification. We also subscribe to the
1156 * detach event so we know when our peer goes away, and we do that
1157 * before attaching so we don't miss an event. If all this succeeds,
1158 * we update our state and wakeup anything waiting in accept() for a
1159 * connection.
1160 */
1161
1162 /* We don't care about attach since we ensure the other side has
1163 * attached by specifying the ATTACH_ONLY flag below.
1164 */
1165 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1166 vmci_transport_peer_detach_cb,
1167 pending, &detach_sub_id);
1168 if (err < VMCI_SUCCESS) {
1169 vmci_transport_send_reset(pending, pkt);
1170 err = vmci_transport_error_to_vsock_error(err);
1171 skerr = -err;
1172 goto destroy;
1173 }
1174
1175 vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1176
1177 /* Now attach to the queue pair the client created. */
1178 handle = pkt->u.handle;
1179
1180 /* vpending->local_addr always has a context id so we do not need to
1181 * worry about VMADDR_CID_ANY in this case.
1182 */
1183 is_local =
1184 vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1185 flags = VMCI_QPFLAG_ATTACH_ONLY;
1186 flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1187
1188 err = vmci_transport_queue_pair_alloc(
1189 &qpair,
1190 &handle,
1191 vmci_trans(vpending)->produce_size,
1192 vmci_trans(vpending)->consume_size,
1193 pkt->dg.src.context,
1194 flags,
1195 vmci_transport_is_trusted(
1196 vpending,
1197 vpending->remote_addr.svm_cid));
1198 if (err < 0) {
1199 vmci_transport_send_reset(pending, pkt);
1200 skerr = -err;
1201 goto destroy;
1202 }
1203
1204 vmci_trans(vpending)->qp_handle = handle;
1205 vmci_trans(vpending)->qpair = qpair;
1206
1207 /* When we send the attach message, we must be ready to handle incoming
1208 * control messages on the newly connected socket. So we move the
1209 * pending socket to the connected state before sending the attach
1210 * message. Otherwise, an incoming packet triggered by the attach being
1211 * received by the peer may be processed concurrently with what happens
1212 * below after sending the attach message, and that incoming packet
1213 * will find the listening socket instead of the (currently) pending
1214 * socket. Note that enqueueing the socket increments the reference
1215 * count, so even if a reset comes before the connection is accepted,
1216 * the socket will be valid until it is removed from the queue.
1217 *
1218 * If we fail sending the attach below, we remove the socket from the
1219 * connected list and move the socket to SS_UNCONNECTED before
1220 * releasing the lock, so a pending slow path processing of an incoming
1221 * packet will not see the socket in the connected state in that case.
1222 */
1223 pending->sk_state = SS_CONNECTED;
1224
1225 vsock_insert_connected(vpending);
1226
1227 /* Notify our peer of our attach. */
1228 err = vmci_transport_send_attach(pending, handle);
1229 if (err < 0) {
1230 vsock_remove_connected(vpending);
1231 pr_err("Could not send attach\n");
1232 vmci_transport_send_reset(pending, pkt);
1233 err = vmci_transport_error_to_vsock_error(err);
1234 skerr = -err;
1235 goto destroy;
1236 }
1237
1238 /* We have a connection. Move the now connected socket from the
1239 * listener's pending list to the accept queue so callers of accept()
1240 * can find it.
1241 */
1242 vsock_remove_pending(listener, pending);
1243 vsock_enqueue_accept(listener, pending);
1244
1245 /* Callers of accept() will be be waiting on the listening socket, not
1246 * the pending socket.
1247 */
1248 listener->sk_state_change(listener);
1249
1250 return 0;
1251
1252destroy:
1253 pending->sk_err = skerr;
1254 pending->sk_state = SS_UNCONNECTED;
1255 /* As long as we drop our reference, all necessary cleanup will handle
1256 * when the cleanup function drops its reference and our destruct
1257 * implementation is called. Note that since the listen handler will
1258 * remove pending from the pending list upon our failure, the cleanup
1259 * function won't drop the additional reference, which is why we do it
1260 * here.
1261 */
1262 sock_put(pending);
1263
1264 return err;
1265}
1266
1267static int
1268vmci_transport_recv_connecting_client(struct sock *sk,
1269 struct vmci_transport_packet *pkt)
1270{
1271 struct vsock_sock *vsk;
1272 int err;
1273 int skerr;
1274
1275 vsk = vsock_sk(sk);
1276
1277 switch (pkt->type) {
1278 case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1279 if (vmci_handle_is_invalid(pkt->u.handle) ||
1280 !vmci_handle_is_equal(pkt->u.handle,
1281 vmci_trans(vsk)->qp_handle)) {
1282 skerr = EPROTO;
1283 err = -EINVAL;
1284 goto destroy;
1285 }
1286
1287 /* Signify the socket is connected and wakeup the waiter in
1288 * connect(). Also place the socket in the connected table for
1289 * accounting (it can already be found since it's in the bound
1290 * table).
1291 */
1292 sk->sk_state = SS_CONNECTED;
1293 sk->sk_socket->state = SS_CONNECTED;
1294 vsock_insert_connected(vsk);
1295 sk->sk_state_change(sk);
1296
1297 break;
1298 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1299 case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1300 if (pkt->u.size == 0
1301 || pkt->dg.src.context != vsk->remote_addr.svm_cid
1302 || pkt->src_port != vsk->remote_addr.svm_port
1303 || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1304 || vmci_trans(vsk)->qpair
1305 || vmci_trans(vsk)->produce_size != 0
1306 || vmci_trans(vsk)->consume_size != 0
1307 || vmci_trans(vsk)->attach_sub_id != VMCI_INVALID_ID
1308 || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1309 skerr = EPROTO;
1310 err = -EINVAL;
1311
1312 goto destroy;
1313 }
1314
1315 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1316 if (err) {
1317 skerr = -err;
1318 goto destroy;
1319 }
1320
1321 break;
1322 case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1323 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1324 if (err) {
1325 skerr = -err;
1326 goto destroy;
1327 }
1328
1329 break;
1330 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1331 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1332 * continue processing here after they sent an INVALID packet.
1333 * This meant that we got a RST after the INVALID. We ignore a
1334 * RST after an INVALID. The common code doesn't send the RST
1335 * ... so we can hang if an old version of the common code
1336 * fails between getting a REQUEST and sending an OFFER back.
1337 * Not much we can do about it... except hope that it doesn't
1338 * happen.
1339 */
1340 if (vsk->ignore_connecting_rst) {
1341 vsk->ignore_connecting_rst = false;
1342 } else {
1343 skerr = ECONNRESET;
1344 err = 0;
1345 goto destroy;
1346 }
1347
1348 break;
1349 default:
1350 /* Close and cleanup the connection. */
1351 skerr = EPROTO;
1352 err = -EINVAL;
1353 goto destroy;
1354 }
1355
1356 return 0;
1357
1358destroy:
1359 vmci_transport_send_reset(sk, pkt);
1360
1361 sk->sk_state = SS_UNCONNECTED;
1362 sk->sk_err = skerr;
1363 sk->sk_error_report(sk);
1364 return err;
1365}
1366
1367static int vmci_transport_recv_connecting_client_negotiate(
1368 struct sock *sk,
1369 struct vmci_transport_packet *pkt)
1370{
1371 int err;
1372 struct vsock_sock *vsk;
1373 struct vmci_handle handle;
1374 struct vmci_qp *qpair;
1375 u32 attach_sub_id;
1376 u32 detach_sub_id;
1377 bool is_local;
1378 u32 flags;
1379 bool old_proto = true;
1380 bool old_pkt_proto;
1381 u16 version;
1382
1383 vsk = vsock_sk(sk);
1384 handle = VMCI_INVALID_HANDLE;
1385 attach_sub_id = VMCI_INVALID_ID;
1386 detach_sub_id = VMCI_INVALID_ID;
1387
1388 /* If we have gotten here then we should be past the point where old
1389 * linux vsock could have sent the bogus rst.
1390 */
1391 vsk->sent_request = false;
1392 vsk->ignore_connecting_rst = false;
1393
1394 /* Verify that we're OK with the proposed queue pair size */
1395 if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1396 pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1397 err = -EINVAL;
1398 goto destroy;
1399 }
1400
1401 /* At this point we know the CID the peer is using to talk to us. */
1402
1403 if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1404 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1405
1406 /* Setup the notify ops to be the highest supported version that both
1407 * the server and the client support.
1408 */
1409
1410 if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1411 old_proto = old_pkt_proto;
1412 } else {
1413 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1414 old_proto = true;
1415 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1416 old_proto = false;
1417
1418 }
1419
1420 if (old_proto)
1421 version = VSOCK_PROTO_INVALID;
1422 else
1423 version = pkt->proto;
1424
1425 if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1426 err = -EINVAL;
1427 goto destroy;
1428 }
1429
1430 /* Subscribe to attach and detach events first.
1431 *
1432 * XXX We attach once for each queue pair created for now so it is easy
1433 * to find the socket (it's provided), but later we should only
1434 * subscribe once and add a way to lookup sockets by queue pair handle.
1435 */
1436 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_ATTACH,
1437 vmci_transport_peer_attach_cb,
1438 sk, &attach_sub_id);
1439 if (err < VMCI_SUCCESS) {
1440 err = vmci_transport_error_to_vsock_error(err);
1441 goto destroy;
1442 }
1443
1444 err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1445 vmci_transport_peer_detach_cb,
1446 sk, &detach_sub_id);
1447 if (err < VMCI_SUCCESS) {
1448 err = vmci_transport_error_to_vsock_error(err);
1449 goto destroy;
1450 }
1451
1452 /* Make VMCI select the handle for us. */
1453 handle = VMCI_INVALID_HANDLE;
1454 is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1455 flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1456
1457 err = vmci_transport_queue_pair_alloc(&qpair,
1458 &handle,
1459 pkt->u.size,
1460 pkt->u.size,
1461 vsk->remote_addr.svm_cid,
1462 flags,
1463 vmci_transport_is_trusted(
1464 vsk,
1465 vsk->
1466 remote_addr.svm_cid));
1467 if (err < 0)
1468 goto destroy;
1469
1470 err = vmci_transport_send_qp_offer(sk, handle);
1471 if (err < 0) {
1472 err = vmci_transport_error_to_vsock_error(err);
1473 goto destroy;
1474 }
1475
1476 vmci_trans(vsk)->qp_handle = handle;
1477 vmci_trans(vsk)->qpair = qpair;
1478
1479 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1480 pkt->u.size;
1481
1482 vmci_trans(vsk)->attach_sub_id = attach_sub_id;
1483 vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1484
1485 vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1486
1487 return 0;
1488
1489destroy:
1490 if (attach_sub_id != VMCI_INVALID_ID)
1491 vmci_event_unsubscribe(attach_sub_id);
1492
1493 if (detach_sub_id != VMCI_INVALID_ID)
1494 vmci_event_unsubscribe(detach_sub_id);
1495
1496 if (!vmci_handle_is_invalid(handle))
1497 vmci_qpair_detach(&qpair);
1498
1499 return err;
1500}
1501
1502static int
1503vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1504 struct vmci_transport_packet *pkt)
1505{
1506 int err = 0;
1507 struct vsock_sock *vsk = vsock_sk(sk);
1508
1509 if (vsk->sent_request) {
1510 vsk->sent_request = false;
1511 vsk->ignore_connecting_rst = true;
1512
1513 err = vmci_transport_send_conn_request(
1514 sk, vmci_trans(vsk)->queue_pair_size);
1515 if (err < 0)
1516 err = vmci_transport_error_to_vsock_error(err);
1517 else
1518 err = 0;
1519
1520 }
1521
1522 return err;
1523}
1524
1525static int vmci_transport_recv_connected(struct sock *sk,
1526 struct vmci_transport_packet *pkt)
1527{
1528 struct vsock_sock *vsk;
1529 bool pkt_processed = false;
1530
1531 /* In cases where we are closing the connection, it's sufficient to
1532 * mark the state change (and maybe error) and wake up any waiting
1533 * threads. Since this is a connected socket, it's owned by a user
1534 * process and will be cleaned up when the failure is passed back on
1535 * the current or next system call. Our system call implementations
1536 * must therefore check for error and state changes on entry and when
1537 * being awoken.
1538 */
1539 switch (pkt->type) {
1540 case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1541 if (pkt->u.mode) {
1542 vsk = vsock_sk(sk);
1543
1544 vsk->peer_shutdown |= pkt->u.mode;
1545 sk->sk_state_change(sk);
1546 }
1547 break;
1548
1549 case VMCI_TRANSPORT_PACKET_TYPE_RST:
1550 vsk = vsock_sk(sk);
1551 /* It is possible that we sent our peer a message (e.g a
1552 * WAITING_READ) right before we got notified that the peer had
1553 * detached. If that happens then we can get a RST pkt back
1554 * from our peer even though there is data available for us to
1555 * read. In that case, don't shutdown the socket completely but
1556 * instead allow the local client to finish reading data off
1557 * the queuepair. Always treat a RST pkt in connected mode like
1558 * a clean shutdown.
1559 */
1560 sock_set_flag(sk, SOCK_DONE);
1561 vsk->peer_shutdown = SHUTDOWN_MASK;
1562 if (vsock_stream_has_data(vsk) <= 0)
1563 sk->sk_state = SS_DISCONNECTING;
1564
1565 sk->sk_state_change(sk);
1566 break;
1567
1568 default:
1569 vsk = vsock_sk(sk);
1570 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1571 sk, pkt, false, NULL, NULL,
1572 &pkt_processed);
1573 if (!pkt_processed)
1574 return -EINVAL;
1575
1576 break;
1577 }
1578
1579 return 0;
1580}
1581
1582static int vmci_transport_socket_init(struct vsock_sock *vsk,
1583 struct vsock_sock *psk)
1584{
1585 vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1586 if (!vsk->trans)
1587 return -ENOMEM;
1588
1589 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1590 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1591 vmci_trans(vsk)->qpair = NULL;
1592 vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1593 vmci_trans(vsk)->attach_sub_id = vmci_trans(vsk)->detach_sub_id =
1594 VMCI_INVALID_ID;
1595 vmci_trans(vsk)->notify_ops = NULL;
1596 if (psk) {
1597 vmci_trans(vsk)->queue_pair_size =
1598 vmci_trans(psk)->queue_pair_size;
1599 vmci_trans(vsk)->queue_pair_min_size =
1600 vmci_trans(psk)->queue_pair_min_size;
1601 vmci_trans(vsk)->queue_pair_max_size =
1602 vmci_trans(psk)->queue_pair_max_size;
1603 } else {
1604 vmci_trans(vsk)->queue_pair_size =
1605 VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1606 vmci_trans(vsk)->queue_pair_min_size =
1607 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1608 vmci_trans(vsk)->queue_pair_max_size =
1609 VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1610 }
1611
1612 return 0;
1613}
1614
1615static void vmci_transport_destruct(struct vsock_sock *vsk)
1616{
1617 if (vmci_trans(vsk)->attach_sub_id != VMCI_INVALID_ID) {
1618 vmci_event_unsubscribe(vmci_trans(vsk)->attach_sub_id);
1619 vmci_trans(vsk)->attach_sub_id = VMCI_INVALID_ID;
1620 }
1621
1622 if (vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1623 vmci_event_unsubscribe(vmci_trans(vsk)->detach_sub_id);
1624 vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1625 }
1626
1627 if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
1628 vmci_qpair_detach(&vmci_trans(vsk)->qpair);
1629 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1630 vmci_trans(vsk)->produce_size = 0;
1631 vmci_trans(vsk)->consume_size = 0;
1632 }
1633
1634 if (vmci_trans(vsk)->notify_ops)
1635 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1636
1637 kfree(vsk->trans);
1638 vsk->trans = NULL;
1639}
1640
1641static void vmci_transport_release(struct vsock_sock *vsk)
1642{
1643 if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1644 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1645 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1646 }
1647}
1648
1649static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1650 struct sockaddr_vm *addr)
1651{
1652 u32 port;
1653 u32 flags;
1654 int err;
1655
1656 /* VMCI will select a resource ID for us if we provide
1657 * VMCI_INVALID_ID.
1658 */
1659 port = addr->svm_port == VMADDR_PORT_ANY ?
1660 VMCI_INVALID_ID : addr->svm_port;
1661
1662 if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1663 return -EACCES;
1664
1665 flags = addr->svm_cid == VMADDR_CID_ANY ?
1666 VMCI_FLAG_ANYCID_DG_HND : 0;
1667
1668 err = vmci_transport_datagram_create_hnd(port, flags,
1669 vmci_transport_recv_dgram_cb,
1670 &vsk->sk,
1671 &vmci_trans(vsk)->dg_handle);
1672 if (err < VMCI_SUCCESS)
1673 return vmci_transport_error_to_vsock_error(err);
1674 vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1675 vmci_trans(vsk)->dg_handle.resource);
1676
1677 return 0;
1678}
1679
1680static int vmci_transport_dgram_enqueue(
1681 struct vsock_sock *vsk,
1682 struct sockaddr_vm *remote_addr,
1683 struct iovec *iov,
1684 size_t len)
1685{
1686 int err;
1687 struct vmci_datagram *dg;
1688
1689 if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1690 return -EMSGSIZE;
1691
1692 if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1693 return -EPERM;
1694
1695 /* Allocate a buffer for the user's message and our packet header. */
1696 dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1697 if (!dg)
1698 return -ENOMEM;
1699
1700 memcpy_fromiovec(VMCI_DG_PAYLOAD(dg), iov, len);
1701
1702 dg->dst = vmci_make_handle(remote_addr->svm_cid,
1703 remote_addr->svm_port);
1704 dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1705 vsk->local_addr.svm_port);
1706 dg->payload_size = len;
1707
1708 err = vmci_datagram_send(dg);
1709 kfree(dg);
1710 if (err < 0)
1711 return vmci_transport_error_to_vsock_error(err);
1712
1713 return err - sizeof(*dg);
1714}
1715
1716static int vmci_transport_dgram_dequeue(struct kiocb *kiocb,
1717 struct vsock_sock *vsk,
1718 struct msghdr *msg, size_t len,
1719 int flags)
1720{
1721 int err;
1722 int noblock;
1723 struct vmci_datagram *dg;
1724 size_t payload_len;
1725 struct sk_buff *skb;
1726
1727 noblock = flags & MSG_DONTWAIT;
1728
1729 if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1730 return -EOPNOTSUPP;
1731
1732 /* Retrieve the head sk_buff from the socket's receive queue. */
1733 err = 0;
1734 skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1735 if (err)
1736 return err;
1737
1738 if (!skb)
1739 return -EAGAIN;
1740
1741 dg = (struct vmci_datagram *)skb->data;
1742 if (!dg)
1743 /* err is 0, meaning we read zero bytes. */
1744 goto out;
1745
1746 payload_len = dg->payload_size;
1747 /* Ensure the sk_buff matches the payload size claimed in the packet. */
1748 if (payload_len != skb->len - sizeof(*dg)) {
1749 err = -EINVAL;
1750 goto out;
1751 }
1752
1753 if (payload_len > len) {
1754 payload_len = len;
1755 msg->msg_flags |= MSG_TRUNC;
1756 }
1757
1758 /* Place the datagram payload in the user's iovec. */
1759 err = skb_copy_datagram_iovec(skb, sizeof(*dg), msg->msg_iov,
1760 payload_len);
1761 if (err)
1762 goto out;
1763
1764 msg->msg_namelen = 0;
1765 if (msg->msg_name) {
1766 struct sockaddr_vm *vm_addr;
1767
1768 /* Provide the address of the sender. */
1769 vm_addr = (struct sockaddr_vm *)msg->msg_name;
1770 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1771 msg->msg_namelen = sizeof(*vm_addr);
1772 }
1773 err = payload_len;
1774
1775out:
1776 skb_free_datagram(&vsk->sk, skb);
1777 return err;
1778}
1779
1780static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1781{
1782 if (cid == VMADDR_CID_HYPERVISOR) {
1783 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1784 * state and are allowed.
1785 */
1786 return port == VMCI_UNITY_PBRPC_REGISTER;
1787 }
1788
1789 return true;
1790}
1791
1792static int vmci_transport_connect(struct vsock_sock *vsk)
1793{
1794 int err;
1795 bool old_pkt_proto = false;
1796 struct sock *sk = &vsk->sk;
1797
1798 if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1799 old_pkt_proto) {
1800 err = vmci_transport_send_conn_request(
1801 sk, vmci_trans(vsk)->queue_pair_size);
1802 if (err < 0) {
1803 sk->sk_state = SS_UNCONNECTED;
1804 return err;
1805 }
1806 } else {
1807 int supported_proto_versions =
1808 vmci_transport_new_proto_supported_versions();
1809 err = vmci_transport_send_conn_request2(
1810 sk, vmci_trans(vsk)->queue_pair_size,
1811 supported_proto_versions);
1812 if (err < 0) {
1813 sk->sk_state = SS_UNCONNECTED;
1814 return err;
1815 }
1816
1817 vsk->sent_request = true;
1818 }
1819
1820 return err;
1821}
1822
1823static ssize_t vmci_transport_stream_dequeue(
1824 struct vsock_sock *vsk,
1825 struct iovec *iov,
1826 size_t len,
1827 int flags)
1828{
1829 if (flags & MSG_PEEK)
1830 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, iov, len, 0);
1831 else
1832 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, iov, len, 0);
1833}
1834
1835static ssize_t vmci_transport_stream_enqueue(
1836 struct vsock_sock *vsk,
1837 struct iovec *iov,
1838 size_t len)
1839{
1840 return vmci_qpair_enquev(vmci_trans(vsk)->qpair, iov, len, 0);
1841}
1842
1843static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1844{
1845 return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1846}
1847
1848static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1849{
1850 return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1851}
1852
1853static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1854{
1855 return vmci_trans(vsk)->consume_size;
1856}
1857
1858static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1859{
1860 return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1861}
1862
1863static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1864{
1865 return vmci_trans(vsk)->queue_pair_size;
1866}
1867
1868static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1869{
1870 return vmci_trans(vsk)->queue_pair_min_size;
1871}
1872
1873static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1874{
1875 return vmci_trans(vsk)->queue_pair_max_size;
1876}
1877
1878static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1879{
1880 if (val < vmci_trans(vsk)->queue_pair_min_size)
1881 vmci_trans(vsk)->queue_pair_min_size = val;
1882 if (val > vmci_trans(vsk)->queue_pair_max_size)
1883 vmci_trans(vsk)->queue_pair_max_size = val;
1884 vmci_trans(vsk)->queue_pair_size = val;
1885}
1886
1887static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1888 u64 val)
1889{
1890 if (val > vmci_trans(vsk)->queue_pair_size)
1891 vmci_trans(vsk)->queue_pair_size = val;
1892 vmci_trans(vsk)->queue_pair_min_size = val;
1893}
1894
1895static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1896 u64 val)
1897{
1898 if (val < vmci_trans(vsk)->queue_pair_size)
1899 vmci_trans(vsk)->queue_pair_size = val;
1900 vmci_trans(vsk)->queue_pair_max_size = val;
1901}
1902
1903static int vmci_transport_notify_poll_in(
1904 struct vsock_sock *vsk,
1905 size_t target,
1906 bool *data_ready_now)
1907{
1908 return vmci_trans(vsk)->notify_ops->poll_in(
1909 &vsk->sk, target, data_ready_now);
1910}
1911
1912static int vmci_transport_notify_poll_out(
1913 struct vsock_sock *vsk,
1914 size_t target,
1915 bool *space_available_now)
1916{
1917 return vmci_trans(vsk)->notify_ops->poll_out(
1918 &vsk->sk, target, space_available_now);
1919}
1920
1921static int vmci_transport_notify_recv_init(
1922 struct vsock_sock *vsk,
1923 size_t target,
1924 struct vsock_transport_recv_notify_data *data)
1925{
1926 return vmci_trans(vsk)->notify_ops->recv_init(
1927 &vsk->sk, target,
1928 (struct vmci_transport_recv_notify_data *)data);
1929}
1930
1931static int vmci_transport_notify_recv_pre_block(
1932 struct vsock_sock *vsk,
1933 size_t target,
1934 struct vsock_transport_recv_notify_data *data)
1935{
1936 return vmci_trans(vsk)->notify_ops->recv_pre_block(
1937 &vsk->sk, target,
1938 (struct vmci_transport_recv_notify_data *)data);
1939}
1940
1941static int vmci_transport_notify_recv_pre_dequeue(
1942 struct vsock_sock *vsk,
1943 size_t target,
1944 struct vsock_transport_recv_notify_data *data)
1945{
1946 return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1947 &vsk->sk, target,
1948 (struct vmci_transport_recv_notify_data *)data);
1949}
1950
1951static int vmci_transport_notify_recv_post_dequeue(
1952 struct vsock_sock *vsk,
1953 size_t target,
1954 ssize_t copied,
1955 bool data_read,
1956 struct vsock_transport_recv_notify_data *data)
1957{
1958 return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1959 &vsk->sk, target, copied, data_read,
1960 (struct vmci_transport_recv_notify_data *)data);
1961}
1962
1963static int vmci_transport_notify_send_init(
1964 struct vsock_sock *vsk,
1965 struct vsock_transport_send_notify_data *data)
1966{
1967 return vmci_trans(vsk)->notify_ops->send_init(
1968 &vsk->sk,
1969 (struct vmci_transport_send_notify_data *)data);
1970}
1971
1972static int vmci_transport_notify_send_pre_block(
1973 struct vsock_sock *vsk,
1974 struct vsock_transport_send_notify_data *data)
1975{
1976 return vmci_trans(vsk)->notify_ops->send_pre_block(
1977 &vsk->sk,
1978 (struct vmci_transport_send_notify_data *)data);
1979}
1980
1981static int vmci_transport_notify_send_pre_enqueue(
1982 struct vsock_sock *vsk,
1983 struct vsock_transport_send_notify_data *data)
1984{
1985 return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1986 &vsk->sk,
1987 (struct vmci_transport_send_notify_data *)data);
1988}
1989
1990static int vmci_transport_notify_send_post_enqueue(
1991 struct vsock_sock *vsk,
1992 ssize_t written,
1993 struct vsock_transport_send_notify_data *data)
1994{
1995 return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1996 &vsk->sk, written,
1997 (struct vmci_transport_send_notify_data *)data);
1998}
1999
2000static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2001{
2002 if (PROTOCOL_OVERRIDE != -1) {
2003 if (PROTOCOL_OVERRIDE == 0)
2004 *old_pkt_proto = true;
2005 else
2006 *old_pkt_proto = false;
2007
2008 pr_info("Proto override in use\n");
2009 return true;
2010 }
2011
2012 return false;
2013}
2014
2015static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2016 u16 *proto,
2017 bool old_pkt_proto)
2018{
2019 struct vsock_sock *vsk = vsock_sk(sk);
2020
2021 if (old_pkt_proto) {
2022 if (*proto != VSOCK_PROTO_INVALID) {
2023 pr_err("Can't set both an old and new protocol\n");
2024 return false;
2025 }
2026 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2027 goto exit;
2028 }
2029
2030 switch (*proto) {
2031 case VSOCK_PROTO_PKT_ON_NOTIFY:
2032 vmci_trans(vsk)->notify_ops =
2033 &vmci_transport_notify_pkt_q_state_ops;
2034 break;
2035 default:
2036 pr_err("Unknown notify protocol version\n");
2037 return false;
2038 }
2039
2040exit:
2041 vmci_trans(vsk)->notify_ops->socket_init(sk);
2042 return true;
2043}
2044
2045static u16 vmci_transport_new_proto_supported_versions(void)
2046{
2047 if (PROTOCOL_OVERRIDE != -1)
2048 return PROTOCOL_OVERRIDE;
2049
2050 return VSOCK_PROTO_ALL_SUPPORTED;
2051}
2052
2053static u32 vmci_transport_get_local_cid(void)
2054{
2055 return vmci_get_context_id();
2056}
2057
2058static struct vsock_transport vmci_transport = {
2059 .init = vmci_transport_socket_init,
2060 .destruct = vmci_transport_destruct,
2061 .release = vmci_transport_release,
2062 .connect = vmci_transport_connect,
2063 .dgram_bind = vmci_transport_dgram_bind,
2064 .dgram_dequeue = vmci_transport_dgram_dequeue,
2065 .dgram_enqueue = vmci_transport_dgram_enqueue,
2066 .dgram_allow = vmci_transport_dgram_allow,
2067 .stream_dequeue = vmci_transport_stream_dequeue,
2068 .stream_enqueue = vmci_transport_stream_enqueue,
2069 .stream_has_data = vmci_transport_stream_has_data,
2070 .stream_has_space = vmci_transport_stream_has_space,
2071 .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2072 .stream_is_active = vmci_transport_stream_is_active,
2073 .stream_allow = vmci_transport_stream_allow,
2074 .notify_poll_in = vmci_transport_notify_poll_in,
2075 .notify_poll_out = vmci_transport_notify_poll_out,
2076 .notify_recv_init = vmci_transport_notify_recv_init,
2077 .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2078 .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2079 .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2080 .notify_send_init = vmci_transport_notify_send_init,
2081 .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2082 .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2083 .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2084 .shutdown = vmci_transport_shutdown,
2085 .set_buffer_size = vmci_transport_set_buffer_size,
2086 .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2087 .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2088 .get_buffer_size = vmci_transport_get_buffer_size,
2089 .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2090 .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2091 .get_local_cid = vmci_transport_get_local_cid,
2092};
2093
2094static int __init vmci_transport_init(void)
2095{
2096 int err;
2097
2098 /* Create the datagram handle that we will use to send and receive all
2099 * VSocket control messages for this context.
2100 */
2101 err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2102 VMCI_FLAG_ANYCID_DG_HND,
2103 vmci_transport_recv_stream_cb,
2104 NULL,
2105 &vmci_transport_stream_handle);
2106 if (err < VMCI_SUCCESS) {
2107 pr_err("Unable to create datagram handle. (%d)\n", err);
2108 return vmci_transport_error_to_vsock_error(err);
2109 }
2110
2111 err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2112 vmci_transport_qp_resumed_cb,
2113 NULL, &vmci_transport_qp_resumed_sub_id);
2114 if (err < VMCI_SUCCESS) {
2115 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2116 err = vmci_transport_error_to_vsock_error(err);
2117 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2118 goto err_destroy_stream_handle;
2119 }
2120
2121 err = vsock_core_init(&vmci_transport);
2122 if (err < 0)
2123 goto err_unsubscribe;
2124
2125 return 0;
2126
2127err_unsubscribe:
2128 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2129err_destroy_stream_handle:
2130 vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2131 return err;
2132}
2133module_init(vmci_transport_init);
2134
2135static void __exit vmci_transport_exit(void)
2136{
2137 if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2138 if (vmci_datagram_destroy_handle(
2139 vmci_transport_stream_handle) != VMCI_SUCCESS)
2140 pr_err("Couldn't destroy datagram handle\n");
2141 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2142 }
2143
2144 if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2145 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2146 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2147 }
2148
2149 vsock_core_exit();
2150}
2151module_exit(vmci_transport_exit);
2152
2153MODULE_AUTHOR("VMware, Inc.");
2154MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2155MODULE_LICENSE("GPL v2");
2156MODULE_ALIAS("vmware_vsock");
2157MODULE_ALIAS_NETPROTO(PF_VSOCK);
diff --git a/net/vmw_vsock/vmci_transport.h b/net/vmw_vsock/vmci_transport.h
new file mode 100644
index 000000000000..1bf991803ec0
--- /dev/null
+++ b/net/vmw_vsock/vmci_transport.h
@@ -0,0 +1,139 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#ifndef _VMCI_TRANSPORT_H_
17#define _VMCI_TRANSPORT_H_
18
19#include <linux/vmw_vmci_defs.h>
20#include <linux/vmw_vmci_api.h>
21
22#include "vsock_addr.h"
23#include "af_vsock.h"
24
25/* If the packet format changes in a release then this should change too. */
26#define VMCI_TRANSPORT_PACKET_VERSION 1
27
28/* The resource ID on which control packets are sent. */
29#define VMCI_TRANSPORT_PACKET_RID 1
30
31#define VSOCK_PROTO_INVALID 0
32#define VSOCK_PROTO_PKT_ON_NOTIFY (1 << 0)
33#define VSOCK_PROTO_ALL_SUPPORTED (VSOCK_PROTO_PKT_ON_NOTIFY)
34
35#define vmci_trans(_vsk) ((struct vmci_transport *)((_vsk)->trans))
36
37enum vmci_transport_packet_type {
38 VMCI_TRANSPORT_PACKET_TYPE_INVALID = 0,
39 VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
40 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
41 VMCI_TRANSPORT_PACKET_TYPE_OFFER,
42 VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
43 VMCI_TRANSPORT_PACKET_TYPE_WROTE,
44 VMCI_TRANSPORT_PACKET_TYPE_READ,
45 VMCI_TRANSPORT_PACKET_TYPE_RST,
46 VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
47 VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
48 VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
49 VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
50 VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
51 VMCI_TRANSPORT_PACKET_TYPE_MAX
52};
53
54struct vmci_transport_waiting_info {
55 u64 generation;
56 u64 offset;
57};
58
59/* Control packet type for STREAM sockets. DGRAMs have no control packets nor
60 * special packet header for data packets, they are just raw VMCI DGRAM
61 * messages. For STREAMs, control packets are sent over the control channel
62 * while data is written and read directly from queue pairs with no packet
63 * format.
64 */
65struct vmci_transport_packet {
66 struct vmci_datagram dg;
67 u8 version;
68 u8 type;
69 u16 proto;
70 u32 src_port;
71 u32 dst_port;
72 u32 _reserved2;
73 union {
74 u64 size;
75 u64 mode;
76 struct vmci_handle handle;
77 struct vmci_transport_waiting_info wait;
78 } u;
79};
80
81struct vmci_transport_notify_pkt {
82 u64 write_notify_window;
83 u64 write_notify_min_window;
84 bool peer_waiting_read;
85 bool peer_waiting_write;
86 bool peer_waiting_write_detected;
87 bool sent_waiting_read;
88 bool sent_waiting_write;
89 struct vmci_transport_waiting_info peer_waiting_read_info;
90 struct vmci_transport_waiting_info peer_waiting_write_info;
91 u64 produce_q_generation;
92 u64 consume_q_generation;
93};
94
95struct vmci_transport_notify_pkt_q_state {
96 u64 write_notify_window;
97 u64 write_notify_min_window;
98 bool peer_waiting_write;
99 bool peer_waiting_write_detected;
100};
101
102union vmci_transport_notify {
103 struct vmci_transport_notify_pkt pkt;
104 struct vmci_transport_notify_pkt_q_state pkt_q_state;
105};
106
107/* Our transport-specific data. */
108struct vmci_transport {
109 /* For DGRAMs. */
110 struct vmci_handle dg_handle;
111 /* For STREAMs. */
112 struct vmci_handle qp_handle;
113 struct vmci_qp *qpair;
114 u64 produce_size;
115 u64 consume_size;
116 u64 queue_pair_size;
117 u64 queue_pair_min_size;
118 u64 queue_pair_max_size;
119 u32 attach_sub_id;
120 u32 detach_sub_id;
121 union vmci_transport_notify notify;
122 struct vmci_transport_notify_ops *notify_ops;
123};
124
125int vmci_transport_register(void);
126void vmci_transport_unregister(void);
127
128int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
129 struct sockaddr_vm *src);
130int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
131 struct sockaddr_vm *src);
132int vmci_transport_send_wrote(struct sock *sk);
133int vmci_transport_send_read(struct sock *sk);
134int vmci_transport_send_waiting_write(struct sock *sk,
135 struct vmci_transport_waiting_info *wait);
136int vmci_transport_send_waiting_read(struct sock *sk,
137 struct vmci_transport_waiting_info *wait);
138
139#endif
diff --git a/net/vmw_vsock/vmci_transport_notify.c b/net/vmw_vsock/vmci_transport_notify.c
new file mode 100644
index 000000000000..9a730744e7bc
--- /dev/null
+++ b/net/vmw_vsock/vmci_transport_notify.c
@@ -0,0 +1,680 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2009-2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#include <linux/types.h>
17#include <linux/socket.h>
18#include <linux/stddef.h>
19#include <net/sock.h>
20
21#include "vmci_transport_notify.h"
22
23#define PKT_FIELD(vsk, field_name) (vmci_trans(vsk)->notify.pkt.field_name)
24
25static bool vmci_transport_notify_waiting_write(struct vsock_sock *vsk)
26{
27#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
28 bool retval;
29 u64 notify_limit;
30
31 if (!PKT_FIELD(vsk, peer_waiting_write))
32 return false;
33
34#ifdef VSOCK_OPTIMIZATION_FLOW_CONTROL
35 /* When the sender blocks, we take that as a sign that the sender is
36 * faster than the receiver. To reduce the transmit rate of the sender,
37 * we delay the sending of the read notification by decreasing the
38 * write_notify_window. The notification is delayed until the number of
39 * bytes used in the queue drops below the write_notify_window.
40 */
41
42 if (!PKT_FIELD(vsk, peer_waiting_write_detected)) {
43 PKT_FIELD(vsk, peer_waiting_write_detected) = true;
44 if (PKT_FIELD(vsk, write_notify_window) < PAGE_SIZE) {
45 PKT_FIELD(vsk, write_notify_window) =
46 PKT_FIELD(vsk, write_notify_min_window);
47 } else {
48 PKT_FIELD(vsk, write_notify_window) -= PAGE_SIZE;
49 if (PKT_FIELD(vsk, write_notify_window) <
50 PKT_FIELD(vsk, write_notify_min_window))
51 PKT_FIELD(vsk, write_notify_window) =
52 PKT_FIELD(vsk, write_notify_min_window);
53
54 }
55 }
56 notify_limit = vmci_trans(vsk)->consume_size -
57 PKT_FIELD(vsk, write_notify_window);
58#else
59 notify_limit = 0;
60#endif
61
62 /* For now we ignore the wait information and just see if the free
63 * space exceeds the notify limit. Note that improving this function
64 * to be more intelligent will not require a protocol change and will
65 * retain compatibility between endpoints with mixed versions of this
66 * function.
67 *
68 * The notify_limit is used to delay notifications in the case where
69 * flow control is enabled. Below the test is expressed in terms of
70 * free space in the queue: if free_space > ConsumeSize -
71 * write_notify_window then notify An alternate way of expressing this
72 * is to rewrite the expression to use the data ready in the receive
73 * queue: if write_notify_window > bufferReady then notify as
74 * free_space == ConsumeSize - bufferReady.
75 */
76 retval = vmci_qpair_consume_free_space(vmci_trans(vsk)->qpair) >
77 notify_limit;
78#ifdef VSOCK_OPTIMIZATION_FLOW_CONTROL
79 if (retval) {
80 /*
81 * Once we notify the peer, we reset the detected flag so the
82 * next wait will again cause a decrease in the window size.
83 */
84
85 PKT_FIELD(vsk, peer_waiting_write_detected) = false;
86 }
87#endif
88 return retval;
89#else
90 return true;
91#endif
92}
93
94static bool vmci_transport_notify_waiting_read(struct vsock_sock *vsk)
95{
96#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
97 if (!PKT_FIELD(vsk, peer_waiting_read))
98 return false;
99
100 /* For now we ignore the wait information and just see if there is any
101 * data for our peer to read. Note that improving this function to be
102 * more intelligent will not require a protocol change and will retain
103 * compatibility between endpoints with mixed versions of this
104 * function.
105 */
106 return vmci_qpair_produce_buf_ready(vmci_trans(vsk)->qpair) > 0;
107#else
108 return true;
109#endif
110}
111
112static void
113vmci_transport_handle_waiting_read(struct sock *sk,
114 struct vmci_transport_packet *pkt,
115 bool bottom_half,
116 struct sockaddr_vm *dst,
117 struct sockaddr_vm *src)
118{
119#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
120 struct vsock_sock *vsk;
121
122 vsk = vsock_sk(sk);
123
124 PKT_FIELD(vsk, peer_waiting_read) = true;
125 memcpy(&PKT_FIELD(vsk, peer_waiting_read_info), &pkt->u.wait,
126 sizeof(PKT_FIELD(vsk, peer_waiting_read_info)));
127
128 if (vmci_transport_notify_waiting_read(vsk)) {
129 bool sent;
130
131 if (bottom_half)
132 sent = vmci_transport_send_wrote_bh(dst, src) > 0;
133 else
134 sent = vmci_transport_send_wrote(sk) > 0;
135
136 if (sent)
137 PKT_FIELD(vsk, peer_waiting_read) = false;
138 }
139#endif
140}
141
142static void
143vmci_transport_handle_waiting_write(struct sock *sk,
144 struct vmci_transport_packet *pkt,
145 bool bottom_half,
146 struct sockaddr_vm *dst,
147 struct sockaddr_vm *src)
148{
149#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
150 struct vsock_sock *vsk;
151
152 vsk = vsock_sk(sk);
153
154 PKT_FIELD(vsk, peer_waiting_write) = true;
155 memcpy(&PKT_FIELD(vsk, peer_waiting_write_info), &pkt->u.wait,
156 sizeof(PKT_FIELD(vsk, peer_waiting_write_info)));
157
158 if (vmci_transport_notify_waiting_write(vsk)) {
159 bool sent;
160
161 if (bottom_half)
162 sent = vmci_transport_send_read_bh(dst, src) > 0;
163 else
164 sent = vmci_transport_send_read(sk) > 0;
165
166 if (sent)
167 PKT_FIELD(vsk, peer_waiting_write) = false;
168 }
169#endif
170}
171
172static void
173vmci_transport_handle_read(struct sock *sk,
174 struct vmci_transport_packet *pkt,
175 bool bottom_half,
176 struct sockaddr_vm *dst, struct sockaddr_vm *src)
177{
178#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
179 struct vsock_sock *vsk;
180
181 vsk = vsock_sk(sk);
182 PKT_FIELD(vsk, sent_waiting_write) = false;
183#endif
184
185 sk->sk_write_space(sk);
186}
187
188static bool send_waiting_read(struct sock *sk, u64 room_needed)
189{
190#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
191 struct vsock_sock *vsk;
192 struct vmci_transport_waiting_info waiting_info;
193 u64 tail;
194 u64 head;
195 u64 room_left;
196 bool ret;
197
198 vsk = vsock_sk(sk);
199
200 if (PKT_FIELD(vsk, sent_waiting_read))
201 return true;
202
203 if (PKT_FIELD(vsk, write_notify_window) <
204 vmci_trans(vsk)->consume_size)
205 PKT_FIELD(vsk, write_notify_window) =
206 min(PKT_FIELD(vsk, write_notify_window) + PAGE_SIZE,
207 vmci_trans(vsk)->consume_size);
208
209 vmci_qpair_get_consume_indexes(vmci_trans(vsk)->qpair, &tail, &head);
210 room_left = vmci_trans(vsk)->consume_size - head;
211 if (room_needed >= room_left) {
212 waiting_info.offset = room_needed - room_left;
213 waiting_info.generation =
214 PKT_FIELD(vsk, consume_q_generation) + 1;
215 } else {
216 waiting_info.offset = head + room_needed;
217 waiting_info.generation = PKT_FIELD(vsk, consume_q_generation);
218 }
219
220 ret = vmci_transport_send_waiting_read(sk, &waiting_info) > 0;
221 if (ret)
222 PKT_FIELD(vsk, sent_waiting_read) = true;
223
224 return ret;
225#else
226 return true;
227#endif
228}
229
230static bool send_waiting_write(struct sock *sk, u64 room_needed)
231{
232#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
233 struct vsock_sock *vsk;
234 struct vmci_transport_waiting_info waiting_info;
235 u64 tail;
236 u64 head;
237 u64 room_left;
238 bool ret;
239
240 vsk = vsock_sk(sk);
241
242 if (PKT_FIELD(vsk, sent_waiting_write))
243 return true;
244
245 vmci_qpair_get_produce_indexes(vmci_trans(vsk)->qpair, &tail, &head);
246 room_left = vmci_trans(vsk)->produce_size - tail;
247 if (room_needed + 1 >= room_left) {
248 /* Wraps around to current generation. */
249 waiting_info.offset = room_needed + 1 - room_left;
250 waiting_info.generation = PKT_FIELD(vsk, produce_q_generation);
251 } else {
252 waiting_info.offset = tail + room_needed + 1;
253 waiting_info.generation =
254 PKT_FIELD(vsk, produce_q_generation) - 1;
255 }
256
257 ret = vmci_transport_send_waiting_write(sk, &waiting_info) > 0;
258 if (ret)
259 PKT_FIELD(vsk, sent_waiting_write) = true;
260
261 return ret;
262#else
263 return true;
264#endif
265}
266
267static int vmci_transport_send_read_notification(struct sock *sk)
268{
269 struct vsock_sock *vsk;
270 bool sent_read;
271 unsigned int retries;
272 int err;
273
274 vsk = vsock_sk(sk);
275 sent_read = false;
276 retries = 0;
277 err = 0;
278
279 if (vmci_transport_notify_waiting_write(vsk)) {
280 /* Notify the peer that we have read, retrying the send on
281 * failure up to our maximum value. XXX For now we just log
282 * the failure, but later we should schedule a work item to
283 * handle the resend until it succeeds. That would require
284 * keeping track of work items in the vsk and cleaning them up
285 * upon socket close.
286 */
287 while (!(vsk->peer_shutdown & RCV_SHUTDOWN) &&
288 !sent_read &&
289 retries < VMCI_TRANSPORT_MAX_DGRAM_RESENDS) {
290 err = vmci_transport_send_read(sk);
291 if (err >= 0)
292 sent_read = true;
293
294 retries++;
295 }
296
297 if (retries >= VMCI_TRANSPORT_MAX_DGRAM_RESENDS)
298 pr_err("%p unable to send read notify to peer\n", sk);
299 else
300#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
301 PKT_FIELD(vsk, peer_waiting_write) = false;
302#endif
303
304 }
305 return err;
306}
307
308static void
309vmci_transport_handle_wrote(struct sock *sk,
310 struct vmci_transport_packet *pkt,
311 bool bottom_half,
312 struct sockaddr_vm *dst, struct sockaddr_vm *src)
313{
314#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
315 struct vsock_sock *vsk = vsock_sk(sk);
316 PKT_FIELD(vsk, sent_waiting_read) = false;
317#endif
318 sk->sk_data_ready(sk, 0);
319}
320
321static void vmci_transport_notify_pkt_socket_init(struct sock *sk)
322{
323 struct vsock_sock *vsk = vsock_sk(sk);
324
325 PKT_FIELD(vsk, write_notify_window) = PAGE_SIZE;
326 PKT_FIELD(vsk, write_notify_min_window) = PAGE_SIZE;
327 PKT_FIELD(vsk, peer_waiting_read) = false;
328 PKT_FIELD(vsk, peer_waiting_write) = false;
329 PKT_FIELD(vsk, peer_waiting_write_detected) = false;
330 PKT_FIELD(vsk, sent_waiting_read) = false;
331 PKT_FIELD(vsk, sent_waiting_write) = false;
332 PKT_FIELD(vsk, produce_q_generation) = 0;
333 PKT_FIELD(vsk, consume_q_generation) = 0;
334
335 memset(&PKT_FIELD(vsk, peer_waiting_read_info), 0,
336 sizeof(PKT_FIELD(vsk, peer_waiting_read_info)));
337 memset(&PKT_FIELD(vsk, peer_waiting_write_info), 0,
338 sizeof(PKT_FIELD(vsk, peer_waiting_write_info)));
339}
340
341static void vmci_transport_notify_pkt_socket_destruct(struct vsock_sock *vsk)
342{
343}
344
345static int
346vmci_transport_notify_pkt_poll_in(struct sock *sk,
347 size_t target, bool *data_ready_now)
348{
349 struct vsock_sock *vsk = vsock_sk(sk);
350
351 if (vsock_stream_has_data(vsk)) {
352 *data_ready_now = true;
353 } else {
354 /* We can't read right now because there is nothing in the
355 * queue. Ask for notifications when there is something to
356 * read.
357 */
358 if (sk->sk_state == SS_CONNECTED) {
359 if (!send_waiting_read(sk, 1))
360 return -1;
361
362 }
363 *data_ready_now = false;
364 }
365
366 return 0;
367}
368
369static int
370vmci_transport_notify_pkt_poll_out(struct sock *sk,
371 size_t target, bool *space_avail_now)
372{
373 s64 produce_q_free_space;
374 struct vsock_sock *vsk = vsock_sk(sk);
375
376 produce_q_free_space = vsock_stream_has_space(vsk);
377 if (produce_q_free_space > 0) {
378 *space_avail_now = true;
379 return 0;
380 } else if (produce_q_free_space == 0) {
381 /* This is a connected socket but we can't currently send data.
382 * Notify the peer that we are waiting if the queue is full. We
383 * only send a waiting write if the queue is full because
384 * otherwise we end up in an infinite WAITING_WRITE, READ,
385 * WAITING_WRITE, READ, etc. loop. Treat failing to send the
386 * notification as a socket error, passing that back through
387 * the mask.
388 */
389 if (!send_waiting_write(sk, 1))
390 return -1;
391
392 *space_avail_now = false;
393 }
394
395 return 0;
396}
397
398static int
399vmci_transport_notify_pkt_recv_init(
400 struct sock *sk,
401 size_t target,
402 struct vmci_transport_recv_notify_data *data)
403{
404 struct vsock_sock *vsk = vsock_sk(sk);
405
406#ifdef VSOCK_OPTIMIZATION_WAITING_NOTIFY
407 data->consume_head = 0;
408 data->produce_tail = 0;
409#ifdef VSOCK_OPTIMIZATION_FLOW_CONTROL
410 data->notify_on_block = false;
411
412 if (PKT_FIELD(vsk, write_notify_min_window) < target + 1) {
413 PKT_FIELD(vsk, write_notify_min_window) = target + 1;
414 if (PKT_FIELD(vsk, write_notify_window) <
415 PKT_FIELD(vsk, write_notify_min_window)) {
416 /* If the current window is smaller than the new
417 * minimal window size, we need to reevaluate whether
418 * we need to notify the sender. If the number of ready
419 * bytes are smaller than the new window, we need to
420 * send a notification to the sender before we block.
421 */
422
423 PKT_FIELD(vsk, write_notify_window) =
424 PKT_FIELD(vsk, write_notify_min_window);
425 data->notify_on_block = true;
426 }
427 }
428#endif
429#endif
430
431 return 0;
432}
433
434static int
435vmci_transport_notify_pkt_recv_pre_block(
436 struct sock *sk,
437 size_t target,
438 struct vmci_transport_recv_notify_data *data)
439{
440 int err = 0;
441
442 /* Notify our peer that we are waiting for data to read. */
443 if (!send_waiting_read(sk, target)) {
444 err = -EHOSTUNREACH;
445 return err;
446 }
447#ifdef VSOCK_OPTIMIZATION_FLOW_CONTROL
448 if (data->notify_on_block) {
449 err = vmci_transport_send_read_notification(sk);
450 if (err < 0)
451 return err;
452
453 data->notify_on_block = false;
454 }
455#endif
456
457 return err;
458}
459
460static int
461vmci_transport_notify_pkt_recv_pre_dequeue(
462 struct sock *sk,
463 size_t target,
464 struct vmci_transport_recv_notify_data *data)
465{
466 struct vsock_sock *vsk = vsock_sk(sk);
467
468 /* Now consume up to len bytes from the queue. Note that since we have
469 * the socket locked we should copy at least ready bytes.
470 */
471#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
472 vmci_qpair_get_consume_indexes(vmci_trans(vsk)->qpair,
473 &data->produce_tail,
474 &data->consume_head);
475#endif
476
477 return 0;
478}
479
480static int
481vmci_transport_notify_pkt_recv_post_dequeue(
482 struct sock *sk,
483 size_t target,
484 ssize_t copied,
485 bool data_read,
486 struct vmci_transport_recv_notify_data *data)
487{
488 struct vsock_sock *vsk;
489 int err;
490
491 vsk = vsock_sk(sk);
492 err = 0;
493
494 if (data_read) {
495#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
496 /* Detect a wrap-around to maintain queue generation. Note
497 * that this is safe since we hold the socket lock across the
498 * two queue pair operations.
499 */
500 if (copied >=
501 vmci_trans(vsk)->consume_size - data->consume_head)
502 PKT_FIELD(vsk, consume_q_generation)++;
503#endif
504
505 err = vmci_transport_send_read_notification(sk);
506 if (err < 0)
507 return err;
508
509 }
510 return err;
511}
512
513static int
514vmci_transport_notify_pkt_send_init(
515 struct sock *sk,
516 struct vmci_transport_send_notify_data *data)
517{
518#ifdef VSOCK_OPTIMIZATION_WAITING_NOTIFY
519 data->consume_head = 0;
520 data->produce_tail = 0;
521#endif
522
523 return 0;
524}
525
526static int
527vmci_transport_notify_pkt_send_pre_block(
528 struct sock *sk,
529 struct vmci_transport_send_notify_data *data)
530{
531 /* Notify our peer that we are waiting for room to write. */
532 if (!send_waiting_write(sk, 1))
533 return -EHOSTUNREACH;
534
535 return 0;
536}
537
538static int
539vmci_transport_notify_pkt_send_pre_enqueue(
540 struct sock *sk,
541 struct vmci_transport_send_notify_data *data)
542{
543 struct vsock_sock *vsk = vsock_sk(sk);
544
545#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
546 vmci_qpair_get_produce_indexes(vmci_trans(vsk)->qpair,
547 &data->produce_tail,
548 &data->consume_head);
549#endif
550
551 return 0;
552}
553
554static int
555vmci_transport_notify_pkt_send_post_enqueue(
556 struct sock *sk,
557 ssize_t written,
558 struct vmci_transport_send_notify_data *data)
559{
560 int err = 0;
561 struct vsock_sock *vsk;
562 bool sent_wrote = false;
563 int retries = 0;
564
565 vsk = vsock_sk(sk);
566
567#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
568 /* Detect a wrap-around to maintain queue generation. Note that this
569 * is safe since we hold the socket lock across the two queue pair
570 * operations.
571 */
572 if (written >= vmci_trans(vsk)->produce_size - data->produce_tail)
573 PKT_FIELD(vsk, produce_q_generation)++;
574
575#endif
576
577 if (vmci_transport_notify_waiting_read(vsk)) {
578 /* Notify the peer that we have written, retrying the send on
579 * failure up to our maximum value. See the XXX comment for the
580 * corresponding piece of code in StreamRecvmsg() for potential
581 * improvements.
582 */
583 while (!(vsk->peer_shutdown & RCV_SHUTDOWN) &&
584 !sent_wrote &&
585 retries < VMCI_TRANSPORT_MAX_DGRAM_RESENDS) {
586 err = vmci_transport_send_wrote(sk);
587 if (err >= 0)
588 sent_wrote = true;
589
590 retries++;
591 }
592
593 if (retries >= VMCI_TRANSPORT_MAX_DGRAM_RESENDS) {
594 pr_err("%p unable to send wrote notify to peer\n", sk);
595 return err;
596 } else {
597#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
598 PKT_FIELD(vsk, peer_waiting_read) = false;
599#endif
600 }
601 }
602 return err;
603}
604
605static void
606vmci_transport_notify_pkt_handle_pkt(
607 struct sock *sk,
608 struct vmci_transport_packet *pkt,
609 bool bottom_half,
610 struct sockaddr_vm *dst,
611 struct sockaddr_vm *src, bool *pkt_processed)
612{
613 bool processed = false;
614
615 switch (pkt->type) {
616 case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
617 vmci_transport_handle_wrote(sk, pkt, bottom_half, dst, src);
618 processed = true;
619 break;
620 case VMCI_TRANSPORT_PACKET_TYPE_READ:
621 vmci_transport_handle_read(sk, pkt, bottom_half, dst, src);
622 processed = true;
623 break;
624 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
625 vmci_transport_handle_waiting_write(sk, pkt, bottom_half,
626 dst, src);
627 processed = true;
628 break;
629
630 case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
631 vmci_transport_handle_waiting_read(sk, pkt, bottom_half,
632 dst, src);
633 processed = true;
634 break;
635 }
636
637 if (pkt_processed)
638 *pkt_processed = processed;
639}
640
641static void vmci_transport_notify_pkt_process_request(struct sock *sk)
642{
643 struct vsock_sock *vsk = vsock_sk(sk);
644
645 PKT_FIELD(vsk, write_notify_window) = vmci_trans(vsk)->consume_size;
646 if (vmci_trans(vsk)->consume_size <
647 PKT_FIELD(vsk, write_notify_min_window))
648 PKT_FIELD(vsk, write_notify_min_window) =
649 vmci_trans(vsk)->consume_size;
650}
651
652static void vmci_transport_notify_pkt_process_negotiate(struct sock *sk)
653{
654 struct vsock_sock *vsk = vsock_sk(sk);
655
656 PKT_FIELD(vsk, write_notify_window) = vmci_trans(vsk)->consume_size;
657 if (vmci_trans(vsk)->consume_size <
658 PKT_FIELD(vsk, write_notify_min_window))
659 PKT_FIELD(vsk, write_notify_min_window) =
660 vmci_trans(vsk)->consume_size;
661}
662
663/* Socket control packet based operations. */
664struct vmci_transport_notify_ops vmci_transport_notify_pkt_ops = {
665 vmci_transport_notify_pkt_socket_init,
666 vmci_transport_notify_pkt_socket_destruct,
667 vmci_transport_notify_pkt_poll_in,
668 vmci_transport_notify_pkt_poll_out,
669 vmci_transport_notify_pkt_handle_pkt,
670 vmci_transport_notify_pkt_recv_init,
671 vmci_transport_notify_pkt_recv_pre_block,
672 vmci_transport_notify_pkt_recv_pre_dequeue,
673 vmci_transport_notify_pkt_recv_post_dequeue,
674 vmci_transport_notify_pkt_send_init,
675 vmci_transport_notify_pkt_send_pre_block,
676 vmci_transport_notify_pkt_send_pre_enqueue,
677 vmci_transport_notify_pkt_send_post_enqueue,
678 vmci_transport_notify_pkt_process_request,
679 vmci_transport_notify_pkt_process_negotiate,
680};
diff --git a/net/vmw_vsock/vmci_transport_notify.h b/net/vmw_vsock/vmci_transport_notify.h
new file mode 100644
index 000000000000..7df793249b6c
--- /dev/null
+++ b/net/vmw_vsock/vmci_transport_notify.h
@@ -0,0 +1,83 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2009-2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#ifndef __VMCI_TRANSPORT_NOTIFY_H__
17#define __VMCI_TRANSPORT_NOTIFY_H__
18
19#include <linux/types.h>
20#include <linux/vmw_vmci_defs.h>
21#include <linux/vmw_vmci_api.h>
22#include <linux/vm_sockets.h>
23
24#include "vmci_transport.h"
25
26/* Comment this out to compare with old protocol. */
27#define VSOCK_OPTIMIZATION_WAITING_NOTIFY 1
28#if defined(VSOCK_OPTIMIZATION_WAITING_NOTIFY)
29/* Comment this out to remove flow control for "new" protocol */
30#define VSOCK_OPTIMIZATION_FLOW_CONTROL 1
31#endif
32
33#define VMCI_TRANSPORT_MAX_DGRAM_RESENDS 10
34
35struct vmci_transport_recv_notify_data {
36 u64 consume_head;
37 u64 produce_tail;
38 bool notify_on_block;
39};
40
41struct vmci_transport_send_notify_data {
42 u64 consume_head;
43 u64 produce_tail;
44};
45
46/* Socket notification callbacks. */
47struct vmci_transport_notify_ops {
48 void (*socket_init) (struct sock *sk);
49 void (*socket_destruct) (struct vsock_sock *vsk);
50 int (*poll_in) (struct sock *sk, size_t target,
51 bool *data_ready_now);
52 int (*poll_out) (struct sock *sk, size_t target,
53 bool *space_avail_now);
54 void (*handle_notify_pkt) (struct sock *sk,
55 struct vmci_transport_packet *pkt,
56 bool bottom_half, struct sockaddr_vm *dst,
57 struct sockaddr_vm *src,
58 bool *pkt_processed);
59 int (*recv_init) (struct sock *sk, size_t target,
60 struct vmci_transport_recv_notify_data *data);
61 int (*recv_pre_block) (struct sock *sk, size_t target,
62 struct vmci_transport_recv_notify_data *data);
63 int (*recv_pre_dequeue) (struct sock *sk, size_t target,
64 struct vmci_transport_recv_notify_data *data);
65 int (*recv_post_dequeue) (struct sock *sk, size_t target,
66 ssize_t copied, bool data_read,
67 struct vmci_transport_recv_notify_data *data);
68 int (*send_init) (struct sock *sk,
69 struct vmci_transport_send_notify_data *data);
70 int (*send_pre_block) (struct sock *sk,
71 struct vmci_transport_send_notify_data *data);
72 int (*send_pre_enqueue) (struct sock *sk,
73 struct vmci_transport_send_notify_data *data);
74 int (*send_post_enqueue) (struct sock *sk, ssize_t written,
75 struct vmci_transport_send_notify_data *data);
76 void (*process_request) (struct sock *sk);
77 void (*process_negotiate) (struct sock *sk);
78};
79
80extern struct vmci_transport_notify_ops vmci_transport_notify_pkt_ops;
81extern struct vmci_transport_notify_ops vmci_transport_notify_pkt_q_state_ops;
82
83#endif /* __VMCI_TRANSPORT_NOTIFY_H__ */
diff --git a/net/vmw_vsock/vmci_transport_notify_qstate.c b/net/vmw_vsock/vmci_transport_notify_qstate.c
new file mode 100644
index 000000000000..622bd7aa1016
--- /dev/null
+++ b/net/vmw_vsock/vmci_transport_notify_qstate.c
@@ -0,0 +1,438 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2009-2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#include <linux/types.h>
17#include <linux/socket.h>
18#include <linux/stddef.h>
19#include <net/sock.h>
20
21#include "vmci_transport_notify.h"
22
23#define PKT_FIELD(vsk, field_name) \
24 (vmci_trans(vsk)->notify.pkt_q_state.field_name)
25
26static bool vmci_transport_notify_waiting_write(struct vsock_sock *vsk)
27{
28 bool retval;
29 u64 notify_limit;
30
31 if (!PKT_FIELD(vsk, peer_waiting_write))
32 return false;
33
34 /* When the sender blocks, we take that as a sign that the sender is
35 * faster than the receiver. To reduce the transmit rate of the sender,
36 * we delay the sending of the read notification by decreasing the
37 * write_notify_window. The notification is delayed until the number of
38 * bytes used in the queue drops below the write_notify_window.
39 */
40
41 if (!PKT_FIELD(vsk, peer_waiting_write_detected)) {
42 PKT_FIELD(vsk, peer_waiting_write_detected) = true;
43 if (PKT_FIELD(vsk, write_notify_window) < PAGE_SIZE) {
44 PKT_FIELD(vsk, write_notify_window) =
45 PKT_FIELD(vsk, write_notify_min_window);
46 } else {
47 PKT_FIELD(vsk, write_notify_window) -= PAGE_SIZE;
48 if (PKT_FIELD(vsk, write_notify_window) <
49 PKT_FIELD(vsk, write_notify_min_window))
50 PKT_FIELD(vsk, write_notify_window) =
51 PKT_FIELD(vsk, write_notify_min_window);
52
53 }
54 }
55 notify_limit = vmci_trans(vsk)->consume_size -
56 PKT_FIELD(vsk, write_notify_window);
57
58 /* The notify_limit is used to delay notifications in the case where
59 * flow control is enabled. Below the test is expressed in terms of
60 * free space in the queue: if free_space > ConsumeSize -
61 * write_notify_window then notify An alternate way of expressing this
62 * is to rewrite the expression to use the data ready in the receive
63 * queue: if write_notify_window > bufferReady then notify as
64 * free_space == ConsumeSize - bufferReady.
65 */
66
67 retval = vmci_qpair_consume_free_space(vmci_trans(vsk)->qpair) >
68 notify_limit;
69
70 if (retval) {
71 /* Once we notify the peer, we reset the detected flag so the
72 * next wait will again cause a decrease in the window size.
73 */
74
75 PKT_FIELD(vsk, peer_waiting_write_detected) = false;
76 }
77 return retval;
78}
79
80static void
81vmci_transport_handle_read(struct sock *sk,
82 struct vmci_transport_packet *pkt,
83 bool bottom_half,
84 struct sockaddr_vm *dst, struct sockaddr_vm *src)
85{
86 sk->sk_write_space(sk);
87}
88
89static void
90vmci_transport_handle_wrote(struct sock *sk,
91 struct vmci_transport_packet *pkt,
92 bool bottom_half,
93 struct sockaddr_vm *dst, struct sockaddr_vm *src)
94{
95 sk->sk_data_ready(sk, 0);
96}
97
98static void vsock_block_update_write_window(struct sock *sk)
99{
100 struct vsock_sock *vsk = vsock_sk(sk);
101
102 if (PKT_FIELD(vsk, write_notify_window) < vmci_trans(vsk)->consume_size)
103 PKT_FIELD(vsk, write_notify_window) =
104 min(PKT_FIELD(vsk, write_notify_window) + PAGE_SIZE,
105 vmci_trans(vsk)->consume_size);
106}
107
108static int vmci_transport_send_read_notification(struct sock *sk)
109{
110 struct vsock_sock *vsk;
111 bool sent_read;
112 unsigned int retries;
113 int err;
114
115 vsk = vsock_sk(sk);
116 sent_read = false;
117 retries = 0;
118 err = 0;
119
120 if (vmci_transport_notify_waiting_write(vsk)) {
121 /* Notify the peer that we have read, retrying the send on
122 * failure up to our maximum value. XXX For now we just log
123 * the failure, but later we should schedule a work item to
124 * handle the resend until it succeeds. That would require
125 * keeping track of work items in the vsk and cleaning them up
126 * upon socket close.
127 */
128 while (!(vsk->peer_shutdown & RCV_SHUTDOWN) &&
129 !sent_read &&
130 retries < VMCI_TRANSPORT_MAX_DGRAM_RESENDS) {
131 err = vmci_transport_send_read(sk);
132 if (err >= 0)
133 sent_read = true;
134
135 retries++;
136 }
137
138 if (retries >= VMCI_TRANSPORT_MAX_DGRAM_RESENDS && !sent_read)
139 pr_err("%p unable to send read notification to peer\n",
140 sk);
141 else
142 PKT_FIELD(vsk, peer_waiting_write) = false;
143
144 }
145 return err;
146}
147
148static void vmci_transport_notify_pkt_socket_init(struct sock *sk)
149{
150 struct vsock_sock *vsk = vsock_sk(sk);
151
152 PKT_FIELD(vsk, write_notify_window) = PAGE_SIZE;
153 PKT_FIELD(vsk, write_notify_min_window) = PAGE_SIZE;
154 PKT_FIELD(vsk, peer_waiting_write) = false;
155 PKT_FIELD(vsk, peer_waiting_write_detected) = false;
156}
157
158static void vmci_transport_notify_pkt_socket_destruct(struct vsock_sock *vsk)
159{
160 PKT_FIELD(vsk, write_notify_window) = PAGE_SIZE;
161 PKT_FIELD(vsk, write_notify_min_window) = PAGE_SIZE;
162 PKT_FIELD(vsk, peer_waiting_write) = false;
163 PKT_FIELD(vsk, peer_waiting_write_detected) = false;
164}
165
166static int
167vmci_transport_notify_pkt_poll_in(struct sock *sk,
168 size_t target, bool *data_ready_now)
169{
170 struct vsock_sock *vsk = vsock_sk(sk);
171
172 if (vsock_stream_has_data(vsk)) {
173 *data_ready_now = true;
174 } else {
175 /* We can't read right now because there is nothing in the
176 * queue. Ask for notifications when there is something to
177 * read.
178 */
179 if (sk->sk_state == SS_CONNECTED)
180 vsock_block_update_write_window(sk);
181 *data_ready_now = false;
182 }
183
184 return 0;
185}
186
187static int
188vmci_transport_notify_pkt_poll_out(struct sock *sk,
189 size_t target, bool *space_avail_now)
190{
191 s64 produce_q_free_space;
192 struct vsock_sock *vsk = vsock_sk(sk);
193
194 produce_q_free_space = vsock_stream_has_space(vsk);
195 if (produce_q_free_space > 0) {
196 *space_avail_now = true;
197 return 0;
198 } else if (produce_q_free_space == 0) {
199 /* This is a connected socket but we can't currently send data.
200 * Nothing else to do.
201 */
202 *space_avail_now = false;
203 }
204
205 return 0;
206}
207
208static int
209vmci_transport_notify_pkt_recv_init(
210 struct sock *sk,
211 size_t target,
212 struct vmci_transport_recv_notify_data *data)
213{
214 struct vsock_sock *vsk = vsock_sk(sk);
215
216 data->consume_head = 0;
217 data->produce_tail = 0;
218 data->notify_on_block = false;
219
220 if (PKT_FIELD(vsk, write_notify_min_window) < target + 1) {
221 PKT_FIELD(vsk, write_notify_min_window) = target + 1;
222 if (PKT_FIELD(vsk, write_notify_window) <
223 PKT_FIELD(vsk, write_notify_min_window)) {
224 /* If the current window is smaller than the new
225 * minimal window size, we need to reevaluate whether
226 * we need to notify the sender. If the number of ready
227 * bytes are smaller than the new window, we need to
228 * send a notification to the sender before we block.
229 */
230
231 PKT_FIELD(vsk, write_notify_window) =
232 PKT_FIELD(vsk, write_notify_min_window);
233 data->notify_on_block = true;
234 }
235 }
236
237 return 0;
238}
239
240static int
241vmci_transport_notify_pkt_recv_pre_block(
242 struct sock *sk,
243 size_t target,
244 struct vmci_transport_recv_notify_data *data)
245{
246 int err = 0;
247
248 vsock_block_update_write_window(sk);
249
250 if (data->notify_on_block) {
251 err = vmci_transport_send_read_notification(sk);
252 if (err < 0)
253 return err;
254 data->notify_on_block = false;
255 }
256
257 return err;
258}
259
260static int
261vmci_transport_notify_pkt_recv_post_dequeue(
262 struct sock *sk,
263 size_t target,
264 ssize_t copied,
265 bool data_read,
266 struct vmci_transport_recv_notify_data *data)
267{
268 struct vsock_sock *vsk;
269 int err;
270 bool was_full = false;
271 u64 free_space;
272
273 vsk = vsock_sk(sk);
274 err = 0;
275
276 if (data_read) {
277 smp_mb();
278
279 free_space =
280 vmci_qpair_consume_free_space(vmci_trans(vsk)->qpair);
281 was_full = free_space == copied;
282
283 if (was_full)
284 PKT_FIELD(vsk, peer_waiting_write) = true;
285
286 err = vmci_transport_send_read_notification(sk);
287 if (err < 0)
288 return err;
289
290 /* See the comment in
291 * vmci_transport_notify_pkt_send_post_enqueue().
292 */
293 sk->sk_data_ready(sk, 0);
294 }
295
296 return err;
297}
298
299static int
300vmci_transport_notify_pkt_send_init(
301 struct sock *sk,
302 struct vmci_transport_send_notify_data *data)
303{
304 data->consume_head = 0;
305 data->produce_tail = 0;
306
307 return 0;
308}
309
310static int
311vmci_transport_notify_pkt_send_post_enqueue(
312 struct sock *sk,
313 ssize_t written,
314 struct vmci_transport_send_notify_data *data)
315{
316 int err = 0;
317 struct vsock_sock *vsk;
318 bool sent_wrote = false;
319 bool was_empty;
320 int retries = 0;
321
322 vsk = vsock_sk(sk);
323
324 smp_mb();
325
326 was_empty =
327 vmci_qpair_produce_buf_ready(vmci_trans(vsk)->qpair) == written;
328 if (was_empty) {
329 while (!(vsk->peer_shutdown & RCV_SHUTDOWN) &&
330 !sent_wrote &&
331 retries < VMCI_TRANSPORT_MAX_DGRAM_RESENDS) {
332 err = vmci_transport_send_wrote(sk);
333 if (err >= 0)
334 sent_wrote = true;
335
336 retries++;
337 }
338 }
339
340 if (retries >= VMCI_TRANSPORT_MAX_DGRAM_RESENDS && !sent_wrote) {
341 pr_err("%p unable to send wrote notification to peer\n",
342 sk);
343 return err;
344 }
345
346 return err;
347}
348
349static void
350vmci_transport_notify_pkt_handle_pkt(
351 struct sock *sk,
352 struct vmci_transport_packet *pkt,
353 bool bottom_half,
354 struct sockaddr_vm *dst,
355 struct sockaddr_vm *src, bool *pkt_processed)
356{
357 bool processed = false;
358
359 switch (pkt->type) {
360 case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
361 vmci_transport_handle_wrote(sk, pkt, bottom_half, dst, src);
362 processed = true;
363 break;
364 case VMCI_TRANSPORT_PACKET_TYPE_READ:
365 vmci_transport_handle_read(sk, pkt, bottom_half, dst, src);
366 processed = true;
367 break;
368 }
369
370 if (pkt_processed)
371 *pkt_processed = processed;
372}
373
374static void vmci_transport_notify_pkt_process_request(struct sock *sk)
375{
376 struct vsock_sock *vsk = vsock_sk(sk);
377
378 PKT_FIELD(vsk, write_notify_window) = vmci_trans(vsk)->consume_size;
379 if (vmci_trans(vsk)->consume_size <
380 PKT_FIELD(vsk, write_notify_min_window))
381 PKT_FIELD(vsk, write_notify_min_window) =
382 vmci_trans(vsk)->consume_size;
383}
384
385static void vmci_transport_notify_pkt_process_negotiate(struct sock *sk)
386{
387 struct vsock_sock *vsk = vsock_sk(sk);
388
389 PKT_FIELD(vsk, write_notify_window) = vmci_trans(vsk)->consume_size;
390 if (vmci_trans(vsk)->consume_size <
391 PKT_FIELD(vsk, write_notify_min_window))
392 PKT_FIELD(vsk, write_notify_min_window) =
393 vmci_trans(vsk)->consume_size;
394}
395
396static int
397vmci_transport_notify_pkt_recv_pre_dequeue(
398 struct sock *sk,
399 size_t target,
400 struct vmci_transport_recv_notify_data *data)
401{
402 return 0; /* NOP for QState. */
403}
404
405static int
406vmci_transport_notify_pkt_send_pre_block(
407 struct sock *sk,
408 struct vmci_transport_send_notify_data *data)
409{
410 return 0; /* NOP for QState. */
411}
412
413static int
414vmci_transport_notify_pkt_send_pre_enqueue(
415 struct sock *sk,
416 struct vmci_transport_send_notify_data *data)
417{
418 return 0; /* NOP for QState. */
419}
420
421/* Socket always on control packet based operations. */
422struct vmci_transport_notify_ops vmci_transport_notify_pkt_q_state_ops = {
423 vmci_transport_notify_pkt_socket_init,
424 vmci_transport_notify_pkt_socket_destruct,
425 vmci_transport_notify_pkt_poll_in,
426 vmci_transport_notify_pkt_poll_out,
427 vmci_transport_notify_pkt_handle_pkt,
428 vmci_transport_notify_pkt_recv_init,
429 vmci_transport_notify_pkt_recv_pre_block,
430 vmci_transport_notify_pkt_recv_pre_dequeue,
431 vmci_transport_notify_pkt_recv_post_dequeue,
432 vmci_transport_notify_pkt_send_init,
433 vmci_transport_notify_pkt_send_pre_block,
434 vmci_transport_notify_pkt_send_pre_enqueue,
435 vmci_transport_notify_pkt_send_post_enqueue,
436 vmci_transport_notify_pkt_process_request,
437 vmci_transport_notify_pkt_process_negotiate,
438};
diff --git a/net/vmw_vsock/vsock_addr.c b/net/vmw_vsock/vsock_addr.c
new file mode 100644
index 000000000000..b7df1aea7c59
--- /dev/null
+++ b/net/vmw_vsock/vsock_addr.c
@@ -0,0 +1,86 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2007-2012 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#include <linux/types.h>
17#include <linux/socket.h>
18#include <linux/stddef.h>
19#include <net/sock.h>
20
21#include "vsock_addr.h"
22
23void vsock_addr_init(struct sockaddr_vm *addr, u32 cid, u32 port)
24{
25 memset(addr, 0, sizeof(*addr));
26 addr->svm_family = AF_VSOCK;
27 addr->svm_cid = cid;
28 addr->svm_port = port;
29}
30EXPORT_SYMBOL_GPL(vsock_addr_init);
31
32int vsock_addr_validate(const struct sockaddr_vm *addr)
33{
34 if (!addr)
35 return -EFAULT;
36
37 if (addr->svm_family != AF_VSOCK)
38 return -EAFNOSUPPORT;
39
40 if (addr->svm_zero[0] != 0)
41 return -EINVAL;
42
43 return 0;
44}
45EXPORT_SYMBOL_GPL(vsock_addr_validate);
46
47bool vsock_addr_bound(const struct sockaddr_vm *addr)
48{
49 return addr->svm_port != VMADDR_PORT_ANY;
50}
51EXPORT_SYMBOL_GPL(vsock_addr_bound);
52
53void vsock_addr_unbind(struct sockaddr_vm *addr)
54{
55 vsock_addr_init(addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
56}
57EXPORT_SYMBOL_GPL(vsock_addr_unbind);
58
59bool vsock_addr_equals_addr(const struct sockaddr_vm *addr,
60 const struct sockaddr_vm *other)
61{
62 return addr->svm_cid == other->svm_cid &&
63 addr->svm_port == other->svm_port;
64}
65EXPORT_SYMBOL_GPL(vsock_addr_equals_addr);
66
67bool vsock_addr_equals_addr_any(const struct sockaddr_vm *addr,
68 const struct sockaddr_vm *other)
69{
70 return (addr->svm_cid == VMADDR_CID_ANY ||
71 other->svm_cid == VMADDR_CID_ANY ||
72 addr->svm_cid == other->svm_cid) &&
73 addr->svm_port == other->svm_port;
74}
75EXPORT_SYMBOL_GPL(vsock_addr_equals_addr_any);
76
77int vsock_addr_cast(const struct sockaddr *addr,
78 size_t len, struct sockaddr_vm **out_addr)
79{
80 if (len < sizeof(**out_addr))
81 return -EFAULT;
82
83 *out_addr = (struct sockaddr_vm *)addr;
84 return vsock_addr_validate(*out_addr);
85}
86EXPORT_SYMBOL_GPL(vsock_addr_cast);
diff --git a/net/vmw_vsock/vsock_addr.h b/net/vmw_vsock/vsock_addr.h
new file mode 100644
index 000000000000..cdfbcefdf843
--- /dev/null
+++ b/net/vmw_vsock/vsock_addr.h
@@ -0,0 +1,32 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#ifndef _VSOCK_ADDR_H_
17#define _VSOCK_ADDR_H_
18
19#include <linux/vm_sockets.h>
20
21void vsock_addr_init(struct sockaddr_vm *addr, u32 cid, u32 port);
22int vsock_addr_validate(const struct sockaddr_vm *addr);
23bool vsock_addr_bound(const struct sockaddr_vm *addr);
24void vsock_addr_unbind(struct sockaddr_vm *addr);
25bool vsock_addr_equals_addr(const struct sockaddr_vm *addr,
26 const struct sockaddr_vm *other);
27bool vsock_addr_equals_addr_any(const struct sockaddr_vm *addr,
28 const struct sockaddr_vm *other);
29int vsock_addr_cast(const struct sockaddr *addr, size_t len,
30 struct sockaddr_vm **out_addr);
31
32#endif
diff --git a/net/vmw_vsock/vsock_version.h b/net/vmw_vsock/vsock_version.h
new file mode 100644
index 000000000000..4df7f5e2151c
--- /dev/null
+++ b/net/vmw_vsock/vsock_version.h
@@ -0,0 +1,22 @@
1/*
2 * VMware vSockets Driver
3 *
4 * Copyright (C) 2011-2012 VMware, Inc. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 */
15
16#ifndef _VSOCK_VERSION_H_
17#define _VSOCK_VERSION_H_
18
19#define VSOCK_DRIVER_VERSION_PARTS { 1, 0, 0, 0 }
20#define VSOCK_DRIVER_VERSION_STRING "1.0.0.0-k"
21
22#endif /* _VSOCK_VERSION_H_ */
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-21 01:33:31 -0500