1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET An implementation of the SOCKET network access protocol.
5 * Version: @(#)socket.c 1.1.93 18/02/95
7 * Authors: Orest Zborowski, <obz@Kodak.COM>
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
14 * Alan Cox : verify_area() fixes
15 * Alan Cox : Removed DDI
16 * Jonathan Kamens : SOCK_DGRAM reconnect bug
17 * Alan Cox : Moved a load of checks to the very
19 * Alan Cox : Move address structures to/from user
20 * mode above the protocol layers.
21 * Rob Janssen : Allow 0 length sends.
22 * Alan Cox : Asynchronous I/O support (cribbed from the
24 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
25 * Jeff Uphoff : Made max number of sockets command-line
27 * Matti Aarnio : Made the number of sockets dynamic,
28 * to be allocated when needed, and mr.
29 * Uphoff's max is used as max to be
30 * allowed to allocate.
31 * Linus : Argh. removed all the socket allocation
32 * altogether: it's in the inode now.
33 * Alan Cox : Made sock_alloc()/sock_release() public
34 * for NetROM and future kernel nfsd type
36 * Alan Cox : sendmsg/recvmsg basics.
37 * Tom Dyas : Export net symbols.
38 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
39 * Alan Cox : Added thread locking to sys_* calls
40 * for sockets. May have errors at the
42 * Kevin Buhr : Fixed the dumb errors in the above.
43 * Andi Kleen : Some small cleanups, optimizations,
44 * and fixed a copy_from_user() bug.
45 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
46 * Tigran Aivazian : Made listen(2) backlog sanity checks
47 * protocol-independent
49 * This module is effectively the top level interface to the BSD socket
52 * Based upon Swansea University Computer Society NET3.039
55 #include <linux/bpf-cgroup.h>
56 #include <linux/ethtool.h>
58 #include <linux/socket.h>
59 #include <linux/file.h>
60 #include <linux/splice.h>
61 #include <linux/net.h>
62 #include <linux/interrupt.h>
63 #include <linux/thread_info.h>
64 #include <linux/rcupdate.h>
65 #include <linux/netdevice.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
68 #include <linux/mutex.h>
69 #include <linux/if_bridge.h>
70 #include <linux/if_vlan.h>
71 #include <linux/ptp_classify.h>
72 #include <linux/init.h>
73 #include <linux/poll.h>
74 #include <linux/cache.h>
75 #include <linux/module.h>
76 #include <linux/highmem.h>
77 #include <linux/mount.h>
78 #include <linux/pseudo_fs.h>
79 #include <linux/security.h>
80 #include <linux/syscalls.h>
81 #include <linux/compat.h>
82 #include <linux/kmod.h>
83 #include <linux/audit.h>
84 #include <linux/wireless.h>
85 #include <linux/nsproxy.h>
86 #include <linux/magic.h>
87 #include <linux/slab.h>
88 #include <linux/xattr.h>
89 #include <linux/nospec.h>
90 #include <linux/indirect_call_wrapper.h>
92 #include <linux/uaccess.h>
93 #include <asm/unistd.h>
95 #include <net/compat.h>
97 #include <net/cls_cgroup.h>
100 #include <linux/netfilter.h>
102 #include <linux/if_tun.h>
103 #include <linux/ipv6_route.h>
104 #include <linux/route.h>
105 #include <linux/termios.h>
106 #include <linux/sockios.h>
107 #include <net/busy_poll.h>
108 #include <linux/errqueue.h>
109 #include <linux/ptp_clock_kernel.h>
110 #include <trace/events/sock.h>
112 #ifdef CONFIG_NET_RX_BUSY_POLL
113 unsigned int sysctl_net_busy_read __read_mostly;
114 unsigned int sysctl_net_busy_poll __read_mostly;
117 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
118 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
119 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
121 static int sock_close(struct inode *inode, struct file *file);
122 static __poll_t sock_poll(struct file *file,
123 struct poll_table_struct *wait);
124 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
126 static long compat_sock_ioctl(struct file *file,
127 unsigned int cmd, unsigned long arg);
129 static int sock_fasync(int fd, struct file *filp, int on);
130 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
131 struct pipe_inode_info *pipe, size_t len,
133 static void sock_splice_eof(struct file *file);
135 #ifdef CONFIG_PROC_FS
136 static void sock_show_fdinfo(struct seq_file *m, struct file *f)
138 struct socket *sock = f->private_data;
139 const struct proto_ops *ops = READ_ONCE(sock->ops);
141 if (ops->show_fdinfo)
142 ops->show_fdinfo(m, sock);
145 #define sock_show_fdinfo NULL
149 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
150 * in the operation structures but are done directly via the socketcall() multiplexor.
153 static const struct file_operations socket_file_ops = {
154 .owner = THIS_MODULE,
156 .read_iter = sock_read_iter,
157 .write_iter = sock_write_iter,
159 .unlocked_ioctl = sock_ioctl,
161 .compat_ioctl = compat_sock_ioctl,
164 .release = sock_close,
165 .fasync = sock_fasync,
166 .splice_write = splice_to_socket,
167 .splice_read = sock_splice_read,
168 .splice_eof = sock_splice_eof,
169 .show_fdinfo = sock_show_fdinfo,
172 static const char * const pf_family_names[] = {
173 [PF_UNSPEC] = "PF_UNSPEC",
174 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
175 [PF_INET] = "PF_INET",
176 [PF_AX25] = "PF_AX25",
178 [PF_APPLETALK] = "PF_APPLETALK",
179 [PF_NETROM] = "PF_NETROM",
180 [PF_BRIDGE] = "PF_BRIDGE",
181 [PF_ATMPVC] = "PF_ATMPVC",
183 [PF_INET6] = "PF_INET6",
184 [PF_ROSE] = "PF_ROSE",
185 [PF_DECnet] = "PF_DECnet",
186 [PF_NETBEUI] = "PF_NETBEUI",
187 [PF_SECURITY] = "PF_SECURITY",
189 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
190 [PF_PACKET] = "PF_PACKET",
192 [PF_ECONET] = "PF_ECONET",
193 [PF_ATMSVC] = "PF_ATMSVC",
196 [PF_IRDA] = "PF_IRDA",
197 [PF_PPPOX] = "PF_PPPOX",
198 [PF_WANPIPE] = "PF_WANPIPE",
201 [PF_MPLS] = "PF_MPLS",
203 [PF_TIPC] = "PF_TIPC",
204 [PF_BLUETOOTH] = "PF_BLUETOOTH",
205 [PF_IUCV] = "PF_IUCV",
206 [PF_RXRPC] = "PF_RXRPC",
207 [PF_ISDN] = "PF_ISDN",
208 [PF_PHONET] = "PF_PHONET",
209 [PF_IEEE802154] = "PF_IEEE802154",
210 [PF_CAIF] = "PF_CAIF",
213 [PF_VSOCK] = "PF_VSOCK",
215 [PF_QIPCRTR] = "PF_QIPCRTR",
218 [PF_MCTP] = "PF_MCTP",
222 * The protocol list. Each protocol is registered in here.
225 static DEFINE_SPINLOCK(net_family_lock);
226 static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
230 * Move socket addresses back and forth across the kernel/user
231 * divide and look after the messy bits.
235 * move_addr_to_kernel - copy a socket address into kernel space
236 * @uaddr: Address in user space
237 * @kaddr: Address in kernel space
238 * @ulen: Length in user space
240 * The address is copied into kernel space. If the provided address is
241 * too long an error code of -EINVAL is returned. If the copy gives
242 * invalid addresses -EFAULT is returned. On a success 0 is returned.
245 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
247 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
251 if (copy_from_user(kaddr, uaddr, ulen))
253 return audit_sockaddr(ulen, kaddr);
257 * move_addr_to_user - copy an address to user space
258 * @kaddr: kernel space address
259 * @klen: length of address in kernel
260 * @uaddr: user space address
261 * @ulen: pointer to user length field
263 * The value pointed to by ulen on entry is the buffer length available.
264 * This is overwritten with the buffer space used. -EINVAL is returned
265 * if an overlong buffer is specified or a negative buffer size. -EFAULT
266 * is returned if either the buffer or the length field are not
268 * After copying the data up to the limit the user specifies, the true
269 * length of the data is written over the length limit the user
270 * specified. Zero is returned for a success.
273 static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
274 void __user *uaddr, int __user *ulen)
279 BUG_ON(klen > sizeof(struct sockaddr_storage));
280 err = get_user(len, ulen);
288 if (audit_sockaddr(klen, kaddr))
290 if (copy_to_user(uaddr, kaddr, len))
294 * "fromlen shall refer to the value before truncation.."
297 return __put_user(klen, ulen);
300 static struct kmem_cache *sock_inode_cachep __ro_after_init;
302 static struct inode *sock_alloc_inode(struct super_block *sb)
304 struct socket_alloc *ei;
306 ei = alloc_inode_sb(sb, sock_inode_cachep, GFP_KERNEL);
309 init_waitqueue_head(&ei->socket.wq.wait);
310 ei->socket.wq.fasync_list = NULL;
311 ei->socket.wq.flags = 0;
313 ei->socket.state = SS_UNCONNECTED;
314 ei->socket.flags = 0;
315 ei->socket.ops = NULL;
316 ei->socket.sk = NULL;
317 ei->socket.file = NULL;
319 return &ei->vfs_inode;
322 static void sock_free_inode(struct inode *inode)
324 struct socket_alloc *ei;
326 ei = container_of(inode, struct socket_alloc, vfs_inode);
327 kmem_cache_free(sock_inode_cachep, ei);
330 static void init_once(void *foo)
332 struct socket_alloc *ei = (struct socket_alloc *)foo;
334 inode_init_once(&ei->vfs_inode);
337 static void init_inodecache(void)
339 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
340 sizeof(struct socket_alloc),
342 (SLAB_HWCACHE_ALIGN |
343 SLAB_RECLAIM_ACCOUNT |
344 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
346 BUG_ON(sock_inode_cachep == NULL);
349 static const struct super_operations sockfs_ops = {
350 .alloc_inode = sock_alloc_inode,
351 .free_inode = sock_free_inode,
352 .statfs = simple_statfs,
356 * sockfs_dname() is called from d_path().
358 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
360 return dynamic_dname(buffer, buflen, "socket:[%lu]",
361 d_inode(dentry)->i_ino);
364 static const struct dentry_operations sockfs_dentry_operations = {
365 .d_dname = sockfs_dname,
368 static int sockfs_xattr_get(const struct xattr_handler *handler,
369 struct dentry *dentry, struct inode *inode,
370 const char *suffix, void *value, size_t size)
373 if (dentry->d_name.len + 1 > size)
375 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
377 return dentry->d_name.len + 1;
380 #define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
381 #define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
382 #define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
384 static const struct xattr_handler sockfs_xattr_handler = {
385 .name = XATTR_NAME_SOCKPROTONAME,
386 .get = sockfs_xattr_get,
389 static int sockfs_security_xattr_set(const struct xattr_handler *handler,
390 struct mnt_idmap *idmap,
391 struct dentry *dentry, struct inode *inode,
392 const char *suffix, const void *value,
393 size_t size, int flags)
395 /* Handled by LSM. */
399 static const struct xattr_handler sockfs_security_xattr_handler = {
400 .prefix = XATTR_SECURITY_PREFIX,
401 .set = sockfs_security_xattr_set,
404 static const struct xattr_handler *sockfs_xattr_handlers[] = {
405 &sockfs_xattr_handler,
406 &sockfs_security_xattr_handler,
410 static int sockfs_init_fs_context(struct fs_context *fc)
412 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
415 ctx->ops = &sockfs_ops;
416 ctx->dops = &sockfs_dentry_operations;
417 ctx->xattr = sockfs_xattr_handlers;
421 static struct vfsmount *sock_mnt __read_mostly;
423 static struct file_system_type sock_fs_type = {
425 .init_fs_context = sockfs_init_fs_context,
426 .kill_sb = kill_anon_super,
430 * Obtains the first available file descriptor and sets it up for use.
432 * These functions create file structures and maps them to fd space
433 * of the current process. On success it returns file descriptor
434 * and file struct implicitly stored in sock->file.
435 * Note that another thread may close file descriptor before we return
436 * from this function. We use the fact that now we do not refer
437 * to socket after mapping. If one day we will need it, this
438 * function will increment ref. count on file by 1.
440 * In any case returned fd MAY BE not valid!
441 * This race condition is unavoidable
442 * with shared fd spaces, we cannot solve it inside kernel,
443 * but we take care of internal coherence yet.
447 * sock_alloc_file - Bind a &socket to a &file
449 * @flags: file status flags
450 * @dname: protocol name
452 * Returns the &file bound with @sock, implicitly storing it
453 * in sock->file. If dname is %NULL, sets to "".
455 * On failure @sock is released, and an ERR pointer is returned.
457 * This function uses GFP_KERNEL internally.
460 struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
465 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
467 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
468 O_RDWR | (flags & O_NONBLOCK),
475 file->f_mode |= FMODE_NOWAIT;
477 file->private_data = sock;
478 stream_open(SOCK_INODE(sock), file);
481 EXPORT_SYMBOL(sock_alloc_file);
483 static int sock_map_fd(struct socket *sock, int flags)
485 struct file *newfile;
486 int fd = get_unused_fd_flags(flags);
487 if (unlikely(fd < 0)) {
492 newfile = sock_alloc_file(sock, flags, NULL);
493 if (!IS_ERR(newfile)) {
494 fd_install(fd, newfile);
499 return PTR_ERR(newfile);
503 * sock_from_file - Return the &socket bounded to @file.
506 * On failure returns %NULL.
509 struct socket *sock_from_file(struct file *file)
511 if (file->f_op == &socket_file_ops)
512 return file->private_data; /* set in sock_alloc_file */
516 EXPORT_SYMBOL(sock_from_file);
519 * sockfd_lookup - Go from a file number to its socket slot
521 * @err: pointer to an error code return
523 * The file handle passed in is locked and the socket it is bound
524 * to is returned. If an error occurs the err pointer is overwritten
525 * with a negative errno code and NULL is returned. The function checks
526 * for both invalid handles and passing a handle which is not a socket.
528 * On a success the socket object pointer is returned.
531 struct socket *sockfd_lookup(int fd, int *err)
542 sock = sock_from_file(file);
549 EXPORT_SYMBOL(sockfd_lookup);
551 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
553 struct fd f = fdget(fd);
558 sock = sock_from_file(f.file);
560 *fput_needed = f.flags & FDPUT_FPUT;
569 static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
575 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
585 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
590 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
597 static int sockfs_setattr(struct mnt_idmap *idmap,
598 struct dentry *dentry, struct iattr *iattr)
600 int err = simple_setattr(&nop_mnt_idmap, dentry, iattr);
602 if (!err && (iattr->ia_valid & ATTR_UID)) {
603 struct socket *sock = SOCKET_I(d_inode(dentry));
606 sock->sk->sk_uid = iattr->ia_uid;
614 static const struct inode_operations sockfs_inode_ops = {
615 .listxattr = sockfs_listxattr,
616 .setattr = sockfs_setattr,
620 * sock_alloc - allocate a socket
622 * Allocate a new inode and socket object. The two are bound together
623 * and initialised. The socket is then returned. If we are out of inodes
624 * NULL is returned. This functions uses GFP_KERNEL internally.
627 struct socket *sock_alloc(void)
632 inode = new_inode_pseudo(sock_mnt->mnt_sb);
636 sock = SOCKET_I(inode);
638 inode->i_ino = get_next_ino();
639 inode->i_mode = S_IFSOCK | S_IRWXUGO;
640 inode->i_uid = current_fsuid();
641 inode->i_gid = current_fsgid();
642 inode->i_op = &sockfs_inode_ops;
646 EXPORT_SYMBOL(sock_alloc);
648 static void __sock_release(struct socket *sock, struct inode *inode)
650 const struct proto_ops *ops = READ_ONCE(sock->ops);
653 struct module *owner = ops->owner;
665 if (sock->wq.fasync_list)
666 pr_err("%s: fasync list not empty!\n", __func__);
669 iput(SOCK_INODE(sock));
676 * sock_release - close a socket
677 * @sock: socket to close
679 * The socket is released from the protocol stack if it has a release
680 * callback, and the inode is then released if the socket is bound to
681 * an inode not a file.
683 void sock_release(struct socket *sock)
685 __sock_release(sock, NULL);
687 EXPORT_SYMBOL(sock_release);
689 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
691 u8 flags = *tx_flags;
693 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE) {
694 flags |= SKBTX_HW_TSTAMP;
696 /* PTP hardware clocks can provide a free running cycle counter
697 * as a time base for virtual clocks. Tell driver to use the
698 * free running cycle counter for timestamp if socket is bound
701 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
702 flags |= SKBTX_HW_TSTAMP_USE_CYCLES;
705 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
706 flags |= SKBTX_SW_TSTAMP;
708 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
709 flags |= SKBTX_SCHED_TSTAMP;
713 EXPORT_SYMBOL(__sock_tx_timestamp);
715 INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
717 INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
720 static noinline void call_trace_sock_send_length(struct sock *sk, int ret,
723 trace_sock_send_length(sk, ret, 0);
726 static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
728 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->sendmsg, inet6_sendmsg,
729 inet_sendmsg, sock, msg,
731 BUG_ON(ret == -EIOCBQUEUED);
733 if (trace_sock_send_length_enabled())
734 call_trace_sock_send_length(sock->sk, ret, 0);
739 * sock_sendmsg - send a message through @sock
741 * @msg: message to send
743 * Sends @msg through @sock, passing through LSM.
744 * Returns the number of bytes sent, or an error code.
746 int sock_sendmsg(struct socket *sock, struct msghdr *msg)
748 int err = security_socket_sendmsg(sock, msg,
751 return err ?: sock_sendmsg_nosec(sock, msg);
753 EXPORT_SYMBOL(sock_sendmsg);
756 * kernel_sendmsg - send a message through @sock (kernel-space)
758 * @msg: message header
760 * @num: vec array length
761 * @size: total message data size
763 * Builds the message data with @vec and sends it through @sock.
764 * Returns the number of bytes sent, or an error code.
767 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
768 struct kvec *vec, size_t num, size_t size)
770 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
771 return sock_sendmsg(sock, msg);
773 EXPORT_SYMBOL(kernel_sendmsg);
776 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
778 * @msg: message header
779 * @vec: output s/g array
780 * @num: output s/g array length
781 * @size: total message data size
783 * Builds the message data with @vec and sends it through @sock.
784 * Returns the number of bytes sent, or an error code.
785 * Caller must hold @sk.
788 int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
789 struct kvec *vec, size_t num, size_t size)
791 struct socket *sock = sk->sk_socket;
792 const struct proto_ops *ops = READ_ONCE(sock->ops);
794 if (!ops->sendmsg_locked)
795 return sock_no_sendmsg_locked(sk, msg, size);
797 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
799 return ops->sendmsg_locked(sk, msg, msg_data_left(msg));
801 EXPORT_SYMBOL(kernel_sendmsg_locked);
803 static bool skb_is_err_queue(const struct sk_buff *skb)
805 /* pkt_type of skbs enqueued on the error queue are set to
806 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
807 * in recvmsg, since skbs received on a local socket will never
808 * have a pkt_type of PACKET_OUTGOING.
810 return skb->pkt_type == PACKET_OUTGOING;
813 /* On transmit, software and hardware timestamps are returned independently.
814 * As the two skb clones share the hardware timestamp, which may be updated
815 * before the software timestamp is received, a hardware TX timestamp may be
816 * returned only if there is no software TX timestamp. Ignore false software
817 * timestamps, which may be made in the __sock_recv_timestamp() call when the
818 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
819 * hardware timestamp.
821 static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
823 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
826 static ktime_t get_timestamp(struct sock *sk, struct sk_buff *skb, int *if_index)
828 bool cycles = sk->sk_tsflags & SOF_TIMESTAMPING_BIND_PHC;
829 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
830 struct net_device *orig_dev;
834 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
836 *if_index = orig_dev->ifindex;
837 hwtstamp = netdev_get_tstamp(orig_dev, shhwtstamps, cycles);
839 hwtstamp = shhwtstamps->hwtstamp;
846 static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb,
849 struct scm_ts_pktinfo ts_pktinfo;
850 struct net_device *orig_dev;
852 if (!skb_mac_header_was_set(skb))
855 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
859 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
861 if_index = orig_dev->ifindex;
864 ts_pktinfo.if_index = if_index;
866 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
867 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
868 sizeof(ts_pktinfo), &ts_pktinfo);
872 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
874 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
877 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
878 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
879 struct scm_timestamping_internal tss;
881 int empty = 1, false_tstamp = 0;
882 struct skb_shared_hwtstamps *shhwtstamps =
887 /* Race occurred between timestamp enabling and packet
888 receiving. Fill in the current time for now. */
889 if (need_software_tstamp && skb->tstamp == 0) {
890 __net_timestamp(skb);
894 if (need_software_tstamp) {
895 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
897 struct __kernel_sock_timeval tv;
899 skb_get_new_timestamp(skb, &tv);
900 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
903 struct __kernel_old_timeval tv;
905 skb_get_timestamp(skb, &tv);
906 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
911 struct __kernel_timespec ts;
913 skb_get_new_timestampns(skb, &ts);
914 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
917 struct __kernel_old_timespec ts;
919 skb_get_timestampns(skb, &ts);
920 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
926 memset(&tss, 0, sizeof(tss));
927 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
928 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
931 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
932 !skb_is_swtx_tstamp(skb, false_tstamp)) {
934 if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP_NETDEV)
935 hwtstamp = get_timestamp(sk, skb, &if_index);
937 hwtstamp = shhwtstamps->hwtstamp;
939 if (sk->sk_tsflags & SOF_TIMESTAMPING_BIND_PHC)
940 hwtstamp = ptp_convert_timestamp(&hwtstamp,
943 if (ktime_to_timespec64_cond(hwtstamp, tss.ts + 2)) {
946 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
947 !skb_is_err_queue(skb))
948 put_ts_pktinfo(msg, skb, if_index);
952 if (sock_flag(sk, SOCK_TSTAMP_NEW))
953 put_cmsg_scm_timestamping64(msg, &tss);
955 put_cmsg_scm_timestamping(msg, &tss);
957 if (skb_is_err_queue(skb) && skb->len &&
958 SKB_EXT_ERR(skb)->opt_stats)
959 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
960 skb->len, skb->data);
963 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
965 #ifdef CONFIG_WIRELESS
966 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
971 if (!sock_flag(sk, SOCK_WIFI_STATUS))
973 if (!skb->wifi_acked_valid)
976 ack = skb->wifi_acked;
978 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
980 EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
983 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
986 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
987 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
988 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
991 static void sock_recv_mark(struct msghdr *msg, struct sock *sk,
994 if (sock_flag(sk, SOCK_RCVMARK) && skb) {
995 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */
996 __u32 mark = skb->mark;
998 put_cmsg(msg, SOL_SOCKET, SO_MARK, sizeof(__u32), &mark);
1002 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
1003 struct sk_buff *skb)
1005 sock_recv_timestamp(msg, sk, skb);
1006 sock_recv_drops(msg, sk, skb);
1007 sock_recv_mark(msg, sk, skb);
1009 EXPORT_SYMBOL_GPL(__sock_recv_cmsgs);
1011 INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
1013 INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
1016 static noinline void call_trace_sock_recv_length(struct sock *sk, int ret, int flags)
1018 trace_sock_recv_length(sk, ret, flags);
1021 static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
1024 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->recvmsg,
1026 inet_recvmsg, sock, msg,
1027 msg_data_left(msg), flags);
1028 if (trace_sock_recv_length_enabled())
1029 call_trace_sock_recv_length(sock->sk, ret, flags);
1034 * sock_recvmsg - receive a message from @sock
1036 * @msg: message to receive
1037 * @flags: message flags
1039 * Receives @msg from @sock, passing through LSM. Returns the total number
1040 * of bytes received, or an error.
1042 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
1044 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
1046 return err ?: sock_recvmsg_nosec(sock, msg, flags);
1048 EXPORT_SYMBOL(sock_recvmsg);
1051 * kernel_recvmsg - Receive a message from a socket (kernel space)
1052 * @sock: The socket to receive the message from
1053 * @msg: Received message
1054 * @vec: Input s/g array for message data
1055 * @num: Size of input s/g array
1056 * @size: Number of bytes to read
1057 * @flags: Message flags (MSG_DONTWAIT, etc...)
1059 * On return the msg structure contains the scatter/gather array passed in the
1060 * vec argument. The array is modified so that it consists of the unfilled
1061 * portion of the original array.
1063 * The returned value is the total number of bytes received, or an error.
1066 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
1067 struct kvec *vec, size_t num, size_t size, int flags)
1069 msg->msg_control_is_user = false;
1070 iov_iter_kvec(&msg->msg_iter, ITER_DEST, vec, num, size);
1071 return sock_recvmsg(sock, msg, flags);
1073 EXPORT_SYMBOL(kernel_recvmsg);
1075 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1076 struct pipe_inode_info *pipe, size_t len,
1079 struct socket *sock = file->private_data;
1080 const struct proto_ops *ops;
1082 ops = READ_ONCE(sock->ops);
1083 if (unlikely(!ops->splice_read))
1084 return copy_splice_read(file, ppos, pipe, len, flags);
1086 return ops->splice_read(sock, ppos, pipe, len, flags);
1089 static void sock_splice_eof(struct file *file)
1091 struct socket *sock = file->private_data;
1092 const struct proto_ops *ops;
1094 ops = READ_ONCE(sock->ops);
1095 if (ops->splice_eof)
1096 ops->splice_eof(sock);
1099 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1101 struct file *file = iocb->ki_filp;
1102 struct socket *sock = file->private_data;
1103 struct msghdr msg = {.msg_iter = *to,
1107 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1108 msg.msg_flags = MSG_DONTWAIT;
1110 if (iocb->ki_pos != 0)
1113 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1116 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1121 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1123 struct file *file = iocb->ki_filp;
1124 struct socket *sock = file->private_data;
1125 struct msghdr msg = {.msg_iter = *from,
1129 if (iocb->ki_pos != 0)
1132 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1133 msg.msg_flags = MSG_DONTWAIT;
1135 if (sock->type == SOCK_SEQPACKET)
1136 msg.msg_flags |= MSG_EOR;
1138 res = sock_sendmsg(sock, &msg);
1139 *from = msg.msg_iter;
1144 * Atomic setting of ioctl hooks to avoid race
1145 * with module unload.
1148 static DEFINE_MUTEX(br_ioctl_mutex);
1149 static int (*br_ioctl_hook)(struct net *net, struct net_bridge *br,
1150 unsigned int cmd, struct ifreq *ifr,
1153 void brioctl_set(int (*hook)(struct net *net, struct net_bridge *br,
1154 unsigned int cmd, struct ifreq *ifr,
1157 mutex_lock(&br_ioctl_mutex);
1158 br_ioctl_hook = hook;
1159 mutex_unlock(&br_ioctl_mutex);
1161 EXPORT_SYMBOL(brioctl_set);
1163 int br_ioctl_call(struct net *net, struct net_bridge *br, unsigned int cmd,
1164 struct ifreq *ifr, void __user *uarg)
1169 request_module("bridge");
1171 mutex_lock(&br_ioctl_mutex);
1173 err = br_ioctl_hook(net, br, cmd, ifr, uarg);
1174 mutex_unlock(&br_ioctl_mutex);
1179 static DEFINE_MUTEX(vlan_ioctl_mutex);
1180 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1182 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1184 mutex_lock(&vlan_ioctl_mutex);
1185 vlan_ioctl_hook = hook;
1186 mutex_unlock(&vlan_ioctl_mutex);
1188 EXPORT_SYMBOL(vlan_ioctl_set);
1190 static long sock_do_ioctl(struct net *net, struct socket *sock,
1191 unsigned int cmd, unsigned long arg)
1193 const struct proto_ops *ops = READ_ONCE(sock->ops);
1197 void __user *argp = (void __user *)arg;
1200 err = ops->ioctl(sock, cmd, arg);
1203 * If this ioctl is unknown try to hand it down
1204 * to the NIC driver.
1206 if (err != -ENOIOCTLCMD)
1209 if (!is_socket_ioctl_cmd(cmd))
1212 if (get_user_ifreq(&ifr, &data, argp))
1214 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1215 if (!err && need_copyout)
1216 if (put_user_ifreq(&ifr, argp))
1223 * With an ioctl, arg may well be a user mode pointer, but we don't know
1224 * what to do with it - that's up to the protocol still.
1227 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1229 const struct proto_ops *ops;
1230 struct socket *sock;
1232 void __user *argp = (void __user *)arg;
1236 sock = file->private_data;
1237 ops = READ_ONCE(sock->ops);
1240 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1244 if (get_user_ifreq(&ifr, &data, argp))
1246 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1247 if (!err && need_copyout)
1248 if (put_user_ifreq(&ifr, argp))
1251 #ifdef CONFIG_WEXT_CORE
1252 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1253 err = wext_handle_ioctl(net, cmd, argp);
1260 if (get_user(pid, (int __user *)argp))
1262 err = f_setown(sock->file, pid, 1);
1266 err = put_user(f_getown(sock->file),
1267 (int __user *)argp);
1273 err = br_ioctl_call(net, NULL, cmd, NULL, argp);
1278 if (!vlan_ioctl_hook)
1279 request_module("8021q");
1281 mutex_lock(&vlan_ioctl_mutex);
1282 if (vlan_ioctl_hook)
1283 err = vlan_ioctl_hook(net, argp);
1284 mutex_unlock(&vlan_ioctl_mutex);
1288 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1291 err = open_related_ns(&net->ns, get_net_ns);
1293 case SIOCGSTAMP_OLD:
1294 case SIOCGSTAMPNS_OLD:
1295 if (!ops->gettstamp) {
1299 err = ops->gettstamp(sock, argp,
1300 cmd == SIOCGSTAMP_OLD,
1301 !IS_ENABLED(CONFIG_64BIT));
1303 case SIOCGSTAMP_NEW:
1304 case SIOCGSTAMPNS_NEW:
1305 if (!ops->gettstamp) {
1309 err = ops->gettstamp(sock, argp,
1310 cmd == SIOCGSTAMP_NEW,
1315 err = dev_ifconf(net, argp);
1319 err = sock_do_ioctl(net, sock, cmd, arg);
1326 * sock_create_lite - creates a socket
1327 * @family: protocol family (AF_INET, ...)
1328 * @type: communication type (SOCK_STREAM, ...)
1329 * @protocol: protocol (0, ...)
1332 * Creates a new socket and assigns it to @res, passing through LSM.
1333 * The new socket initialization is not complete, see kernel_accept().
1334 * Returns 0 or an error. On failure @res is set to %NULL.
1335 * This function internally uses GFP_KERNEL.
1338 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1341 struct socket *sock = NULL;
1343 err = security_socket_create(family, type, protocol, 1);
1347 sock = sock_alloc();
1354 err = security_socket_post_create(sock, family, type, protocol, 1);
1366 EXPORT_SYMBOL(sock_create_lite);
1368 /* No kernel lock held - perfect */
1369 static __poll_t sock_poll(struct file *file, poll_table *wait)
1371 struct socket *sock = file->private_data;
1372 const struct proto_ops *ops = READ_ONCE(sock->ops);
1373 __poll_t events = poll_requested_events(wait), flag = 0;
1378 if (sk_can_busy_loop(sock->sk)) {
1379 /* poll once if requested by the syscall */
1380 if (events & POLL_BUSY_LOOP)
1381 sk_busy_loop(sock->sk, 1);
1383 /* if this socket can poll_ll, tell the system call */
1384 flag = POLL_BUSY_LOOP;
1387 return ops->poll(file, sock, wait) | flag;
1390 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1392 struct socket *sock = file->private_data;
1394 return READ_ONCE(sock->ops)->mmap(file, sock, vma);
1397 static int sock_close(struct inode *inode, struct file *filp)
1399 __sock_release(SOCKET_I(inode), inode);
1404 * Update the socket async list
1406 * Fasync_list locking strategy.
1408 * 1. fasync_list is modified only under process context socket lock
1409 * i.e. under semaphore.
1410 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1411 * or under socket lock
1414 static int sock_fasync(int fd, struct file *filp, int on)
1416 struct socket *sock = filp->private_data;
1417 struct sock *sk = sock->sk;
1418 struct socket_wq *wq = &sock->wq;
1424 fasync_helper(fd, filp, on, &wq->fasync_list);
1426 if (!wq->fasync_list)
1427 sock_reset_flag(sk, SOCK_FASYNC);
1429 sock_set_flag(sk, SOCK_FASYNC);
1435 /* This function may be called only under rcu_lock */
1437 int sock_wake_async(struct socket_wq *wq, int how, int band)
1439 if (!wq || !wq->fasync_list)
1443 case SOCK_WAKE_WAITD:
1444 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1447 case SOCK_WAKE_SPACE:
1448 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1453 kill_fasync(&wq->fasync_list, SIGIO, band);
1456 kill_fasync(&wq->fasync_list, SIGURG, band);
1461 EXPORT_SYMBOL(sock_wake_async);
1464 * __sock_create - creates a socket
1465 * @net: net namespace
1466 * @family: protocol family (AF_INET, ...)
1467 * @type: communication type (SOCK_STREAM, ...)
1468 * @protocol: protocol (0, ...)
1470 * @kern: boolean for kernel space sockets
1472 * Creates a new socket and assigns it to @res, passing through LSM.
1473 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1474 * be set to true if the socket resides in kernel space.
1475 * This function internally uses GFP_KERNEL.
1478 int __sock_create(struct net *net, int family, int type, int protocol,
1479 struct socket **res, int kern)
1482 struct socket *sock;
1483 const struct net_proto_family *pf;
1486 * Check protocol is in range
1488 if (family < 0 || family >= NPROTO)
1489 return -EAFNOSUPPORT;
1490 if (type < 0 || type >= SOCK_MAX)
1495 This uglymoron is moved from INET layer to here to avoid
1496 deadlock in module load.
1498 if (family == PF_INET && type == SOCK_PACKET) {
1499 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1504 err = security_socket_create(family, type, protocol, kern);
1509 * Allocate the socket and allow the family to set things up. if
1510 * the protocol is 0, the family is instructed to select an appropriate
1513 sock = sock_alloc();
1515 net_warn_ratelimited("socket: no more sockets\n");
1516 return -ENFILE; /* Not exactly a match, but its the
1517 closest posix thing */
1522 #ifdef CONFIG_MODULES
1523 /* Attempt to load a protocol module if the find failed.
1525 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1526 * requested real, full-featured networking support upon configuration.
1527 * Otherwise module support will break!
1529 if (rcu_access_pointer(net_families[family]) == NULL)
1530 request_module("net-pf-%d", family);
1534 pf = rcu_dereference(net_families[family]);
1535 err = -EAFNOSUPPORT;
1540 * We will call the ->create function, that possibly is in a loadable
1541 * module, so we have to bump that loadable module refcnt first.
1543 if (!try_module_get(pf->owner))
1546 /* Now protected by module ref count */
1549 err = pf->create(net, sock, protocol, kern);
1551 goto out_module_put;
1554 * Now to bump the refcnt of the [loadable] module that owns this
1555 * socket at sock_release time we decrement its refcnt.
1557 if (!try_module_get(sock->ops->owner))
1558 goto out_module_busy;
1561 * Now that we're done with the ->create function, the [loadable]
1562 * module can have its refcnt decremented
1564 module_put(pf->owner);
1565 err = security_socket_post_create(sock, family, type, protocol, kern);
1567 goto out_sock_release;
1573 err = -EAFNOSUPPORT;
1576 module_put(pf->owner);
1583 goto out_sock_release;
1585 EXPORT_SYMBOL(__sock_create);
1588 * sock_create - creates a socket
1589 * @family: protocol family (AF_INET, ...)
1590 * @type: communication type (SOCK_STREAM, ...)
1591 * @protocol: protocol (0, ...)
1594 * A wrapper around __sock_create().
1595 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1598 int sock_create(int family, int type, int protocol, struct socket **res)
1600 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1602 EXPORT_SYMBOL(sock_create);
1605 * sock_create_kern - creates a socket (kernel space)
1606 * @net: net namespace
1607 * @family: protocol family (AF_INET, ...)
1608 * @type: communication type (SOCK_STREAM, ...)
1609 * @protocol: protocol (0, ...)
1612 * A wrapper around __sock_create().
1613 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1616 int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1618 return __sock_create(net, family, type, protocol, res, 1);
1620 EXPORT_SYMBOL(sock_create_kern);
1622 static struct socket *__sys_socket_create(int family, int type, int protocol)
1624 struct socket *sock;
1627 /* Check the SOCK_* constants for consistency. */
1628 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1629 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1630 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1631 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1633 if ((type & ~SOCK_TYPE_MASK) & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1634 return ERR_PTR(-EINVAL);
1635 type &= SOCK_TYPE_MASK;
1637 retval = sock_create(family, type, protocol, &sock);
1639 return ERR_PTR(retval);
1644 struct file *__sys_socket_file(int family, int type, int protocol)
1646 struct socket *sock;
1649 sock = __sys_socket_create(family, type, protocol);
1651 return ERR_CAST(sock);
1653 flags = type & ~SOCK_TYPE_MASK;
1654 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1655 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1657 return sock_alloc_file(sock, flags, NULL);
1660 int __sys_socket(int family, int type, int protocol)
1662 struct socket *sock;
1665 sock = __sys_socket_create(family, type, protocol);
1667 return PTR_ERR(sock);
1669 flags = type & ~SOCK_TYPE_MASK;
1670 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1671 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1673 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1676 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1678 return __sys_socket(family, type, protocol);
1682 * Create a pair of connected sockets.
1685 int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1687 struct socket *sock1, *sock2;
1689 struct file *newfile1, *newfile2;
1692 flags = type & ~SOCK_TYPE_MASK;
1693 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1695 type &= SOCK_TYPE_MASK;
1697 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1698 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1701 * reserve descriptors and make sure we won't fail
1702 * to return them to userland.
1704 fd1 = get_unused_fd_flags(flags);
1705 if (unlikely(fd1 < 0))
1708 fd2 = get_unused_fd_flags(flags);
1709 if (unlikely(fd2 < 0)) {
1714 err = put_user(fd1, &usockvec[0]);
1718 err = put_user(fd2, &usockvec[1]);
1723 * Obtain the first socket and check if the underlying protocol
1724 * supports the socketpair call.
1727 err = sock_create(family, type, protocol, &sock1);
1728 if (unlikely(err < 0))
1731 err = sock_create(family, type, protocol, &sock2);
1732 if (unlikely(err < 0)) {
1733 sock_release(sock1);
1737 err = security_socket_socketpair(sock1, sock2);
1738 if (unlikely(err)) {
1739 sock_release(sock2);
1740 sock_release(sock1);
1744 err = READ_ONCE(sock1->ops)->socketpair(sock1, sock2);
1745 if (unlikely(err < 0)) {
1746 sock_release(sock2);
1747 sock_release(sock1);
1751 newfile1 = sock_alloc_file(sock1, flags, NULL);
1752 if (IS_ERR(newfile1)) {
1753 err = PTR_ERR(newfile1);
1754 sock_release(sock2);
1758 newfile2 = sock_alloc_file(sock2, flags, NULL);
1759 if (IS_ERR(newfile2)) {
1760 err = PTR_ERR(newfile2);
1765 audit_fd_pair(fd1, fd2);
1767 fd_install(fd1, newfile1);
1768 fd_install(fd2, newfile2);
1777 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1778 int __user *, usockvec)
1780 return __sys_socketpair(family, type, protocol, usockvec);
1784 * Bind a name to a socket. Nothing much to do here since it's
1785 * the protocol's responsibility to handle the local address.
1787 * We move the socket address to kernel space before we call
1788 * the protocol layer (having also checked the address is ok).
1791 int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1793 struct socket *sock;
1794 struct sockaddr_storage address;
1795 int err, fput_needed;
1797 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1799 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1801 err = security_socket_bind(sock,
1802 (struct sockaddr *)&address,
1805 err = READ_ONCE(sock->ops)->bind(sock,
1809 fput_light(sock->file, fput_needed);
1814 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1816 return __sys_bind(fd, umyaddr, addrlen);
1820 * Perform a listen. Basically, we allow the protocol to do anything
1821 * necessary for a listen, and if that works, we mark the socket as
1822 * ready for listening.
1825 int __sys_listen(int fd, int backlog)
1827 struct socket *sock;
1828 int err, fput_needed;
1831 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1833 somaxconn = READ_ONCE(sock_net(sock->sk)->core.sysctl_somaxconn);
1834 if ((unsigned int)backlog > somaxconn)
1835 backlog = somaxconn;
1837 err = security_socket_listen(sock, backlog);
1839 err = READ_ONCE(sock->ops)->listen(sock, backlog);
1841 fput_light(sock->file, fput_needed);
1846 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1848 return __sys_listen(fd, backlog);
1851 struct file *do_accept(struct file *file, unsigned file_flags,
1852 struct sockaddr __user *upeer_sockaddr,
1853 int __user *upeer_addrlen, int flags)
1855 struct socket *sock, *newsock;
1856 struct file *newfile;
1858 struct sockaddr_storage address;
1859 const struct proto_ops *ops;
1861 sock = sock_from_file(file);
1863 return ERR_PTR(-ENOTSOCK);
1865 newsock = sock_alloc();
1867 return ERR_PTR(-ENFILE);
1868 ops = READ_ONCE(sock->ops);
1870 newsock->type = sock->type;
1874 * We don't need try_module_get here, as the listening socket (sock)
1875 * has the protocol module (sock->ops->owner) held.
1877 __module_get(ops->owner);
1879 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1880 if (IS_ERR(newfile))
1883 err = security_socket_accept(sock, newsock);
1887 err = ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1892 if (upeer_sockaddr) {
1893 len = ops->getname(newsock, (struct sockaddr *)&address, 2);
1895 err = -ECONNABORTED;
1898 err = move_addr_to_user(&address,
1899 len, upeer_sockaddr, upeer_addrlen);
1904 /* File flags are not inherited via accept() unlike another OSes. */
1908 return ERR_PTR(err);
1911 static int __sys_accept4_file(struct file *file, struct sockaddr __user *upeer_sockaddr,
1912 int __user *upeer_addrlen, int flags)
1914 struct file *newfile;
1917 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1920 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1921 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1923 newfd = get_unused_fd_flags(flags);
1924 if (unlikely(newfd < 0))
1927 newfile = do_accept(file, 0, upeer_sockaddr, upeer_addrlen,
1929 if (IS_ERR(newfile)) {
1930 put_unused_fd(newfd);
1931 return PTR_ERR(newfile);
1933 fd_install(newfd, newfile);
1938 * For accept, we attempt to create a new socket, set up the link
1939 * with the client, wake up the client, then return the new
1940 * connected fd. We collect the address of the connector in kernel
1941 * space and move it to user at the very end. This is unclean because
1942 * we open the socket then return an error.
1944 * 1003.1g adds the ability to recvmsg() to query connection pending
1945 * status to recvmsg. We need to add that support in a way thats
1946 * clean when we restructure accept also.
1949 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1950 int __user *upeer_addrlen, int flags)
1957 ret = __sys_accept4_file(f.file, upeer_sockaddr,
1958 upeer_addrlen, flags);
1965 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1966 int __user *, upeer_addrlen, int, flags)
1968 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1971 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1972 int __user *, upeer_addrlen)
1974 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1978 * Attempt to connect to a socket with the server address. The address
1979 * is in user space so we verify it is OK and move it to kernel space.
1981 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1984 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1985 * other SEQPACKET protocols that take time to connect() as it doesn't
1986 * include the -EINPROGRESS status for such sockets.
1989 int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
1990 int addrlen, int file_flags)
1992 struct socket *sock;
1995 sock = sock_from_file(file);
2002 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
2006 err = READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)address,
2007 addrlen, sock->file->f_flags | file_flags);
2012 int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
2019 struct sockaddr_storage address;
2021 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
2023 ret = __sys_connect_file(f.file, &address, addrlen, 0);
2030 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
2033 return __sys_connect(fd, uservaddr, addrlen);
2037 * Get the local address ('name') of a socket object. Move the obtained
2038 * name to user space.
2041 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
2042 int __user *usockaddr_len)
2044 struct socket *sock;
2045 struct sockaddr_storage address;
2046 int err, fput_needed;
2048 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2052 err = security_socket_getsockname(sock);
2056 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 0);
2059 /* "err" is actually length in this case */
2060 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
2063 fput_light(sock->file, fput_needed);
2068 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
2069 int __user *, usockaddr_len)
2071 return __sys_getsockname(fd, usockaddr, usockaddr_len);
2075 * Get the remote address ('name') of a socket object. Move the obtained
2076 * name to user space.
2079 int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
2080 int __user *usockaddr_len)
2082 struct socket *sock;
2083 struct sockaddr_storage address;
2084 int err, fput_needed;
2086 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2088 const struct proto_ops *ops = READ_ONCE(sock->ops);
2090 err = security_socket_getpeername(sock);
2092 fput_light(sock->file, fput_needed);
2096 err = ops->getname(sock, (struct sockaddr *)&address, 1);
2098 /* "err" is actually length in this case */
2099 err = move_addr_to_user(&address, err, usockaddr,
2101 fput_light(sock->file, fput_needed);
2106 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
2107 int __user *, usockaddr_len)
2109 return __sys_getpeername(fd, usockaddr, usockaddr_len);
2113 * Send a datagram to a given address. We move the address into kernel
2114 * space and check the user space data area is readable before invoking
2117 int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
2118 struct sockaddr __user *addr, int addr_len)
2120 struct socket *sock;
2121 struct sockaddr_storage address;
2127 err = import_single_range(ITER_SOURCE, buff, len, &iov, &msg.msg_iter);
2130 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2134 msg.msg_name = NULL;
2135 msg.msg_control = NULL;
2136 msg.msg_controllen = 0;
2137 msg.msg_namelen = 0;
2138 msg.msg_ubuf = NULL;
2140 err = move_addr_to_kernel(addr, addr_len, &address);
2143 msg.msg_name = (struct sockaddr *)&address;
2144 msg.msg_namelen = addr_len;
2146 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2147 if (sock->file->f_flags & O_NONBLOCK)
2148 flags |= MSG_DONTWAIT;
2149 msg.msg_flags = flags;
2150 err = sock_sendmsg(sock, &msg);
2153 fput_light(sock->file, fput_needed);
2158 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2159 unsigned int, flags, struct sockaddr __user *, addr,
2162 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2166 * Send a datagram down a socket.
2169 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2170 unsigned int, flags)
2172 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2176 * Receive a frame from the socket and optionally record the address of the
2177 * sender. We verify the buffers are writable and if needed move the
2178 * sender address from kernel to user space.
2180 int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2181 struct sockaddr __user *addr, int __user *addr_len)
2183 struct sockaddr_storage address;
2184 struct msghdr msg = {
2185 /* Save some cycles and don't copy the address if not needed */
2186 .msg_name = addr ? (struct sockaddr *)&address : NULL,
2188 struct socket *sock;
2193 err = import_single_range(ITER_DEST, ubuf, size, &iov, &msg.msg_iter);
2196 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2200 if (sock->file->f_flags & O_NONBLOCK)
2201 flags |= MSG_DONTWAIT;
2202 err = sock_recvmsg(sock, &msg, flags);
2204 if (err >= 0 && addr != NULL) {
2205 err2 = move_addr_to_user(&address,
2206 msg.msg_namelen, addr, addr_len);
2211 fput_light(sock->file, fput_needed);
2216 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2217 unsigned int, flags, struct sockaddr __user *, addr,
2218 int __user *, addr_len)
2220 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2224 * Receive a datagram from a socket.
2227 SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2228 unsigned int, flags)
2230 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2233 static bool sock_use_custom_sol_socket(const struct socket *sock)
2235 return test_bit(SOCK_CUSTOM_SOCKOPT, &sock->flags);
2239 * Set a socket option. Because we don't know the option lengths we have
2240 * to pass the user mode parameter for the protocols to sort out.
2242 int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2245 sockptr_t optval = USER_SOCKPTR(user_optval);
2246 const struct proto_ops *ops;
2247 char *kernel_optval = NULL;
2248 int err, fput_needed;
2249 struct socket *sock;
2254 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2258 err = security_socket_setsockopt(sock, level, optname);
2262 if (!in_compat_syscall())
2263 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2264 user_optval, &optlen,
2274 optval = KERNEL_SOCKPTR(kernel_optval);
2275 ops = READ_ONCE(sock->ops);
2276 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2277 err = sock_setsockopt(sock, level, optname, optval, optlen);
2278 else if (unlikely(!ops->setsockopt))
2281 err = ops->setsockopt(sock, level, optname, optval,
2283 kfree(kernel_optval);
2285 fput_light(sock->file, fput_needed);
2289 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2290 char __user *, optval, int, optlen)
2292 return __sys_setsockopt(fd, level, optname, optval, optlen);
2295 INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2299 * Get a socket option. Because we don't know the option lengths we have
2300 * to pass a user mode parameter for the protocols to sort out.
2302 int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2305 int max_optlen __maybe_unused;
2306 const struct proto_ops *ops;
2307 int err, fput_needed;
2308 struct socket *sock;
2310 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2314 err = security_socket_getsockopt(sock, level, optname);
2318 if (!in_compat_syscall())
2319 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2321 ops = READ_ONCE(sock->ops);
2322 if (level == SOL_SOCKET)
2323 err = sock_getsockopt(sock, level, optname, optval, optlen);
2324 else if (unlikely(!ops->getsockopt))
2327 err = ops->getsockopt(sock, level, optname, optval,
2330 if (!in_compat_syscall())
2331 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2332 optval, optlen, max_optlen,
2335 fput_light(sock->file, fput_needed);
2339 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2340 char __user *, optval, int __user *, optlen)
2342 return __sys_getsockopt(fd, level, optname, optval, optlen);
2346 * Shutdown a socket.
2349 int __sys_shutdown_sock(struct socket *sock, int how)
2353 err = security_socket_shutdown(sock, how);
2355 err = READ_ONCE(sock->ops)->shutdown(sock, how);
2360 int __sys_shutdown(int fd, int how)
2362 int err, fput_needed;
2363 struct socket *sock;
2365 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2367 err = __sys_shutdown_sock(sock, how);
2368 fput_light(sock->file, fput_needed);
2373 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2375 return __sys_shutdown(fd, how);
2378 /* A couple of helpful macros for getting the address of the 32/64 bit
2379 * fields which are the same type (int / unsigned) on our platforms.
2381 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2382 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2383 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2385 struct used_address {
2386 struct sockaddr_storage name;
2387 unsigned int name_len;
2390 int __copy_msghdr(struct msghdr *kmsg,
2391 struct user_msghdr *msg,
2392 struct sockaddr __user **save_addr)
2396 kmsg->msg_control_is_user = true;
2397 kmsg->msg_get_inq = 0;
2398 kmsg->msg_control_user = msg->msg_control;
2399 kmsg->msg_controllen = msg->msg_controllen;
2400 kmsg->msg_flags = msg->msg_flags;
2402 kmsg->msg_namelen = msg->msg_namelen;
2404 kmsg->msg_namelen = 0;
2406 if (kmsg->msg_namelen < 0)
2409 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2410 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2413 *save_addr = msg->msg_name;
2415 if (msg->msg_name && kmsg->msg_namelen) {
2417 err = move_addr_to_kernel(msg->msg_name,
2424 kmsg->msg_name = NULL;
2425 kmsg->msg_namelen = 0;
2428 if (msg->msg_iovlen > UIO_MAXIOV)
2431 kmsg->msg_iocb = NULL;
2432 kmsg->msg_ubuf = NULL;
2436 static int copy_msghdr_from_user(struct msghdr *kmsg,
2437 struct user_msghdr __user *umsg,
2438 struct sockaddr __user **save_addr,
2441 struct user_msghdr msg;
2444 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2447 err = __copy_msghdr(kmsg, &msg, save_addr);
2451 err = import_iovec(save_addr ? ITER_DEST : ITER_SOURCE,
2452 msg.msg_iov, msg.msg_iovlen,
2453 UIO_FASTIOV, iov, &kmsg->msg_iter);
2454 return err < 0 ? err : 0;
2457 static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2458 unsigned int flags, struct used_address *used_address,
2459 unsigned int allowed_msghdr_flags)
2461 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2462 __aligned(sizeof(__kernel_size_t));
2463 /* 20 is size of ipv6_pktinfo */
2464 unsigned char *ctl_buf = ctl;
2470 if (msg_sys->msg_controllen > INT_MAX)
2472 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2473 ctl_len = msg_sys->msg_controllen;
2474 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2476 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2480 ctl_buf = msg_sys->msg_control;
2481 ctl_len = msg_sys->msg_controllen;
2482 } else if (ctl_len) {
2483 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2484 CMSG_ALIGN(sizeof(struct cmsghdr)));
2485 if (ctl_len > sizeof(ctl)) {
2486 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2487 if (ctl_buf == NULL)
2491 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2493 msg_sys->msg_control = ctl_buf;
2494 msg_sys->msg_control_is_user = false;
2496 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2497 msg_sys->msg_flags = flags;
2499 if (sock->file->f_flags & O_NONBLOCK)
2500 msg_sys->msg_flags |= MSG_DONTWAIT;
2502 * If this is sendmmsg() and current destination address is same as
2503 * previously succeeded address, omit asking LSM's decision.
2504 * used_address->name_len is initialized to UINT_MAX so that the first
2505 * destination address never matches.
2507 if (used_address && msg_sys->msg_name &&
2508 used_address->name_len == msg_sys->msg_namelen &&
2509 !memcmp(&used_address->name, msg_sys->msg_name,
2510 used_address->name_len)) {
2511 err = sock_sendmsg_nosec(sock, msg_sys);
2514 err = sock_sendmsg(sock, msg_sys);
2516 * If this is sendmmsg() and sending to current destination address was
2517 * successful, remember it.
2519 if (used_address && err >= 0) {
2520 used_address->name_len = msg_sys->msg_namelen;
2521 if (msg_sys->msg_name)
2522 memcpy(&used_address->name, msg_sys->msg_name,
2523 used_address->name_len);
2528 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2533 int sendmsg_copy_msghdr(struct msghdr *msg,
2534 struct user_msghdr __user *umsg, unsigned flags,
2539 if (flags & MSG_CMSG_COMPAT) {
2540 struct compat_msghdr __user *msg_compat;
2542 msg_compat = (struct compat_msghdr __user *) umsg;
2543 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2545 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2553 static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2554 struct msghdr *msg_sys, unsigned int flags,
2555 struct used_address *used_address,
2556 unsigned int allowed_msghdr_flags)
2558 struct sockaddr_storage address;
2559 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2562 msg_sys->msg_name = &address;
2564 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2568 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2569 allowed_msghdr_flags);
2575 * BSD sendmsg interface
2577 long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2580 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2583 long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2584 bool forbid_cmsg_compat)
2586 int fput_needed, err;
2587 struct msghdr msg_sys;
2588 struct socket *sock;
2590 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2593 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2597 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2599 fput_light(sock->file, fput_needed);
2604 SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2606 return __sys_sendmsg(fd, msg, flags, true);
2610 * Linux sendmmsg interface
2613 int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2614 unsigned int flags, bool forbid_cmsg_compat)
2616 int fput_needed, err, datagrams;
2617 struct socket *sock;
2618 struct mmsghdr __user *entry;
2619 struct compat_mmsghdr __user *compat_entry;
2620 struct msghdr msg_sys;
2621 struct used_address used_address;
2622 unsigned int oflags = flags;
2624 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2627 if (vlen > UIO_MAXIOV)
2632 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2636 used_address.name_len = UINT_MAX;
2638 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2642 while (datagrams < vlen) {
2643 if (datagrams == vlen - 1)
2646 if (MSG_CMSG_COMPAT & flags) {
2647 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2648 &msg_sys, flags, &used_address, MSG_EOR);
2651 err = __put_user(err, &compat_entry->msg_len);
2654 err = ___sys_sendmsg(sock,
2655 (struct user_msghdr __user *)entry,
2656 &msg_sys, flags, &used_address, MSG_EOR);
2659 err = put_user(err, &entry->msg_len);
2666 if (msg_data_left(&msg_sys))
2671 fput_light(sock->file, fput_needed);
2673 /* We only return an error if no datagrams were able to be sent */
2680 SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2681 unsigned int, vlen, unsigned int, flags)
2683 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2686 int recvmsg_copy_msghdr(struct msghdr *msg,
2687 struct user_msghdr __user *umsg, unsigned flags,
2688 struct sockaddr __user **uaddr,
2693 if (MSG_CMSG_COMPAT & flags) {
2694 struct compat_msghdr __user *msg_compat;
2696 msg_compat = (struct compat_msghdr __user *) umsg;
2697 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2699 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2707 static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2708 struct user_msghdr __user *msg,
2709 struct sockaddr __user *uaddr,
2710 unsigned int flags, int nosec)
2712 struct compat_msghdr __user *msg_compat =
2713 (struct compat_msghdr __user *) msg;
2714 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2715 struct sockaddr_storage addr;
2716 unsigned long cmsg_ptr;
2720 msg_sys->msg_name = &addr;
2721 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2722 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2724 /* We assume all kernel code knows the size of sockaddr_storage */
2725 msg_sys->msg_namelen = 0;
2727 if (sock->file->f_flags & O_NONBLOCK)
2728 flags |= MSG_DONTWAIT;
2730 if (unlikely(nosec))
2731 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2733 err = sock_recvmsg(sock, msg_sys, flags);
2739 if (uaddr != NULL) {
2740 err = move_addr_to_user(&addr,
2741 msg_sys->msg_namelen, uaddr,
2746 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2750 if (MSG_CMSG_COMPAT & flags)
2751 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2752 &msg_compat->msg_controllen);
2754 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2755 &msg->msg_controllen);
2763 static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2764 struct msghdr *msg_sys, unsigned int flags, int nosec)
2766 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2767 /* user mode address pointers */
2768 struct sockaddr __user *uaddr;
2771 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2775 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2781 * BSD recvmsg interface
2784 long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2785 struct user_msghdr __user *umsg,
2786 struct sockaddr __user *uaddr, unsigned int flags)
2788 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2791 long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2792 bool forbid_cmsg_compat)
2794 int fput_needed, err;
2795 struct msghdr msg_sys;
2796 struct socket *sock;
2798 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2801 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2805 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2807 fput_light(sock->file, fput_needed);
2812 SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2813 unsigned int, flags)
2815 return __sys_recvmsg(fd, msg, flags, true);
2819 * Linux recvmmsg interface
2822 static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2823 unsigned int vlen, unsigned int flags,
2824 struct timespec64 *timeout)
2826 int fput_needed, err, datagrams;
2827 struct socket *sock;
2828 struct mmsghdr __user *entry;
2829 struct compat_mmsghdr __user *compat_entry;
2830 struct msghdr msg_sys;
2831 struct timespec64 end_time;
2832 struct timespec64 timeout64;
2835 poll_select_set_timeout(&end_time, timeout->tv_sec,
2841 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2845 if (likely(!(flags & MSG_ERRQUEUE))) {
2846 err = sock_error(sock->sk);
2854 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2856 while (datagrams < vlen) {
2858 * No need to ask LSM for more than the first datagram.
2860 if (MSG_CMSG_COMPAT & flags) {
2861 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2862 &msg_sys, flags & ~MSG_WAITFORONE,
2866 err = __put_user(err, &compat_entry->msg_len);
2869 err = ___sys_recvmsg(sock,
2870 (struct user_msghdr __user *)entry,
2871 &msg_sys, flags & ~MSG_WAITFORONE,
2875 err = put_user(err, &entry->msg_len);
2883 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2884 if (flags & MSG_WAITFORONE)
2885 flags |= MSG_DONTWAIT;
2888 ktime_get_ts64(&timeout64);
2889 *timeout = timespec64_sub(end_time, timeout64);
2890 if (timeout->tv_sec < 0) {
2891 timeout->tv_sec = timeout->tv_nsec = 0;
2895 /* Timeout, return less than vlen datagrams */
2896 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2900 /* Out of band data, return right away */
2901 if (msg_sys.msg_flags & MSG_OOB)
2909 if (datagrams == 0) {
2915 * We may return less entries than requested (vlen) if the
2916 * sock is non block and there aren't enough datagrams...
2918 if (err != -EAGAIN) {
2920 * ... or if recvmsg returns an error after we
2921 * received some datagrams, where we record the
2922 * error to return on the next call or if the
2923 * app asks about it using getsockopt(SO_ERROR).
2925 WRITE_ONCE(sock->sk->sk_err, -err);
2928 fput_light(sock->file, fput_needed);
2933 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2934 unsigned int vlen, unsigned int flags,
2935 struct __kernel_timespec __user *timeout,
2936 struct old_timespec32 __user *timeout32)
2939 struct timespec64 timeout_sys;
2941 if (timeout && get_timespec64(&timeout_sys, timeout))
2944 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2947 if (!timeout && !timeout32)
2948 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2950 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2955 if (timeout && put_timespec64(&timeout_sys, timeout))
2956 datagrams = -EFAULT;
2958 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2959 datagrams = -EFAULT;
2964 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2965 unsigned int, vlen, unsigned int, flags,
2966 struct __kernel_timespec __user *, timeout)
2968 if (flags & MSG_CMSG_COMPAT)
2971 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2974 #ifdef CONFIG_COMPAT_32BIT_TIME
2975 SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2976 unsigned int, vlen, unsigned int, flags,
2977 struct old_timespec32 __user *, timeout)
2979 if (flags & MSG_CMSG_COMPAT)
2982 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2986 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2987 /* Argument list sizes for sys_socketcall */
2988 #define AL(x) ((x) * sizeof(unsigned long))
2989 static const unsigned char nargs[21] = {
2990 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2991 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2992 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2999 * System call vectors.
3001 * Argument checking cleaned up. Saved 20% in size.
3002 * This function doesn't need to set the kernel lock because
3003 * it is set by the callees.
3006 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
3008 unsigned long a[AUDITSC_ARGS];
3009 unsigned long a0, a1;
3013 if (call < 1 || call > SYS_SENDMMSG)
3015 call = array_index_nospec(call, SYS_SENDMMSG + 1);
3018 if (len > sizeof(a))
3021 /* copy_from_user should be SMP safe. */
3022 if (copy_from_user(a, args, len))
3025 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
3034 err = __sys_socket(a0, a1, a[2]);
3037 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
3040 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
3043 err = __sys_listen(a0, a1);
3046 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3047 (int __user *)a[2], 0);
3049 case SYS_GETSOCKNAME:
3051 __sys_getsockname(a0, (struct sockaddr __user *)a1,
3052 (int __user *)a[2]);
3054 case SYS_GETPEERNAME:
3056 __sys_getpeername(a0, (struct sockaddr __user *)a1,
3057 (int __user *)a[2]);
3059 case SYS_SOCKETPAIR:
3060 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
3063 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3067 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3068 (struct sockaddr __user *)a[4], a[5]);
3071 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3075 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3076 (struct sockaddr __user *)a[4],
3077 (int __user *)a[5]);
3080 err = __sys_shutdown(a0, a1);
3082 case SYS_SETSOCKOPT:
3083 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
3086 case SYS_GETSOCKOPT:
3088 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
3089 (int __user *)a[4]);
3092 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
3096 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
3100 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
3104 if (IS_ENABLED(CONFIG_64BIT))
3105 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3107 (struct __kernel_timespec __user *)a[4],
3110 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3112 (struct old_timespec32 __user *)a[4]);
3115 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3116 (int __user *)a[2], a[3]);
3125 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
3128 * sock_register - add a socket protocol handler
3129 * @ops: description of protocol
3131 * This function is called by a protocol handler that wants to
3132 * advertise its address family, and have it linked into the
3133 * socket interface. The value ops->family corresponds to the
3134 * socket system call protocol family.
3136 int sock_register(const struct net_proto_family *ops)
3140 if (ops->family >= NPROTO) {
3141 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3145 spin_lock(&net_family_lock);
3146 if (rcu_dereference_protected(net_families[ops->family],
3147 lockdep_is_held(&net_family_lock)))
3150 rcu_assign_pointer(net_families[ops->family], ops);
3153 spin_unlock(&net_family_lock);
3155 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3158 EXPORT_SYMBOL(sock_register);
3161 * sock_unregister - remove a protocol handler
3162 * @family: protocol family to remove
3164 * This function is called by a protocol handler that wants to
3165 * remove its address family, and have it unlinked from the
3166 * new socket creation.
3168 * If protocol handler is a module, then it can use module reference
3169 * counts to protect against new references. If protocol handler is not
3170 * a module then it needs to provide its own protection in
3171 * the ops->create routine.
3173 void sock_unregister(int family)
3175 BUG_ON(family < 0 || family >= NPROTO);
3177 spin_lock(&net_family_lock);
3178 RCU_INIT_POINTER(net_families[family], NULL);
3179 spin_unlock(&net_family_lock);
3183 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3185 EXPORT_SYMBOL(sock_unregister);
3187 bool sock_is_registered(int family)
3189 return family < NPROTO && rcu_access_pointer(net_families[family]);
3192 static int __init sock_init(void)
3196 * Initialize the network sysctl infrastructure.
3198 err = net_sysctl_init();
3203 * Initialize skbuff SLAB cache
3208 * Initialize the protocols module.
3213 err = register_filesystem(&sock_fs_type);
3216 sock_mnt = kern_mount(&sock_fs_type);
3217 if (IS_ERR(sock_mnt)) {
3218 err = PTR_ERR(sock_mnt);
3222 /* The real protocol initialization is performed in later initcalls.
3225 #ifdef CONFIG_NETFILTER
3226 err = netfilter_init();
3231 ptp_classifier_init();
3237 unregister_filesystem(&sock_fs_type);
3241 core_initcall(sock_init); /* early initcall */
3243 #ifdef CONFIG_PROC_FS
3244 void socket_seq_show(struct seq_file *seq)
3246 seq_printf(seq, "sockets: used %d\n",
3247 sock_inuse_get(seq->private));
3249 #endif /* CONFIG_PROC_FS */
3251 /* Handle the fact that while struct ifreq has the same *layout* on
3252 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3253 * which are handled elsewhere, it still has different *size* due to
3254 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3255 * resulting in struct ifreq being 32 and 40 bytes respectively).
3256 * As a result, if the struct happens to be at the end of a page and
3257 * the next page isn't readable/writable, we get a fault. To prevent
3258 * that, copy back and forth to the full size.
3260 int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3262 if (in_compat_syscall()) {
3263 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3265 memset(ifr, 0, sizeof(*ifr));
3266 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3270 *ifrdata = compat_ptr(ifr32->ifr_data);
3275 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3279 *ifrdata = ifr->ifr_data;
3283 EXPORT_SYMBOL(get_user_ifreq);
3285 int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3287 size_t size = sizeof(*ifr);
3289 if (in_compat_syscall())
3290 size = sizeof(struct compat_ifreq);
3292 if (copy_to_user(arg, ifr, size))
3297 EXPORT_SYMBOL(put_user_ifreq);
3299 #ifdef CONFIG_COMPAT
3300 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3302 compat_uptr_t uptr32;
3307 if (get_user_ifreq(&ifr, NULL, uifr32))
3310 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3313 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3314 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3316 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3318 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3319 if (put_user_ifreq(&ifr, uifr32))
3325 /* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3326 static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3327 struct compat_ifreq __user *u_ifreq32)
3332 if (!is_socket_ioctl_cmd(cmd))
3334 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3336 ifreq.ifr_data = data;
3338 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3341 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3342 unsigned int cmd, unsigned long arg)
3344 void __user *argp = compat_ptr(arg);
3345 struct sock *sk = sock->sk;
3346 struct net *net = sock_net(sk);
3347 const struct proto_ops *ops;
3349 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3350 return sock_ioctl(file, cmd, (unsigned long)argp);
3354 return compat_siocwandev(net, argp);
3355 case SIOCGSTAMP_OLD:
3356 case SIOCGSTAMPNS_OLD:
3357 ops = READ_ONCE(sock->ops);
3358 if (!ops->gettstamp)
3359 return -ENOIOCTLCMD;
3360 return ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3361 !COMPAT_USE_64BIT_TIME);
3364 case SIOCBONDSLAVEINFOQUERY:
3365 case SIOCBONDINFOQUERY:
3368 return compat_ifr_data_ioctl(net, cmd, argp);
3379 case SIOCGSTAMP_NEW:
3380 case SIOCGSTAMPNS_NEW:
3384 return sock_ioctl(file, cmd, arg);
3403 case SIOCSIFHWBROADCAST:
3405 case SIOCGIFBRDADDR:
3406 case SIOCSIFBRDADDR:
3407 case SIOCGIFDSTADDR:
3408 case SIOCSIFDSTADDR:
3409 case SIOCGIFNETMASK:
3410 case SIOCSIFNETMASK:
3422 case SIOCBONDENSLAVE:
3423 case SIOCBONDRELEASE:
3424 case SIOCBONDSETHWADDR:
3425 case SIOCBONDCHANGEACTIVE:
3432 return sock_do_ioctl(net, sock, cmd, arg);
3435 return -ENOIOCTLCMD;
3438 static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3441 struct socket *sock = file->private_data;
3442 const struct proto_ops *ops = READ_ONCE(sock->ops);
3443 int ret = -ENOIOCTLCMD;
3450 if (ops->compat_ioctl)
3451 ret = ops->compat_ioctl(sock, cmd, arg);
3453 if (ret == -ENOIOCTLCMD &&
3454 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3455 ret = compat_wext_handle_ioctl(net, cmd, arg);
3457 if (ret == -ENOIOCTLCMD)
3458 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3465 * kernel_bind - bind an address to a socket (kernel space)
3468 * @addrlen: length of address
3470 * Returns 0 or an error.
3473 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3475 return READ_ONCE(sock->ops)->bind(sock, addr, addrlen);
3477 EXPORT_SYMBOL(kernel_bind);
3480 * kernel_listen - move socket to listening state (kernel space)
3482 * @backlog: pending connections queue size
3484 * Returns 0 or an error.
3487 int kernel_listen(struct socket *sock, int backlog)
3489 return READ_ONCE(sock->ops)->listen(sock, backlog);
3491 EXPORT_SYMBOL(kernel_listen);
3494 * kernel_accept - accept a connection (kernel space)
3495 * @sock: listening socket
3496 * @newsock: new connected socket
3499 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3500 * If it fails, @newsock is guaranteed to be %NULL.
3501 * Returns 0 or an error.
3504 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3506 struct sock *sk = sock->sk;
3507 const struct proto_ops *ops = READ_ONCE(sock->ops);
3510 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3515 err = ops->accept(sock, *newsock, flags, true);
3517 sock_release(*newsock);
3522 (*newsock)->ops = ops;
3523 __module_get(ops->owner);
3528 EXPORT_SYMBOL(kernel_accept);
3531 * kernel_connect - connect a socket (kernel space)
3534 * @addrlen: address length
3535 * @flags: flags (O_NONBLOCK, ...)
3537 * For datagram sockets, @addr is the address to which datagrams are sent
3538 * by default, and the only address from which datagrams are received.
3539 * For stream sockets, attempts to connect to @addr.
3540 * Returns 0 or an error code.
3543 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3546 return READ_ONCE(sock->ops)->connect(sock, addr, addrlen, flags);
3548 EXPORT_SYMBOL(kernel_connect);
3551 * kernel_getsockname - get the address which the socket is bound (kernel space)
3553 * @addr: address holder
3555 * Fills the @addr pointer with the address which the socket is bound.
3556 * Returns the length of the address in bytes or an error code.
3559 int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3561 return READ_ONCE(sock->ops)->getname(sock, addr, 0);
3563 EXPORT_SYMBOL(kernel_getsockname);
3566 * kernel_getpeername - get the address which the socket is connected (kernel space)
3568 * @addr: address holder
3570 * Fills the @addr pointer with the address which the socket is connected.
3571 * Returns the length of the address in bytes or an error code.
3574 int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3576 return READ_ONCE(sock->ops)->getname(sock, addr, 1);
3578 EXPORT_SYMBOL(kernel_getpeername);
3581 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3583 * @how: connection part
3585 * Returns 0 or an error.
3588 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3590 return READ_ONCE(sock->ops)->shutdown(sock, how);
3592 EXPORT_SYMBOL(kernel_sock_shutdown);
3595 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3598 * This routine returns the IP overhead imposed by a socket i.e.
3599 * the length of the underlying IP header, depending on whether
3600 * this is an IPv4 or IPv6 socket and the length from IP options turned
3601 * on at the socket. Assumes that the caller has a lock on the socket.
3604 u32 kernel_sock_ip_overhead(struct sock *sk)
3606 struct inet_sock *inet;
3607 struct ip_options_rcu *opt;
3609 #if IS_ENABLED(CONFIG_IPV6)
3610 struct ipv6_pinfo *np;
3611 struct ipv6_txoptions *optv6 = NULL;
3612 #endif /* IS_ENABLED(CONFIG_IPV6) */
3617 switch (sk->sk_family) {
3620 overhead += sizeof(struct iphdr);
3621 opt = rcu_dereference_protected(inet->inet_opt,
3622 sock_owned_by_user(sk));
3624 overhead += opt->opt.optlen;
3626 #if IS_ENABLED(CONFIG_IPV6)
3629 overhead += sizeof(struct ipv6hdr);
3631 optv6 = rcu_dereference_protected(np->opt,
3632 sock_owned_by_user(sk));
3634 overhead += (optv6->opt_flen + optv6->opt_nflen);
3636 #endif /* IS_ENABLED(CONFIG_IPV6) */
3637 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3641 EXPORT_SYMBOL(kernel_sock_ip_overhead);