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>
91 #include <linux/io_uring.h>
93 #include <linux/uaccess.h>
94 #include <asm/unistd.h>
96 #include <net/compat.h>
98 #include <net/cls_cgroup.h>
100 #include <net/sock.h>
101 #include <linux/netfilter.h>
103 #include <linux/if_tun.h>
104 #include <linux/ipv6_route.h>
105 #include <linux/route.h>
106 #include <linux/termios.h>
107 #include <linux/sockios.h>
108 #include <net/busy_poll.h>
109 #include <linux/errqueue.h>
110 #include <linux/ptp_clock_kernel.h>
111 #include <trace/events/sock.h>
113 #ifdef CONFIG_NET_RX_BUSY_POLL
114 unsigned int sysctl_net_busy_read __read_mostly;
115 unsigned int sysctl_net_busy_poll __read_mostly;
118 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
119 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
120 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
122 static int sock_close(struct inode *inode, struct file *file);
123 static __poll_t sock_poll(struct file *file,
124 struct poll_table_struct *wait);
125 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
127 static long compat_sock_ioctl(struct file *file,
128 unsigned int cmd, unsigned long arg);
130 static int sock_fasync(int fd, struct file *filp, int on);
131 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
132 struct pipe_inode_info *pipe, size_t len,
134 static void sock_splice_eof(struct file *file);
136 #ifdef CONFIG_PROC_FS
137 static void sock_show_fdinfo(struct seq_file *m, struct file *f)
139 struct socket *sock = f->private_data;
140 const struct proto_ops *ops = READ_ONCE(sock->ops);
142 if (ops->show_fdinfo)
143 ops->show_fdinfo(m, sock);
146 #define sock_show_fdinfo NULL
150 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
151 * in the operation structures but are done directly via the socketcall() multiplexor.
154 static const struct file_operations socket_file_ops = {
155 .owner = THIS_MODULE,
157 .read_iter = sock_read_iter,
158 .write_iter = sock_write_iter,
160 .unlocked_ioctl = sock_ioctl,
162 .compat_ioctl = compat_sock_ioctl,
164 .uring_cmd = io_uring_cmd_sock,
166 .release = sock_close,
167 .fasync = sock_fasync,
168 .splice_write = splice_to_socket,
169 .splice_read = sock_splice_read,
170 .splice_eof = sock_splice_eof,
171 .show_fdinfo = sock_show_fdinfo,
174 static const char * const pf_family_names[] = {
175 [PF_UNSPEC] = "PF_UNSPEC",
176 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
177 [PF_INET] = "PF_INET",
178 [PF_AX25] = "PF_AX25",
180 [PF_APPLETALK] = "PF_APPLETALK",
181 [PF_NETROM] = "PF_NETROM",
182 [PF_BRIDGE] = "PF_BRIDGE",
183 [PF_ATMPVC] = "PF_ATMPVC",
185 [PF_INET6] = "PF_INET6",
186 [PF_ROSE] = "PF_ROSE",
187 [PF_DECnet] = "PF_DECnet",
188 [PF_NETBEUI] = "PF_NETBEUI",
189 [PF_SECURITY] = "PF_SECURITY",
191 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
192 [PF_PACKET] = "PF_PACKET",
194 [PF_ECONET] = "PF_ECONET",
195 [PF_ATMSVC] = "PF_ATMSVC",
198 [PF_IRDA] = "PF_IRDA",
199 [PF_PPPOX] = "PF_PPPOX",
200 [PF_WANPIPE] = "PF_WANPIPE",
203 [PF_MPLS] = "PF_MPLS",
205 [PF_TIPC] = "PF_TIPC",
206 [PF_BLUETOOTH] = "PF_BLUETOOTH",
207 [PF_IUCV] = "PF_IUCV",
208 [PF_RXRPC] = "PF_RXRPC",
209 [PF_ISDN] = "PF_ISDN",
210 [PF_PHONET] = "PF_PHONET",
211 [PF_IEEE802154] = "PF_IEEE802154",
212 [PF_CAIF] = "PF_CAIF",
215 [PF_VSOCK] = "PF_VSOCK",
217 [PF_QIPCRTR] = "PF_QIPCRTR",
220 [PF_MCTP] = "PF_MCTP",
224 * The protocol list. Each protocol is registered in here.
227 static DEFINE_SPINLOCK(net_family_lock);
228 static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
232 * Move socket addresses back and forth across the kernel/user
233 * divide and look after the messy bits.
237 * move_addr_to_kernel - copy a socket address into kernel space
238 * @uaddr: Address in user space
239 * @kaddr: Address in kernel space
240 * @ulen: Length in user space
242 * The address is copied into kernel space. If the provided address is
243 * too long an error code of -EINVAL is returned. If the copy gives
244 * invalid addresses -EFAULT is returned. On a success 0 is returned.
247 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
249 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
253 if (copy_from_user(kaddr, uaddr, ulen))
255 return audit_sockaddr(ulen, kaddr);
259 * move_addr_to_user - copy an address to user space
260 * @kaddr: kernel space address
261 * @klen: length of address in kernel
262 * @uaddr: user space address
263 * @ulen: pointer to user length field
265 * The value pointed to by ulen on entry is the buffer length available.
266 * This is overwritten with the buffer space used. -EINVAL is returned
267 * if an overlong buffer is specified or a negative buffer size. -EFAULT
268 * is returned if either the buffer or the length field are not
270 * After copying the data up to the limit the user specifies, the true
271 * length of the data is written over the length limit the user
272 * specified. Zero is returned for a success.
275 static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
276 void __user *uaddr, int __user *ulen)
281 BUG_ON(klen > sizeof(struct sockaddr_storage));
282 err = get_user(len, ulen);
290 if (audit_sockaddr(klen, kaddr))
292 if (copy_to_user(uaddr, kaddr, len))
296 * "fromlen shall refer to the value before truncation.."
299 return __put_user(klen, ulen);
302 static struct kmem_cache *sock_inode_cachep __ro_after_init;
304 static struct inode *sock_alloc_inode(struct super_block *sb)
306 struct socket_alloc *ei;
308 ei = alloc_inode_sb(sb, sock_inode_cachep, GFP_KERNEL);
311 init_waitqueue_head(&ei->socket.wq.wait);
312 ei->socket.wq.fasync_list = NULL;
313 ei->socket.wq.flags = 0;
315 ei->socket.state = SS_UNCONNECTED;
316 ei->socket.flags = 0;
317 ei->socket.ops = NULL;
318 ei->socket.sk = NULL;
319 ei->socket.file = NULL;
321 return &ei->vfs_inode;
324 static void sock_free_inode(struct inode *inode)
326 struct socket_alloc *ei;
328 ei = container_of(inode, struct socket_alloc, vfs_inode);
329 kmem_cache_free(sock_inode_cachep, ei);
332 static void init_once(void *foo)
334 struct socket_alloc *ei = (struct socket_alloc *)foo;
336 inode_init_once(&ei->vfs_inode);
339 static void init_inodecache(void)
341 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
342 sizeof(struct socket_alloc),
344 (SLAB_HWCACHE_ALIGN |
345 SLAB_RECLAIM_ACCOUNT |
346 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
348 BUG_ON(sock_inode_cachep == NULL);
351 static const struct super_operations sockfs_ops = {
352 .alloc_inode = sock_alloc_inode,
353 .free_inode = sock_free_inode,
354 .statfs = simple_statfs,
358 * sockfs_dname() is called from d_path().
360 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
362 return dynamic_dname(buffer, buflen, "socket:[%lu]",
363 d_inode(dentry)->i_ino);
366 static const struct dentry_operations sockfs_dentry_operations = {
367 .d_dname = sockfs_dname,
370 static int sockfs_xattr_get(const struct xattr_handler *handler,
371 struct dentry *dentry, struct inode *inode,
372 const char *suffix, void *value, size_t size)
375 if (dentry->d_name.len + 1 > size)
377 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
379 return dentry->d_name.len + 1;
382 #define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
383 #define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
384 #define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
386 static const struct xattr_handler sockfs_xattr_handler = {
387 .name = XATTR_NAME_SOCKPROTONAME,
388 .get = sockfs_xattr_get,
391 static int sockfs_security_xattr_set(const struct xattr_handler *handler,
392 struct mnt_idmap *idmap,
393 struct dentry *dentry, struct inode *inode,
394 const char *suffix, const void *value,
395 size_t size, int flags)
397 /* Handled by LSM. */
401 static const struct xattr_handler sockfs_security_xattr_handler = {
402 .prefix = XATTR_SECURITY_PREFIX,
403 .set = sockfs_security_xattr_set,
406 static const struct xattr_handler *sockfs_xattr_handlers[] = {
407 &sockfs_xattr_handler,
408 &sockfs_security_xattr_handler,
412 static int sockfs_init_fs_context(struct fs_context *fc)
414 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
417 ctx->ops = &sockfs_ops;
418 ctx->dops = &sockfs_dentry_operations;
419 ctx->xattr = sockfs_xattr_handlers;
423 static struct vfsmount *sock_mnt __read_mostly;
425 static struct file_system_type sock_fs_type = {
427 .init_fs_context = sockfs_init_fs_context,
428 .kill_sb = kill_anon_super,
432 * Obtains the first available file descriptor and sets it up for use.
434 * These functions create file structures and maps them to fd space
435 * of the current process. On success it returns file descriptor
436 * and file struct implicitly stored in sock->file.
437 * Note that another thread may close file descriptor before we return
438 * from this function. We use the fact that now we do not refer
439 * to socket after mapping. If one day we will need it, this
440 * function will increment ref. count on file by 1.
442 * In any case returned fd MAY BE not valid!
443 * This race condition is unavoidable
444 * with shared fd spaces, we cannot solve it inside kernel,
445 * but we take care of internal coherence yet.
449 * sock_alloc_file - Bind a &socket to a &file
451 * @flags: file status flags
452 * @dname: protocol name
454 * Returns the &file bound with @sock, implicitly storing it
455 * in sock->file. If dname is %NULL, sets to "".
457 * On failure @sock is released, and an ERR pointer is returned.
459 * This function uses GFP_KERNEL internally.
462 struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
467 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
469 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
470 O_RDWR | (flags & O_NONBLOCK),
477 file->f_mode |= FMODE_NOWAIT;
479 file->private_data = sock;
480 stream_open(SOCK_INODE(sock), file);
483 EXPORT_SYMBOL(sock_alloc_file);
485 static int sock_map_fd(struct socket *sock, int flags)
487 struct file *newfile;
488 int fd = get_unused_fd_flags(flags);
489 if (unlikely(fd < 0)) {
494 newfile = sock_alloc_file(sock, flags, NULL);
495 if (!IS_ERR(newfile)) {
496 fd_install(fd, newfile);
501 return PTR_ERR(newfile);
505 * sock_from_file - Return the &socket bounded to @file.
508 * On failure returns %NULL.
511 struct socket *sock_from_file(struct file *file)
513 if (file->f_op == &socket_file_ops)
514 return file->private_data; /* set in sock_alloc_file */
518 EXPORT_SYMBOL(sock_from_file);
521 * sockfd_lookup - Go from a file number to its socket slot
523 * @err: pointer to an error code return
525 * The file handle passed in is locked and the socket it is bound
526 * to is returned. If an error occurs the err pointer is overwritten
527 * with a negative errno code and NULL is returned. The function checks
528 * for both invalid handles and passing a handle which is not a socket.
530 * On a success the socket object pointer is returned.
533 struct socket *sockfd_lookup(int fd, int *err)
544 sock = sock_from_file(file);
551 EXPORT_SYMBOL(sockfd_lookup);
553 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
555 struct fd f = fdget(fd);
560 sock = sock_from_file(f.file);
562 *fput_needed = f.flags & FDPUT_FPUT;
571 static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
577 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
587 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
592 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
599 static int sockfs_setattr(struct mnt_idmap *idmap,
600 struct dentry *dentry, struct iattr *iattr)
602 int err = simple_setattr(&nop_mnt_idmap, dentry, iattr);
604 if (!err && (iattr->ia_valid & ATTR_UID)) {
605 struct socket *sock = SOCKET_I(d_inode(dentry));
608 sock->sk->sk_uid = iattr->ia_uid;
616 static const struct inode_operations sockfs_inode_ops = {
617 .listxattr = sockfs_listxattr,
618 .setattr = sockfs_setattr,
622 * sock_alloc - allocate a socket
624 * Allocate a new inode and socket object. The two are bound together
625 * and initialised. The socket is then returned. If we are out of inodes
626 * NULL is returned. This functions uses GFP_KERNEL internally.
629 struct socket *sock_alloc(void)
634 inode = new_inode_pseudo(sock_mnt->mnt_sb);
638 sock = SOCKET_I(inode);
640 inode->i_ino = get_next_ino();
641 inode->i_mode = S_IFSOCK | S_IRWXUGO;
642 inode->i_uid = current_fsuid();
643 inode->i_gid = current_fsgid();
644 inode->i_op = &sockfs_inode_ops;
648 EXPORT_SYMBOL(sock_alloc);
650 static void __sock_release(struct socket *sock, struct inode *inode)
652 const struct proto_ops *ops = READ_ONCE(sock->ops);
655 struct module *owner = ops->owner;
667 if (sock->wq.fasync_list)
668 pr_err("%s: fasync list not empty!\n", __func__);
671 iput(SOCK_INODE(sock));
678 * sock_release - close a socket
679 * @sock: socket to close
681 * The socket is released from the protocol stack if it has a release
682 * callback, and the inode is then released if the socket is bound to
683 * an inode not a file.
685 void sock_release(struct socket *sock)
687 __sock_release(sock, NULL);
689 EXPORT_SYMBOL(sock_release);
691 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
693 u8 flags = *tx_flags;
695 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE) {
696 flags |= SKBTX_HW_TSTAMP;
698 /* PTP hardware clocks can provide a free running cycle counter
699 * as a time base for virtual clocks. Tell driver to use the
700 * free running cycle counter for timestamp if socket is bound
703 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
704 flags |= SKBTX_HW_TSTAMP_USE_CYCLES;
707 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
708 flags |= SKBTX_SW_TSTAMP;
710 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
711 flags |= SKBTX_SCHED_TSTAMP;
715 EXPORT_SYMBOL(__sock_tx_timestamp);
717 INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
719 INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
722 static noinline void call_trace_sock_send_length(struct sock *sk, int ret,
725 trace_sock_send_length(sk, ret, 0);
728 static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
730 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->sendmsg, inet6_sendmsg,
731 inet_sendmsg, sock, msg,
733 BUG_ON(ret == -EIOCBQUEUED);
735 if (trace_sock_send_length_enabled())
736 call_trace_sock_send_length(sock->sk, ret, 0);
741 * sock_sendmsg - send a message through @sock
743 * @msg: message to send
745 * Sends @msg through @sock, passing through LSM.
746 * Returns the number of bytes sent, or an error code.
748 int sock_sendmsg(struct socket *sock, struct msghdr *msg)
750 int err = security_socket_sendmsg(sock, msg,
753 return err ?: sock_sendmsg_nosec(sock, msg);
755 EXPORT_SYMBOL(sock_sendmsg);
758 * kernel_sendmsg - send a message through @sock (kernel-space)
760 * @msg: message header
762 * @num: vec array length
763 * @size: total message data size
765 * Builds the message data with @vec and sends it through @sock.
766 * Returns the number of bytes sent, or an error code.
769 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
770 struct kvec *vec, size_t num, size_t size)
772 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
773 return sock_sendmsg(sock, msg);
775 EXPORT_SYMBOL(kernel_sendmsg);
778 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
780 * @msg: message header
781 * @vec: output s/g array
782 * @num: output s/g array length
783 * @size: total message data size
785 * Builds the message data with @vec and sends it through @sock.
786 * Returns the number of bytes sent, or an error code.
787 * Caller must hold @sk.
790 int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
791 struct kvec *vec, size_t num, size_t size)
793 struct socket *sock = sk->sk_socket;
794 const struct proto_ops *ops = READ_ONCE(sock->ops);
796 if (!ops->sendmsg_locked)
797 return sock_no_sendmsg_locked(sk, msg, size);
799 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
801 return ops->sendmsg_locked(sk, msg, msg_data_left(msg));
803 EXPORT_SYMBOL(kernel_sendmsg_locked);
805 static bool skb_is_err_queue(const struct sk_buff *skb)
807 /* pkt_type of skbs enqueued on the error queue are set to
808 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
809 * in recvmsg, since skbs received on a local socket will never
810 * have a pkt_type of PACKET_OUTGOING.
812 return skb->pkt_type == PACKET_OUTGOING;
815 /* On transmit, software and hardware timestamps are returned independently.
816 * As the two skb clones share the hardware timestamp, which may be updated
817 * before the software timestamp is received, a hardware TX timestamp may be
818 * returned only if there is no software TX timestamp. Ignore false software
819 * timestamps, which may be made in the __sock_recv_timestamp() call when the
820 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
821 * hardware timestamp.
823 static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
825 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
828 static ktime_t get_timestamp(struct sock *sk, struct sk_buff *skb, int *if_index)
830 bool cycles = sk->sk_tsflags & SOF_TIMESTAMPING_BIND_PHC;
831 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
832 struct net_device *orig_dev;
836 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
838 *if_index = orig_dev->ifindex;
839 hwtstamp = netdev_get_tstamp(orig_dev, shhwtstamps, cycles);
841 hwtstamp = shhwtstamps->hwtstamp;
848 static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb,
851 struct scm_ts_pktinfo ts_pktinfo;
852 struct net_device *orig_dev;
854 if (!skb_mac_header_was_set(skb))
857 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
861 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
863 if_index = orig_dev->ifindex;
866 ts_pktinfo.if_index = if_index;
868 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
869 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
870 sizeof(ts_pktinfo), &ts_pktinfo);
874 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
876 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
879 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
880 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
881 struct scm_timestamping_internal tss;
883 int empty = 1, false_tstamp = 0;
884 struct skb_shared_hwtstamps *shhwtstamps =
889 /* Race occurred between timestamp enabling and packet
890 receiving. Fill in the current time for now. */
891 if (need_software_tstamp && skb->tstamp == 0) {
892 __net_timestamp(skb);
896 if (need_software_tstamp) {
897 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
899 struct __kernel_sock_timeval tv;
901 skb_get_new_timestamp(skb, &tv);
902 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
905 struct __kernel_old_timeval tv;
907 skb_get_timestamp(skb, &tv);
908 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
913 struct __kernel_timespec ts;
915 skb_get_new_timestampns(skb, &ts);
916 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
919 struct __kernel_old_timespec ts;
921 skb_get_timestampns(skb, &ts);
922 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
928 memset(&tss, 0, sizeof(tss));
929 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
930 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
933 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
934 !skb_is_swtx_tstamp(skb, false_tstamp)) {
936 if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP_NETDEV)
937 hwtstamp = get_timestamp(sk, skb, &if_index);
939 hwtstamp = shhwtstamps->hwtstamp;
941 if (sk->sk_tsflags & SOF_TIMESTAMPING_BIND_PHC)
942 hwtstamp = ptp_convert_timestamp(&hwtstamp,
945 if (ktime_to_timespec64_cond(hwtstamp, tss.ts + 2)) {
948 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
949 !skb_is_err_queue(skb))
950 put_ts_pktinfo(msg, skb, if_index);
954 if (sock_flag(sk, SOCK_TSTAMP_NEW))
955 put_cmsg_scm_timestamping64(msg, &tss);
957 put_cmsg_scm_timestamping(msg, &tss);
959 if (skb_is_err_queue(skb) && skb->len &&
960 SKB_EXT_ERR(skb)->opt_stats)
961 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
962 skb->len, skb->data);
965 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
967 #ifdef CONFIG_WIRELESS
968 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
973 if (!sock_flag(sk, SOCK_WIFI_STATUS))
975 if (!skb->wifi_acked_valid)
978 ack = skb->wifi_acked;
980 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
982 EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
985 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
988 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
989 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
990 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
993 static void sock_recv_mark(struct msghdr *msg, struct sock *sk,
996 if (sock_flag(sk, SOCK_RCVMARK) && skb) {
997 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */
998 __u32 mark = skb->mark;
1000 put_cmsg(msg, SOL_SOCKET, SO_MARK, sizeof(__u32), &mark);
1004 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
1005 struct sk_buff *skb)
1007 sock_recv_timestamp(msg, sk, skb);
1008 sock_recv_drops(msg, sk, skb);
1009 sock_recv_mark(msg, sk, skb);
1011 EXPORT_SYMBOL_GPL(__sock_recv_cmsgs);
1013 INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
1015 INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
1018 static noinline void call_trace_sock_recv_length(struct sock *sk, int ret, int flags)
1020 trace_sock_recv_length(sk, ret, flags);
1023 static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
1026 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->recvmsg,
1028 inet_recvmsg, sock, msg,
1029 msg_data_left(msg), flags);
1030 if (trace_sock_recv_length_enabled())
1031 call_trace_sock_recv_length(sock->sk, ret, flags);
1036 * sock_recvmsg - receive a message from @sock
1038 * @msg: message to receive
1039 * @flags: message flags
1041 * Receives @msg from @sock, passing through LSM. Returns the total number
1042 * of bytes received, or an error.
1044 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
1046 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
1048 return err ?: sock_recvmsg_nosec(sock, msg, flags);
1050 EXPORT_SYMBOL(sock_recvmsg);
1053 * kernel_recvmsg - Receive a message from a socket (kernel space)
1054 * @sock: The socket to receive the message from
1055 * @msg: Received message
1056 * @vec: Input s/g array for message data
1057 * @num: Size of input s/g array
1058 * @size: Number of bytes to read
1059 * @flags: Message flags (MSG_DONTWAIT, etc...)
1061 * On return the msg structure contains the scatter/gather array passed in the
1062 * vec argument. The array is modified so that it consists of the unfilled
1063 * portion of the original array.
1065 * The returned value is the total number of bytes received, or an error.
1068 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
1069 struct kvec *vec, size_t num, size_t size, int flags)
1071 msg->msg_control_is_user = false;
1072 iov_iter_kvec(&msg->msg_iter, ITER_DEST, vec, num, size);
1073 return sock_recvmsg(sock, msg, flags);
1075 EXPORT_SYMBOL(kernel_recvmsg);
1077 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1078 struct pipe_inode_info *pipe, size_t len,
1081 struct socket *sock = file->private_data;
1082 const struct proto_ops *ops;
1084 ops = READ_ONCE(sock->ops);
1085 if (unlikely(!ops->splice_read))
1086 return copy_splice_read(file, ppos, pipe, len, flags);
1088 return ops->splice_read(sock, ppos, pipe, len, flags);
1091 static void sock_splice_eof(struct file *file)
1093 struct socket *sock = file->private_data;
1094 const struct proto_ops *ops;
1096 ops = READ_ONCE(sock->ops);
1097 if (ops->splice_eof)
1098 ops->splice_eof(sock);
1101 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1103 struct file *file = iocb->ki_filp;
1104 struct socket *sock = file->private_data;
1105 struct msghdr msg = {.msg_iter = *to,
1109 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1110 msg.msg_flags = MSG_DONTWAIT;
1112 if (iocb->ki_pos != 0)
1115 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1118 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1123 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1125 struct file *file = iocb->ki_filp;
1126 struct socket *sock = file->private_data;
1127 struct msghdr msg = {.msg_iter = *from,
1131 if (iocb->ki_pos != 0)
1134 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1135 msg.msg_flags = MSG_DONTWAIT;
1137 if (sock->type == SOCK_SEQPACKET)
1138 msg.msg_flags |= MSG_EOR;
1140 res = sock_sendmsg(sock, &msg);
1141 *from = msg.msg_iter;
1146 * Atomic setting of ioctl hooks to avoid race
1147 * with module unload.
1150 static DEFINE_MUTEX(br_ioctl_mutex);
1151 static int (*br_ioctl_hook)(struct net *net, struct net_bridge *br,
1152 unsigned int cmd, struct ifreq *ifr,
1155 void brioctl_set(int (*hook)(struct net *net, struct net_bridge *br,
1156 unsigned int cmd, struct ifreq *ifr,
1159 mutex_lock(&br_ioctl_mutex);
1160 br_ioctl_hook = hook;
1161 mutex_unlock(&br_ioctl_mutex);
1163 EXPORT_SYMBOL(brioctl_set);
1165 int br_ioctl_call(struct net *net, struct net_bridge *br, unsigned int cmd,
1166 struct ifreq *ifr, void __user *uarg)
1171 request_module("bridge");
1173 mutex_lock(&br_ioctl_mutex);
1175 err = br_ioctl_hook(net, br, cmd, ifr, uarg);
1176 mutex_unlock(&br_ioctl_mutex);
1181 static DEFINE_MUTEX(vlan_ioctl_mutex);
1182 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1184 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1186 mutex_lock(&vlan_ioctl_mutex);
1187 vlan_ioctl_hook = hook;
1188 mutex_unlock(&vlan_ioctl_mutex);
1190 EXPORT_SYMBOL(vlan_ioctl_set);
1192 static long sock_do_ioctl(struct net *net, struct socket *sock,
1193 unsigned int cmd, unsigned long arg)
1195 const struct proto_ops *ops = READ_ONCE(sock->ops);
1199 void __user *argp = (void __user *)arg;
1202 err = ops->ioctl(sock, cmd, arg);
1205 * If this ioctl is unknown try to hand it down
1206 * to the NIC driver.
1208 if (err != -ENOIOCTLCMD)
1211 if (!is_socket_ioctl_cmd(cmd))
1214 if (get_user_ifreq(&ifr, &data, argp))
1216 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1217 if (!err && need_copyout)
1218 if (put_user_ifreq(&ifr, argp))
1225 * With an ioctl, arg may well be a user mode pointer, but we don't know
1226 * what to do with it - that's up to the protocol still.
1229 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1231 const struct proto_ops *ops;
1232 struct socket *sock;
1234 void __user *argp = (void __user *)arg;
1238 sock = file->private_data;
1239 ops = READ_ONCE(sock->ops);
1242 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1246 if (get_user_ifreq(&ifr, &data, argp))
1248 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1249 if (!err && need_copyout)
1250 if (put_user_ifreq(&ifr, argp))
1253 #ifdef CONFIG_WEXT_CORE
1254 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1255 err = wext_handle_ioctl(net, cmd, argp);
1262 if (get_user(pid, (int __user *)argp))
1264 err = f_setown(sock->file, pid, 1);
1268 err = put_user(f_getown(sock->file),
1269 (int __user *)argp);
1275 err = br_ioctl_call(net, NULL, cmd, NULL, argp);
1280 if (!vlan_ioctl_hook)
1281 request_module("8021q");
1283 mutex_lock(&vlan_ioctl_mutex);
1284 if (vlan_ioctl_hook)
1285 err = vlan_ioctl_hook(net, argp);
1286 mutex_unlock(&vlan_ioctl_mutex);
1290 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1293 err = open_related_ns(&net->ns, get_net_ns);
1295 case SIOCGSTAMP_OLD:
1296 case SIOCGSTAMPNS_OLD:
1297 if (!ops->gettstamp) {
1301 err = ops->gettstamp(sock, argp,
1302 cmd == SIOCGSTAMP_OLD,
1303 !IS_ENABLED(CONFIG_64BIT));
1305 case SIOCGSTAMP_NEW:
1306 case SIOCGSTAMPNS_NEW:
1307 if (!ops->gettstamp) {
1311 err = ops->gettstamp(sock, argp,
1312 cmd == SIOCGSTAMP_NEW,
1317 err = dev_ifconf(net, argp);
1321 err = sock_do_ioctl(net, sock, cmd, arg);
1328 * sock_create_lite - creates a socket
1329 * @family: protocol family (AF_INET, ...)
1330 * @type: communication type (SOCK_STREAM, ...)
1331 * @protocol: protocol (0, ...)
1334 * Creates a new socket and assigns it to @res, passing through LSM.
1335 * The new socket initialization is not complete, see kernel_accept().
1336 * Returns 0 or an error. On failure @res is set to %NULL.
1337 * This function internally uses GFP_KERNEL.
1340 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1343 struct socket *sock = NULL;
1345 err = security_socket_create(family, type, protocol, 1);
1349 sock = sock_alloc();
1356 err = security_socket_post_create(sock, family, type, protocol, 1);
1368 EXPORT_SYMBOL(sock_create_lite);
1370 /* No kernel lock held - perfect */
1371 static __poll_t sock_poll(struct file *file, poll_table *wait)
1373 struct socket *sock = file->private_data;
1374 const struct proto_ops *ops = READ_ONCE(sock->ops);
1375 __poll_t events = poll_requested_events(wait), flag = 0;
1380 if (sk_can_busy_loop(sock->sk)) {
1381 /* poll once if requested by the syscall */
1382 if (events & POLL_BUSY_LOOP)
1383 sk_busy_loop(sock->sk, 1);
1385 /* if this socket can poll_ll, tell the system call */
1386 flag = POLL_BUSY_LOOP;
1389 return ops->poll(file, sock, wait) | flag;
1392 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1394 struct socket *sock = file->private_data;
1396 return READ_ONCE(sock->ops)->mmap(file, sock, vma);
1399 static int sock_close(struct inode *inode, struct file *filp)
1401 __sock_release(SOCKET_I(inode), inode);
1406 * Update the socket async list
1408 * Fasync_list locking strategy.
1410 * 1. fasync_list is modified only under process context socket lock
1411 * i.e. under semaphore.
1412 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1413 * or under socket lock
1416 static int sock_fasync(int fd, struct file *filp, int on)
1418 struct socket *sock = filp->private_data;
1419 struct sock *sk = sock->sk;
1420 struct socket_wq *wq = &sock->wq;
1426 fasync_helper(fd, filp, on, &wq->fasync_list);
1428 if (!wq->fasync_list)
1429 sock_reset_flag(sk, SOCK_FASYNC);
1431 sock_set_flag(sk, SOCK_FASYNC);
1437 /* This function may be called only under rcu_lock */
1439 int sock_wake_async(struct socket_wq *wq, int how, int band)
1441 if (!wq || !wq->fasync_list)
1445 case SOCK_WAKE_WAITD:
1446 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1449 case SOCK_WAKE_SPACE:
1450 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1455 kill_fasync(&wq->fasync_list, SIGIO, band);
1458 kill_fasync(&wq->fasync_list, SIGURG, band);
1463 EXPORT_SYMBOL(sock_wake_async);
1466 * __sock_create - creates a socket
1467 * @net: net namespace
1468 * @family: protocol family (AF_INET, ...)
1469 * @type: communication type (SOCK_STREAM, ...)
1470 * @protocol: protocol (0, ...)
1472 * @kern: boolean for kernel space sockets
1474 * Creates a new socket and assigns it to @res, passing through LSM.
1475 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1476 * be set to true if the socket resides in kernel space.
1477 * This function internally uses GFP_KERNEL.
1480 int __sock_create(struct net *net, int family, int type, int protocol,
1481 struct socket **res, int kern)
1484 struct socket *sock;
1485 const struct net_proto_family *pf;
1488 * Check protocol is in range
1490 if (family < 0 || family >= NPROTO)
1491 return -EAFNOSUPPORT;
1492 if (type < 0 || type >= SOCK_MAX)
1497 This uglymoron is moved from INET layer to here to avoid
1498 deadlock in module load.
1500 if (family == PF_INET && type == SOCK_PACKET) {
1501 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1506 err = security_socket_create(family, type, protocol, kern);
1511 * Allocate the socket and allow the family to set things up. if
1512 * the protocol is 0, the family is instructed to select an appropriate
1515 sock = sock_alloc();
1517 net_warn_ratelimited("socket: no more sockets\n");
1518 return -ENFILE; /* Not exactly a match, but its the
1519 closest posix thing */
1524 #ifdef CONFIG_MODULES
1525 /* Attempt to load a protocol module if the find failed.
1527 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1528 * requested real, full-featured networking support upon configuration.
1529 * Otherwise module support will break!
1531 if (rcu_access_pointer(net_families[family]) == NULL)
1532 request_module("net-pf-%d", family);
1536 pf = rcu_dereference(net_families[family]);
1537 err = -EAFNOSUPPORT;
1542 * We will call the ->create function, that possibly is in a loadable
1543 * module, so we have to bump that loadable module refcnt first.
1545 if (!try_module_get(pf->owner))
1548 /* Now protected by module ref count */
1551 err = pf->create(net, sock, protocol, kern);
1553 goto out_module_put;
1556 * Now to bump the refcnt of the [loadable] module that owns this
1557 * socket at sock_release time we decrement its refcnt.
1559 if (!try_module_get(sock->ops->owner))
1560 goto out_module_busy;
1563 * Now that we're done with the ->create function, the [loadable]
1564 * module can have its refcnt decremented
1566 module_put(pf->owner);
1567 err = security_socket_post_create(sock, family, type, protocol, kern);
1569 goto out_sock_release;
1575 err = -EAFNOSUPPORT;
1578 module_put(pf->owner);
1585 goto out_sock_release;
1587 EXPORT_SYMBOL(__sock_create);
1590 * sock_create - creates a socket
1591 * @family: protocol family (AF_INET, ...)
1592 * @type: communication type (SOCK_STREAM, ...)
1593 * @protocol: protocol (0, ...)
1596 * A wrapper around __sock_create().
1597 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1600 int sock_create(int family, int type, int protocol, struct socket **res)
1602 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1604 EXPORT_SYMBOL(sock_create);
1607 * sock_create_kern - creates a socket (kernel space)
1608 * @net: net namespace
1609 * @family: protocol family (AF_INET, ...)
1610 * @type: communication type (SOCK_STREAM, ...)
1611 * @protocol: protocol (0, ...)
1614 * A wrapper around __sock_create().
1615 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1618 int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1620 return __sock_create(net, family, type, protocol, res, 1);
1622 EXPORT_SYMBOL(sock_create_kern);
1624 static struct socket *__sys_socket_create(int family, int type, int protocol)
1626 struct socket *sock;
1629 /* Check the SOCK_* constants for consistency. */
1630 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1631 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1632 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1633 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1635 if ((type & ~SOCK_TYPE_MASK) & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1636 return ERR_PTR(-EINVAL);
1637 type &= SOCK_TYPE_MASK;
1639 retval = sock_create(family, type, protocol, &sock);
1641 return ERR_PTR(retval);
1646 struct file *__sys_socket_file(int family, int type, int protocol)
1648 struct socket *sock;
1651 sock = __sys_socket_create(family, type, protocol);
1653 return ERR_CAST(sock);
1655 flags = type & ~SOCK_TYPE_MASK;
1656 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1657 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1659 return sock_alloc_file(sock, flags, NULL);
1662 /* A hook for bpf progs to attach to and update socket protocol.
1664 * A static noinline declaration here could cause the compiler to
1665 * optimize away the function. A global noinline declaration will
1666 * keep the definition, but may optimize away the callsite.
1667 * Therefore, __weak is needed to ensure that the call is still
1668 * emitted, by telling the compiler that we don't know what the
1669 * function might eventually be.
1671 * __diag_* below are needed to dismiss the missing prototype warning.
1675 __diag_ignore_all("-Wmissing-prototypes",
1676 "A fmod_ret entry point for BPF programs");
1678 __weak noinline int update_socket_protocol(int family, int type, int protocol)
1685 int __sys_socket(int family, int type, int protocol)
1687 struct socket *sock;
1690 sock = __sys_socket_create(family, type,
1691 update_socket_protocol(family, type, protocol));
1693 return PTR_ERR(sock);
1695 flags = type & ~SOCK_TYPE_MASK;
1696 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1697 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1699 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1702 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1704 return __sys_socket(family, type, protocol);
1708 * Create a pair of connected sockets.
1711 int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1713 struct socket *sock1, *sock2;
1715 struct file *newfile1, *newfile2;
1718 flags = type & ~SOCK_TYPE_MASK;
1719 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1721 type &= SOCK_TYPE_MASK;
1723 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1724 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1727 * reserve descriptors and make sure we won't fail
1728 * to return them to userland.
1730 fd1 = get_unused_fd_flags(flags);
1731 if (unlikely(fd1 < 0))
1734 fd2 = get_unused_fd_flags(flags);
1735 if (unlikely(fd2 < 0)) {
1740 err = put_user(fd1, &usockvec[0]);
1744 err = put_user(fd2, &usockvec[1]);
1749 * Obtain the first socket and check if the underlying protocol
1750 * supports the socketpair call.
1753 err = sock_create(family, type, protocol, &sock1);
1754 if (unlikely(err < 0))
1757 err = sock_create(family, type, protocol, &sock2);
1758 if (unlikely(err < 0)) {
1759 sock_release(sock1);
1763 err = security_socket_socketpair(sock1, sock2);
1764 if (unlikely(err)) {
1765 sock_release(sock2);
1766 sock_release(sock1);
1770 err = READ_ONCE(sock1->ops)->socketpair(sock1, sock2);
1771 if (unlikely(err < 0)) {
1772 sock_release(sock2);
1773 sock_release(sock1);
1777 newfile1 = sock_alloc_file(sock1, flags, NULL);
1778 if (IS_ERR(newfile1)) {
1779 err = PTR_ERR(newfile1);
1780 sock_release(sock2);
1784 newfile2 = sock_alloc_file(sock2, flags, NULL);
1785 if (IS_ERR(newfile2)) {
1786 err = PTR_ERR(newfile2);
1791 audit_fd_pair(fd1, fd2);
1793 fd_install(fd1, newfile1);
1794 fd_install(fd2, newfile2);
1803 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1804 int __user *, usockvec)
1806 return __sys_socketpair(family, type, protocol, usockvec);
1810 * Bind a name to a socket. Nothing much to do here since it's
1811 * the protocol's responsibility to handle the local address.
1813 * We move the socket address to kernel space before we call
1814 * the protocol layer (having also checked the address is ok).
1817 int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1819 struct socket *sock;
1820 struct sockaddr_storage address;
1821 int err, fput_needed;
1823 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1825 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1827 err = security_socket_bind(sock,
1828 (struct sockaddr *)&address,
1831 err = READ_ONCE(sock->ops)->bind(sock,
1835 fput_light(sock->file, fput_needed);
1840 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1842 return __sys_bind(fd, umyaddr, addrlen);
1846 * Perform a listen. Basically, we allow the protocol to do anything
1847 * necessary for a listen, and if that works, we mark the socket as
1848 * ready for listening.
1851 int __sys_listen(int fd, int backlog)
1853 struct socket *sock;
1854 int err, fput_needed;
1857 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1859 somaxconn = READ_ONCE(sock_net(sock->sk)->core.sysctl_somaxconn);
1860 if ((unsigned int)backlog > somaxconn)
1861 backlog = somaxconn;
1863 err = security_socket_listen(sock, backlog);
1865 err = READ_ONCE(sock->ops)->listen(sock, backlog);
1867 fput_light(sock->file, fput_needed);
1872 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1874 return __sys_listen(fd, backlog);
1877 struct file *do_accept(struct file *file, unsigned file_flags,
1878 struct sockaddr __user *upeer_sockaddr,
1879 int __user *upeer_addrlen, int flags)
1881 struct socket *sock, *newsock;
1882 struct file *newfile;
1884 struct sockaddr_storage address;
1885 const struct proto_ops *ops;
1887 sock = sock_from_file(file);
1889 return ERR_PTR(-ENOTSOCK);
1891 newsock = sock_alloc();
1893 return ERR_PTR(-ENFILE);
1894 ops = READ_ONCE(sock->ops);
1896 newsock->type = sock->type;
1900 * We don't need try_module_get here, as the listening socket (sock)
1901 * has the protocol module (sock->ops->owner) held.
1903 __module_get(ops->owner);
1905 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1906 if (IS_ERR(newfile))
1909 err = security_socket_accept(sock, newsock);
1913 err = ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1918 if (upeer_sockaddr) {
1919 len = ops->getname(newsock, (struct sockaddr *)&address, 2);
1921 err = -ECONNABORTED;
1924 err = move_addr_to_user(&address,
1925 len, upeer_sockaddr, upeer_addrlen);
1930 /* File flags are not inherited via accept() unlike another OSes. */
1934 return ERR_PTR(err);
1937 static int __sys_accept4_file(struct file *file, struct sockaddr __user *upeer_sockaddr,
1938 int __user *upeer_addrlen, int flags)
1940 struct file *newfile;
1943 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1946 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1947 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1949 newfd = get_unused_fd_flags(flags);
1950 if (unlikely(newfd < 0))
1953 newfile = do_accept(file, 0, upeer_sockaddr, upeer_addrlen,
1955 if (IS_ERR(newfile)) {
1956 put_unused_fd(newfd);
1957 return PTR_ERR(newfile);
1959 fd_install(newfd, newfile);
1964 * For accept, we attempt to create a new socket, set up the link
1965 * with the client, wake up the client, then return the new
1966 * connected fd. We collect the address of the connector in kernel
1967 * space and move it to user at the very end. This is unclean because
1968 * we open the socket then return an error.
1970 * 1003.1g adds the ability to recvmsg() to query connection pending
1971 * status to recvmsg. We need to add that support in a way thats
1972 * clean when we restructure accept also.
1975 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1976 int __user *upeer_addrlen, int flags)
1983 ret = __sys_accept4_file(f.file, upeer_sockaddr,
1984 upeer_addrlen, flags);
1991 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1992 int __user *, upeer_addrlen, int, flags)
1994 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1997 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1998 int __user *, upeer_addrlen)
2000 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
2004 * Attempt to connect to a socket with the server address. The address
2005 * is in user space so we verify it is OK and move it to kernel space.
2007 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
2010 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
2011 * other SEQPACKET protocols that take time to connect() as it doesn't
2012 * include the -EINPROGRESS status for such sockets.
2015 int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
2016 int addrlen, int file_flags)
2018 struct socket *sock;
2021 sock = sock_from_file(file);
2028 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
2032 err = READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)address,
2033 addrlen, sock->file->f_flags | file_flags);
2038 int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
2045 struct sockaddr_storage address;
2047 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
2049 ret = __sys_connect_file(f.file, &address, addrlen, 0);
2056 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
2059 return __sys_connect(fd, uservaddr, addrlen);
2063 * Get the local address ('name') of a socket object. Move the obtained
2064 * name to user space.
2067 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
2068 int __user *usockaddr_len)
2070 struct socket *sock;
2071 struct sockaddr_storage address;
2072 int err, fput_needed;
2074 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2078 err = security_socket_getsockname(sock);
2082 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 0);
2085 /* "err" is actually length in this case */
2086 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
2089 fput_light(sock->file, fput_needed);
2094 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
2095 int __user *, usockaddr_len)
2097 return __sys_getsockname(fd, usockaddr, usockaddr_len);
2101 * Get the remote address ('name') of a socket object. Move the obtained
2102 * name to user space.
2105 int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
2106 int __user *usockaddr_len)
2108 struct socket *sock;
2109 struct sockaddr_storage address;
2110 int err, fput_needed;
2112 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2114 const struct proto_ops *ops = READ_ONCE(sock->ops);
2116 err = security_socket_getpeername(sock);
2118 fput_light(sock->file, fput_needed);
2122 err = ops->getname(sock, (struct sockaddr *)&address, 1);
2124 /* "err" is actually length in this case */
2125 err = move_addr_to_user(&address, err, usockaddr,
2127 fput_light(sock->file, fput_needed);
2132 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
2133 int __user *, usockaddr_len)
2135 return __sys_getpeername(fd, usockaddr, usockaddr_len);
2139 * Send a datagram to a given address. We move the address into kernel
2140 * space and check the user space data area is readable before invoking
2143 int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
2144 struct sockaddr __user *addr, int addr_len)
2146 struct socket *sock;
2147 struct sockaddr_storage address;
2153 err = import_single_range(ITER_SOURCE, buff, len, &iov, &msg.msg_iter);
2156 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2160 msg.msg_name = NULL;
2161 msg.msg_control = NULL;
2162 msg.msg_controllen = 0;
2163 msg.msg_namelen = 0;
2164 msg.msg_ubuf = NULL;
2166 err = move_addr_to_kernel(addr, addr_len, &address);
2169 msg.msg_name = (struct sockaddr *)&address;
2170 msg.msg_namelen = addr_len;
2172 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2173 if (sock->file->f_flags & O_NONBLOCK)
2174 flags |= MSG_DONTWAIT;
2175 msg.msg_flags = flags;
2176 err = sock_sendmsg(sock, &msg);
2179 fput_light(sock->file, fput_needed);
2184 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2185 unsigned int, flags, struct sockaddr __user *, addr,
2188 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2192 * Send a datagram down a socket.
2195 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2196 unsigned int, flags)
2198 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2202 * Receive a frame from the socket and optionally record the address of the
2203 * sender. We verify the buffers are writable and if needed move the
2204 * sender address from kernel to user space.
2206 int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2207 struct sockaddr __user *addr, int __user *addr_len)
2209 struct sockaddr_storage address;
2210 struct msghdr msg = {
2211 /* Save some cycles and don't copy the address if not needed */
2212 .msg_name = addr ? (struct sockaddr *)&address : NULL,
2214 struct socket *sock;
2219 err = import_single_range(ITER_DEST, ubuf, size, &iov, &msg.msg_iter);
2222 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2226 if (sock->file->f_flags & O_NONBLOCK)
2227 flags |= MSG_DONTWAIT;
2228 err = sock_recvmsg(sock, &msg, flags);
2230 if (err >= 0 && addr != NULL) {
2231 err2 = move_addr_to_user(&address,
2232 msg.msg_namelen, addr, addr_len);
2237 fput_light(sock->file, fput_needed);
2242 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2243 unsigned int, flags, struct sockaddr __user *, addr,
2244 int __user *, addr_len)
2246 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2250 * Receive a datagram from a socket.
2253 SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2254 unsigned int, flags)
2256 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2259 static bool sock_use_custom_sol_socket(const struct socket *sock)
2261 return test_bit(SOCK_CUSTOM_SOCKOPT, &sock->flags);
2265 * Set a socket option. Because we don't know the option lengths we have
2266 * to pass the user mode parameter for the protocols to sort out.
2268 int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2271 sockptr_t optval = USER_SOCKPTR(user_optval);
2272 const struct proto_ops *ops;
2273 char *kernel_optval = NULL;
2274 int err, fput_needed;
2275 struct socket *sock;
2280 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2284 err = security_socket_setsockopt(sock, level, optname);
2288 if (!in_compat_syscall())
2289 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2290 user_optval, &optlen,
2300 optval = KERNEL_SOCKPTR(kernel_optval);
2301 ops = READ_ONCE(sock->ops);
2302 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2303 err = sock_setsockopt(sock, level, optname, optval, optlen);
2304 else if (unlikely(!ops->setsockopt))
2307 err = ops->setsockopt(sock, level, optname, optval,
2309 kfree(kernel_optval);
2311 fput_light(sock->file, fput_needed);
2315 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2316 char __user *, optval, int, optlen)
2318 return __sys_setsockopt(fd, level, optname, optval, optlen);
2321 INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2325 * Get a socket option. Because we don't know the option lengths we have
2326 * to pass a user mode parameter for the protocols to sort out.
2328 int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2331 int max_optlen __maybe_unused;
2332 const struct proto_ops *ops;
2333 int err, fput_needed;
2334 struct socket *sock;
2336 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2340 err = security_socket_getsockopt(sock, level, optname);
2344 if (!in_compat_syscall())
2345 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2347 ops = READ_ONCE(sock->ops);
2348 if (level == SOL_SOCKET)
2349 err = sock_getsockopt(sock, level, optname, optval, optlen);
2350 else if (unlikely(!ops->getsockopt))
2353 err = ops->getsockopt(sock, level, optname, optval,
2356 if (!in_compat_syscall())
2357 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2358 optval, optlen, max_optlen,
2361 fput_light(sock->file, fput_needed);
2365 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2366 char __user *, optval, int __user *, optlen)
2368 return __sys_getsockopt(fd, level, optname, optval, optlen);
2372 * Shutdown a socket.
2375 int __sys_shutdown_sock(struct socket *sock, int how)
2379 err = security_socket_shutdown(sock, how);
2381 err = READ_ONCE(sock->ops)->shutdown(sock, how);
2386 int __sys_shutdown(int fd, int how)
2388 int err, fput_needed;
2389 struct socket *sock;
2391 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2393 err = __sys_shutdown_sock(sock, how);
2394 fput_light(sock->file, fput_needed);
2399 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2401 return __sys_shutdown(fd, how);
2404 /* A couple of helpful macros for getting the address of the 32/64 bit
2405 * fields which are the same type (int / unsigned) on our platforms.
2407 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2408 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2409 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2411 struct used_address {
2412 struct sockaddr_storage name;
2413 unsigned int name_len;
2416 int __copy_msghdr(struct msghdr *kmsg,
2417 struct user_msghdr *msg,
2418 struct sockaddr __user **save_addr)
2422 kmsg->msg_control_is_user = true;
2423 kmsg->msg_get_inq = 0;
2424 kmsg->msg_control_user = msg->msg_control;
2425 kmsg->msg_controllen = msg->msg_controllen;
2426 kmsg->msg_flags = msg->msg_flags;
2428 kmsg->msg_namelen = msg->msg_namelen;
2430 kmsg->msg_namelen = 0;
2432 if (kmsg->msg_namelen < 0)
2435 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2436 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2439 *save_addr = msg->msg_name;
2441 if (msg->msg_name && kmsg->msg_namelen) {
2443 err = move_addr_to_kernel(msg->msg_name,
2450 kmsg->msg_name = NULL;
2451 kmsg->msg_namelen = 0;
2454 if (msg->msg_iovlen > UIO_MAXIOV)
2457 kmsg->msg_iocb = NULL;
2458 kmsg->msg_ubuf = NULL;
2462 static int copy_msghdr_from_user(struct msghdr *kmsg,
2463 struct user_msghdr __user *umsg,
2464 struct sockaddr __user **save_addr,
2467 struct user_msghdr msg;
2470 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2473 err = __copy_msghdr(kmsg, &msg, save_addr);
2477 err = import_iovec(save_addr ? ITER_DEST : ITER_SOURCE,
2478 msg.msg_iov, msg.msg_iovlen,
2479 UIO_FASTIOV, iov, &kmsg->msg_iter);
2480 return err < 0 ? err : 0;
2483 static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2484 unsigned int flags, struct used_address *used_address,
2485 unsigned int allowed_msghdr_flags)
2487 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2488 __aligned(sizeof(__kernel_size_t));
2489 /* 20 is size of ipv6_pktinfo */
2490 unsigned char *ctl_buf = ctl;
2496 if (msg_sys->msg_controllen > INT_MAX)
2498 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2499 ctl_len = msg_sys->msg_controllen;
2500 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2502 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2506 ctl_buf = msg_sys->msg_control;
2507 ctl_len = msg_sys->msg_controllen;
2508 } else if (ctl_len) {
2509 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2510 CMSG_ALIGN(sizeof(struct cmsghdr)));
2511 if (ctl_len > sizeof(ctl)) {
2512 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2513 if (ctl_buf == NULL)
2517 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2519 msg_sys->msg_control = ctl_buf;
2520 msg_sys->msg_control_is_user = false;
2522 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2523 msg_sys->msg_flags = flags;
2525 if (sock->file->f_flags & O_NONBLOCK)
2526 msg_sys->msg_flags |= MSG_DONTWAIT;
2528 * If this is sendmmsg() and current destination address is same as
2529 * previously succeeded address, omit asking LSM's decision.
2530 * used_address->name_len is initialized to UINT_MAX so that the first
2531 * destination address never matches.
2533 if (used_address && msg_sys->msg_name &&
2534 used_address->name_len == msg_sys->msg_namelen &&
2535 !memcmp(&used_address->name, msg_sys->msg_name,
2536 used_address->name_len)) {
2537 err = sock_sendmsg_nosec(sock, msg_sys);
2540 err = sock_sendmsg(sock, msg_sys);
2542 * If this is sendmmsg() and sending to current destination address was
2543 * successful, remember it.
2545 if (used_address && err >= 0) {
2546 used_address->name_len = msg_sys->msg_namelen;
2547 if (msg_sys->msg_name)
2548 memcpy(&used_address->name, msg_sys->msg_name,
2549 used_address->name_len);
2554 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2559 int sendmsg_copy_msghdr(struct msghdr *msg,
2560 struct user_msghdr __user *umsg, unsigned flags,
2565 if (flags & MSG_CMSG_COMPAT) {
2566 struct compat_msghdr __user *msg_compat;
2568 msg_compat = (struct compat_msghdr __user *) umsg;
2569 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2571 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2579 static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2580 struct msghdr *msg_sys, unsigned int flags,
2581 struct used_address *used_address,
2582 unsigned int allowed_msghdr_flags)
2584 struct sockaddr_storage address;
2585 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2588 msg_sys->msg_name = &address;
2590 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2594 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2595 allowed_msghdr_flags);
2601 * BSD sendmsg interface
2603 long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2606 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2609 long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2610 bool forbid_cmsg_compat)
2612 int fput_needed, err;
2613 struct msghdr msg_sys;
2614 struct socket *sock;
2616 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2619 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2623 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2625 fput_light(sock->file, fput_needed);
2630 SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2632 return __sys_sendmsg(fd, msg, flags, true);
2636 * Linux sendmmsg interface
2639 int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2640 unsigned int flags, bool forbid_cmsg_compat)
2642 int fput_needed, err, datagrams;
2643 struct socket *sock;
2644 struct mmsghdr __user *entry;
2645 struct compat_mmsghdr __user *compat_entry;
2646 struct msghdr msg_sys;
2647 struct used_address used_address;
2648 unsigned int oflags = flags;
2650 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2653 if (vlen > UIO_MAXIOV)
2658 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2662 used_address.name_len = UINT_MAX;
2664 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2668 while (datagrams < vlen) {
2669 if (datagrams == vlen - 1)
2672 if (MSG_CMSG_COMPAT & flags) {
2673 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2674 &msg_sys, flags, &used_address, MSG_EOR);
2677 err = __put_user(err, &compat_entry->msg_len);
2680 err = ___sys_sendmsg(sock,
2681 (struct user_msghdr __user *)entry,
2682 &msg_sys, flags, &used_address, MSG_EOR);
2685 err = put_user(err, &entry->msg_len);
2692 if (msg_data_left(&msg_sys))
2697 fput_light(sock->file, fput_needed);
2699 /* We only return an error if no datagrams were able to be sent */
2706 SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2707 unsigned int, vlen, unsigned int, flags)
2709 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2712 int recvmsg_copy_msghdr(struct msghdr *msg,
2713 struct user_msghdr __user *umsg, unsigned flags,
2714 struct sockaddr __user **uaddr,
2719 if (MSG_CMSG_COMPAT & flags) {
2720 struct compat_msghdr __user *msg_compat;
2722 msg_compat = (struct compat_msghdr __user *) umsg;
2723 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2725 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2733 static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2734 struct user_msghdr __user *msg,
2735 struct sockaddr __user *uaddr,
2736 unsigned int flags, int nosec)
2738 struct compat_msghdr __user *msg_compat =
2739 (struct compat_msghdr __user *) msg;
2740 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2741 struct sockaddr_storage addr;
2742 unsigned long cmsg_ptr;
2746 msg_sys->msg_name = &addr;
2747 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2748 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2750 /* We assume all kernel code knows the size of sockaddr_storage */
2751 msg_sys->msg_namelen = 0;
2753 if (sock->file->f_flags & O_NONBLOCK)
2754 flags |= MSG_DONTWAIT;
2756 if (unlikely(nosec))
2757 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2759 err = sock_recvmsg(sock, msg_sys, flags);
2765 if (uaddr != NULL) {
2766 err = move_addr_to_user(&addr,
2767 msg_sys->msg_namelen, uaddr,
2772 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2776 if (MSG_CMSG_COMPAT & flags)
2777 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2778 &msg_compat->msg_controllen);
2780 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2781 &msg->msg_controllen);
2789 static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2790 struct msghdr *msg_sys, unsigned int flags, int nosec)
2792 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2793 /* user mode address pointers */
2794 struct sockaddr __user *uaddr;
2797 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2801 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2807 * BSD recvmsg interface
2810 long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2811 struct user_msghdr __user *umsg,
2812 struct sockaddr __user *uaddr, unsigned int flags)
2814 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2817 long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2818 bool forbid_cmsg_compat)
2820 int fput_needed, err;
2821 struct msghdr msg_sys;
2822 struct socket *sock;
2824 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2827 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2831 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2833 fput_light(sock->file, fput_needed);
2838 SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2839 unsigned int, flags)
2841 return __sys_recvmsg(fd, msg, flags, true);
2845 * Linux recvmmsg interface
2848 static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2849 unsigned int vlen, unsigned int flags,
2850 struct timespec64 *timeout)
2852 int fput_needed, err, datagrams;
2853 struct socket *sock;
2854 struct mmsghdr __user *entry;
2855 struct compat_mmsghdr __user *compat_entry;
2856 struct msghdr msg_sys;
2857 struct timespec64 end_time;
2858 struct timespec64 timeout64;
2861 poll_select_set_timeout(&end_time, timeout->tv_sec,
2867 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2871 if (likely(!(flags & MSG_ERRQUEUE))) {
2872 err = sock_error(sock->sk);
2880 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2882 while (datagrams < vlen) {
2884 * No need to ask LSM for more than the first datagram.
2886 if (MSG_CMSG_COMPAT & flags) {
2887 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2888 &msg_sys, flags & ~MSG_WAITFORONE,
2892 err = __put_user(err, &compat_entry->msg_len);
2895 err = ___sys_recvmsg(sock,
2896 (struct user_msghdr __user *)entry,
2897 &msg_sys, flags & ~MSG_WAITFORONE,
2901 err = put_user(err, &entry->msg_len);
2909 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2910 if (flags & MSG_WAITFORONE)
2911 flags |= MSG_DONTWAIT;
2914 ktime_get_ts64(&timeout64);
2915 *timeout = timespec64_sub(end_time, timeout64);
2916 if (timeout->tv_sec < 0) {
2917 timeout->tv_sec = timeout->tv_nsec = 0;
2921 /* Timeout, return less than vlen datagrams */
2922 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2926 /* Out of band data, return right away */
2927 if (msg_sys.msg_flags & MSG_OOB)
2935 if (datagrams == 0) {
2941 * We may return less entries than requested (vlen) if the
2942 * sock is non block and there aren't enough datagrams...
2944 if (err != -EAGAIN) {
2946 * ... or if recvmsg returns an error after we
2947 * received some datagrams, where we record the
2948 * error to return on the next call or if the
2949 * app asks about it using getsockopt(SO_ERROR).
2951 WRITE_ONCE(sock->sk->sk_err, -err);
2954 fput_light(sock->file, fput_needed);
2959 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2960 unsigned int vlen, unsigned int flags,
2961 struct __kernel_timespec __user *timeout,
2962 struct old_timespec32 __user *timeout32)
2965 struct timespec64 timeout_sys;
2967 if (timeout && get_timespec64(&timeout_sys, timeout))
2970 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2973 if (!timeout && !timeout32)
2974 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2976 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2981 if (timeout && put_timespec64(&timeout_sys, timeout))
2982 datagrams = -EFAULT;
2984 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2985 datagrams = -EFAULT;
2990 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2991 unsigned int, vlen, unsigned int, flags,
2992 struct __kernel_timespec __user *, timeout)
2994 if (flags & MSG_CMSG_COMPAT)
2997 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
3000 #ifdef CONFIG_COMPAT_32BIT_TIME
3001 SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
3002 unsigned int, vlen, unsigned int, flags,
3003 struct old_timespec32 __user *, timeout)
3005 if (flags & MSG_CMSG_COMPAT)
3008 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
3012 #ifdef __ARCH_WANT_SYS_SOCKETCALL
3013 /* Argument list sizes for sys_socketcall */
3014 #define AL(x) ((x) * sizeof(unsigned long))
3015 static const unsigned char nargs[21] = {
3016 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
3017 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
3018 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
3025 * System call vectors.
3027 * Argument checking cleaned up. Saved 20% in size.
3028 * This function doesn't need to set the kernel lock because
3029 * it is set by the callees.
3032 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
3034 unsigned long a[AUDITSC_ARGS];
3035 unsigned long a0, a1;
3039 if (call < 1 || call > SYS_SENDMMSG)
3041 call = array_index_nospec(call, SYS_SENDMMSG + 1);
3044 if (len > sizeof(a))
3047 /* copy_from_user should be SMP safe. */
3048 if (copy_from_user(a, args, len))
3051 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
3060 err = __sys_socket(a0, a1, a[2]);
3063 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
3066 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
3069 err = __sys_listen(a0, a1);
3072 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3073 (int __user *)a[2], 0);
3075 case SYS_GETSOCKNAME:
3077 __sys_getsockname(a0, (struct sockaddr __user *)a1,
3078 (int __user *)a[2]);
3080 case SYS_GETPEERNAME:
3082 __sys_getpeername(a0, (struct sockaddr __user *)a1,
3083 (int __user *)a[2]);
3085 case SYS_SOCKETPAIR:
3086 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
3089 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3093 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3094 (struct sockaddr __user *)a[4], a[5]);
3097 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3101 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3102 (struct sockaddr __user *)a[4],
3103 (int __user *)a[5]);
3106 err = __sys_shutdown(a0, a1);
3108 case SYS_SETSOCKOPT:
3109 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
3112 case SYS_GETSOCKOPT:
3114 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
3115 (int __user *)a[4]);
3118 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
3122 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
3126 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
3130 if (IS_ENABLED(CONFIG_64BIT))
3131 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3133 (struct __kernel_timespec __user *)a[4],
3136 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3138 (struct old_timespec32 __user *)a[4]);
3141 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3142 (int __user *)a[2], a[3]);
3151 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
3154 * sock_register - add a socket protocol handler
3155 * @ops: description of protocol
3157 * This function is called by a protocol handler that wants to
3158 * advertise its address family, and have it linked into the
3159 * socket interface. The value ops->family corresponds to the
3160 * socket system call protocol family.
3162 int sock_register(const struct net_proto_family *ops)
3166 if (ops->family >= NPROTO) {
3167 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3171 spin_lock(&net_family_lock);
3172 if (rcu_dereference_protected(net_families[ops->family],
3173 lockdep_is_held(&net_family_lock)))
3176 rcu_assign_pointer(net_families[ops->family], ops);
3179 spin_unlock(&net_family_lock);
3181 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3184 EXPORT_SYMBOL(sock_register);
3187 * sock_unregister - remove a protocol handler
3188 * @family: protocol family to remove
3190 * This function is called by a protocol handler that wants to
3191 * remove its address family, and have it unlinked from the
3192 * new socket creation.
3194 * If protocol handler is a module, then it can use module reference
3195 * counts to protect against new references. If protocol handler is not
3196 * a module then it needs to provide its own protection in
3197 * the ops->create routine.
3199 void sock_unregister(int family)
3201 BUG_ON(family < 0 || family >= NPROTO);
3203 spin_lock(&net_family_lock);
3204 RCU_INIT_POINTER(net_families[family], NULL);
3205 spin_unlock(&net_family_lock);
3209 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3211 EXPORT_SYMBOL(sock_unregister);
3213 bool sock_is_registered(int family)
3215 return family < NPROTO && rcu_access_pointer(net_families[family]);
3218 static int __init sock_init(void)
3222 * Initialize the network sysctl infrastructure.
3224 err = net_sysctl_init();
3229 * Initialize skbuff SLAB cache
3234 * Initialize the protocols module.
3239 err = register_filesystem(&sock_fs_type);
3242 sock_mnt = kern_mount(&sock_fs_type);
3243 if (IS_ERR(sock_mnt)) {
3244 err = PTR_ERR(sock_mnt);
3248 /* The real protocol initialization is performed in later initcalls.
3251 #ifdef CONFIG_NETFILTER
3252 err = netfilter_init();
3257 ptp_classifier_init();
3263 unregister_filesystem(&sock_fs_type);
3267 core_initcall(sock_init); /* early initcall */
3269 #ifdef CONFIG_PROC_FS
3270 void socket_seq_show(struct seq_file *seq)
3272 seq_printf(seq, "sockets: used %d\n",
3273 sock_inuse_get(seq->private));
3275 #endif /* CONFIG_PROC_FS */
3277 /* Handle the fact that while struct ifreq has the same *layout* on
3278 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3279 * which are handled elsewhere, it still has different *size* due to
3280 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3281 * resulting in struct ifreq being 32 and 40 bytes respectively).
3282 * As a result, if the struct happens to be at the end of a page and
3283 * the next page isn't readable/writable, we get a fault. To prevent
3284 * that, copy back and forth to the full size.
3286 int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3288 if (in_compat_syscall()) {
3289 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3291 memset(ifr, 0, sizeof(*ifr));
3292 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3296 *ifrdata = compat_ptr(ifr32->ifr_data);
3301 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3305 *ifrdata = ifr->ifr_data;
3309 EXPORT_SYMBOL(get_user_ifreq);
3311 int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3313 size_t size = sizeof(*ifr);
3315 if (in_compat_syscall())
3316 size = sizeof(struct compat_ifreq);
3318 if (copy_to_user(arg, ifr, size))
3323 EXPORT_SYMBOL(put_user_ifreq);
3325 #ifdef CONFIG_COMPAT
3326 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3328 compat_uptr_t uptr32;
3333 if (get_user_ifreq(&ifr, NULL, uifr32))
3336 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3339 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3340 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3342 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3344 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3345 if (put_user_ifreq(&ifr, uifr32))
3351 /* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3352 static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3353 struct compat_ifreq __user *u_ifreq32)
3358 if (!is_socket_ioctl_cmd(cmd))
3360 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3362 ifreq.ifr_data = data;
3364 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3367 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3368 unsigned int cmd, unsigned long arg)
3370 void __user *argp = compat_ptr(arg);
3371 struct sock *sk = sock->sk;
3372 struct net *net = sock_net(sk);
3373 const struct proto_ops *ops;
3375 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3376 return sock_ioctl(file, cmd, (unsigned long)argp);
3380 return compat_siocwandev(net, argp);
3381 case SIOCGSTAMP_OLD:
3382 case SIOCGSTAMPNS_OLD:
3383 ops = READ_ONCE(sock->ops);
3384 if (!ops->gettstamp)
3385 return -ENOIOCTLCMD;
3386 return ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3387 !COMPAT_USE_64BIT_TIME);
3390 case SIOCBONDSLAVEINFOQUERY:
3391 case SIOCBONDINFOQUERY:
3394 return compat_ifr_data_ioctl(net, cmd, argp);
3405 case SIOCGSTAMP_NEW:
3406 case SIOCGSTAMPNS_NEW:
3410 return sock_ioctl(file, cmd, arg);
3429 case SIOCSIFHWBROADCAST:
3431 case SIOCGIFBRDADDR:
3432 case SIOCSIFBRDADDR:
3433 case SIOCGIFDSTADDR:
3434 case SIOCSIFDSTADDR:
3435 case SIOCGIFNETMASK:
3436 case SIOCSIFNETMASK:
3448 case SIOCBONDENSLAVE:
3449 case SIOCBONDRELEASE:
3450 case SIOCBONDSETHWADDR:
3451 case SIOCBONDCHANGEACTIVE:
3458 return sock_do_ioctl(net, sock, cmd, arg);
3461 return -ENOIOCTLCMD;
3464 static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3467 struct socket *sock = file->private_data;
3468 const struct proto_ops *ops = READ_ONCE(sock->ops);
3469 int ret = -ENOIOCTLCMD;
3476 if (ops->compat_ioctl)
3477 ret = ops->compat_ioctl(sock, cmd, arg);
3479 if (ret == -ENOIOCTLCMD &&
3480 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3481 ret = compat_wext_handle_ioctl(net, cmd, arg);
3483 if (ret == -ENOIOCTLCMD)
3484 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3491 * kernel_bind - bind an address to a socket (kernel space)
3494 * @addrlen: length of address
3496 * Returns 0 or an error.
3499 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3501 return READ_ONCE(sock->ops)->bind(sock, addr, addrlen);
3503 EXPORT_SYMBOL(kernel_bind);
3506 * kernel_listen - move socket to listening state (kernel space)
3508 * @backlog: pending connections queue size
3510 * Returns 0 or an error.
3513 int kernel_listen(struct socket *sock, int backlog)
3515 return READ_ONCE(sock->ops)->listen(sock, backlog);
3517 EXPORT_SYMBOL(kernel_listen);
3520 * kernel_accept - accept a connection (kernel space)
3521 * @sock: listening socket
3522 * @newsock: new connected socket
3525 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3526 * If it fails, @newsock is guaranteed to be %NULL.
3527 * Returns 0 or an error.
3530 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3532 struct sock *sk = sock->sk;
3533 const struct proto_ops *ops = READ_ONCE(sock->ops);
3536 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3541 err = ops->accept(sock, *newsock, flags, true);
3543 sock_release(*newsock);
3548 (*newsock)->ops = ops;
3549 __module_get(ops->owner);
3554 EXPORT_SYMBOL(kernel_accept);
3557 * kernel_connect - connect a socket (kernel space)
3560 * @addrlen: address length
3561 * @flags: flags (O_NONBLOCK, ...)
3563 * For datagram sockets, @addr is the address to which datagrams are sent
3564 * by default, and the only address from which datagrams are received.
3565 * For stream sockets, attempts to connect to @addr.
3566 * Returns 0 or an error code.
3569 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3572 struct sockaddr_storage address;
3574 memcpy(&address, addr, addrlen);
3576 return READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)&address,
3579 EXPORT_SYMBOL(kernel_connect);
3582 * kernel_getsockname - get the address which the socket is bound (kernel space)
3584 * @addr: address holder
3586 * Fills the @addr pointer with the address which the socket is bound.
3587 * Returns the length of the address in bytes or an error code.
3590 int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3592 return READ_ONCE(sock->ops)->getname(sock, addr, 0);
3594 EXPORT_SYMBOL(kernel_getsockname);
3597 * kernel_getpeername - get the address which the socket is connected (kernel space)
3599 * @addr: address holder
3601 * Fills the @addr pointer with the address which the socket is connected.
3602 * Returns the length of the address in bytes or an error code.
3605 int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3607 return READ_ONCE(sock->ops)->getname(sock, addr, 1);
3609 EXPORT_SYMBOL(kernel_getpeername);
3612 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3614 * @how: connection part
3616 * Returns 0 or an error.
3619 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3621 return READ_ONCE(sock->ops)->shutdown(sock, how);
3623 EXPORT_SYMBOL(kernel_sock_shutdown);
3626 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3629 * This routine returns the IP overhead imposed by a socket i.e.
3630 * the length of the underlying IP header, depending on whether
3631 * this is an IPv4 or IPv6 socket and the length from IP options turned
3632 * on at the socket. Assumes that the caller has a lock on the socket.
3635 u32 kernel_sock_ip_overhead(struct sock *sk)
3637 struct inet_sock *inet;
3638 struct ip_options_rcu *opt;
3640 #if IS_ENABLED(CONFIG_IPV6)
3641 struct ipv6_pinfo *np;
3642 struct ipv6_txoptions *optv6 = NULL;
3643 #endif /* IS_ENABLED(CONFIG_IPV6) */
3648 switch (sk->sk_family) {
3651 overhead += sizeof(struct iphdr);
3652 opt = rcu_dereference_protected(inet->inet_opt,
3653 sock_owned_by_user(sk));
3655 overhead += opt->opt.optlen;
3657 #if IS_ENABLED(CONFIG_IPV6)
3660 overhead += sizeof(struct ipv6hdr);
3662 optv6 = rcu_dereference_protected(np->opt,
3663 sock_owned_by_user(sk));
3665 overhead += (optv6->opt_flen + optv6->opt_nflen);
3667 #endif /* IS_ENABLED(CONFIG_IPV6) */
3668 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3672 EXPORT_SYMBOL(kernel_sock_ip_overhead);