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/ethtool.h>
57 #include <linux/socket.h>
58 #include <linux/file.h>
59 #include <linux/net.h>
60 #include <linux/interrupt.h>
61 #include <linux/thread_info.h>
62 #include <linux/rcupdate.h>
63 #include <linux/netdevice.h>
64 #include <linux/proc_fs.h>
65 #include <linux/seq_file.h>
66 #include <linux/mutex.h>
67 #include <linux/if_bridge.h>
68 #include <linux/if_vlan.h>
69 #include <linux/ptp_classify.h>
70 #include <linux/init.h>
71 #include <linux/poll.h>
72 #include <linux/cache.h>
73 #include <linux/module.h>
74 #include <linux/highmem.h>
75 #include <linux/mount.h>
76 #include <linux/pseudo_fs.h>
77 #include <linux/security.h>
78 #include <linux/syscalls.h>
79 #include <linux/compat.h>
80 #include <linux/kmod.h>
81 #include <linux/audit.h>
82 #include <linux/wireless.h>
83 #include <linux/nsproxy.h>
84 #include <linux/magic.h>
85 #include <linux/slab.h>
86 #include <linux/xattr.h>
87 #include <linux/nospec.h>
88 #include <linux/indirect_call_wrapper.h>
90 #include <linux/uaccess.h>
91 #include <asm/unistd.h>
93 #include <net/compat.h>
95 #include <net/cls_cgroup.h>
98 #include <linux/netfilter.h>
100 #include <linux/if_tun.h>
101 #include <linux/ipv6_route.h>
102 #include <linux/route.h>
103 #include <linux/termios.h>
104 #include <linux/sockios.h>
105 #include <net/busy_poll.h>
106 #include <linux/errqueue.h>
107 #include <linux/ptp_clock_kernel.h>
109 #ifdef CONFIG_NET_RX_BUSY_POLL
110 unsigned int sysctl_net_busy_read __read_mostly;
111 unsigned int sysctl_net_busy_poll __read_mostly;
114 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
115 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
116 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
118 static int sock_close(struct inode *inode, struct file *file);
119 static __poll_t sock_poll(struct file *file,
120 struct poll_table_struct *wait);
121 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
123 static long compat_sock_ioctl(struct file *file,
124 unsigned int cmd, unsigned long arg);
126 static int sock_fasync(int fd, struct file *filp, int on);
127 static ssize_t sock_sendpage(struct file *file, struct page *page,
128 int offset, size_t size, loff_t *ppos, int more);
129 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
130 struct pipe_inode_info *pipe, size_t len,
133 #ifdef CONFIG_PROC_FS
134 static void sock_show_fdinfo(struct seq_file *m, struct file *f)
136 struct socket *sock = f->private_data;
138 if (sock->ops->show_fdinfo)
139 sock->ops->show_fdinfo(m, sock);
142 #define sock_show_fdinfo NULL
146 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
147 * in the operation structures but are done directly via the socketcall() multiplexor.
150 static const struct file_operations socket_file_ops = {
151 .owner = THIS_MODULE,
153 .read_iter = sock_read_iter,
154 .write_iter = sock_write_iter,
156 .unlocked_ioctl = sock_ioctl,
158 .compat_ioctl = compat_sock_ioctl,
161 .release = sock_close,
162 .fasync = sock_fasync,
163 .sendpage = sock_sendpage,
164 .splice_write = generic_splice_sendpage,
165 .splice_read = sock_splice_read,
166 .show_fdinfo = sock_show_fdinfo,
169 static const char * const pf_family_names[] = {
170 [PF_UNSPEC] = "PF_UNSPEC",
171 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
172 [PF_INET] = "PF_INET",
173 [PF_AX25] = "PF_AX25",
175 [PF_APPLETALK] = "PF_APPLETALK",
176 [PF_NETROM] = "PF_NETROM",
177 [PF_BRIDGE] = "PF_BRIDGE",
178 [PF_ATMPVC] = "PF_ATMPVC",
180 [PF_INET6] = "PF_INET6",
181 [PF_ROSE] = "PF_ROSE",
182 [PF_DECnet] = "PF_DECnet",
183 [PF_NETBEUI] = "PF_NETBEUI",
184 [PF_SECURITY] = "PF_SECURITY",
186 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
187 [PF_PACKET] = "PF_PACKET",
189 [PF_ECONET] = "PF_ECONET",
190 [PF_ATMSVC] = "PF_ATMSVC",
193 [PF_IRDA] = "PF_IRDA",
194 [PF_PPPOX] = "PF_PPPOX",
195 [PF_WANPIPE] = "PF_WANPIPE",
198 [PF_MPLS] = "PF_MPLS",
200 [PF_TIPC] = "PF_TIPC",
201 [PF_BLUETOOTH] = "PF_BLUETOOTH",
202 [PF_IUCV] = "PF_IUCV",
203 [PF_RXRPC] = "PF_RXRPC",
204 [PF_ISDN] = "PF_ISDN",
205 [PF_PHONET] = "PF_PHONET",
206 [PF_IEEE802154] = "PF_IEEE802154",
207 [PF_CAIF] = "PF_CAIF",
210 [PF_VSOCK] = "PF_VSOCK",
212 [PF_QIPCRTR] = "PF_QIPCRTR",
218 * The protocol list. Each protocol is registered in here.
221 static DEFINE_SPINLOCK(net_family_lock);
222 static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
226 * Move socket addresses back and forth across the kernel/user
227 * divide and look after the messy bits.
231 * move_addr_to_kernel - copy a socket address into kernel space
232 * @uaddr: Address in user space
233 * @kaddr: Address in kernel space
234 * @ulen: Length in user space
236 * The address is copied into kernel space. If the provided address is
237 * too long an error code of -EINVAL is returned. If the copy gives
238 * invalid addresses -EFAULT is returned. On a success 0 is returned.
241 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
243 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
247 if (copy_from_user(kaddr, uaddr, ulen))
249 return audit_sockaddr(ulen, kaddr);
253 * move_addr_to_user - copy an address to user space
254 * @kaddr: kernel space address
255 * @klen: length of address in kernel
256 * @uaddr: user space address
257 * @ulen: pointer to user length field
259 * The value pointed to by ulen on entry is the buffer length available.
260 * This is overwritten with the buffer space used. -EINVAL is returned
261 * if an overlong buffer is specified or a negative buffer size. -EFAULT
262 * is returned if either the buffer or the length field are not
264 * After copying the data up to the limit the user specifies, the true
265 * length of the data is written over the length limit the user
266 * specified. Zero is returned for a success.
269 static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
270 void __user *uaddr, int __user *ulen)
275 BUG_ON(klen > sizeof(struct sockaddr_storage));
276 err = get_user(len, ulen);
284 if (audit_sockaddr(klen, kaddr))
286 if (copy_to_user(uaddr, kaddr, len))
290 * "fromlen shall refer to the value before truncation.."
293 return __put_user(klen, ulen);
296 static struct kmem_cache *sock_inode_cachep __ro_after_init;
298 static struct inode *sock_alloc_inode(struct super_block *sb)
300 struct socket_alloc *ei;
302 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
305 init_waitqueue_head(&ei->socket.wq.wait);
306 ei->socket.wq.fasync_list = NULL;
307 ei->socket.wq.flags = 0;
309 ei->socket.state = SS_UNCONNECTED;
310 ei->socket.flags = 0;
311 ei->socket.ops = NULL;
312 ei->socket.sk = NULL;
313 ei->socket.file = NULL;
315 return &ei->vfs_inode;
318 static void sock_free_inode(struct inode *inode)
320 struct socket_alloc *ei;
322 ei = container_of(inode, struct socket_alloc, vfs_inode);
323 kmem_cache_free(sock_inode_cachep, ei);
326 static void init_once(void *foo)
328 struct socket_alloc *ei = (struct socket_alloc *)foo;
330 inode_init_once(&ei->vfs_inode);
333 static void init_inodecache(void)
335 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
336 sizeof(struct socket_alloc),
338 (SLAB_HWCACHE_ALIGN |
339 SLAB_RECLAIM_ACCOUNT |
340 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
342 BUG_ON(sock_inode_cachep == NULL);
345 static const struct super_operations sockfs_ops = {
346 .alloc_inode = sock_alloc_inode,
347 .free_inode = sock_free_inode,
348 .statfs = simple_statfs,
352 * sockfs_dname() is called from d_path().
354 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
356 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
357 d_inode(dentry)->i_ino);
360 static const struct dentry_operations sockfs_dentry_operations = {
361 .d_dname = sockfs_dname,
364 static int sockfs_xattr_get(const struct xattr_handler *handler,
365 struct dentry *dentry, struct inode *inode,
366 const char *suffix, void *value, size_t size)
369 if (dentry->d_name.len + 1 > size)
371 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
373 return dentry->d_name.len + 1;
376 #define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
377 #define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
378 #define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
380 static const struct xattr_handler sockfs_xattr_handler = {
381 .name = XATTR_NAME_SOCKPROTONAME,
382 .get = sockfs_xattr_get,
385 static int sockfs_security_xattr_set(const struct xattr_handler *handler,
386 struct user_namespace *mnt_userns,
387 struct dentry *dentry, struct inode *inode,
388 const char *suffix, const void *value,
389 size_t size, int flags)
391 /* Handled by LSM. */
395 static const struct xattr_handler sockfs_security_xattr_handler = {
396 .prefix = XATTR_SECURITY_PREFIX,
397 .set = sockfs_security_xattr_set,
400 static const struct xattr_handler *sockfs_xattr_handlers[] = {
401 &sockfs_xattr_handler,
402 &sockfs_security_xattr_handler,
406 static int sockfs_init_fs_context(struct fs_context *fc)
408 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
411 ctx->ops = &sockfs_ops;
412 ctx->dops = &sockfs_dentry_operations;
413 ctx->xattr = sockfs_xattr_handlers;
417 static struct vfsmount *sock_mnt __read_mostly;
419 static struct file_system_type sock_fs_type = {
421 .init_fs_context = sockfs_init_fs_context,
422 .kill_sb = kill_anon_super,
426 * Obtains the first available file descriptor and sets it up for use.
428 * These functions create file structures and maps them to fd space
429 * of the current process. On success it returns file descriptor
430 * and file struct implicitly stored in sock->file.
431 * Note that another thread may close file descriptor before we return
432 * from this function. We use the fact that now we do not refer
433 * to socket after mapping. If one day we will need it, this
434 * function will increment ref. count on file by 1.
436 * In any case returned fd MAY BE not valid!
437 * This race condition is unavoidable
438 * with shared fd spaces, we cannot solve it inside kernel,
439 * but we take care of internal coherence yet.
443 * sock_alloc_file - Bind a &socket to a &file
445 * @flags: file status flags
446 * @dname: protocol name
448 * Returns the &file bound with @sock, implicitly storing it
449 * in sock->file. If dname is %NULL, sets to "".
450 * On failure the return is a ERR pointer (see linux/err.h).
451 * This function uses GFP_KERNEL internally.
454 struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
459 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
461 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
462 O_RDWR | (flags & O_NONBLOCK),
470 file->private_data = sock;
471 stream_open(SOCK_INODE(sock), file);
474 EXPORT_SYMBOL(sock_alloc_file);
476 static int sock_map_fd(struct socket *sock, int flags)
478 struct file *newfile;
479 int fd = get_unused_fd_flags(flags);
480 if (unlikely(fd < 0)) {
485 newfile = sock_alloc_file(sock, flags, NULL);
486 if (!IS_ERR(newfile)) {
487 fd_install(fd, newfile);
492 return PTR_ERR(newfile);
496 * sock_from_file - Return the &socket bounded to @file.
499 * On failure returns %NULL.
502 struct socket *sock_from_file(struct file *file)
504 if (file->f_op == &socket_file_ops)
505 return file->private_data; /* set in sock_map_fd */
509 EXPORT_SYMBOL(sock_from_file);
512 * sockfd_lookup - Go from a file number to its socket slot
514 * @err: pointer to an error code return
516 * The file handle passed in is locked and the socket it is bound
517 * to is returned. If an error occurs the err pointer is overwritten
518 * with a negative errno code and NULL is returned. The function checks
519 * for both invalid handles and passing a handle which is not a socket.
521 * On a success the socket object pointer is returned.
524 struct socket *sockfd_lookup(int fd, int *err)
535 sock = sock_from_file(file);
542 EXPORT_SYMBOL(sockfd_lookup);
544 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
546 struct fd f = fdget(fd);
551 sock = sock_from_file(f.file);
553 *fput_needed = f.flags & FDPUT_FPUT;
562 static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
568 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
578 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
583 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
590 static int sockfs_setattr(struct user_namespace *mnt_userns,
591 struct dentry *dentry, struct iattr *iattr)
593 int err = simple_setattr(&init_user_ns, dentry, iattr);
595 if (!err && (iattr->ia_valid & ATTR_UID)) {
596 struct socket *sock = SOCKET_I(d_inode(dentry));
599 sock->sk->sk_uid = iattr->ia_uid;
607 static const struct inode_operations sockfs_inode_ops = {
608 .listxattr = sockfs_listxattr,
609 .setattr = sockfs_setattr,
613 * sock_alloc - allocate a socket
615 * Allocate a new inode and socket object. The two are bound together
616 * and initialised. The socket is then returned. If we are out of inodes
617 * NULL is returned. This functions uses GFP_KERNEL internally.
620 struct socket *sock_alloc(void)
625 inode = new_inode_pseudo(sock_mnt->mnt_sb);
629 sock = SOCKET_I(inode);
631 inode->i_ino = get_next_ino();
632 inode->i_mode = S_IFSOCK | S_IRWXUGO;
633 inode->i_uid = current_fsuid();
634 inode->i_gid = current_fsgid();
635 inode->i_op = &sockfs_inode_ops;
639 EXPORT_SYMBOL(sock_alloc);
641 static void __sock_release(struct socket *sock, struct inode *inode)
644 struct module *owner = sock->ops->owner;
648 sock->ops->release(sock);
656 if (sock->wq.fasync_list)
657 pr_err("%s: fasync list not empty!\n", __func__);
660 iput(SOCK_INODE(sock));
667 * sock_release - close a socket
668 * @sock: socket to close
670 * The socket is released from the protocol stack if it has a release
671 * callback, and the inode is then released if the socket is bound to
672 * an inode not a file.
674 void sock_release(struct socket *sock)
676 __sock_release(sock, NULL);
678 EXPORT_SYMBOL(sock_release);
680 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
682 u8 flags = *tx_flags;
684 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
685 flags |= SKBTX_HW_TSTAMP;
687 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
688 flags |= SKBTX_SW_TSTAMP;
690 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
691 flags |= SKBTX_SCHED_TSTAMP;
695 EXPORT_SYMBOL(__sock_tx_timestamp);
697 INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
699 INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
701 static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
703 int ret = INDIRECT_CALL_INET(sock->ops->sendmsg, inet6_sendmsg,
704 inet_sendmsg, sock, msg,
706 BUG_ON(ret == -EIOCBQUEUED);
711 * sock_sendmsg - send a message through @sock
713 * @msg: message to send
715 * Sends @msg through @sock, passing through LSM.
716 * Returns the number of bytes sent, or an error code.
718 int sock_sendmsg(struct socket *sock, struct msghdr *msg)
720 int err = security_socket_sendmsg(sock, msg,
723 return err ?: sock_sendmsg_nosec(sock, msg);
725 EXPORT_SYMBOL(sock_sendmsg);
728 * kernel_sendmsg - send a message through @sock (kernel-space)
730 * @msg: message header
732 * @num: vec array length
733 * @size: total message data size
735 * Builds the message data with @vec and sends it through @sock.
736 * Returns the number of bytes sent, or an error code.
739 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
740 struct kvec *vec, size_t num, size_t size)
742 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
743 return sock_sendmsg(sock, msg);
745 EXPORT_SYMBOL(kernel_sendmsg);
748 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
750 * @msg: message header
751 * @vec: output s/g array
752 * @num: output s/g array length
753 * @size: total message data size
755 * Builds the message data with @vec and sends it through @sock.
756 * Returns the number of bytes sent, or an error code.
757 * Caller must hold @sk.
760 int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
761 struct kvec *vec, size_t num, size_t size)
763 struct socket *sock = sk->sk_socket;
765 if (!sock->ops->sendmsg_locked)
766 return sock_no_sendmsg_locked(sk, msg, size);
768 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
770 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
772 EXPORT_SYMBOL(kernel_sendmsg_locked);
774 static bool skb_is_err_queue(const struct sk_buff *skb)
776 /* pkt_type of skbs enqueued on the error queue are set to
777 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
778 * in recvmsg, since skbs received on a local socket will never
779 * have a pkt_type of PACKET_OUTGOING.
781 return skb->pkt_type == PACKET_OUTGOING;
784 /* On transmit, software and hardware timestamps are returned independently.
785 * As the two skb clones share the hardware timestamp, which may be updated
786 * before the software timestamp is received, a hardware TX timestamp may be
787 * returned only if there is no software TX timestamp. Ignore false software
788 * timestamps, which may be made in the __sock_recv_timestamp() call when the
789 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
790 * hardware timestamp.
792 static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
794 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
797 static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
799 struct scm_ts_pktinfo ts_pktinfo;
800 struct net_device *orig_dev;
802 if (!skb_mac_header_was_set(skb))
805 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
808 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
810 ts_pktinfo.if_index = orig_dev->ifindex;
813 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
814 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
815 sizeof(ts_pktinfo), &ts_pktinfo);
819 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
821 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
824 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
825 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
826 struct scm_timestamping_internal tss;
828 int empty = 1, false_tstamp = 0;
829 struct skb_shared_hwtstamps *shhwtstamps =
832 /* Race occurred between timestamp enabling and packet
833 receiving. Fill in the current time for now. */
834 if (need_software_tstamp && skb->tstamp == 0) {
835 __net_timestamp(skb);
839 if (need_software_tstamp) {
840 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
842 struct __kernel_sock_timeval tv;
844 skb_get_new_timestamp(skb, &tv);
845 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
848 struct __kernel_old_timeval tv;
850 skb_get_timestamp(skb, &tv);
851 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
856 struct __kernel_timespec ts;
858 skb_get_new_timestampns(skb, &ts);
859 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
862 struct __kernel_old_timespec ts;
864 skb_get_timestampns(skb, &ts);
865 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
871 memset(&tss, 0, sizeof(tss));
872 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
873 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
876 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
877 !skb_is_swtx_tstamp(skb, false_tstamp)) {
878 if (sk->sk_tsflags & SOF_TIMESTAMPING_BIND_PHC)
879 ptp_convert_timestamp(shhwtstamps, sk->sk_bind_phc);
881 if (ktime_to_timespec64_cond(shhwtstamps->hwtstamp,
885 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
886 !skb_is_err_queue(skb))
887 put_ts_pktinfo(msg, skb);
891 if (sock_flag(sk, SOCK_TSTAMP_NEW))
892 put_cmsg_scm_timestamping64(msg, &tss);
894 put_cmsg_scm_timestamping(msg, &tss);
896 if (skb_is_err_queue(skb) && skb->len &&
897 SKB_EXT_ERR(skb)->opt_stats)
898 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
899 skb->len, skb->data);
902 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
904 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
909 if (!sock_flag(sk, SOCK_WIFI_STATUS))
911 if (!skb->wifi_acked_valid)
914 ack = skb->wifi_acked;
916 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
918 EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
920 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
923 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
924 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
925 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
928 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
931 sock_recv_timestamp(msg, sk, skb);
932 sock_recv_drops(msg, sk, skb);
934 EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
936 INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
938 INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
940 static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
943 return INDIRECT_CALL_INET(sock->ops->recvmsg, inet6_recvmsg,
944 inet_recvmsg, sock, msg, msg_data_left(msg),
949 * sock_recvmsg - receive a message from @sock
951 * @msg: message to receive
952 * @flags: message flags
954 * Receives @msg from @sock, passing through LSM. Returns the total number
955 * of bytes received, or an error.
957 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
959 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
961 return err ?: sock_recvmsg_nosec(sock, msg, flags);
963 EXPORT_SYMBOL(sock_recvmsg);
966 * kernel_recvmsg - Receive a message from a socket (kernel space)
967 * @sock: The socket to receive the message from
968 * @msg: Received message
969 * @vec: Input s/g array for message data
970 * @num: Size of input s/g array
971 * @size: Number of bytes to read
972 * @flags: Message flags (MSG_DONTWAIT, etc...)
974 * On return the msg structure contains the scatter/gather array passed in the
975 * vec argument. The array is modified so that it consists of the unfilled
976 * portion of the original array.
978 * The returned value is the total number of bytes received, or an error.
981 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
982 struct kvec *vec, size_t num, size_t size, int flags)
984 msg->msg_control_is_user = false;
985 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
986 return sock_recvmsg(sock, msg, flags);
988 EXPORT_SYMBOL(kernel_recvmsg);
990 static ssize_t sock_sendpage(struct file *file, struct page *page,
991 int offset, size_t size, loff_t *ppos, int more)
996 sock = file->private_data;
998 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
999 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
1002 return kernel_sendpage(sock, page, offset, size, flags);
1005 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1006 struct pipe_inode_info *pipe, size_t len,
1009 struct socket *sock = file->private_data;
1011 if (unlikely(!sock->ops->splice_read))
1012 return generic_file_splice_read(file, ppos, pipe, len, flags);
1014 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
1017 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1019 struct file *file = iocb->ki_filp;
1020 struct socket *sock = file->private_data;
1021 struct msghdr msg = {.msg_iter = *to,
1025 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1026 msg.msg_flags = MSG_DONTWAIT;
1028 if (iocb->ki_pos != 0)
1031 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1034 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1039 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1041 struct file *file = iocb->ki_filp;
1042 struct socket *sock = file->private_data;
1043 struct msghdr msg = {.msg_iter = *from,
1047 if (iocb->ki_pos != 0)
1050 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1051 msg.msg_flags = MSG_DONTWAIT;
1053 if (sock->type == SOCK_SEQPACKET)
1054 msg.msg_flags |= MSG_EOR;
1056 res = sock_sendmsg(sock, &msg);
1057 *from = msg.msg_iter;
1062 * Atomic setting of ioctl hooks to avoid race
1063 * with module unload.
1066 static DEFINE_MUTEX(br_ioctl_mutex);
1067 static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
1069 void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
1071 mutex_lock(&br_ioctl_mutex);
1072 br_ioctl_hook = hook;
1073 mutex_unlock(&br_ioctl_mutex);
1075 EXPORT_SYMBOL(brioctl_set);
1077 static DEFINE_MUTEX(vlan_ioctl_mutex);
1078 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1080 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1082 mutex_lock(&vlan_ioctl_mutex);
1083 vlan_ioctl_hook = hook;
1084 mutex_unlock(&vlan_ioctl_mutex);
1086 EXPORT_SYMBOL(vlan_ioctl_set);
1088 static long sock_do_ioctl(struct net *net, struct socket *sock,
1089 unsigned int cmd, unsigned long arg)
1094 void __user *argp = (void __user *)arg;
1097 err = sock->ops->ioctl(sock, cmd, arg);
1100 * If this ioctl is unknown try to hand it down
1101 * to the NIC driver.
1103 if (err != -ENOIOCTLCMD)
1106 if (get_user_ifreq(&ifr, &data, argp))
1108 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1109 if (!err && need_copyout)
1110 if (put_user_ifreq(&ifr, argp))
1117 * With an ioctl, arg may well be a user mode pointer, but we don't know
1118 * what to do with it - that's up to the protocol still.
1121 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1123 struct socket *sock;
1125 void __user *argp = (void __user *)arg;
1129 sock = file->private_data;
1132 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1136 if (get_user_ifreq(&ifr, &data, argp))
1138 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1139 if (!err && need_copyout)
1140 if (put_user_ifreq(&ifr, argp))
1143 #ifdef CONFIG_WEXT_CORE
1144 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1145 err = wext_handle_ioctl(net, cmd, argp);
1152 if (get_user(pid, (int __user *)argp))
1154 err = f_setown(sock->file, pid, 1);
1158 err = put_user(f_getown(sock->file),
1159 (int __user *)argp);
1167 request_module("bridge");
1169 mutex_lock(&br_ioctl_mutex);
1171 err = br_ioctl_hook(net, cmd, argp);
1172 mutex_unlock(&br_ioctl_mutex);
1177 if (!vlan_ioctl_hook)
1178 request_module("8021q");
1180 mutex_lock(&vlan_ioctl_mutex);
1181 if (vlan_ioctl_hook)
1182 err = vlan_ioctl_hook(net, argp);
1183 mutex_unlock(&vlan_ioctl_mutex);
1187 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1190 err = open_related_ns(&net->ns, get_net_ns);
1192 case SIOCGSTAMP_OLD:
1193 case SIOCGSTAMPNS_OLD:
1194 if (!sock->ops->gettstamp) {
1198 err = sock->ops->gettstamp(sock, argp,
1199 cmd == SIOCGSTAMP_OLD,
1200 !IS_ENABLED(CONFIG_64BIT));
1202 case SIOCGSTAMP_NEW:
1203 case SIOCGSTAMPNS_NEW:
1204 if (!sock->ops->gettstamp) {
1208 err = sock->ops->gettstamp(sock, argp,
1209 cmd == SIOCGSTAMP_NEW,
1214 err = dev_ifconf(net, argp);
1218 err = sock_do_ioctl(net, sock, cmd, arg);
1225 * sock_create_lite - creates a socket
1226 * @family: protocol family (AF_INET, ...)
1227 * @type: communication type (SOCK_STREAM, ...)
1228 * @protocol: protocol (0, ...)
1231 * Creates a new socket and assigns it to @res, passing through LSM.
1232 * The new socket initialization is not complete, see kernel_accept().
1233 * Returns 0 or an error. On failure @res is set to %NULL.
1234 * This function internally uses GFP_KERNEL.
1237 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1240 struct socket *sock = NULL;
1242 err = security_socket_create(family, type, protocol, 1);
1246 sock = sock_alloc();
1253 err = security_socket_post_create(sock, family, type, protocol, 1);
1265 EXPORT_SYMBOL(sock_create_lite);
1267 /* No kernel lock held - perfect */
1268 static __poll_t sock_poll(struct file *file, poll_table *wait)
1270 struct socket *sock = file->private_data;
1271 __poll_t events = poll_requested_events(wait), flag = 0;
1273 if (!sock->ops->poll)
1276 if (sk_can_busy_loop(sock->sk)) {
1277 /* poll once if requested by the syscall */
1278 if (events & POLL_BUSY_LOOP)
1279 sk_busy_loop(sock->sk, 1);
1281 /* if this socket can poll_ll, tell the system call */
1282 flag = POLL_BUSY_LOOP;
1285 return sock->ops->poll(file, sock, wait) | flag;
1288 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1290 struct socket *sock = file->private_data;
1292 return sock->ops->mmap(file, sock, vma);
1295 static int sock_close(struct inode *inode, struct file *filp)
1297 __sock_release(SOCKET_I(inode), inode);
1302 * Update the socket async list
1304 * Fasync_list locking strategy.
1306 * 1. fasync_list is modified only under process context socket lock
1307 * i.e. under semaphore.
1308 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1309 * or under socket lock
1312 static int sock_fasync(int fd, struct file *filp, int on)
1314 struct socket *sock = filp->private_data;
1315 struct sock *sk = sock->sk;
1316 struct socket_wq *wq = &sock->wq;
1322 fasync_helper(fd, filp, on, &wq->fasync_list);
1324 if (!wq->fasync_list)
1325 sock_reset_flag(sk, SOCK_FASYNC);
1327 sock_set_flag(sk, SOCK_FASYNC);
1333 /* This function may be called only under rcu_lock */
1335 int sock_wake_async(struct socket_wq *wq, int how, int band)
1337 if (!wq || !wq->fasync_list)
1341 case SOCK_WAKE_WAITD:
1342 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1345 case SOCK_WAKE_SPACE:
1346 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1351 kill_fasync(&wq->fasync_list, SIGIO, band);
1354 kill_fasync(&wq->fasync_list, SIGURG, band);
1359 EXPORT_SYMBOL(sock_wake_async);
1362 * __sock_create - creates a socket
1363 * @net: net namespace
1364 * @family: protocol family (AF_INET, ...)
1365 * @type: communication type (SOCK_STREAM, ...)
1366 * @protocol: protocol (0, ...)
1368 * @kern: boolean for kernel space sockets
1370 * Creates a new socket and assigns it to @res, passing through LSM.
1371 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1372 * be set to true if the socket resides in kernel space.
1373 * This function internally uses GFP_KERNEL.
1376 int __sock_create(struct net *net, int family, int type, int protocol,
1377 struct socket **res, int kern)
1380 struct socket *sock;
1381 const struct net_proto_family *pf;
1384 * Check protocol is in range
1386 if (family < 0 || family >= NPROTO)
1387 return -EAFNOSUPPORT;
1388 if (type < 0 || type >= SOCK_MAX)
1393 This uglymoron is moved from INET layer to here to avoid
1394 deadlock in module load.
1396 if (family == PF_INET && type == SOCK_PACKET) {
1397 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1402 err = security_socket_create(family, type, protocol, kern);
1407 * Allocate the socket and allow the family to set things up. if
1408 * the protocol is 0, the family is instructed to select an appropriate
1411 sock = sock_alloc();
1413 net_warn_ratelimited("socket: no more sockets\n");
1414 return -ENFILE; /* Not exactly a match, but its the
1415 closest posix thing */
1420 #ifdef CONFIG_MODULES
1421 /* Attempt to load a protocol module if the find failed.
1423 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1424 * requested real, full-featured networking support upon configuration.
1425 * Otherwise module support will break!
1427 if (rcu_access_pointer(net_families[family]) == NULL)
1428 request_module("net-pf-%d", family);
1432 pf = rcu_dereference(net_families[family]);
1433 err = -EAFNOSUPPORT;
1438 * We will call the ->create function, that possibly is in a loadable
1439 * module, so we have to bump that loadable module refcnt first.
1441 if (!try_module_get(pf->owner))
1444 /* Now protected by module ref count */
1447 err = pf->create(net, sock, protocol, kern);
1449 goto out_module_put;
1452 * Now to bump the refcnt of the [loadable] module that owns this
1453 * socket at sock_release time we decrement its refcnt.
1455 if (!try_module_get(sock->ops->owner))
1456 goto out_module_busy;
1459 * Now that we're done with the ->create function, the [loadable]
1460 * module can have its refcnt decremented
1462 module_put(pf->owner);
1463 err = security_socket_post_create(sock, family, type, protocol, kern);
1465 goto out_sock_release;
1471 err = -EAFNOSUPPORT;
1474 module_put(pf->owner);
1481 goto out_sock_release;
1483 EXPORT_SYMBOL(__sock_create);
1486 * sock_create - creates a socket
1487 * @family: protocol family (AF_INET, ...)
1488 * @type: communication type (SOCK_STREAM, ...)
1489 * @protocol: protocol (0, ...)
1492 * A wrapper around __sock_create().
1493 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1496 int sock_create(int family, int type, int protocol, struct socket **res)
1498 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1500 EXPORT_SYMBOL(sock_create);
1503 * sock_create_kern - creates a socket (kernel space)
1504 * @net: net namespace
1505 * @family: protocol family (AF_INET, ...)
1506 * @type: communication type (SOCK_STREAM, ...)
1507 * @protocol: protocol (0, ...)
1510 * A wrapper around __sock_create().
1511 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1514 int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1516 return __sock_create(net, family, type, protocol, res, 1);
1518 EXPORT_SYMBOL(sock_create_kern);
1520 int __sys_socket(int family, int type, int protocol)
1523 struct socket *sock;
1526 /* Check the SOCK_* constants for consistency. */
1527 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1528 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1529 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1530 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1532 flags = type & ~SOCK_TYPE_MASK;
1533 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1535 type &= SOCK_TYPE_MASK;
1537 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1538 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1540 retval = sock_create(family, type, protocol, &sock);
1544 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1547 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1549 return __sys_socket(family, type, protocol);
1553 * Create a pair of connected sockets.
1556 int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1558 struct socket *sock1, *sock2;
1560 struct file *newfile1, *newfile2;
1563 flags = type & ~SOCK_TYPE_MASK;
1564 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1566 type &= SOCK_TYPE_MASK;
1568 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1569 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1572 * reserve descriptors and make sure we won't fail
1573 * to return them to userland.
1575 fd1 = get_unused_fd_flags(flags);
1576 if (unlikely(fd1 < 0))
1579 fd2 = get_unused_fd_flags(flags);
1580 if (unlikely(fd2 < 0)) {
1585 err = put_user(fd1, &usockvec[0]);
1589 err = put_user(fd2, &usockvec[1]);
1594 * Obtain the first socket and check if the underlying protocol
1595 * supports the socketpair call.
1598 err = sock_create(family, type, protocol, &sock1);
1599 if (unlikely(err < 0))
1602 err = sock_create(family, type, protocol, &sock2);
1603 if (unlikely(err < 0)) {
1604 sock_release(sock1);
1608 err = security_socket_socketpair(sock1, sock2);
1609 if (unlikely(err)) {
1610 sock_release(sock2);
1611 sock_release(sock1);
1615 err = sock1->ops->socketpair(sock1, sock2);
1616 if (unlikely(err < 0)) {
1617 sock_release(sock2);
1618 sock_release(sock1);
1622 newfile1 = sock_alloc_file(sock1, flags, NULL);
1623 if (IS_ERR(newfile1)) {
1624 err = PTR_ERR(newfile1);
1625 sock_release(sock2);
1629 newfile2 = sock_alloc_file(sock2, flags, NULL);
1630 if (IS_ERR(newfile2)) {
1631 err = PTR_ERR(newfile2);
1636 audit_fd_pair(fd1, fd2);
1638 fd_install(fd1, newfile1);
1639 fd_install(fd2, newfile2);
1648 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1649 int __user *, usockvec)
1651 return __sys_socketpair(family, type, protocol, usockvec);
1655 * Bind a name to a socket. Nothing much to do here since it's
1656 * the protocol's responsibility to handle the local address.
1658 * We move the socket address to kernel space before we call
1659 * the protocol layer (having also checked the address is ok).
1662 int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1664 struct socket *sock;
1665 struct sockaddr_storage address;
1666 int err, fput_needed;
1668 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1670 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1672 err = security_socket_bind(sock,
1673 (struct sockaddr *)&address,
1676 err = sock->ops->bind(sock,
1680 fput_light(sock->file, fput_needed);
1685 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1687 return __sys_bind(fd, umyaddr, addrlen);
1691 * Perform a listen. Basically, we allow the protocol to do anything
1692 * necessary for a listen, and if that works, we mark the socket as
1693 * ready for listening.
1696 int __sys_listen(int fd, int backlog)
1698 struct socket *sock;
1699 int err, fput_needed;
1702 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1704 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1705 if ((unsigned int)backlog > somaxconn)
1706 backlog = somaxconn;
1708 err = security_socket_listen(sock, backlog);
1710 err = sock->ops->listen(sock, backlog);
1712 fput_light(sock->file, fput_needed);
1717 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1719 return __sys_listen(fd, backlog);
1722 int __sys_accept4_file(struct file *file, unsigned file_flags,
1723 struct sockaddr __user *upeer_sockaddr,
1724 int __user *upeer_addrlen, int flags,
1725 unsigned long nofile)
1727 struct socket *sock, *newsock;
1728 struct file *newfile;
1729 int err, len, newfd;
1730 struct sockaddr_storage address;
1732 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1735 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1736 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1738 sock = sock_from_file(file);
1745 newsock = sock_alloc();
1749 newsock->type = sock->type;
1750 newsock->ops = sock->ops;
1753 * We don't need try_module_get here, as the listening socket (sock)
1754 * has the protocol module (sock->ops->owner) held.
1756 __module_get(newsock->ops->owner);
1758 newfd = __get_unused_fd_flags(flags, nofile);
1759 if (unlikely(newfd < 0)) {
1761 sock_release(newsock);
1764 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1765 if (IS_ERR(newfile)) {
1766 err = PTR_ERR(newfile);
1767 put_unused_fd(newfd);
1771 err = security_socket_accept(sock, newsock);
1775 err = sock->ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1780 if (upeer_sockaddr) {
1781 len = newsock->ops->getname(newsock,
1782 (struct sockaddr *)&address, 2);
1784 err = -ECONNABORTED;
1787 err = move_addr_to_user(&address,
1788 len, upeer_sockaddr, upeer_addrlen);
1793 /* File flags are not inherited via accept() unlike another OSes. */
1795 fd_install(newfd, newfile);
1801 put_unused_fd(newfd);
1807 * For accept, we attempt to create a new socket, set up the link
1808 * with the client, wake up the client, then return the new
1809 * connected fd. We collect the address of the connector in kernel
1810 * space and move it to user at the very end. This is unclean because
1811 * we open the socket then return an error.
1813 * 1003.1g adds the ability to recvmsg() to query connection pending
1814 * status to recvmsg. We need to add that support in a way thats
1815 * clean when we restructure accept also.
1818 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1819 int __user *upeer_addrlen, int flags)
1826 ret = __sys_accept4_file(f.file, 0, upeer_sockaddr,
1827 upeer_addrlen, flags,
1828 rlimit(RLIMIT_NOFILE));
1835 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1836 int __user *, upeer_addrlen, int, flags)
1838 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1841 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1842 int __user *, upeer_addrlen)
1844 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1848 * Attempt to connect to a socket with the server address. The address
1849 * is in user space so we verify it is OK and move it to kernel space.
1851 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1854 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1855 * other SEQPACKET protocols that take time to connect() as it doesn't
1856 * include the -EINPROGRESS status for such sockets.
1859 int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
1860 int addrlen, int file_flags)
1862 struct socket *sock;
1865 sock = sock_from_file(file);
1872 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1876 err = sock->ops->connect(sock, (struct sockaddr *)address, addrlen,
1877 sock->file->f_flags | file_flags);
1882 int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1889 struct sockaddr_storage address;
1891 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
1893 ret = __sys_connect_file(f.file, &address, addrlen, 0);
1900 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1903 return __sys_connect(fd, uservaddr, addrlen);
1907 * Get the local address ('name') of a socket object. Move the obtained
1908 * name to user space.
1911 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1912 int __user *usockaddr_len)
1914 struct socket *sock;
1915 struct sockaddr_storage address;
1916 int err, fput_needed;
1918 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1922 err = security_socket_getsockname(sock);
1926 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1929 /* "err" is actually length in this case */
1930 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1933 fput_light(sock->file, fput_needed);
1938 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1939 int __user *, usockaddr_len)
1941 return __sys_getsockname(fd, usockaddr, usockaddr_len);
1945 * Get the remote address ('name') of a socket object. Move the obtained
1946 * name to user space.
1949 int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1950 int __user *usockaddr_len)
1952 struct socket *sock;
1953 struct sockaddr_storage address;
1954 int err, fput_needed;
1956 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1958 err = security_socket_getpeername(sock);
1960 fput_light(sock->file, fput_needed);
1964 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
1966 /* "err" is actually length in this case */
1967 err = move_addr_to_user(&address, err, usockaddr,
1969 fput_light(sock->file, fput_needed);
1974 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1975 int __user *, usockaddr_len)
1977 return __sys_getpeername(fd, usockaddr, usockaddr_len);
1981 * Send a datagram to a given address. We move the address into kernel
1982 * space and check the user space data area is readable before invoking
1985 int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
1986 struct sockaddr __user *addr, int addr_len)
1988 struct socket *sock;
1989 struct sockaddr_storage address;
1995 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
1998 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2002 msg.msg_name = NULL;
2003 msg.msg_control = NULL;
2004 msg.msg_controllen = 0;
2005 msg.msg_namelen = 0;
2007 err = move_addr_to_kernel(addr, addr_len, &address);
2010 msg.msg_name = (struct sockaddr *)&address;
2011 msg.msg_namelen = addr_len;
2013 if (sock->file->f_flags & O_NONBLOCK)
2014 flags |= MSG_DONTWAIT;
2015 msg.msg_flags = flags;
2016 err = sock_sendmsg(sock, &msg);
2019 fput_light(sock->file, fput_needed);
2024 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2025 unsigned int, flags, struct sockaddr __user *, addr,
2028 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2032 * Send a datagram down a socket.
2035 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2036 unsigned int, flags)
2038 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2042 * Receive a frame from the socket and optionally record the address of the
2043 * sender. We verify the buffers are writable and if needed move the
2044 * sender address from kernel to user space.
2046 int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2047 struct sockaddr __user *addr, int __user *addr_len)
2049 struct socket *sock;
2052 struct sockaddr_storage address;
2056 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
2059 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2063 msg.msg_control = NULL;
2064 msg.msg_controllen = 0;
2065 /* Save some cycles and don't copy the address if not needed */
2066 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
2067 /* We assume all kernel code knows the size of sockaddr_storage */
2068 msg.msg_namelen = 0;
2069 msg.msg_iocb = NULL;
2071 if (sock->file->f_flags & O_NONBLOCK)
2072 flags |= MSG_DONTWAIT;
2073 err = sock_recvmsg(sock, &msg, flags);
2075 if (err >= 0 && addr != NULL) {
2076 err2 = move_addr_to_user(&address,
2077 msg.msg_namelen, addr, addr_len);
2082 fput_light(sock->file, fput_needed);
2087 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2088 unsigned int, flags, struct sockaddr __user *, addr,
2089 int __user *, addr_len)
2091 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2095 * Receive a datagram from a socket.
2098 SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2099 unsigned int, flags)
2101 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2104 static bool sock_use_custom_sol_socket(const struct socket *sock)
2106 const struct sock *sk = sock->sk;
2108 /* Use sock->ops->setsockopt() for MPTCP */
2109 return IS_ENABLED(CONFIG_MPTCP) &&
2110 sk->sk_protocol == IPPROTO_MPTCP &&
2111 sk->sk_type == SOCK_STREAM &&
2112 (sk->sk_family == AF_INET || sk->sk_family == AF_INET6);
2116 * Set a socket option. Because we don't know the option lengths we have
2117 * to pass the user mode parameter for the protocols to sort out.
2119 int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2122 sockptr_t optval = USER_SOCKPTR(user_optval);
2123 char *kernel_optval = NULL;
2124 int err, fput_needed;
2125 struct socket *sock;
2130 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2134 err = security_socket_setsockopt(sock, level, optname);
2138 if (!in_compat_syscall())
2139 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2140 user_optval, &optlen,
2150 optval = KERNEL_SOCKPTR(kernel_optval);
2151 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2152 err = sock_setsockopt(sock, level, optname, optval, optlen);
2153 else if (unlikely(!sock->ops->setsockopt))
2156 err = sock->ops->setsockopt(sock, level, optname, optval,
2158 kfree(kernel_optval);
2160 fput_light(sock->file, fput_needed);
2164 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2165 char __user *, optval, int, optlen)
2167 return __sys_setsockopt(fd, level, optname, optval, optlen);
2170 INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2174 * Get a socket option. Because we don't know the option lengths we have
2175 * to pass a user mode parameter for the protocols to sort out.
2177 int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2180 int err, fput_needed;
2181 struct socket *sock;
2184 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2188 err = security_socket_getsockopt(sock, level, optname);
2192 if (!in_compat_syscall())
2193 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2195 if (level == SOL_SOCKET)
2196 err = sock_getsockopt(sock, level, optname, optval, optlen);
2197 else if (unlikely(!sock->ops->getsockopt))
2200 err = sock->ops->getsockopt(sock, level, optname, optval,
2203 if (!in_compat_syscall())
2204 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2205 optval, optlen, max_optlen,
2208 fput_light(sock->file, fput_needed);
2212 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2213 char __user *, optval, int __user *, optlen)
2215 return __sys_getsockopt(fd, level, optname, optval, optlen);
2219 * Shutdown a socket.
2222 int __sys_shutdown_sock(struct socket *sock, int how)
2226 err = security_socket_shutdown(sock, how);
2228 err = sock->ops->shutdown(sock, how);
2233 int __sys_shutdown(int fd, int how)
2235 int err, fput_needed;
2236 struct socket *sock;
2238 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2240 err = __sys_shutdown_sock(sock, how);
2241 fput_light(sock->file, fput_needed);
2246 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2248 return __sys_shutdown(fd, how);
2251 /* A couple of helpful macros for getting the address of the 32/64 bit
2252 * fields which are the same type (int / unsigned) on our platforms.
2254 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2255 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2256 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2258 struct used_address {
2259 struct sockaddr_storage name;
2260 unsigned int name_len;
2263 int __copy_msghdr_from_user(struct msghdr *kmsg,
2264 struct user_msghdr __user *umsg,
2265 struct sockaddr __user **save_addr,
2266 struct iovec __user **uiov, size_t *nsegs)
2268 struct user_msghdr msg;
2271 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2274 kmsg->msg_control_is_user = true;
2275 kmsg->msg_control_user = msg.msg_control;
2276 kmsg->msg_controllen = msg.msg_controllen;
2277 kmsg->msg_flags = msg.msg_flags;
2279 kmsg->msg_namelen = msg.msg_namelen;
2281 kmsg->msg_namelen = 0;
2283 if (kmsg->msg_namelen < 0)
2286 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2287 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2290 *save_addr = msg.msg_name;
2292 if (msg.msg_name && kmsg->msg_namelen) {
2294 err = move_addr_to_kernel(msg.msg_name,
2301 kmsg->msg_name = NULL;
2302 kmsg->msg_namelen = 0;
2305 if (msg.msg_iovlen > UIO_MAXIOV)
2308 kmsg->msg_iocb = NULL;
2309 *uiov = msg.msg_iov;
2310 *nsegs = msg.msg_iovlen;
2314 static int copy_msghdr_from_user(struct msghdr *kmsg,
2315 struct user_msghdr __user *umsg,
2316 struct sockaddr __user **save_addr,
2319 struct user_msghdr msg;
2322 err = __copy_msghdr_from_user(kmsg, umsg, save_addr, &msg.msg_iov,
2327 err = import_iovec(save_addr ? READ : WRITE,
2328 msg.msg_iov, msg.msg_iovlen,
2329 UIO_FASTIOV, iov, &kmsg->msg_iter);
2330 return err < 0 ? err : 0;
2333 static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2334 unsigned int flags, struct used_address *used_address,
2335 unsigned int allowed_msghdr_flags)
2337 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2338 __aligned(sizeof(__kernel_size_t));
2339 /* 20 is size of ipv6_pktinfo */
2340 unsigned char *ctl_buf = ctl;
2346 if (msg_sys->msg_controllen > INT_MAX)
2348 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2349 ctl_len = msg_sys->msg_controllen;
2350 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2352 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2356 ctl_buf = msg_sys->msg_control;
2357 ctl_len = msg_sys->msg_controllen;
2358 } else if (ctl_len) {
2359 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2360 CMSG_ALIGN(sizeof(struct cmsghdr)));
2361 if (ctl_len > sizeof(ctl)) {
2362 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2363 if (ctl_buf == NULL)
2367 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2369 msg_sys->msg_control = ctl_buf;
2370 msg_sys->msg_control_is_user = false;
2372 msg_sys->msg_flags = flags;
2374 if (sock->file->f_flags & O_NONBLOCK)
2375 msg_sys->msg_flags |= MSG_DONTWAIT;
2377 * If this is sendmmsg() and current destination address is same as
2378 * previously succeeded address, omit asking LSM's decision.
2379 * used_address->name_len is initialized to UINT_MAX so that the first
2380 * destination address never matches.
2382 if (used_address && msg_sys->msg_name &&
2383 used_address->name_len == msg_sys->msg_namelen &&
2384 !memcmp(&used_address->name, msg_sys->msg_name,
2385 used_address->name_len)) {
2386 err = sock_sendmsg_nosec(sock, msg_sys);
2389 err = sock_sendmsg(sock, msg_sys);
2391 * If this is sendmmsg() and sending to current destination address was
2392 * successful, remember it.
2394 if (used_address && err >= 0) {
2395 used_address->name_len = msg_sys->msg_namelen;
2396 if (msg_sys->msg_name)
2397 memcpy(&used_address->name, msg_sys->msg_name,
2398 used_address->name_len);
2403 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2408 int sendmsg_copy_msghdr(struct msghdr *msg,
2409 struct user_msghdr __user *umsg, unsigned flags,
2414 if (flags & MSG_CMSG_COMPAT) {
2415 struct compat_msghdr __user *msg_compat;
2417 msg_compat = (struct compat_msghdr __user *) umsg;
2418 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2420 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2428 static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2429 struct msghdr *msg_sys, unsigned int flags,
2430 struct used_address *used_address,
2431 unsigned int allowed_msghdr_flags)
2433 struct sockaddr_storage address;
2434 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2437 msg_sys->msg_name = &address;
2439 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2443 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2444 allowed_msghdr_flags);
2450 * BSD sendmsg interface
2452 long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2455 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2458 long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2459 bool forbid_cmsg_compat)
2461 int fput_needed, err;
2462 struct msghdr msg_sys;
2463 struct socket *sock;
2465 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2468 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2472 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2474 fput_light(sock->file, fput_needed);
2479 SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2481 return __sys_sendmsg(fd, msg, flags, true);
2485 * Linux sendmmsg interface
2488 int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2489 unsigned int flags, bool forbid_cmsg_compat)
2491 int fput_needed, err, datagrams;
2492 struct socket *sock;
2493 struct mmsghdr __user *entry;
2494 struct compat_mmsghdr __user *compat_entry;
2495 struct msghdr msg_sys;
2496 struct used_address used_address;
2497 unsigned int oflags = flags;
2499 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2502 if (vlen > UIO_MAXIOV)
2507 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2511 used_address.name_len = UINT_MAX;
2513 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2517 while (datagrams < vlen) {
2518 if (datagrams == vlen - 1)
2521 if (MSG_CMSG_COMPAT & flags) {
2522 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2523 &msg_sys, flags, &used_address, MSG_EOR);
2526 err = __put_user(err, &compat_entry->msg_len);
2529 err = ___sys_sendmsg(sock,
2530 (struct user_msghdr __user *)entry,
2531 &msg_sys, flags, &used_address, MSG_EOR);
2534 err = put_user(err, &entry->msg_len);
2541 if (msg_data_left(&msg_sys))
2546 fput_light(sock->file, fput_needed);
2548 /* We only return an error if no datagrams were able to be sent */
2555 SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2556 unsigned int, vlen, unsigned int, flags)
2558 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2561 int recvmsg_copy_msghdr(struct msghdr *msg,
2562 struct user_msghdr __user *umsg, unsigned flags,
2563 struct sockaddr __user **uaddr,
2568 if (MSG_CMSG_COMPAT & flags) {
2569 struct compat_msghdr __user *msg_compat;
2571 msg_compat = (struct compat_msghdr __user *) umsg;
2572 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2574 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2582 static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2583 struct user_msghdr __user *msg,
2584 struct sockaddr __user *uaddr,
2585 unsigned int flags, int nosec)
2587 struct compat_msghdr __user *msg_compat =
2588 (struct compat_msghdr __user *) msg;
2589 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2590 struct sockaddr_storage addr;
2591 unsigned long cmsg_ptr;
2595 msg_sys->msg_name = &addr;
2596 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2597 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2599 /* We assume all kernel code knows the size of sockaddr_storage */
2600 msg_sys->msg_namelen = 0;
2602 if (sock->file->f_flags & O_NONBLOCK)
2603 flags |= MSG_DONTWAIT;
2605 if (unlikely(nosec))
2606 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2608 err = sock_recvmsg(sock, msg_sys, flags);
2614 if (uaddr != NULL) {
2615 err = move_addr_to_user(&addr,
2616 msg_sys->msg_namelen, uaddr,
2621 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2625 if (MSG_CMSG_COMPAT & flags)
2626 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2627 &msg_compat->msg_controllen);
2629 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2630 &msg->msg_controllen);
2638 static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2639 struct msghdr *msg_sys, unsigned int flags, int nosec)
2641 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2642 /* user mode address pointers */
2643 struct sockaddr __user *uaddr;
2646 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2650 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2656 * BSD recvmsg interface
2659 long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2660 struct user_msghdr __user *umsg,
2661 struct sockaddr __user *uaddr, unsigned int flags)
2663 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2666 long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2667 bool forbid_cmsg_compat)
2669 int fput_needed, err;
2670 struct msghdr msg_sys;
2671 struct socket *sock;
2673 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2676 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2680 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2682 fput_light(sock->file, fput_needed);
2687 SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2688 unsigned int, flags)
2690 return __sys_recvmsg(fd, msg, flags, true);
2694 * Linux recvmmsg interface
2697 static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2698 unsigned int vlen, unsigned int flags,
2699 struct timespec64 *timeout)
2701 int fput_needed, err, datagrams;
2702 struct socket *sock;
2703 struct mmsghdr __user *entry;
2704 struct compat_mmsghdr __user *compat_entry;
2705 struct msghdr msg_sys;
2706 struct timespec64 end_time;
2707 struct timespec64 timeout64;
2710 poll_select_set_timeout(&end_time, timeout->tv_sec,
2716 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2720 if (likely(!(flags & MSG_ERRQUEUE))) {
2721 err = sock_error(sock->sk);
2729 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2731 while (datagrams < vlen) {
2733 * No need to ask LSM for more than the first datagram.
2735 if (MSG_CMSG_COMPAT & flags) {
2736 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2737 &msg_sys, flags & ~MSG_WAITFORONE,
2741 err = __put_user(err, &compat_entry->msg_len);
2744 err = ___sys_recvmsg(sock,
2745 (struct user_msghdr __user *)entry,
2746 &msg_sys, flags & ~MSG_WAITFORONE,
2750 err = put_user(err, &entry->msg_len);
2758 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2759 if (flags & MSG_WAITFORONE)
2760 flags |= MSG_DONTWAIT;
2763 ktime_get_ts64(&timeout64);
2764 *timeout = timespec64_sub(end_time, timeout64);
2765 if (timeout->tv_sec < 0) {
2766 timeout->tv_sec = timeout->tv_nsec = 0;
2770 /* Timeout, return less than vlen datagrams */
2771 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2775 /* Out of band data, return right away */
2776 if (msg_sys.msg_flags & MSG_OOB)
2784 if (datagrams == 0) {
2790 * We may return less entries than requested (vlen) if the
2791 * sock is non block and there aren't enough datagrams...
2793 if (err != -EAGAIN) {
2795 * ... or if recvmsg returns an error after we
2796 * received some datagrams, where we record the
2797 * error to return on the next call or if the
2798 * app asks about it using getsockopt(SO_ERROR).
2800 sock->sk->sk_err = -err;
2803 fput_light(sock->file, fput_needed);
2808 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2809 unsigned int vlen, unsigned int flags,
2810 struct __kernel_timespec __user *timeout,
2811 struct old_timespec32 __user *timeout32)
2814 struct timespec64 timeout_sys;
2816 if (timeout && get_timespec64(&timeout_sys, timeout))
2819 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2822 if (!timeout && !timeout32)
2823 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2825 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2830 if (timeout && put_timespec64(&timeout_sys, timeout))
2831 datagrams = -EFAULT;
2833 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2834 datagrams = -EFAULT;
2839 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2840 unsigned int, vlen, unsigned int, flags,
2841 struct __kernel_timespec __user *, timeout)
2843 if (flags & MSG_CMSG_COMPAT)
2846 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2849 #ifdef CONFIG_COMPAT_32BIT_TIME
2850 SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2851 unsigned int, vlen, unsigned int, flags,
2852 struct old_timespec32 __user *, timeout)
2854 if (flags & MSG_CMSG_COMPAT)
2857 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2861 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2862 /* Argument list sizes for sys_socketcall */
2863 #define AL(x) ((x) * sizeof(unsigned long))
2864 static const unsigned char nargs[21] = {
2865 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2866 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2867 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2874 * System call vectors.
2876 * Argument checking cleaned up. Saved 20% in size.
2877 * This function doesn't need to set the kernel lock because
2878 * it is set by the callees.
2881 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2883 unsigned long a[AUDITSC_ARGS];
2884 unsigned long a0, a1;
2888 if (call < 1 || call > SYS_SENDMMSG)
2890 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2893 if (len > sizeof(a))
2896 /* copy_from_user should be SMP safe. */
2897 if (copy_from_user(a, args, len))
2900 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2909 err = __sys_socket(a0, a1, a[2]);
2912 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2915 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2918 err = __sys_listen(a0, a1);
2921 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2922 (int __user *)a[2], 0);
2924 case SYS_GETSOCKNAME:
2926 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2927 (int __user *)a[2]);
2929 case SYS_GETPEERNAME:
2931 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2932 (int __user *)a[2]);
2934 case SYS_SOCKETPAIR:
2935 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2938 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2942 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2943 (struct sockaddr __user *)a[4], a[5]);
2946 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2950 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2951 (struct sockaddr __user *)a[4],
2952 (int __user *)a[5]);
2955 err = __sys_shutdown(a0, a1);
2957 case SYS_SETSOCKOPT:
2958 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
2961 case SYS_GETSOCKOPT:
2963 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2964 (int __user *)a[4]);
2967 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
2971 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
2975 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
2979 if (IS_ENABLED(CONFIG_64BIT))
2980 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2982 (struct __kernel_timespec __user *)a[4],
2985 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2987 (struct old_timespec32 __user *)a[4]);
2990 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2991 (int __user *)a[2], a[3]);
3000 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
3003 * sock_register - add a socket protocol handler
3004 * @ops: description of protocol
3006 * This function is called by a protocol handler that wants to
3007 * advertise its address family, and have it linked into the
3008 * socket interface. The value ops->family corresponds to the
3009 * socket system call protocol family.
3011 int sock_register(const struct net_proto_family *ops)
3015 if (ops->family >= NPROTO) {
3016 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3020 spin_lock(&net_family_lock);
3021 if (rcu_dereference_protected(net_families[ops->family],
3022 lockdep_is_held(&net_family_lock)))
3025 rcu_assign_pointer(net_families[ops->family], ops);
3028 spin_unlock(&net_family_lock);
3030 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3033 EXPORT_SYMBOL(sock_register);
3036 * sock_unregister - remove a protocol handler
3037 * @family: protocol family to remove
3039 * This function is called by a protocol handler that wants to
3040 * remove its address family, and have it unlinked from the
3041 * new socket creation.
3043 * If protocol handler is a module, then it can use module reference
3044 * counts to protect against new references. If protocol handler is not
3045 * a module then it needs to provide its own protection in
3046 * the ops->create routine.
3048 void sock_unregister(int family)
3050 BUG_ON(family < 0 || family >= NPROTO);
3052 spin_lock(&net_family_lock);
3053 RCU_INIT_POINTER(net_families[family], NULL);
3054 spin_unlock(&net_family_lock);
3058 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3060 EXPORT_SYMBOL(sock_unregister);
3062 bool sock_is_registered(int family)
3064 return family < NPROTO && rcu_access_pointer(net_families[family]);
3067 static int __init sock_init(void)
3071 * Initialize the network sysctl infrastructure.
3073 err = net_sysctl_init();
3078 * Initialize skbuff SLAB cache
3083 * Initialize the protocols module.
3088 err = register_filesystem(&sock_fs_type);
3091 sock_mnt = kern_mount(&sock_fs_type);
3092 if (IS_ERR(sock_mnt)) {
3093 err = PTR_ERR(sock_mnt);
3097 /* The real protocol initialization is performed in later initcalls.
3100 #ifdef CONFIG_NETFILTER
3101 err = netfilter_init();
3106 ptp_classifier_init();
3112 unregister_filesystem(&sock_fs_type);
3116 core_initcall(sock_init); /* early initcall */
3118 #ifdef CONFIG_PROC_FS
3119 void socket_seq_show(struct seq_file *seq)
3121 seq_printf(seq, "sockets: used %d\n",
3122 sock_inuse_get(seq->private));
3124 #endif /* CONFIG_PROC_FS */
3126 /* Handle the fact that while struct ifreq has the same *layout* on
3127 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3128 * which are handled elsewhere, it still has different *size* due to
3129 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3130 * resulting in struct ifreq being 32 and 40 bytes respectively).
3131 * As a result, if the struct happens to be at the end of a page and
3132 * the next page isn't readable/writable, we get a fault. To prevent
3133 * that, copy back and forth to the full size.
3135 int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3137 if (in_compat_syscall()) {
3138 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3140 memset(ifr, 0, sizeof(*ifr));
3141 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3145 *ifrdata = compat_ptr(ifr32->ifr_data);
3150 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3154 *ifrdata = ifr->ifr_data;
3158 EXPORT_SYMBOL(get_user_ifreq);
3160 int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3162 size_t size = sizeof(*ifr);
3164 if (in_compat_syscall())
3165 size = sizeof(struct compat_ifreq);
3167 if (copy_to_user(arg, ifr, size))
3172 EXPORT_SYMBOL(put_user_ifreq);
3174 #ifdef CONFIG_COMPAT
3175 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3177 compat_uptr_t uptr32;
3182 if (get_user_ifreq(&ifr, NULL, uifr32))
3185 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3188 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3189 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3191 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3193 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3194 if (put_user_ifreq(&ifr, uifr32))
3200 /* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3201 static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3202 struct compat_ifreq __user *u_ifreq32)
3207 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3209 ifreq.ifr_data = data;
3211 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3214 /* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3215 * for some operations; this forces use of the newer bridge-utils that
3216 * use compatible ioctls
3218 static int old_bridge_ioctl(compat_ulong_t __user *argp)
3222 if (get_user(tmp, argp))
3224 if (tmp == BRCTL_GET_VERSION)
3225 return BRCTL_VERSION + 1;
3229 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3230 unsigned int cmd, unsigned long arg)
3232 void __user *argp = compat_ptr(arg);
3233 struct sock *sk = sock->sk;
3234 struct net *net = sock_net(sk);
3236 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3237 return compat_ifr_data_ioctl(net, cmd, argp);
3242 return old_bridge_ioctl(argp);
3244 return compat_siocwandev(net, argp);
3245 case SIOCGSTAMP_OLD:
3246 case SIOCGSTAMPNS_OLD:
3247 if (!sock->ops->gettstamp)
3248 return -ENOIOCTLCMD;
3249 return sock->ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3250 !COMPAT_USE_64BIT_TIME);
3253 case SIOCBONDSLAVEINFOQUERY:
3254 case SIOCBONDINFOQUERY:
3257 return compat_ifr_data_ioctl(net, cmd, argp);
3268 case SIOCGSTAMP_NEW:
3269 case SIOCGSTAMPNS_NEW:
3271 return sock_ioctl(file, cmd, arg);
3290 case SIOCSIFHWBROADCAST:
3292 case SIOCGIFBRDADDR:
3293 case SIOCSIFBRDADDR:
3294 case SIOCGIFDSTADDR:
3295 case SIOCSIFDSTADDR:
3296 case SIOCGIFNETMASK:
3297 case SIOCSIFNETMASK:
3309 case SIOCBONDENSLAVE:
3310 case SIOCBONDRELEASE:
3311 case SIOCBONDSETHWADDR:
3312 case SIOCBONDCHANGEACTIVE:
3319 return sock_do_ioctl(net, sock, cmd, arg);
3322 return -ENOIOCTLCMD;
3325 static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3328 struct socket *sock = file->private_data;
3329 int ret = -ENOIOCTLCMD;
3336 if (sock->ops->compat_ioctl)
3337 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3339 if (ret == -ENOIOCTLCMD &&
3340 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3341 ret = compat_wext_handle_ioctl(net, cmd, arg);
3343 if (ret == -ENOIOCTLCMD)
3344 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3351 * kernel_bind - bind an address to a socket (kernel space)
3354 * @addrlen: length of address
3356 * Returns 0 or an error.
3359 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3361 return sock->ops->bind(sock, addr, addrlen);
3363 EXPORT_SYMBOL(kernel_bind);
3366 * kernel_listen - move socket to listening state (kernel space)
3368 * @backlog: pending connections queue size
3370 * Returns 0 or an error.
3373 int kernel_listen(struct socket *sock, int backlog)
3375 return sock->ops->listen(sock, backlog);
3377 EXPORT_SYMBOL(kernel_listen);
3380 * kernel_accept - accept a connection (kernel space)
3381 * @sock: listening socket
3382 * @newsock: new connected socket
3385 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3386 * If it fails, @newsock is guaranteed to be %NULL.
3387 * Returns 0 or an error.
3390 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3392 struct sock *sk = sock->sk;
3395 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3400 err = sock->ops->accept(sock, *newsock, flags, true);
3402 sock_release(*newsock);
3407 (*newsock)->ops = sock->ops;
3408 __module_get((*newsock)->ops->owner);
3413 EXPORT_SYMBOL(kernel_accept);
3416 * kernel_connect - connect a socket (kernel space)
3419 * @addrlen: address length
3420 * @flags: flags (O_NONBLOCK, ...)
3422 * For datagram sockets, @addr is the address to which datagrams are sent
3423 * by default, and the only address from which datagrams are received.
3424 * For stream sockets, attempts to connect to @addr.
3425 * Returns 0 or an error code.
3428 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3431 return sock->ops->connect(sock, addr, addrlen, flags);
3433 EXPORT_SYMBOL(kernel_connect);
3436 * kernel_getsockname - get the address which the socket is bound (kernel space)
3438 * @addr: address holder
3440 * Fills the @addr pointer with the address which the socket is bound.
3441 * Returns 0 or an error code.
3444 int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3446 return sock->ops->getname(sock, addr, 0);
3448 EXPORT_SYMBOL(kernel_getsockname);
3451 * kernel_getpeername - get the address which the socket is connected (kernel space)
3453 * @addr: address holder
3455 * Fills the @addr pointer with the address which the socket is connected.
3456 * Returns 0 or an error code.
3459 int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3461 return sock->ops->getname(sock, addr, 1);
3463 EXPORT_SYMBOL(kernel_getpeername);
3466 * kernel_sendpage - send a &page through a socket (kernel space)
3469 * @offset: page offset
3470 * @size: total size in bytes
3471 * @flags: flags (MSG_DONTWAIT, ...)
3473 * Returns the total amount sent in bytes or an error.
3476 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3477 size_t size, int flags)
3479 if (sock->ops->sendpage) {
3480 /* Warn in case the improper page to zero-copy send */
3481 WARN_ONCE(!sendpage_ok(page), "improper page for zero-copy send");
3482 return sock->ops->sendpage(sock, page, offset, size, flags);
3484 return sock_no_sendpage(sock, page, offset, size, flags);
3486 EXPORT_SYMBOL(kernel_sendpage);
3489 * kernel_sendpage_locked - send a &page through the locked sock (kernel space)
3492 * @offset: page offset
3493 * @size: total size in bytes
3494 * @flags: flags (MSG_DONTWAIT, ...)
3496 * Returns the total amount sent in bytes or an error.
3497 * Caller must hold @sk.
3500 int kernel_sendpage_locked(struct sock *sk, struct page *page, int offset,
3501 size_t size, int flags)
3503 struct socket *sock = sk->sk_socket;
3505 if (sock->ops->sendpage_locked)
3506 return sock->ops->sendpage_locked(sk, page, offset, size,
3509 return sock_no_sendpage_locked(sk, page, offset, size, flags);
3511 EXPORT_SYMBOL(kernel_sendpage_locked);
3514 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3516 * @how: connection part
3518 * Returns 0 or an error.
3521 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3523 return sock->ops->shutdown(sock, how);
3525 EXPORT_SYMBOL(kernel_sock_shutdown);
3528 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3531 * This routine returns the IP overhead imposed by a socket i.e.
3532 * the length of the underlying IP header, depending on whether
3533 * this is an IPv4 or IPv6 socket and the length from IP options turned
3534 * on at the socket. Assumes that the caller has a lock on the socket.
3537 u32 kernel_sock_ip_overhead(struct sock *sk)
3539 struct inet_sock *inet;
3540 struct ip_options_rcu *opt;
3542 #if IS_ENABLED(CONFIG_IPV6)
3543 struct ipv6_pinfo *np;
3544 struct ipv6_txoptions *optv6 = NULL;
3545 #endif /* IS_ENABLED(CONFIG_IPV6) */
3550 switch (sk->sk_family) {
3553 overhead += sizeof(struct iphdr);
3554 opt = rcu_dereference_protected(inet->inet_opt,
3555 sock_owned_by_user(sk));
3557 overhead += opt->opt.optlen;
3559 #if IS_ENABLED(CONFIG_IPV6)
3562 overhead += sizeof(struct ipv6hdr);
3564 optv6 = rcu_dereference_protected(np->opt,
3565 sock_owned_by_user(sk));
3567 overhead += (optv6->opt_flen + optv6->opt_nflen);
3569 #endif /* IS_ENABLED(CONFIG_IPV6) */
3570 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3574 EXPORT_SYMBOL(kernel_sock_ip_overhead);