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/net.h>
61 #include <linux/interrupt.h>
62 #include <linux/thread_info.h>
63 #include <linux/rcupdate.h>
64 #include <linux/netdevice.h>
65 #include <linux/proc_fs.h>
66 #include <linux/seq_file.h>
67 #include <linux/mutex.h>
68 #include <linux/if_bridge.h>
69 #include <linux/if_vlan.h>
70 #include <linux/ptp_classify.h>
71 #include <linux/init.h>
72 #include <linux/poll.h>
73 #include <linux/cache.h>
74 #include <linux/module.h>
75 #include <linux/highmem.h>
76 #include <linux/mount.h>
77 #include <linux/pseudo_fs.h>
78 #include <linux/security.h>
79 #include <linux/syscalls.h>
80 #include <linux/compat.h>
81 #include <linux/kmod.h>
82 #include <linux/audit.h>
83 #include <linux/wireless.h>
84 #include <linux/nsproxy.h>
85 #include <linux/magic.h>
86 #include <linux/slab.h>
87 #include <linux/xattr.h>
88 #include <linux/nospec.h>
89 #include <linux/indirect_call_wrapper.h>
91 #include <linux/uaccess.h>
92 #include <asm/unistd.h>
94 #include <net/compat.h>
96 #include <net/cls_cgroup.h>
99 #include <linux/netfilter.h>
101 #include <linux/if_tun.h>
102 #include <linux/ipv6_route.h>
103 #include <linux/route.h>
104 #include <linux/termios.h>
105 #include <linux/sockios.h>
106 #include <net/busy_poll.h>
107 #include <linux/errqueue.h>
108 #include <linux/ptp_clock_kernel.h>
110 #ifdef CONFIG_NET_RX_BUSY_POLL
111 unsigned int sysctl_net_busy_read __read_mostly;
112 unsigned int sysctl_net_busy_poll __read_mostly;
115 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
116 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
117 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
119 static int sock_close(struct inode *inode, struct file *file);
120 static __poll_t sock_poll(struct file *file,
121 struct poll_table_struct *wait);
122 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
124 static long compat_sock_ioctl(struct file *file,
125 unsigned int cmd, unsigned long arg);
127 static int sock_fasync(int fd, struct file *filp, int on);
128 static ssize_t sock_sendpage(struct file *file, struct page *page,
129 int offset, size_t size, loff_t *ppos, int more);
130 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
131 struct pipe_inode_info *pipe, size_t len,
134 #ifdef CONFIG_PROC_FS
135 static void sock_show_fdinfo(struct seq_file *m, struct file *f)
137 struct socket *sock = f->private_data;
139 if (sock->ops->show_fdinfo)
140 sock->ops->show_fdinfo(m, sock);
143 #define sock_show_fdinfo NULL
147 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
148 * in the operation structures but are done directly via the socketcall() multiplexor.
151 static const struct file_operations socket_file_ops = {
152 .owner = THIS_MODULE,
154 .read_iter = sock_read_iter,
155 .write_iter = sock_write_iter,
157 .unlocked_ioctl = sock_ioctl,
159 .compat_ioctl = compat_sock_ioctl,
162 .release = sock_close,
163 .fasync = sock_fasync,
164 .sendpage = sock_sendpage,
165 .splice_write = generic_splice_sendpage,
166 .splice_read = sock_splice_read,
167 .show_fdinfo = sock_show_fdinfo,
170 static const char * const pf_family_names[] = {
171 [PF_UNSPEC] = "PF_UNSPEC",
172 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
173 [PF_INET] = "PF_INET",
174 [PF_AX25] = "PF_AX25",
176 [PF_APPLETALK] = "PF_APPLETALK",
177 [PF_NETROM] = "PF_NETROM",
178 [PF_BRIDGE] = "PF_BRIDGE",
179 [PF_ATMPVC] = "PF_ATMPVC",
181 [PF_INET6] = "PF_INET6",
182 [PF_ROSE] = "PF_ROSE",
183 [PF_DECnet] = "PF_DECnet",
184 [PF_NETBEUI] = "PF_NETBEUI",
185 [PF_SECURITY] = "PF_SECURITY",
187 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
188 [PF_PACKET] = "PF_PACKET",
190 [PF_ECONET] = "PF_ECONET",
191 [PF_ATMSVC] = "PF_ATMSVC",
194 [PF_IRDA] = "PF_IRDA",
195 [PF_PPPOX] = "PF_PPPOX",
196 [PF_WANPIPE] = "PF_WANPIPE",
199 [PF_MPLS] = "PF_MPLS",
201 [PF_TIPC] = "PF_TIPC",
202 [PF_BLUETOOTH] = "PF_BLUETOOTH",
203 [PF_IUCV] = "PF_IUCV",
204 [PF_RXRPC] = "PF_RXRPC",
205 [PF_ISDN] = "PF_ISDN",
206 [PF_PHONET] = "PF_PHONET",
207 [PF_IEEE802154] = "PF_IEEE802154",
208 [PF_CAIF] = "PF_CAIF",
211 [PF_VSOCK] = "PF_VSOCK",
213 [PF_QIPCRTR] = "PF_QIPCRTR",
216 [PF_MCTP] = "PF_MCTP",
220 * The protocol list. Each protocol is registered in here.
223 static DEFINE_SPINLOCK(net_family_lock);
224 static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
228 * Move socket addresses back and forth across the kernel/user
229 * divide and look after the messy bits.
233 * move_addr_to_kernel - copy a socket address into kernel space
234 * @uaddr: Address in user space
235 * @kaddr: Address in kernel space
236 * @ulen: Length in user space
238 * The address is copied into kernel space. If the provided address is
239 * too long an error code of -EINVAL is returned. If the copy gives
240 * invalid addresses -EFAULT is returned. On a success 0 is returned.
243 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
245 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
249 if (copy_from_user(kaddr, uaddr, ulen))
251 return audit_sockaddr(ulen, kaddr);
255 * move_addr_to_user - copy an address to user space
256 * @kaddr: kernel space address
257 * @klen: length of address in kernel
258 * @uaddr: user space address
259 * @ulen: pointer to user length field
261 * The value pointed to by ulen on entry is the buffer length available.
262 * This is overwritten with the buffer space used. -EINVAL is returned
263 * if an overlong buffer is specified or a negative buffer size. -EFAULT
264 * is returned if either the buffer or the length field are not
266 * After copying the data up to the limit the user specifies, the true
267 * length of the data is written over the length limit the user
268 * specified. Zero is returned for a success.
271 static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
272 void __user *uaddr, int __user *ulen)
277 BUG_ON(klen > sizeof(struct sockaddr_storage));
278 err = get_user(len, ulen);
286 if (audit_sockaddr(klen, kaddr))
288 if (copy_to_user(uaddr, kaddr, len))
292 * "fromlen shall refer to the value before truncation.."
295 return __put_user(klen, ulen);
298 static struct kmem_cache *sock_inode_cachep __ro_after_init;
300 static struct inode *sock_alloc_inode(struct super_block *sb)
302 struct socket_alloc *ei;
304 ei = alloc_inode_sb(sb, sock_inode_cachep, GFP_KERNEL);
307 init_waitqueue_head(&ei->socket.wq.wait);
308 ei->socket.wq.fasync_list = NULL;
309 ei->socket.wq.flags = 0;
311 ei->socket.state = SS_UNCONNECTED;
312 ei->socket.flags = 0;
313 ei->socket.ops = NULL;
314 ei->socket.sk = NULL;
315 ei->socket.file = NULL;
317 return &ei->vfs_inode;
320 static void sock_free_inode(struct inode *inode)
322 struct socket_alloc *ei;
324 ei = container_of(inode, struct socket_alloc, vfs_inode);
325 kmem_cache_free(sock_inode_cachep, ei);
328 static void init_once(void *foo)
330 struct socket_alloc *ei = (struct socket_alloc *)foo;
332 inode_init_once(&ei->vfs_inode);
335 static void init_inodecache(void)
337 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
338 sizeof(struct socket_alloc),
340 (SLAB_HWCACHE_ALIGN |
341 SLAB_RECLAIM_ACCOUNT |
342 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
344 BUG_ON(sock_inode_cachep == NULL);
347 static const struct super_operations sockfs_ops = {
348 .alloc_inode = sock_alloc_inode,
349 .free_inode = sock_free_inode,
350 .statfs = simple_statfs,
354 * sockfs_dname() is called from d_path().
356 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
358 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
359 d_inode(dentry)->i_ino);
362 static const struct dentry_operations sockfs_dentry_operations = {
363 .d_dname = sockfs_dname,
366 static int sockfs_xattr_get(const struct xattr_handler *handler,
367 struct dentry *dentry, struct inode *inode,
368 const char *suffix, void *value, size_t size)
371 if (dentry->d_name.len + 1 > size)
373 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
375 return dentry->d_name.len + 1;
378 #define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
379 #define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
380 #define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
382 static const struct xattr_handler sockfs_xattr_handler = {
383 .name = XATTR_NAME_SOCKPROTONAME,
384 .get = sockfs_xattr_get,
387 static int sockfs_security_xattr_set(const struct xattr_handler *handler,
388 struct user_namespace *mnt_userns,
389 struct dentry *dentry, struct inode *inode,
390 const char *suffix, const void *value,
391 size_t size, int flags)
393 /* Handled by LSM. */
397 static const struct xattr_handler sockfs_security_xattr_handler = {
398 .prefix = XATTR_SECURITY_PREFIX,
399 .set = sockfs_security_xattr_set,
402 static const struct xattr_handler *sockfs_xattr_handlers[] = {
403 &sockfs_xattr_handler,
404 &sockfs_security_xattr_handler,
408 static int sockfs_init_fs_context(struct fs_context *fc)
410 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
413 ctx->ops = &sockfs_ops;
414 ctx->dops = &sockfs_dentry_operations;
415 ctx->xattr = sockfs_xattr_handlers;
419 static struct vfsmount *sock_mnt __read_mostly;
421 static struct file_system_type sock_fs_type = {
423 .init_fs_context = sockfs_init_fs_context,
424 .kill_sb = kill_anon_super,
428 * Obtains the first available file descriptor and sets it up for use.
430 * These functions create file structures and maps them to fd space
431 * of the current process. On success it returns file descriptor
432 * and file struct implicitly stored in sock->file.
433 * Note that another thread may close file descriptor before we return
434 * from this function. We use the fact that now we do not refer
435 * to socket after mapping. If one day we will need it, this
436 * function will increment ref. count on file by 1.
438 * In any case returned fd MAY BE not valid!
439 * This race condition is unavoidable
440 * with shared fd spaces, we cannot solve it inside kernel,
441 * but we take care of internal coherence yet.
445 * sock_alloc_file - Bind a &socket to a &file
447 * @flags: file status flags
448 * @dname: protocol name
450 * Returns the &file bound with @sock, implicitly storing it
451 * in sock->file. If dname is %NULL, sets to "".
452 * On failure the return is a ERR pointer (see linux/err.h).
453 * This function uses GFP_KERNEL internally.
456 struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
461 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
463 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
464 O_RDWR | (flags & O_NONBLOCK),
472 file->private_data = sock;
473 stream_open(SOCK_INODE(sock), file);
476 EXPORT_SYMBOL(sock_alloc_file);
478 static int sock_map_fd(struct socket *sock, int flags)
480 struct file *newfile;
481 int fd = get_unused_fd_flags(flags);
482 if (unlikely(fd < 0)) {
487 newfile = sock_alloc_file(sock, flags, NULL);
488 if (!IS_ERR(newfile)) {
489 fd_install(fd, newfile);
494 return PTR_ERR(newfile);
498 * sock_from_file - Return the &socket bounded to @file.
501 * On failure returns %NULL.
504 struct socket *sock_from_file(struct file *file)
506 if (file->f_op == &socket_file_ops)
507 return file->private_data; /* set in sock_alloc_file */
511 EXPORT_SYMBOL(sock_from_file);
514 * sockfd_lookup - Go from a file number to its socket slot
516 * @err: pointer to an error code return
518 * The file handle passed in is locked and the socket it is bound
519 * to is returned. If an error occurs the err pointer is overwritten
520 * with a negative errno code and NULL is returned. The function checks
521 * for both invalid handles and passing a handle which is not a socket.
523 * On a success the socket object pointer is returned.
526 struct socket *sockfd_lookup(int fd, int *err)
537 sock = sock_from_file(file);
544 EXPORT_SYMBOL(sockfd_lookup);
546 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
548 struct fd f = fdget(fd);
553 sock = sock_from_file(f.file);
555 *fput_needed = f.flags & FDPUT_FPUT;
564 static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
570 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
580 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
585 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
592 static int sockfs_setattr(struct user_namespace *mnt_userns,
593 struct dentry *dentry, struct iattr *iattr)
595 int err = simple_setattr(&init_user_ns, dentry, iattr);
597 if (!err && (iattr->ia_valid & ATTR_UID)) {
598 struct socket *sock = SOCKET_I(d_inode(dentry));
601 sock->sk->sk_uid = iattr->ia_uid;
609 static const struct inode_operations sockfs_inode_ops = {
610 .listxattr = sockfs_listxattr,
611 .setattr = sockfs_setattr,
615 * sock_alloc - allocate a socket
617 * Allocate a new inode and socket object. The two are bound together
618 * and initialised. The socket is then returned. If we are out of inodes
619 * NULL is returned. This functions uses GFP_KERNEL internally.
622 struct socket *sock_alloc(void)
627 inode = new_inode_pseudo(sock_mnt->mnt_sb);
631 sock = SOCKET_I(inode);
633 inode->i_ino = get_next_ino();
634 inode->i_mode = S_IFSOCK | S_IRWXUGO;
635 inode->i_uid = current_fsuid();
636 inode->i_gid = current_fsgid();
637 inode->i_op = &sockfs_inode_ops;
641 EXPORT_SYMBOL(sock_alloc);
643 static void __sock_release(struct socket *sock, struct inode *inode)
646 struct module *owner = sock->ops->owner;
650 sock->ops->release(sock);
658 if (sock->wq.fasync_list)
659 pr_err("%s: fasync list not empty!\n", __func__);
662 iput(SOCK_INODE(sock));
669 * sock_release - close a socket
670 * @sock: socket to close
672 * The socket is released from the protocol stack if it has a release
673 * callback, and the inode is then released if the socket is bound to
674 * an inode not a file.
676 void sock_release(struct socket *sock)
678 __sock_release(sock, NULL);
680 EXPORT_SYMBOL(sock_release);
682 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
684 u8 flags = *tx_flags;
686 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
687 flags |= SKBTX_HW_TSTAMP;
689 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
690 flags |= SKBTX_SW_TSTAMP;
692 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
693 flags |= SKBTX_SCHED_TSTAMP;
697 EXPORT_SYMBOL(__sock_tx_timestamp);
699 INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
701 INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
703 static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
705 int ret = INDIRECT_CALL_INET(sock->ops->sendmsg, inet6_sendmsg,
706 inet_sendmsg, sock, msg,
708 BUG_ON(ret == -EIOCBQUEUED);
713 * sock_sendmsg - send a message through @sock
715 * @msg: message to send
717 * Sends @msg through @sock, passing through LSM.
718 * Returns the number of bytes sent, or an error code.
720 int sock_sendmsg(struct socket *sock, struct msghdr *msg)
722 int err = security_socket_sendmsg(sock, msg,
725 return err ?: sock_sendmsg_nosec(sock, msg);
727 EXPORT_SYMBOL(sock_sendmsg);
730 * kernel_sendmsg - send a message through @sock (kernel-space)
732 * @msg: message header
734 * @num: vec array length
735 * @size: total message data size
737 * Builds the message data with @vec and sends it through @sock.
738 * Returns the number of bytes sent, or an error code.
741 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
742 struct kvec *vec, size_t num, size_t size)
744 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
745 return sock_sendmsg(sock, msg);
747 EXPORT_SYMBOL(kernel_sendmsg);
750 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
752 * @msg: message header
753 * @vec: output s/g array
754 * @num: output s/g array length
755 * @size: total message data size
757 * Builds the message data with @vec and sends it through @sock.
758 * Returns the number of bytes sent, or an error code.
759 * Caller must hold @sk.
762 int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
763 struct kvec *vec, size_t num, size_t size)
765 struct socket *sock = sk->sk_socket;
767 if (!sock->ops->sendmsg_locked)
768 return sock_no_sendmsg_locked(sk, msg, size);
770 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
772 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
774 EXPORT_SYMBOL(kernel_sendmsg_locked);
776 static bool skb_is_err_queue(const struct sk_buff *skb)
778 /* pkt_type of skbs enqueued on the error queue are set to
779 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
780 * in recvmsg, since skbs received on a local socket will never
781 * have a pkt_type of PACKET_OUTGOING.
783 return skb->pkt_type == PACKET_OUTGOING;
786 /* On transmit, software and hardware timestamps are returned independently.
787 * As the two skb clones share the hardware timestamp, which may be updated
788 * before the software timestamp is received, a hardware TX timestamp may be
789 * returned only if there is no software TX timestamp. Ignore false software
790 * timestamps, which may be made in the __sock_recv_timestamp() call when the
791 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
792 * hardware timestamp.
794 static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
796 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
799 static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
801 struct scm_ts_pktinfo ts_pktinfo;
802 struct net_device *orig_dev;
804 if (!skb_mac_header_was_set(skb))
807 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
810 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
812 ts_pktinfo.if_index = orig_dev->ifindex;
815 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
816 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
817 sizeof(ts_pktinfo), &ts_pktinfo);
821 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
823 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
826 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
827 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
828 struct scm_timestamping_internal tss;
830 int empty = 1, false_tstamp = 0;
831 struct skb_shared_hwtstamps *shhwtstamps =
835 /* Race occurred between timestamp enabling and packet
836 receiving. Fill in the current time for now. */
837 if (need_software_tstamp && skb->tstamp == 0) {
838 __net_timestamp(skb);
842 if (need_software_tstamp) {
843 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
845 struct __kernel_sock_timeval tv;
847 skb_get_new_timestamp(skb, &tv);
848 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
851 struct __kernel_old_timeval tv;
853 skb_get_timestamp(skb, &tv);
854 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
859 struct __kernel_timespec ts;
861 skb_get_new_timestampns(skb, &ts);
862 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
865 struct __kernel_old_timespec ts;
867 skb_get_timestampns(skb, &ts);
868 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
874 memset(&tss, 0, sizeof(tss));
875 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
876 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
879 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
880 !skb_is_swtx_tstamp(skb, false_tstamp)) {
881 if (sk->sk_tsflags & SOF_TIMESTAMPING_BIND_PHC)
882 hwtstamp = ptp_convert_timestamp(shhwtstamps,
885 hwtstamp = shhwtstamps->hwtstamp;
887 if (ktime_to_timespec64_cond(hwtstamp, tss.ts + 2)) {
890 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
891 !skb_is_err_queue(skb))
892 put_ts_pktinfo(msg, skb);
896 if (sock_flag(sk, SOCK_TSTAMP_NEW))
897 put_cmsg_scm_timestamping64(msg, &tss);
899 put_cmsg_scm_timestamping(msg, &tss);
901 if (skb_is_err_queue(skb) && skb->len &&
902 SKB_EXT_ERR(skb)->opt_stats)
903 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
904 skb->len, skb->data);
907 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
909 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
914 if (!sock_flag(sk, SOCK_WIFI_STATUS))
916 if (!skb->wifi_acked_valid)
919 ack = skb->wifi_acked;
921 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
923 EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
925 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
928 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
929 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
930 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
933 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
936 sock_recv_timestamp(msg, sk, skb);
937 sock_recv_drops(msg, sk, skb);
939 EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
941 INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
943 INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
945 static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
948 return INDIRECT_CALL_INET(sock->ops->recvmsg, inet6_recvmsg,
949 inet_recvmsg, sock, msg, msg_data_left(msg),
954 * sock_recvmsg - receive a message from @sock
956 * @msg: message to receive
957 * @flags: message flags
959 * Receives @msg from @sock, passing through LSM. Returns the total number
960 * of bytes received, or an error.
962 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
964 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
966 return err ?: sock_recvmsg_nosec(sock, msg, flags);
968 EXPORT_SYMBOL(sock_recvmsg);
971 * kernel_recvmsg - Receive a message from a socket (kernel space)
972 * @sock: The socket to receive the message from
973 * @msg: Received message
974 * @vec: Input s/g array for message data
975 * @num: Size of input s/g array
976 * @size: Number of bytes to read
977 * @flags: Message flags (MSG_DONTWAIT, etc...)
979 * On return the msg structure contains the scatter/gather array passed in the
980 * vec argument. The array is modified so that it consists of the unfilled
981 * portion of the original array.
983 * The returned value is the total number of bytes received, or an error.
986 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
987 struct kvec *vec, size_t num, size_t size, int flags)
989 msg->msg_control_is_user = false;
990 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
991 return sock_recvmsg(sock, msg, flags);
993 EXPORT_SYMBOL(kernel_recvmsg);
995 static ssize_t sock_sendpage(struct file *file, struct page *page,
996 int offset, size_t size, loff_t *ppos, int more)
1001 sock = file->private_data;
1003 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
1004 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
1007 return kernel_sendpage(sock, page, offset, size, flags);
1010 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1011 struct pipe_inode_info *pipe, size_t len,
1014 struct socket *sock = file->private_data;
1016 if (unlikely(!sock->ops->splice_read))
1017 return generic_file_splice_read(file, ppos, pipe, len, flags);
1019 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
1022 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1024 struct file *file = iocb->ki_filp;
1025 struct socket *sock = file->private_data;
1026 struct msghdr msg = {.msg_iter = *to,
1030 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1031 msg.msg_flags = MSG_DONTWAIT;
1033 if (iocb->ki_pos != 0)
1036 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1039 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1044 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1046 struct file *file = iocb->ki_filp;
1047 struct socket *sock = file->private_data;
1048 struct msghdr msg = {.msg_iter = *from,
1052 if (iocb->ki_pos != 0)
1055 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1056 msg.msg_flags = MSG_DONTWAIT;
1058 if (sock->type == SOCK_SEQPACKET)
1059 msg.msg_flags |= MSG_EOR;
1061 res = sock_sendmsg(sock, &msg);
1062 *from = msg.msg_iter;
1067 * Atomic setting of ioctl hooks to avoid race
1068 * with module unload.
1071 static DEFINE_MUTEX(br_ioctl_mutex);
1072 static int (*br_ioctl_hook)(struct net *net, struct net_bridge *br,
1073 unsigned int cmd, struct ifreq *ifr,
1076 void brioctl_set(int (*hook)(struct net *net, struct net_bridge *br,
1077 unsigned int cmd, struct ifreq *ifr,
1080 mutex_lock(&br_ioctl_mutex);
1081 br_ioctl_hook = hook;
1082 mutex_unlock(&br_ioctl_mutex);
1084 EXPORT_SYMBOL(brioctl_set);
1086 int br_ioctl_call(struct net *net, struct net_bridge *br, unsigned int cmd,
1087 struct ifreq *ifr, void __user *uarg)
1092 request_module("bridge");
1094 mutex_lock(&br_ioctl_mutex);
1096 err = br_ioctl_hook(net, br, cmd, ifr, uarg);
1097 mutex_unlock(&br_ioctl_mutex);
1102 static DEFINE_MUTEX(vlan_ioctl_mutex);
1103 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1105 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1107 mutex_lock(&vlan_ioctl_mutex);
1108 vlan_ioctl_hook = hook;
1109 mutex_unlock(&vlan_ioctl_mutex);
1111 EXPORT_SYMBOL(vlan_ioctl_set);
1113 static long sock_do_ioctl(struct net *net, struct socket *sock,
1114 unsigned int cmd, unsigned long arg)
1119 void __user *argp = (void __user *)arg;
1122 err = sock->ops->ioctl(sock, cmd, arg);
1125 * If this ioctl is unknown try to hand it down
1126 * to the NIC driver.
1128 if (err != -ENOIOCTLCMD)
1131 if (!is_socket_ioctl_cmd(cmd))
1134 if (get_user_ifreq(&ifr, &data, argp))
1136 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1137 if (!err && need_copyout)
1138 if (put_user_ifreq(&ifr, argp))
1145 * With an ioctl, arg may well be a user mode pointer, but we don't know
1146 * what to do with it - that's up to the protocol still.
1149 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1151 struct socket *sock;
1153 void __user *argp = (void __user *)arg;
1157 sock = file->private_data;
1160 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1164 if (get_user_ifreq(&ifr, &data, argp))
1166 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1167 if (!err && need_copyout)
1168 if (put_user_ifreq(&ifr, argp))
1171 #ifdef CONFIG_WEXT_CORE
1172 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1173 err = wext_handle_ioctl(net, cmd, argp);
1180 if (get_user(pid, (int __user *)argp))
1182 err = f_setown(sock->file, pid, 1);
1186 err = put_user(f_getown(sock->file),
1187 (int __user *)argp);
1193 err = br_ioctl_call(net, NULL, cmd, NULL, argp);
1198 if (!vlan_ioctl_hook)
1199 request_module("8021q");
1201 mutex_lock(&vlan_ioctl_mutex);
1202 if (vlan_ioctl_hook)
1203 err = vlan_ioctl_hook(net, argp);
1204 mutex_unlock(&vlan_ioctl_mutex);
1208 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1211 err = open_related_ns(&net->ns, get_net_ns);
1213 case SIOCGSTAMP_OLD:
1214 case SIOCGSTAMPNS_OLD:
1215 if (!sock->ops->gettstamp) {
1219 err = sock->ops->gettstamp(sock, argp,
1220 cmd == SIOCGSTAMP_OLD,
1221 !IS_ENABLED(CONFIG_64BIT));
1223 case SIOCGSTAMP_NEW:
1224 case SIOCGSTAMPNS_NEW:
1225 if (!sock->ops->gettstamp) {
1229 err = sock->ops->gettstamp(sock, argp,
1230 cmd == SIOCGSTAMP_NEW,
1235 err = dev_ifconf(net, argp);
1239 err = sock_do_ioctl(net, sock, cmd, arg);
1246 * sock_create_lite - creates a socket
1247 * @family: protocol family (AF_INET, ...)
1248 * @type: communication type (SOCK_STREAM, ...)
1249 * @protocol: protocol (0, ...)
1252 * Creates a new socket and assigns it to @res, passing through LSM.
1253 * The new socket initialization is not complete, see kernel_accept().
1254 * Returns 0 or an error. On failure @res is set to %NULL.
1255 * This function internally uses GFP_KERNEL.
1258 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1261 struct socket *sock = NULL;
1263 err = security_socket_create(family, type, protocol, 1);
1267 sock = sock_alloc();
1274 err = security_socket_post_create(sock, family, type, protocol, 1);
1286 EXPORT_SYMBOL(sock_create_lite);
1288 /* No kernel lock held - perfect */
1289 static __poll_t sock_poll(struct file *file, poll_table *wait)
1291 struct socket *sock = file->private_data;
1292 __poll_t events = poll_requested_events(wait), flag = 0;
1294 if (!sock->ops->poll)
1297 if (sk_can_busy_loop(sock->sk)) {
1298 /* poll once if requested by the syscall */
1299 if (events & POLL_BUSY_LOOP)
1300 sk_busy_loop(sock->sk, 1);
1302 /* if this socket can poll_ll, tell the system call */
1303 flag = POLL_BUSY_LOOP;
1306 return sock->ops->poll(file, sock, wait) | flag;
1309 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1311 struct socket *sock = file->private_data;
1313 return sock->ops->mmap(file, sock, vma);
1316 static int sock_close(struct inode *inode, struct file *filp)
1318 __sock_release(SOCKET_I(inode), inode);
1323 * Update the socket async list
1325 * Fasync_list locking strategy.
1327 * 1. fasync_list is modified only under process context socket lock
1328 * i.e. under semaphore.
1329 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1330 * or under socket lock
1333 static int sock_fasync(int fd, struct file *filp, int on)
1335 struct socket *sock = filp->private_data;
1336 struct sock *sk = sock->sk;
1337 struct socket_wq *wq = &sock->wq;
1343 fasync_helper(fd, filp, on, &wq->fasync_list);
1345 if (!wq->fasync_list)
1346 sock_reset_flag(sk, SOCK_FASYNC);
1348 sock_set_flag(sk, SOCK_FASYNC);
1354 /* This function may be called only under rcu_lock */
1356 int sock_wake_async(struct socket_wq *wq, int how, int band)
1358 if (!wq || !wq->fasync_list)
1362 case SOCK_WAKE_WAITD:
1363 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1366 case SOCK_WAKE_SPACE:
1367 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1372 kill_fasync(&wq->fasync_list, SIGIO, band);
1375 kill_fasync(&wq->fasync_list, SIGURG, band);
1380 EXPORT_SYMBOL(sock_wake_async);
1383 * __sock_create - creates a socket
1384 * @net: net namespace
1385 * @family: protocol family (AF_INET, ...)
1386 * @type: communication type (SOCK_STREAM, ...)
1387 * @protocol: protocol (0, ...)
1389 * @kern: boolean for kernel space sockets
1391 * Creates a new socket and assigns it to @res, passing through LSM.
1392 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1393 * be set to true if the socket resides in kernel space.
1394 * This function internally uses GFP_KERNEL.
1397 int __sock_create(struct net *net, int family, int type, int protocol,
1398 struct socket **res, int kern)
1401 struct socket *sock;
1402 const struct net_proto_family *pf;
1405 * Check protocol is in range
1407 if (family < 0 || family >= NPROTO)
1408 return -EAFNOSUPPORT;
1409 if (type < 0 || type >= SOCK_MAX)
1414 This uglymoron is moved from INET layer to here to avoid
1415 deadlock in module load.
1417 if (family == PF_INET && type == SOCK_PACKET) {
1418 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1423 err = security_socket_create(family, type, protocol, kern);
1428 * Allocate the socket and allow the family to set things up. if
1429 * the protocol is 0, the family is instructed to select an appropriate
1432 sock = sock_alloc();
1434 net_warn_ratelimited("socket: no more sockets\n");
1435 return -ENFILE; /* Not exactly a match, but its the
1436 closest posix thing */
1441 #ifdef CONFIG_MODULES
1442 /* Attempt to load a protocol module if the find failed.
1444 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1445 * requested real, full-featured networking support upon configuration.
1446 * Otherwise module support will break!
1448 if (rcu_access_pointer(net_families[family]) == NULL)
1449 request_module("net-pf-%d", family);
1453 pf = rcu_dereference(net_families[family]);
1454 err = -EAFNOSUPPORT;
1459 * We will call the ->create function, that possibly is in a loadable
1460 * module, so we have to bump that loadable module refcnt first.
1462 if (!try_module_get(pf->owner))
1465 /* Now protected by module ref count */
1468 err = pf->create(net, sock, protocol, kern);
1470 goto out_module_put;
1473 * Now to bump the refcnt of the [loadable] module that owns this
1474 * socket at sock_release time we decrement its refcnt.
1476 if (!try_module_get(sock->ops->owner))
1477 goto out_module_busy;
1480 * Now that we're done with the ->create function, the [loadable]
1481 * module can have its refcnt decremented
1483 module_put(pf->owner);
1484 err = security_socket_post_create(sock, family, type, protocol, kern);
1486 goto out_sock_release;
1492 err = -EAFNOSUPPORT;
1495 module_put(pf->owner);
1502 goto out_sock_release;
1504 EXPORT_SYMBOL(__sock_create);
1507 * sock_create - creates a socket
1508 * @family: protocol family (AF_INET, ...)
1509 * @type: communication type (SOCK_STREAM, ...)
1510 * @protocol: protocol (0, ...)
1513 * A wrapper around __sock_create().
1514 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1517 int sock_create(int family, int type, int protocol, struct socket **res)
1519 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1521 EXPORT_SYMBOL(sock_create);
1524 * sock_create_kern - creates a socket (kernel space)
1525 * @net: net namespace
1526 * @family: protocol family (AF_INET, ...)
1527 * @type: communication type (SOCK_STREAM, ...)
1528 * @protocol: protocol (0, ...)
1531 * A wrapper around __sock_create().
1532 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1535 int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1537 return __sock_create(net, family, type, protocol, res, 1);
1539 EXPORT_SYMBOL(sock_create_kern);
1541 static struct socket *__sys_socket_create(int family, int type, int protocol)
1543 struct socket *sock;
1546 /* Check the SOCK_* constants for consistency. */
1547 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1548 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1549 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1550 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1552 if ((type & ~SOCK_TYPE_MASK) & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1553 return ERR_PTR(-EINVAL);
1554 type &= SOCK_TYPE_MASK;
1556 retval = sock_create(family, type, protocol, &sock);
1558 return ERR_PTR(retval);
1563 struct file *__sys_socket_file(int family, int type, int protocol)
1565 struct socket *sock;
1569 sock = __sys_socket_create(family, type, protocol);
1571 return ERR_CAST(sock);
1573 flags = type & ~SOCK_TYPE_MASK;
1574 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1575 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1577 file = sock_alloc_file(sock, flags, NULL);
1584 int __sys_socket(int family, int type, int protocol)
1586 struct socket *sock;
1589 sock = __sys_socket_create(family, type, protocol);
1591 return PTR_ERR(sock);
1593 flags = type & ~SOCK_TYPE_MASK;
1594 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1595 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1597 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1600 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1602 return __sys_socket(family, type, protocol);
1606 * Create a pair of connected sockets.
1609 int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1611 struct socket *sock1, *sock2;
1613 struct file *newfile1, *newfile2;
1616 flags = type & ~SOCK_TYPE_MASK;
1617 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1619 type &= SOCK_TYPE_MASK;
1621 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1622 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1625 * reserve descriptors and make sure we won't fail
1626 * to return them to userland.
1628 fd1 = get_unused_fd_flags(flags);
1629 if (unlikely(fd1 < 0))
1632 fd2 = get_unused_fd_flags(flags);
1633 if (unlikely(fd2 < 0)) {
1638 err = put_user(fd1, &usockvec[0]);
1642 err = put_user(fd2, &usockvec[1]);
1647 * Obtain the first socket and check if the underlying protocol
1648 * supports the socketpair call.
1651 err = sock_create(family, type, protocol, &sock1);
1652 if (unlikely(err < 0))
1655 err = sock_create(family, type, protocol, &sock2);
1656 if (unlikely(err < 0)) {
1657 sock_release(sock1);
1661 err = security_socket_socketpair(sock1, sock2);
1662 if (unlikely(err)) {
1663 sock_release(sock2);
1664 sock_release(sock1);
1668 err = sock1->ops->socketpair(sock1, sock2);
1669 if (unlikely(err < 0)) {
1670 sock_release(sock2);
1671 sock_release(sock1);
1675 newfile1 = sock_alloc_file(sock1, flags, NULL);
1676 if (IS_ERR(newfile1)) {
1677 err = PTR_ERR(newfile1);
1678 sock_release(sock2);
1682 newfile2 = sock_alloc_file(sock2, flags, NULL);
1683 if (IS_ERR(newfile2)) {
1684 err = PTR_ERR(newfile2);
1689 audit_fd_pair(fd1, fd2);
1691 fd_install(fd1, newfile1);
1692 fd_install(fd2, newfile2);
1701 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1702 int __user *, usockvec)
1704 return __sys_socketpair(family, type, protocol, usockvec);
1708 * Bind a name to a socket. Nothing much to do here since it's
1709 * the protocol's responsibility to handle the local address.
1711 * We move the socket address to kernel space before we call
1712 * the protocol layer (having also checked the address is ok).
1715 int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1717 struct socket *sock;
1718 struct sockaddr_storage address;
1719 int err, fput_needed;
1721 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1723 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1725 err = security_socket_bind(sock,
1726 (struct sockaddr *)&address,
1729 err = sock->ops->bind(sock,
1733 fput_light(sock->file, fput_needed);
1738 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1740 return __sys_bind(fd, umyaddr, addrlen);
1744 * Perform a listen. Basically, we allow the protocol to do anything
1745 * necessary for a listen, and if that works, we mark the socket as
1746 * ready for listening.
1749 int __sys_listen(int fd, int backlog)
1751 struct socket *sock;
1752 int err, fput_needed;
1755 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1757 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1758 if ((unsigned int)backlog > somaxconn)
1759 backlog = somaxconn;
1761 err = security_socket_listen(sock, backlog);
1763 err = sock->ops->listen(sock, backlog);
1765 fput_light(sock->file, fput_needed);
1770 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1772 return __sys_listen(fd, backlog);
1775 struct file *do_accept(struct file *file, unsigned file_flags,
1776 struct sockaddr __user *upeer_sockaddr,
1777 int __user *upeer_addrlen, int flags)
1779 struct socket *sock, *newsock;
1780 struct file *newfile;
1782 struct sockaddr_storage address;
1784 sock = sock_from_file(file);
1786 return ERR_PTR(-ENOTSOCK);
1788 newsock = sock_alloc();
1790 return ERR_PTR(-ENFILE);
1792 newsock->type = sock->type;
1793 newsock->ops = sock->ops;
1796 * We don't need try_module_get here, as the listening socket (sock)
1797 * has the protocol module (sock->ops->owner) held.
1799 __module_get(newsock->ops->owner);
1801 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1802 if (IS_ERR(newfile))
1805 err = security_socket_accept(sock, newsock);
1809 err = sock->ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1814 if (upeer_sockaddr) {
1815 len = newsock->ops->getname(newsock,
1816 (struct sockaddr *)&address, 2);
1818 err = -ECONNABORTED;
1821 err = move_addr_to_user(&address,
1822 len, upeer_sockaddr, upeer_addrlen);
1827 /* File flags are not inherited via accept() unlike another OSes. */
1831 return ERR_PTR(err);
1834 int __sys_accept4_file(struct file *file, unsigned file_flags,
1835 struct sockaddr __user *upeer_sockaddr,
1836 int __user *upeer_addrlen, int flags,
1837 unsigned long nofile)
1839 struct file *newfile;
1842 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1845 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1846 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1848 newfd = __get_unused_fd_flags(flags, nofile);
1849 if (unlikely(newfd < 0))
1852 newfile = do_accept(file, file_flags, upeer_sockaddr, upeer_addrlen,
1854 if (IS_ERR(newfile)) {
1855 put_unused_fd(newfd);
1856 return PTR_ERR(newfile);
1858 fd_install(newfd, newfile);
1863 * For accept, we attempt to create a new socket, set up the link
1864 * with the client, wake up the client, then return the new
1865 * connected fd. We collect the address of the connector in kernel
1866 * space and move it to user at the very end. This is unclean because
1867 * we open the socket then return an error.
1869 * 1003.1g adds the ability to recvmsg() to query connection pending
1870 * status to recvmsg. We need to add that support in a way thats
1871 * clean when we restructure accept also.
1874 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1875 int __user *upeer_addrlen, int flags)
1882 ret = __sys_accept4_file(f.file, 0, upeer_sockaddr,
1883 upeer_addrlen, flags,
1884 rlimit(RLIMIT_NOFILE));
1891 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1892 int __user *, upeer_addrlen, int, flags)
1894 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1897 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1898 int __user *, upeer_addrlen)
1900 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1904 * Attempt to connect to a socket with the server address. The address
1905 * is in user space so we verify it is OK and move it to kernel space.
1907 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1910 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1911 * other SEQPACKET protocols that take time to connect() as it doesn't
1912 * include the -EINPROGRESS status for such sockets.
1915 int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
1916 int addrlen, int file_flags)
1918 struct socket *sock;
1921 sock = sock_from_file(file);
1928 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1932 err = sock->ops->connect(sock, (struct sockaddr *)address, addrlen,
1933 sock->file->f_flags | file_flags);
1938 int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1945 struct sockaddr_storage address;
1947 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
1949 ret = __sys_connect_file(f.file, &address, addrlen, 0);
1956 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1959 return __sys_connect(fd, uservaddr, addrlen);
1963 * Get the local address ('name') of a socket object. Move the obtained
1964 * name to user space.
1967 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1968 int __user *usockaddr_len)
1970 struct socket *sock;
1971 struct sockaddr_storage address;
1972 int err, fput_needed;
1974 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1978 err = security_socket_getsockname(sock);
1982 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1985 /* "err" is actually length in this case */
1986 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1989 fput_light(sock->file, fput_needed);
1994 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1995 int __user *, usockaddr_len)
1997 return __sys_getsockname(fd, usockaddr, usockaddr_len);
2001 * Get the remote address ('name') of a socket object. Move the obtained
2002 * name to user space.
2005 int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
2006 int __user *usockaddr_len)
2008 struct socket *sock;
2009 struct sockaddr_storage address;
2010 int err, fput_needed;
2012 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2014 err = security_socket_getpeername(sock);
2016 fput_light(sock->file, fput_needed);
2020 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
2022 /* "err" is actually length in this case */
2023 err = move_addr_to_user(&address, err, usockaddr,
2025 fput_light(sock->file, fput_needed);
2030 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
2031 int __user *, usockaddr_len)
2033 return __sys_getpeername(fd, usockaddr, usockaddr_len);
2037 * Send a datagram to a given address. We move the address into kernel
2038 * space and check the user space data area is readable before invoking
2041 int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
2042 struct sockaddr __user *addr, int addr_len)
2044 struct socket *sock;
2045 struct sockaddr_storage address;
2051 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
2054 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2058 msg.msg_name = NULL;
2059 msg.msg_control = NULL;
2060 msg.msg_controllen = 0;
2061 msg.msg_namelen = 0;
2063 err = move_addr_to_kernel(addr, addr_len, &address);
2066 msg.msg_name = (struct sockaddr *)&address;
2067 msg.msg_namelen = addr_len;
2069 if (sock->file->f_flags & O_NONBLOCK)
2070 flags |= MSG_DONTWAIT;
2071 msg.msg_flags = flags;
2072 err = sock_sendmsg(sock, &msg);
2075 fput_light(sock->file, fput_needed);
2080 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2081 unsigned int, flags, struct sockaddr __user *, addr,
2084 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2088 * Send a datagram down a socket.
2091 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2092 unsigned int, flags)
2094 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2098 * Receive a frame from the socket and optionally record the address of the
2099 * sender. We verify the buffers are writable and if needed move the
2100 * sender address from kernel to user space.
2102 int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2103 struct sockaddr __user *addr, int __user *addr_len)
2105 struct socket *sock;
2108 struct sockaddr_storage address;
2112 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
2115 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2119 msg.msg_control = NULL;
2120 msg.msg_controllen = 0;
2121 /* Save some cycles and don't copy the address if not needed */
2122 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
2123 /* We assume all kernel code knows the size of sockaddr_storage */
2124 msg.msg_namelen = 0;
2125 msg.msg_iocb = NULL;
2127 if (sock->file->f_flags & O_NONBLOCK)
2128 flags |= MSG_DONTWAIT;
2129 err = sock_recvmsg(sock, &msg, flags);
2131 if (err >= 0 && addr != NULL) {
2132 err2 = move_addr_to_user(&address,
2133 msg.msg_namelen, addr, addr_len);
2138 fput_light(sock->file, fput_needed);
2143 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2144 unsigned int, flags, struct sockaddr __user *, addr,
2145 int __user *, addr_len)
2147 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2151 * Receive a datagram from a socket.
2154 SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2155 unsigned int, flags)
2157 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2160 static bool sock_use_custom_sol_socket(const struct socket *sock)
2162 const struct sock *sk = sock->sk;
2164 /* Use sock->ops->setsockopt() for MPTCP */
2165 return IS_ENABLED(CONFIG_MPTCP) &&
2166 sk->sk_protocol == IPPROTO_MPTCP &&
2167 sk->sk_type == SOCK_STREAM &&
2168 (sk->sk_family == AF_INET || sk->sk_family == AF_INET6);
2172 * Set a socket option. Because we don't know the option lengths we have
2173 * to pass the user mode parameter for the protocols to sort out.
2175 int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2178 sockptr_t optval = USER_SOCKPTR(user_optval);
2179 char *kernel_optval = NULL;
2180 int err, fput_needed;
2181 struct socket *sock;
2186 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2190 err = security_socket_setsockopt(sock, level, optname);
2194 if (!in_compat_syscall())
2195 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2196 user_optval, &optlen,
2206 optval = KERNEL_SOCKPTR(kernel_optval);
2207 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2208 err = sock_setsockopt(sock, level, optname, optval, optlen);
2209 else if (unlikely(!sock->ops->setsockopt))
2212 err = sock->ops->setsockopt(sock, level, optname, optval,
2214 kfree(kernel_optval);
2216 fput_light(sock->file, fput_needed);
2220 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2221 char __user *, optval, int, optlen)
2223 return __sys_setsockopt(fd, level, optname, optval, optlen);
2226 INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2230 * Get a socket option. Because we don't know the option lengths we have
2231 * to pass a user mode parameter for the protocols to sort out.
2233 int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2236 int err, fput_needed;
2237 struct socket *sock;
2240 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2244 err = security_socket_getsockopt(sock, level, optname);
2248 if (!in_compat_syscall())
2249 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2251 if (level == SOL_SOCKET)
2252 err = sock_getsockopt(sock, level, optname, optval, optlen);
2253 else if (unlikely(!sock->ops->getsockopt))
2256 err = sock->ops->getsockopt(sock, level, optname, optval,
2259 if (!in_compat_syscall())
2260 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2261 optval, optlen, max_optlen,
2264 fput_light(sock->file, fput_needed);
2268 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2269 char __user *, optval, int __user *, optlen)
2271 return __sys_getsockopt(fd, level, optname, optval, optlen);
2275 * Shutdown a socket.
2278 int __sys_shutdown_sock(struct socket *sock, int how)
2282 err = security_socket_shutdown(sock, how);
2284 err = sock->ops->shutdown(sock, how);
2289 int __sys_shutdown(int fd, int how)
2291 int err, fput_needed;
2292 struct socket *sock;
2294 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2296 err = __sys_shutdown_sock(sock, how);
2297 fput_light(sock->file, fput_needed);
2302 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2304 return __sys_shutdown(fd, how);
2307 /* A couple of helpful macros for getting the address of the 32/64 bit
2308 * fields which are the same type (int / unsigned) on our platforms.
2310 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2311 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2312 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2314 struct used_address {
2315 struct sockaddr_storage name;
2316 unsigned int name_len;
2319 int __copy_msghdr_from_user(struct msghdr *kmsg,
2320 struct user_msghdr __user *umsg,
2321 struct sockaddr __user **save_addr,
2322 struct iovec __user **uiov, size_t *nsegs)
2324 struct user_msghdr msg;
2327 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2330 kmsg->msg_control_is_user = true;
2331 kmsg->msg_control_user = msg.msg_control;
2332 kmsg->msg_controllen = msg.msg_controllen;
2333 kmsg->msg_flags = msg.msg_flags;
2335 kmsg->msg_namelen = msg.msg_namelen;
2337 kmsg->msg_namelen = 0;
2339 if (kmsg->msg_namelen < 0)
2342 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2343 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2346 *save_addr = msg.msg_name;
2348 if (msg.msg_name && kmsg->msg_namelen) {
2350 err = move_addr_to_kernel(msg.msg_name,
2357 kmsg->msg_name = NULL;
2358 kmsg->msg_namelen = 0;
2361 if (msg.msg_iovlen > UIO_MAXIOV)
2364 kmsg->msg_iocb = NULL;
2365 *uiov = msg.msg_iov;
2366 *nsegs = msg.msg_iovlen;
2370 static int copy_msghdr_from_user(struct msghdr *kmsg,
2371 struct user_msghdr __user *umsg,
2372 struct sockaddr __user **save_addr,
2375 struct user_msghdr msg;
2378 err = __copy_msghdr_from_user(kmsg, umsg, save_addr, &msg.msg_iov,
2383 err = import_iovec(save_addr ? READ : WRITE,
2384 msg.msg_iov, msg.msg_iovlen,
2385 UIO_FASTIOV, iov, &kmsg->msg_iter);
2386 return err < 0 ? err : 0;
2389 static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2390 unsigned int flags, struct used_address *used_address,
2391 unsigned int allowed_msghdr_flags)
2393 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2394 __aligned(sizeof(__kernel_size_t));
2395 /* 20 is size of ipv6_pktinfo */
2396 unsigned char *ctl_buf = ctl;
2402 if (msg_sys->msg_controllen > INT_MAX)
2404 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2405 ctl_len = msg_sys->msg_controllen;
2406 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2408 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2412 ctl_buf = msg_sys->msg_control;
2413 ctl_len = msg_sys->msg_controllen;
2414 } else if (ctl_len) {
2415 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2416 CMSG_ALIGN(sizeof(struct cmsghdr)));
2417 if (ctl_len > sizeof(ctl)) {
2418 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2419 if (ctl_buf == NULL)
2423 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2425 msg_sys->msg_control = ctl_buf;
2426 msg_sys->msg_control_is_user = false;
2428 msg_sys->msg_flags = flags;
2430 if (sock->file->f_flags & O_NONBLOCK)
2431 msg_sys->msg_flags |= MSG_DONTWAIT;
2433 * If this is sendmmsg() and current destination address is same as
2434 * previously succeeded address, omit asking LSM's decision.
2435 * used_address->name_len is initialized to UINT_MAX so that the first
2436 * destination address never matches.
2438 if (used_address && msg_sys->msg_name &&
2439 used_address->name_len == msg_sys->msg_namelen &&
2440 !memcmp(&used_address->name, msg_sys->msg_name,
2441 used_address->name_len)) {
2442 err = sock_sendmsg_nosec(sock, msg_sys);
2445 err = sock_sendmsg(sock, msg_sys);
2447 * If this is sendmmsg() and sending to current destination address was
2448 * successful, remember it.
2450 if (used_address && err >= 0) {
2451 used_address->name_len = msg_sys->msg_namelen;
2452 if (msg_sys->msg_name)
2453 memcpy(&used_address->name, msg_sys->msg_name,
2454 used_address->name_len);
2459 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2464 int sendmsg_copy_msghdr(struct msghdr *msg,
2465 struct user_msghdr __user *umsg, unsigned flags,
2470 if (flags & MSG_CMSG_COMPAT) {
2471 struct compat_msghdr __user *msg_compat;
2473 msg_compat = (struct compat_msghdr __user *) umsg;
2474 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2476 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2484 static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2485 struct msghdr *msg_sys, unsigned int flags,
2486 struct used_address *used_address,
2487 unsigned int allowed_msghdr_flags)
2489 struct sockaddr_storage address;
2490 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2493 msg_sys->msg_name = &address;
2495 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2499 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2500 allowed_msghdr_flags);
2506 * BSD sendmsg interface
2508 long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2511 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2514 long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2515 bool forbid_cmsg_compat)
2517 int fput_needed, err;
2518 struct msghdr msg_sys;
2519 struct socket *sock;
2521 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2524 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2528 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2530 fput_light(sock->file, fput_needed);
2535 SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2537 return __sys_sendmsg(fd, msg, flags, true);
2541 * Linux sendmmsg interface
2544 int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2545 unsigned int flags, bool forbid_cmsg_compat)
2547 int fput_needed, err, datagrams;
2548 struct socket *sock;
2549 struct mmsghdr __user *entry;
2550 struct compat_mmsghdr __user *compat_entry;
2551 struct msghdr msg_sys;
2552 struct used_address used_address;
2553 unsigned int oflags = flags;
2555 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2558 if (vlen > UIO_MAXIOV)
2563 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2567 used_address.name_len = UINT_MAX;
2569 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2573 while (datagrams < vlen) {
2574 if (datagrams == vlen - 1)
2577 if (MSG_CMSG_COMPAT & flags) {
2578 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2579 &msg_sys, flags, &used_address, MSG_EOR);
2582 err = __put_user(err, &compat_entry->msg_len);
2585 err = ___sys_sendmsg(sock,
2586 (struct user_msghdr __user *)entry,
2587 &msg_sys, flags, &used_address, MSG_EOR);
2590 err = put_user(err, &entry->msg_len);
2597 if (msg_data_left(&msg_sys))
2602 fput_light(sock->file, fput_needed);
2604 /* We only return an error if no datagrams were able to be sent */
2611 SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2612 unsigned int, vlen, unsigned int, flags)
2614 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2617 int recvmsg_copy_msghdr(struct msghdr *msg,
2618 struct user_msghdr __user *umsg, unsigned flags,
2619 struct sockaddr __user **uaddr,
2624 if (MSG_CMSG_COMPAT & flags) {
2625 struct compat_msghdr __user *msg_compat;
2627 msg_compat = (struct compat_msghdr __user *) umsg;
2628 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2630 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2638 static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2639 struct user_msghdr __user *msg,
2640 struct sockaddr __user *uaddr,
2641 unsigned int flags, int nosec)
2643 struct compat_msghdr __user *msg_compat =
2644 (struct compat_msghdr __user *) msg;
2645 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2646 struct sockaddr_storage addr;
2647 unsigned long cmsg_ptr;
2651 msg_sys->msg_name = &addr;
2652 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2653 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2655 /* We assume all kernel code knows the size of sockaddr_storage */
2656 msg_sys->msg_namelen = 0;
2658 if (sock->file->f_flags & O_NONBLOCK)
2659 flags |= MSG_DONTWAIT;
2661 if (unlikely(nosec))
2662 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2664 err = sock_recvmsg(sock, msg_sys, flags);
2670 if (uaddr != NULL) {
2671 err = move_addr_to_user(&addr,
2672 msg_sys->msg_namelen, uaddr,
2677 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2681 if (MSG_CMSG_COMPAT & flags)
2682 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2683 &msg_compat->msg_controllen);
2685 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2686 &msg->msg_controllen);
2694 static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2695 struct msghdr *msg_sys, unsigned int flags, int nosec)
2697 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2698 /* user mode address pointers */
2699 struct sockaddr __user *uaddr;
2702 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2706 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2712 * BSD recvmsg interface
2715 long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2716 struct user_msghdr __user *umsg,
2717 struct sockaddr __user *uaddr, unsigned int flags)
2719 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2722 long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2723 bool forbid_cmsg_compat)
2725 int fput_needed, err;
2726 struct msghdr msg_sys;
2727 struct socket *sock;
2729 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2732 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2736 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2738 fput_light(sock->file, fput_needed);
2743 SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2744 unsigned int, flags)
2746 return __sys_recvmsg(fd, msg, flags, true);
2750 * Linux recvmmsg interface
2753 static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2754 unsigned int vlen, unsigned int flags,
2755 struct timespec64 *timeout)
2757 int fput_needed, err, datagrams;
2758 struct socket *sock;
2759 struct mmsghdr __user *entry;
2760 struct compat_mmsghdr __user *compat_entry;
2761 struct msghdr msg_sys;
2762 struct timespec64 end_time;
2763 struct timespec64 timeout64;
2766 poll_select_set_timeout(&end_time, timeout->tv_sec,
2772 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2776 if (likely(!(flags & MSG_ERRQUEUE))) {
2777 err = sock_error(sock->sk);
2785 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2787 while (datagrams < vlen) {
2789 * No need to ask LSM for more than the first datagram.
2791 if (MSG_CMSG_COMPAT & flags) {
2792 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2793 &msg_sys, flags & ~MSG_WAITFORONE,
2797 err = __put_user(err, &compat_entry->msg_len);
2800 err = ___sys_recvmsg(sock,
2801 (struct user_msghdr __user *)entry,
2802 &msg_sys, flags & ~MSG_WAITFORONE,
2806 err = put_user(err, &entry->msg_len);
2814 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2815 if (flags & MSG_WAITFORONE)
2816 flags |= MSG_DONTWAIT;
2819 ktime_get_ts64(&timeout64);
2820 *timeout = timespec64_sub(end_time, timeout64);
2821 if (timeout->tv_sec < 0) {
2822 timeout->tv_sec = timeout->tv_nsec = 0;
2826 /* Timeout, return less than vlen datagrams */
2827 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2831 /* Out of band data, return right away */
2832 if (msg_sys.msg_flags & MSG_OOB)
2840 if (datagrams == 0) {
2846 * We may return less entries than requested (vlen) if the
2847 * sock is non block and there aren't enough datagrams...
2849 if (err != -EAGAIN) {
2851 * ... or if recvmsg returns an error after we
2852 * received some datagrams, where we record the
2853 * error to return on the next call or if the
2854 * app asks about it using getsockopt(SO_ERROR).
2856 sock->sk->sk_err = -err;
2859 fput_light(sock->file, fput_needed);
2864 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2865 unsigned int vlen, unsigned int flags,
2866 struct __kernel_timespec __user *timeout,
2867 struct old_timespec32 __user *timeout32)
2870 struct timespec64 timeout_sys;
2872 if (timeout && get_timespec64(&timeout_sys, timeout))
2875 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2878 if (!timeout && !timeout32)
2879 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2881 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2886 if (timeout && put_timespec64(&timeout_sys, timeout))
2887 datagrams = -EFAULT;
2889 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2890 datagrams = -EFAULT;
2895 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2896 unsigned int, vlen, unsigned int, flags,
2897 struct __kernel_timespec __user *, timeout)
2899 if (flags & MSG_CMSG_COMPAT)
2902 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2905 #ifdef CONFIG_COMPAT_32BIT_TIME
2906 SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2907 unsigned int, vlen, unsigned int, flags,
2908 struct old_timespec32 __user *, timeout)
2910 if (flags & MSG_CMSG_COMPAT)
2913 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2917 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2918 /* Argument list sizes for sys_socketcall */
2919 #define AL(x) ((x) * sizeof(unsigned long))
2920 static const unsigned char nargs[21] = {
2921 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2922 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2923 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2930 * System call vectors.
2932 * Argument checking cleaned up. Saved 20% in size.
2933 * This function doesn't need to set the kernel lock because
2934 * it is set by the callees.
2937 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2939 unsigned long a[AUDITSC_ARGS];
2940 unsigned long a0, a1;
2944 if (call < 1 || call > SYS_SENDMMSG)
2946 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2949 if (len > sizeof(a))
2952 /* copy_from_user should be SMP safe. */
2953 if (copy_from_user(a, args, len))
2956 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2965 err = __sys_socket(a0, a1, a[2]);
2968 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2971 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2974 err = __sys_listen(a0, a1);
2977 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2978 (int __user *)a[2], 0);
2980 case SYS_GETSOCKNAME:
2982 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2983 (int __user *)a[2]);
2985 case SYS_GETPEERNAME:
2987 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2988 (int __user *)a[2]);
2990 case SYS_SOCKETPAIR:
2991 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2994 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2998 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2999 (struct sockaddr __user *)a[4], a[5]);
3002 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3006 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3007 (struct sockaddr __user *)a[4],
3008 (int __user *)a[5]);
3011 err = __sys_shutdown(a0, a1);
3013 case SYS_SETSOCKOPT:
3014 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
3017 case SYS_GETSOCKOPT:
3019 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
3020 (int __user *)a[4]);
3023 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
3027 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
3031 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
3035 if (IS_ENABLED(CONFIG_64BIT))
3036 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3038 (struct __kernel_timespec __user *)a[4],
3041 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3043 (struct old_timespec32 __user *)a[4]);
3046 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3047 (int __user *)a[2], a[3]);
3056 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
3059 * sock_register - add a socket protocol handler
3060 * @ops: description of protocol
3062 * This function is called by a protocol handler that wants to
3063 * advertise its address family, and have it linked into the
3064 * socket interface. The value ops->family corresponds to the
3065 * socket system call protocol family.
3067 int sock_register(const struct net_proto_family *ops)
3071 if (ops->family >= NPROTO) {
3072 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3076 spin_lock(&net_family_lock);
3077 if (rcu_dereference_protected(net_families[ops->family],
3078 lockdep_is_held(&net_family_lock)))
3081 rcu_assign_pointer(net_families[ops->family], ops);
3084 spin_unlock(&net_family_lock);
3086 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3089 EXPORT_SYMBOL(sock_register);
3092 * sock_unregister - remove a protocol handler
3093 * @family: protocol family to remove
3095 * This function is called by a protocol handler that wants to
3096 * remove its address family, and have it unlinked from the
3097 * new socket creation.
3099 * If protocol handler is a module, then it can use module reference
3100 * counts to protect against new references. If protocol handler is not
3101 * a module then it needs to provide its own protection in
3102 * the ops->create routine.
3104 void sock_unregister(int family)
3106 BUG_ON(family < 0 || family >= NPROTO);
3108 spin_lock(&net_family_lock);
3109 RCU_INIT_POINTER(net_families[family], NULL);
3110 spin_unlock(&net_family_lock);
3114 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3116 EXPORT_SYMBOL(sock_unregister);
3118 bool sock_is_registered(int family)
3120 return family < NPROTO && rcu_access_pointer(net_families[family]);
3123 static int __init sock_init(void)
3127 * Initialize the network sysctl infrastructure.
3129 err = net_sysctl_init();
3134 * Initialize skbuff SLAB cache
3139 * Initialize the protocols module.
3144 err = register_filesystem(&sock_fs_type);
3147 sock_mnt = kern_mount(&sock_fs_type);
3148 if (IS_ERR(sock_mnt)) {
3149 err = PTR_ERR(sock_mnt);
3153 /* The real protocol initialization is performed in later initcalls.
3156 #ifdef CONFIG_NETFILTER
3157 err = netfilter_init();
3162 ptp_classifier_init();
3168 unregister_filesystem(&sock_fs_type);
3172 core_initcall(sock_init); /* early initcall */
3174 #ifdef CONFIG_PROC_FS
3175 void socket_seq_show(struct seq_file *seq)
3177 seq_printf(seq, "sockets: used %d\n",
3178 sock_inuse_get(seq->private));
3180 #endif /* CONFIG_PROC_FS */
3182 /* Handle the fact that while struct ifreq has the same *layout* on
3183 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3184 * which are handled elsewhere, it still has different *size* due to
3185 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3186 * resulting in struct ifreq being 32 and 40 bytes respectively).
3187 * As a result, if the struct happens to be at the end of a page and
3188 * the next page isn't readable/writable, we get a fault. To prevent
3189 * that, copy back and forth to the full size.
3191 int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3193 if (in_compat_syscall()) {
3194 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3196 memset(ifr, 0, sizeof(*ifr));
3197 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3201 *ifrdata = compat_ptr(ifr32->ifr_data);
3206 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3210 *ifrdata = ifr->ifr_data;
3214 EXPORT_SYMBOL(get_user_ifreq);
3216 int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3218 size_t size = sizeof(*ifr);
3220 if (in_compat_syscall())
3221 size = sizeof(struct compat_ifreq);
3223 if (copy_to_user(arg, ifr, size))
3228 EXPORT_SYMBOL(put_user_ifreq);
3230 #ifdef CONFIG_COMPAT
3231 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3233 compat_uptr_t uptr32;
3238 if (get_user_ifreq(&ifr, NULL, uifr32))
3241 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3244 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3245 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3247 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3249 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3250 if (put_user_ifreq(&ifr, uifr32))
3256 /* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3257 static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3258 struct compat_ifreq __user *u_ifreq32)
3263 if (!is_socket_ioctl_cmd(cmd))
3265 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3267 ifreq.ifr_data = data;
3269 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3272 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3273 unsigned int cmd, unsigned long arg)
3275 void __user *argp = compat_ptr(arg);
3276 struct sock *sk = sock->sk;
3277 struct net *net = sock_net(sk);
3279 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3280 return sock_ioctl(file, cmd, (unsigned long)argp);
3284 return compat_siocwandev(net, argp);
3285 case SIOCGSTAMP_OLD:
3286 case SIOCGSTAMPNS_OLD:
3287 if (!sock->ops->gettstamp)
3288 return -ENOIOCTLCMD;
3289 return sock->ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3290 !COMPAT_USE_64BIT_TIME);
3293 case SIOCBONDSLAVEINFOQUERY:
3294 case SIOCBONDINFOQUERY:
3297 return compat_ifr_data_ioctl(net, cmd, argp);
3308 case SIOCGSTAMP_NEW:
3309 case SIOCGSTAMPNS_NEW:
3313 return sock_ioctl(file, cmd, arg);
3332 case SIOCSIFHWBROADCAST:
3334 case SIOCGIFBRDADDR:
3335 case SIOCSIFBRDADDR:
3336 case SIOCGIFDSTADDR:
3337 case SIOCSIFDSTADDR:
3338 case SIOCGIFNETMASK:
3339 case SIOCSIFNETMASK:
3351 case SIOCBONDENSLAVE:
3352 case SIOCBONDRELEASE:
3353 case SIOCBONDSETHWADDR:
3354 case SIOCBONDCHANGEACTIVE:
3361 return sock_do_ioctl(net, sock, cmd, arg);
3364 return -ENOIOCTLCMD;
3367 static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3370 struct socket *sock = file->private_data;
3371 int ret = -ENOIOCTLCMD;
3378 if (sock->ops->compat_ioctl)
3379 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3381 if (ret == -ENOIOCTLCMD &&
3382 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3383 ret = compat_wext_handle_ioctl(net, cmd, arg);
3385 if (ret == -ENOIOCTLCMD)
3386 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3393 * kernel_bind - bind an address to a socket (kernel space)
3396 * @addrlen: length of address
3398 * Returns 0 or an error.
3401 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3403 return sock->ops->bind(sock, addr, addrlen);
3405 EXPORT_SYMBOL(kernel_bind);
3408 * kernel_listen - move socket to listening state (kernel space)
3410 * @backlog: pending connections queue size
3412 * Returns 0 or an error.
3415 int kernel_listen(struct socket *sock, int backlog)
3417 return sock->ops->listen(sock, backlog);
3419 EXPORT_SYMBOL(kernel_listen);
3422 * kernel_accept - accept a connection (kernel space)
3423 * @sock: listening socket
3424 * @newsock: new connected socket
3427 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3428 * If it fails, @newsock is guaranteed to be %NULL.
3429 * Returns 0 or an error.
3432 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3434 struct sock *sk = sock->sk;
3437 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3442 err = sock->ops->accept(sock, *newsock, flags, true);
3444 sock_release(*newsock);
3449 (*newsock)->ops = sock->ops;
3450 __module_get((*newsock)->ops->owner);
3455 EXPORT_SYMBOL(kernel_accept);
3458 * kernel_connect - connect a socket (kernel space)
3461 * @addrlen: address length
3462 * @flags: flags (O_NONBLOCK, ...)
3464 * For datagram sockets, @addr is the address to which datagrams are sent
3465 * by default, and the only address from which datagrams are received.
3466 * For stream sockets, attempts to connect to @addr.
3467 * Returns 0 or an error code.
3470 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3473 return sock->ops->connect(sock, addr, addrlen, flags);
3475 EXPORT_SYMBOL(kernel_connect);
3478 * kernel_getsockname - get the address which the socket is bound (kernel space)
3480 * @addr: address holder
3482 * Fills the @addr pointer with the address which the socket is bound.
3483 * Returns the length of the address in bytes or an error code.
3486 int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3488 return sock->ops->getname(sock, addr, 0);
3490 EXPORT_SYMBOL(kernel_getsockname);
3493 * kernel_getpeername - get the address which the socket is connected (kernel space)
3495 * @addr: address holder
3497 * Fills the @addr pointer with the address which the socket is connected.
3498 * Returns the length of the address in bytes or an error code.
3501 int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3503 return sock->ops->getname(sock, addr, 1);
3505 EXPORT_SYMBOL(kernel_getpeername);
3508 * kernel_sendpage - send a &page through a socket (kernel space)
3511 * @offset: page offset
3512 * @size: total size in bytes
3513 * @flags: flags (MSG_DONTWAIT, ...)
3515 * Returns the total amount sent in bytes or an error.
3518 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3519 size_t size, int flags)
3521 if (sock->ops->sendpage) {
3522 /* Warn in case the improper page to zero-copy send */
3523 WARN_ONCE(!sendpage_ok(page), "improper page for zero-copy send");
3524 return sock->ops->sendpage(sock, page, offset, size, flags);
3526 return sock_no_sendpage(sock, page, offset, size, flags);
3528 EXPORT_SYMBOL(kernel_sendpage);
3531 * kernel_sendpage_locked - send a &page through the locked sock (kernel space)
3534 * @offset: page offset
3535 * @size: total size in bytes
3536 * @flags: flags (MSG_DONTWAIT, ...)
3538 * Returns the total amount sent in bytes or an error.
3539 * Caller must hold @sk.
3542 int kernel_sendpage_locked(struct sock *sk, struct page *page, int offset,
3543 size_t size, int flags)
3545 struct socket *sock = sk->sk_socket;
3547 if (sock->ops->sendpage_locked)
3548 return sock->ops->sendpage_locked(sk, page, offset, size,
3551 return sock_no_sendpage_locked(sk, page, offset, size, flags);
3553 EXPORT_SYMBOL(kernel_sendpage_locked);
3556 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3558 * @how: connection part
3560 * Returns 0 or an error.
3563 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3565 return sock->ops->shutdown(sock, how);
3567 EXPORT_SYMBOL(kernel_sock_shutdown);
3570 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3573 * This routine returns the IP overhead imposed by a socket i.e.
3574 * the length of the underlying IP header, depending on whether
3575 * this is an IPv4 or IPv6 socket and the length from IP options turned
3576 * on at the socket. Assumes that the caller has a lock on the socket.
3579 u32 kernel_sock_ip_overhead(struct sock *sk)
3581 struct inet_sock *inet;
3582 struct ip_options_rcu *opt;
3584 #if IS_ENABLED(CONFIG_IPV6)
3585 struct ipv6_pinfo *np;
3586 struct ipv6_txoptions *optv6 = NULL;
3587 #endif /* IS_ENABLED(CONFIG_IPV6) */
3592 switch (sk->sk_family) {
3595 overhead += sizeof(struct iphdr);
3596 opt = rcu_dereference_protected(inet->inet_opt,
3597 sock_owned_by_user(sk));
3599 overhead += opt->opt.optlen;
3601 #if IS_ENABLED(CONFIG_IPV6)
3604 overhead += sizeof(struct ipv6hdr);
3606 optv6 = rcu_dereference_protected(np->opt,
3607 sock_owned_by_user(sk));
3609 overhead += (optv6->opt_flen + optv6->opt_nflen);
3611 #endif /* IS_ENABLED(CONFIG_IPV6) */
3612 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3616 EXPORT_SYMBOL(kernel_sock_ip_overhead);