2 * NET An implementation of the SOCKET network access protocol.
4 * Version: @(#)socket.c 1.1.93 18/02/95
11 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
13 * Alan Cox : verify_area() fixes
14 * Alan Cox : Removed DDI
15 * Jonathan Kamens : SOCK_DGRAM reconnect bug
16 * Alan Cox : Moved a load of checks to the very
18 * Alan Cox : Move address structures to/from user
19 * mode above the protocol layers.
20 * Rob Janssen : Allow 0 length sends.
21 * Alan Cox : Asynchronous I/O support (cribbed from the
23 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
24 * Jeff Uphoff : Made max number of sockets command-line
26 * Matti Aarnio : Made the number of sockets dynamic,
27 * to be allocated when needed, and mr.
28 * Uphoff's max is used as max to be
29 * allowed to allocate.
30 * Linus : Argh. removed all the socket allocation
31 * altogether: it's in the inode now.
32 * Alan Cox : Made sock_alloc()/sock_release() public
33 * for NetROM and future kernel nfsd type
35 * Alan Cox : sendmsg/recvmsg basics.
36 * Tom Dyas : Export net symbols.
37 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
38 * Alan Cox : Added thread locking to sys_* calls
39 * for sockets. May have errors at the
41 * Kevin Buhr : Fixed the dumb errors in the above.
42 * Andi Kleen : Some small cleanups, optimizations,
43 * and fixed a copy_from_user() bug.
44 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
45 * Tigran Aivazian : Made listen(2) backlog sanity checks
46 * protocol-independent
49 * This program is free software; you can redistribute it and/or
50 * modify it under the terms of the GNU General Public License
51 * as published by the Free Software Foundation; either version
52 * 2 of the License, or (at your option) any later version.
55 * This module is effectively the top level interface to the BSD socket
58 * Based upon Swansea University Computer Society NET3.039
62 #include <linux/socket.h>
63 #include <linux/file.h>
64 #include <linux/net.h>
65 #include <linux/interrupt.h>
66 #include <linux/thread_info.h>
67 #include <linux/rcupdate.h>
68 #include <linux/netdevice.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/mutex.h>
72 #include <linux/if_bridge.h>
73 #include <linux/if_frad.h>
74 #include <linux/if_vlan.h>
75 #include <linux/ptp_classify.h>
76 #include <linux/init.h>
77 #include <linux/poll.h>
78 #include <linux/cache.h>
79 #include <linux/module.h>
80 #include <linux/highmem.h>
81 #include <linux/mount.h>
82 #include <linux/security.h>
83 #include <linux/syscalls.h>
84 #include <linux/compat.h>
85 #include <linux/kmod.h>
86 #include <linux/audit.h>
87 #include <linux/wireless.h>
88 #include <linux/nsproxy.h>
89 #include <linux/magic.h>
90 #include <linux/slab.h>
91 #include <linux/xattr.h>
92 #include <linux/nospec.h>
94 #include <linux/uaccess.h>
95 #include <asm/unistd.h>
97 #include <net/compat.h>
99 #include <net/cls_cgroup.h>
101 #include <net/sock.h>
102 #include <linux/netfilter.h>
104 #include <linux/if_tun.h>
105 #include <linux/ipv6_route.h>
106 #include <linux/route.h>
107 #include <linux/sockios.h>
108 #include <net/busy_poll.h>
109 #include <linux/errqueue.h>
111 #ifdef CONFIG_NET_RX_BUSY_POLL
112 unsigned int sysctl_net_busy_read __read_mostly;
113 unsigned int sysctl_net_busy_poll __read_mostly;
116 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
117 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
118 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
120 static int sock_close(struct inode *inode, struct file *file);
121 static __poll_t sock_poll(struct file *file,
122 struct poll_table_struct *wait);
123 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
125 static long compat_sock_ioctl(struct file *file,
126 unsigned int cmd, unsigned long arg);
128 static int sock_fasync(int fd, struct file *filp, int on);
129 static ssize_t sock_sendpage(struct file *file, struct page *page,
130 int offset, size_t size, loff_t *ppos, int more);
131 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
132 struct pipe_inode_info *pipe, size_t len,
136 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
137 * in the operation structures but are done directly via the socketcall() multiplexor.
140 static const struct file_operations socket_file_ops = {
141 .owner = THIS_MODULE,
143 .read_iter = sock_read_iter,
144 .write_iter = sock_write_iter,
146 .unlocked_ioctl = sock_ioctl,
148 .compat_ioctl = compat_sock_ioctl,
151 .release = sock_close,
152 .fasync = sock_fasync,
153 .sendpage = sock_sendpage,
154 .splice_write = generic_splice_sendpage,
155 .splice_read = sock_splice_read,
159 * The protocol list. Each protocol is registered in here.
162 static DEFINE_SPINLOCK(net_family_lock);
163 static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
167 * Move socket addresses back and forth across the kernel/user
168 * divide and look after the messy bits.
172 * move_addr_to_kernel - copy a socket address into kernel space
173 * @uaddr: Address in user space
174 * @kaddr: Address in kernel space
175 * @ulen: Length in user space
177 * The address is copied into kernel space. If the provided address is
178 * too long an error code of -EINVAL is returned. If the copy gives
179 * invalid addresses -EFAULT is returned. On a success 0 is returned.
182 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
184 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
188 if (copy_from_user(kaddr, uaddr, ulen))
190 return audit_sockaddr(ulen, kaddr);
194 * move_addr_to_user - copy an address to user space
195 * @kaddr: kernel space address
196 * @klen: length of address in kernel
197 * @uaddr: user space address
198 * @ulen: pointer to user length field
200 * The value pointed to by ulen on entry is the buffer length available.
201 * This is overwritten with the buffer space used. -EINVAL is returned
202 * if an overlong buffer is specified or a negative buffer size. -EFAULT
203 * is returned if either the buffer or the length field are not
205 * After copying the data up to the limit the user specifies, the true
206 * length of the data is written over the length limit the user
207 * specified. Zero is returned for a success.
210 static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
211 void __user *uaddr, int __user *ulen)
216 BUG_ON(klen > sizeof(struct sockaddr_storage));
217 err = get_user(len, ulen);
225 if (audit_sockaddr(klen, kaddr))
227 if (copy_to_user(uaddr, kaddr, len))
231 * "fromlen shall refer to the value before truncation.."
234 return __put_user(klen, ulen);
237 static struct kmem_cache *sock_inode_cachep __ro_after_init;
239 static struct inode *sock_alloc_inode(struct super_block *sb)
241 struct socket_alloc *ei;
242 struct socket_wq *wq;
244 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
247 wq = kmalloc(sizeof(*wq), GFP_KERNEL);
249 kmem_cache_free(sock_inode_cachep, ei);
252 init_waitqueue_head(&wq->wait);
253 wq->fasync_list = NULL;
257 ei->socket.state = SS_UNCONNECTED;
258 ei->socket.flags = 0;
259 ei->socket.ops = NULL;
260 ei->socket.sk = NULL;
261 ei->socket.file = NULL;
263 return &ei->vfs_inode;
266 static void sock_destroy_inode(struct inode *inode)
268 struct socket_alloc *ei;
270 ei = container_of(inode, struct socket_alloc, vfs_inode);
271 kfree_rcu(ei->socket.wq, rcu);
272 kmem_cache_free(sock_inode_cachep, ei);
275 static void init_once(void *foo)
277 struct socket_alloc *ei = (struct socket_alloc *)foo;
279 inode_init_once(&ei->vfs_inode);
282 static void init_inodecache(void)
284 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
285 sizeof(struct socket_alloc),
287 (SLAB_HWCACHE_ALIGN |
288 SLAB_RECLAIM_ACCOUNT |
289 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
291 BUG_ON(sock_inode_cachep == NULL);
294 static const struct super_operations sockfs_ops = {
295 .alloc_inode = sock_alloc_inode,
296 .destroy_inode = sock_destroy_inode,
297 .statfs = simple_statfs,
301 * sockfs_dname() is called from d_path().
303 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
305 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
306 d_inode(dentry)->i_ino);
309 static const struct dentry_operations sockfs_dentry_operations = {
310 .d_dname = sockfs_dname,
313 static int sockfs_xattr_get(const struct xattr_handler *handler,
314 struct dentry *dentry, struct inode *inode,
315 const char *suffix, void *value, size_t size)
318 if (dentry->d_name.len + 1 > size)
320 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
322 return dentry->d_name.len + 1;
325 #define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
326 #define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
327 #define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
329 static const struct xattr_handler sockfs_xattr_handler = {
330 .name = XATTR_NAME_SOCKPROTONAME,
331 .get = sockfs_xattr_get,
334 static int sockfs_security_xattr_set(const struct xattr_handler *handler,
335 struct dentry *dentry, struct inode *inode,
336 const char *suffix, const void *value,
337 size_t size, int flags)
339 /* Handled by LSM. */
343 static const struct xattr_handler sockfs_security_xattr_handler = {
344 .prefix = XATTR_SECURITY_PREFIX,
345 .set = sockfs_security_xattr_set,
348 static const struct xattr_handler *sockfs_xattr_handlers[] = {
349 &sockfs_xattr_handler,
350 &sockfs_security_xattr_handler,
354 static struct dentry *sockfs_mount(struct file_system_type *fs_type,
355 int flags, const char *dev_name, void *data)
357 return mount_pseudo_xattr(fs_type, "socket:", &sockfs_ops,
358 sockfs_xattr_handlers,
359 &sockfs_dentry_operations, SOCKFS_MAGIC);
362 static struct vfsmount *sock_mnt __read_mostly;
364 static struct file_system_type sock_fs_type = {
366 .mount = sockfs_mount,
367 .kill_sb = kill_anon_super,
371 * Obtains the first available file descriptor and sets it up for use.
373 * These functions create file structures and maps them to fd space
374 * of the current process. On success it returns file descriptor
375 * and file struct implicitly stored in sock->file.
376 * Note that another thread may close file descriptor before we return
377 * from this function. We use the fact that now we do not refer
378 * to socket after mapping. If one day we will need it, this
379 * function will increment ref. count on file by 1.
381 * In any case returned fd MAY BE not valid!
382 * This race condition is unavoidable
383 * with shared fd spaces, we cannot solve it inside kernel,
384 * but we take care of internal coherence yet.
387 struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
392 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
394 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
395 O_RDWR | (flags & O_NONBLOCK),
403 file->private_data = sock;
406 EXPORT_SYMBOL(sock_alloc_file);
408 static int sock_map_fd(struct socket *sock, int flags)
410 struct file *newfile;
411 int fd = get_unused_fd_flags(flags);
412 if (unlikely(fd < 0)) {
417 newfile = sock_alloc_file(sock, flags, NULL);
418 if (likely(!IS_ERR(newfile))) {
419 fd_install(fd, newfile);
424 return PTR_ERR(newfile);
427 struct socket *sock_from_file(struct file *file, int *err)
429 if (file->f_op == &socket_file_ops)
430 return file->private_data; /* set in sock_map_fd */
435 EXPORT_SYMBOL(sock_from_file);
438 * sockfd_lookup - Go from a file number to its socket slot
440 * @err: pointer to an error code return
442 * The file handle passed in is locked and the socket it is bound
443 * to is returned. If an error occurs the err pointer is overwritten
444 * with a negative errno code and NULL is returned. The function checks
445 * for both invalid handles and passing a handle which is not a socket.
447 * On a success the socket object pointer is returned.
450 struct socket *sockfd_lookup(int fd, int *err)
461 sock = sock_from_file(file, err);
466 EXPORT_SYMBOL(sockfd_lookup);
468 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
470 struct fd f = fdget(fd);
475 sock = sock_from_file(f.file, err);
477 *fput_needed = f.flags;
485 static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
491 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
501 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
506 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
513 static int sockfs_setattr(struct dentry *dentry, struct iattr *iattr)
515 int err = simple_setattr(dentry, iattr);
517 if (!err && (iattr->ia_valid & ATTR_UID)) {
518 struct socket *sock = SOCKET_I(d_inode(dentry));
521 sock->sk->sk_uid = iattr->ia_uid;
529 static const struct inode_operations sockfs_inode_ops = {
530 .listxattr = sockfs_listxattr,
531 .setattr = sockfs_setattr,
535 * sock_alloc - allocate a socket
537 * Allocate a new inode and socket object. The two are bound together
538 * and initialised. The socket is then returned. If we are out of inodes
542 struct socket *sock_alloc(void)
547 inode = new_inode_pseudo(sock_mnt->mnt_sb);
551 sock = SOCKET_I(inode);
553 inode->i_ino = get_next_ino();
554 inode->i_mode = S_IFSOCK | S_IRWXUGO;
555 inode->i_uid = current_fsuid();
556 inode->i_gid = current_fsgid();
557 inode->i_op = &sockfs_inode_ops;
561 EXPORT_SYMBOL(sock_alloc);
564 * sock_release - close a socket
565 * @sock: socket to close
567 * The socket is released from the protocol stack if it has a release
568 * callback, and the inode is then released if the socket is bound to
569 * an inode not a file.
572 static void __sock_release(struct socket *sock, struct inode *inode)
575 struct module *owner = sock->ops->owner;
579 sock->ops->release(sock);
587 if (sock->wq->fasync_list)
588 pr_err("%s: fasync list not empty!\n", __func__);
591 iput(SOCK_INODE(sock));
597 void sock_release(struct socket *sock)
599 __sock_release(sock, NULL);
601 EXPORT_SYMBOL(sock_release);
603 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
605 u8 flags = *tx_flags;
607 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
608 flags |= SKBTX_HW_TSTAMP;
610 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
611 flags |= SKBTX_SW_TSTAMP;
613 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
614 flags |= SKBTX_SCHED_TSTAMP;
618 EXPORT_SYMBOL(__sock_tx_timestamp);
620 static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
622 int ret = sock->ops->sendmsg(sock, msg, msg_data_left(msg));
623 BUG_ON(ret == -EIOCBQUEUED);
627 int sock_sendmsg(struct socket *sock, struct msghdr *msg)
629 int err = security_socket_sendmsg(sock, msg,
632 return err ?: sock_sendmsg_nosec(sock, msg);
634 EXPORT_SYMBOL(sock_sendmsg);
636 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
637 struct kvec *vec, size_t num, size_t size)
639 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
640 return sock_sendmsg(sock, msg);
642 EXPORT_SYMBOL(kernel_sendmsg);
644 int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
645 struct kvec *vec, size_t num, size_t size)
647 struct socket *sock = sk->sk_socket;
649 if (!sock->ops->sendmsg_locked)
650 return sock_no_sendmsg_locked(sk, msg, size);
652 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
654 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
656 EXPORT_SYMBOL(kernel_sendmsg_locked);
658 static bool skb_is_err_queue(const struct sk_buff *skb)
660 /* pkt_type of skbs enqueued on the error queue are set to
661 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
662 * in recvmsg, since skbs received on a local socket will never
663 * have a pkt_type of PACKET_OUTGOING.
665 return skb->pkt_type == PACKET_OUTGOING;
668 /* On transmit, software and hardware timestamps are returned independently.
669 * As the two skb clones share the hardware timestamp, which may be updated
670 * before the software timestamp is received, a hardware TX timestamp may be
671 * returned only if there is no software TX timestamp. Ignore false software
672 * timestamps, which may be made in the __sock_recv_timestamp() call when the
673 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
674 * hardware timestamp.
676 static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
678 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
681 static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
683 struct scm_ts_pktinfo ts_pktinfo;
684 struct net_device *orig_dev;
686 if (!skb_mac_header_was_set(skb))
689 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
692 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
694 ts_pktinfo.if_index = orig_dev->ifindex;
697 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
698 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
699 sizeof(ts_pktinfo), &ts_pktinfo);
703 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
705 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
708 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
709 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
710 struct scm_timestamping_internal tss;
712 int empty = 1, false_tstamp = 0;
713 struct skb_shared_hwtstamps *shhwtstamps =
716 /* Race occurred between timestamp enabling and packet
717 receiving. Fill in the current time for now. */
718 if (need_software_tstamp && skb->tstamp == 0) {
719 __net_timestamp(skb);
723 if (need_software_tstamp) {
724 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
726 struct __kernel_sock_timeval tv;
728 skb_get_new_timestamp(skb, &tv);
729 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
732 struct __kernel_old_timeval tv;
734 skb_get_timestamp(skb, &tv);
735 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
740 struct __kernel_timespec ts;
742 skb_get_new_timestampns(skb, &ts);
743 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
748 skb_get_timestampns(skb, &ts);
749 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
755 memset(&tss, 0, sizeof(tss));
756 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
757 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
760 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
761 !skb_is_swtx_tstamp(skb, false_tstamp) &&
762 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, tss.ts + 2)) {
764 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
765 !skb_is_err_queue(skb))
766 put_ts_pktinfo(msg, skb);
769 if (sock_flag(sk, SOCK_TSTAMP_NEW))
770 put_cmsg_scm_timestamping64(msg, &tss);
772 put_cmsg_scm_timestamping(msg, &tss);
774 if (skb_is_err_queue(skb) && skb->len &&
775 SKB_EXT_ERR(skb)->opt_stats)
776 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
777 skb->len, skb->data);
780 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
782 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
787 if (!sock_flag(sk, SOCK_WIFI_STATUS))
789 if (!skb->wifi_acked_valid)
792 ack = skb->wifi_acked;
794 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
796 EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
798 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
801 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
802 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
803 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
806 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
809 sock_recv_timestamp(msg, sk, skb);
810 sock_recv_drops(msg, sk, skb);
812 EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
814 static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
817 return sock->ops->recvmsg(sock, msg, msg_data_left(msg), flags);
820 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
822 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
824 return err ?: sock_recvmsg_nosec(sock, msg, flags);
826 EXPORT_SYMBOL(sock_recvmsg);
829 * kernel_recvmsg - Receive a message from a socket (kernel space)
830 * @sock: The socket to receive the message from
831 * @msg: Received message
832 * @vec: Input s/g array for message data
833 * @num: Size of input s/g array
834 * @size: Number of bytes to read
835 * @flags: Message flags (MSG_DONTWAIT, etc...)
837 * On return the msg structure contains the scatter/gather array passed in the
838 * vec argument. The array is modified so that it consists of the unfilled
839 * portion of the original array.
841 * The returned value is the total number of bytes received, or an error.
843 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
844 struct kvec *vec, size_t num, size_t size, int flags)
846 mm_segment_t oldfs = get_fs();
849 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
851 result = sock_recvmsg(sock, msg, flags);
855 EXPORT_SYMBOL(kernel_recvmsg);
857 static ssize_t sock_sendpage(struct file *file, struct page *page,
858 int offset, size_t size, loff_t *ppos, int more)
863 sock = file->private_data;
865 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
866 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
869 return kernel_sendpage(sock, page, offset, size, flags);
872 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
873 struct pipe_inode_info *pipe, size_t len,
876 struct socket *sock = file->private_data;
878 if (unlikely(!sock->ops->splice_read))
879 return generic_file_splice_read(file, ppos, pipe, len, flags);
881 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
884 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
886 struct file *file = iocb->ki_filp;
887 struct socket *sock = file->private_data;
888 struct msghdr msg = {.msg_iter = *to,
892 if (file->f_flags & O_NONBLOCK)
893 msg.msg_flags = MSG_DONTWAIT;
895 if (iocb->ki_pos != 0)
898 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
901 res = sock_recvmsg(sock, &msg, msg.msg_flags);
906 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
908 struct file *file = iocb->ki_filp;
909 struct socket *sock = file->private_data;
910 struct msghdr msg = {.msg_iter = *from,
914 if (iocb->ki_pos != 0)
917 if (file->f_flags & O_NONBLOCK)
918 msg.msg_flags = MSG_DONTWAIT;
920 if (sock->type == SOCK_SEQPACKET)
921 msg.msg_flags |= MSG_EOR;
923 res = sock_sendmsg(sock, &msg);
924 *from = msg.msg_iter;
929 * Atomic setting of ioctl hooks to avoid race
930 * with module unload.
933 static DEFINE_MUTEX(br_ioctl_mutex);
934 static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
936 void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
938 mutex_lock(&br_ioctl_mutex);
939 br_ioctl_hook = hook;
940 mutex_unlock(&br_ioctl_mutex);
942 EXPORT_SYMBOL(brioctl_set);
944 static DEFINE_MUTEX(vlan_ioctl_mutex);
945 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
947 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
949 mutex_lock(&vlan_ioctl_mutex);
950 vlan_ioctl_hook = hook;
951 mutex_unlock(&vlan_ioctl_mutex);
953 EXPORT_SYMBOL(vlan_ioctl_set);
955 static DEFINE_MUTEX(dlci_ioctl_mutex);
956 static int (*dlci_ioctl_hook) (unsigned int, void __user *);
958 void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
960 mutex_lock(&dlci_ioctl_mutex);
961 dlci_ioctl_hook = hook;
962 mutex_unlock(&dlci_ioctl_mutex);
964 EXPORT_SYMBOL(dlci_ioctl_set);
966 static long sock_do_ioctl(struct net *net, struct socket *sock,
967 unsigned int cmd, unsigned long arg)
970 void __user *argp = (void __user *)arg;
972 err = sock->ops->ioctl(sock, cmd, arg);
975 * If this ioctl is unknown try to hand it down
978 if (err != -ENOIOCTLCMD)
981 if (cmd == SIOCGIFCONF) {
983 if (copy_from_user(&ifc, argp, sizeof(struct ifconf)))
986 err = dev_ifconf(net, &ifc, sizeof(struct ifreq));
988 if (!err && copy_to_user(argp, &ifc, sizeof(struct ifconf)))
993 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
995 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
996 if (!err && need_copyout)
997 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1004 * With an ioctl, arg may well be a user mode pointer, but we don't know
1005 * what to do with it - that's up to the protocol still.
1008 struct ns_common *get_net_ns(struct ns_common *ns)
1010 return &get_net(container_of(ns, struct net, ns))->ns;
1012 EXPORT_SYMBOL_GPL(get_net_ns);
1014 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1016 struct socket *sock;
1018 void __user *argp = (void __user *)arg;
1022 sock = file->private_data;
1025 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1028 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1030 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1031 if (!err && need_copyout)
1032 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1035 #ifdef CONFIG_WEXT_CORE
1036 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1037 err = wext_handle_ioctl(net, cmd, argp);
1044 if (get_user(pid, (int __user *)argp))
1046 err = f_setown(sock->file, pid, 1);
1050 err = put_user(f_getown(sock->file),
1051 (int __user *)argp);
1059 request_module("bridge");
1061 mutex_lock(&br_ioctl_mutex);
1063 err = br_ioctl_hook(net, cmd, argp);
1064 mutex_unlock(&br_ioctl_mutex);
1069 if (!vlan_ioctl_hook)
1070 request_module("8021q");
1072 mutex_lock(&vlan_ioctl_mutex);
1073 if (vlan_ioctl_hook)
1074 err = vlan_ioctl_hook(net, argp);
1075 mutex_unlock(&vlan_ioctl_mutex);
1080 if (!dlci_ioctl_hook)
1081 request_module("dlci");
1083 mutex_lock(&dlci_ioctl_mutex);
1084 if (dlci_ioctl_hook)
1085 err = dlci_ioctl_hook(cmd, argp);
1086 mutex_unlock(&dlci_ioctl_mutex);
1090 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1093 err = open_related_ns(&net->ns, get_net_ns);
1096 err = sock_do_ioctl(net, sock, cmd, arg);
1102 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1105 struct socket *sock = NULL;
1107 err = security_socket_create(family, type, protocol, 1);
1111 sock = sock_alloc();
1118 err = security_socket_post_create(sock, family, type, protocol, 1);
1130 EXPORT_SYMBOL(sock_create_lite);
1132 /* No kernel lock held - perfect */
1133 static __poll_t sock_poll(struct file *file, poll_table *wait)
1135 struct socket *sock = file->private_data;
1136 __poll_t events = poll_requested_events(wait), flag = 0;
1138 if (!sock->ops->poll)
1141 if (sk_can_busy_loop(sock->sk)) {
1142 /* poll once if requested by the syscall */
1143 if (events & POLL_BUSY_LOOP)
1144 sk_busy_loop(sock->sk, 1);
1146 /* if this socket can poll_ll, tell the system call */
1147 flag = POLL_BUSY_LOOP;
1150 return sock->ops->poll(file, sock, wait) | flag;
1153 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1155 struct socket *sock = file->private_data;
1157 return sock->ops->mmap(file, sock, vma);
1160 static int sock_close(struct inode *inode, struct file *filp)
1162 __sock_release(SOCKET_I(inode), inode);
1167 * Update the socket async list
1169 * Fasync_list locking strategy.
1171 * 1. fasync_list is modified only under process context socket lock
1172 * i.e. under semaphore.
1173 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1174 * or under socket lock
1177 static int sock_fasync(int fd, struct file *filp, int on)
1179 struct socket *sock = filp->private_data;
1180 struct sock *sk = sock->sk;
1181 struct socket_wq *wq;
1188 fasync_helper(fd, filp, on, &wq->fasync_list);
1190 if (!wq->fasync_list)
1191 sock_reset_flag(sk, SOCK_FASYNC);
1193 sock_set_flag(sk, SOCK_FASYNC);
1199 /* This function may be called only under rcu_lock */
1201 int sock_wake_async(struct socket_wq *wq, int how, int band)
1203 if (!wq || !wq->fasync_list)
1207 case SOCK_WAKE_WAITD:
1208 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1211 case SOCK_WAKE_SPACE:
1212 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1217 kill_fasync(&wq->fasync_list, SIGIO, band);
1220 kill_fasync(&wq->fasync_list, SIGURG, band);
1225 EXPORT_SYMBOL(sock_wake_async);
1227 int __sock_create(struct net *net, int family, int type, int protocol,
1228 struct socket **res, int kern)
1231 struct socket *sock;
1232 const struct net_proto_family *pf;
1235 * Check protocol is in range
1237 if (family < 0 || family >= NPROTO)
1238 return -EAFNOSUPPORT;
1239 if (type < 0 || type >= SOCK_MAX)
1244 This uglymoron is moved from INET layer to here to avoid
1245 deadlock in module load.
1247 if (family == PF_INET && type == SOCK_PACKET) {
1248 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1253 err = security_socket_create(family, type, protocol, kern);
1258 * Allocate the socket and allow the family to set things up. if
1259 * the protocol is 0, the family is instructed to select an appropriate
1262 sock = sock_alloc();
1264 net_warn_ratelimited("socket: no more sockets\n");
1265 return -ENFILE; /* Not exactly a match, but its the
1266 closest posix thing */
1271 #ifdef CONFIG_MODULES
1272 /* Attempt to load a protocol module if the find failed.
1274 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1275 * requested real, full-featured networking support upon configuration.
1276 * Otherwise module support will break!
1278 if (rcu_access_pointer(net_families[family]) == NULL)
1279 request_module("net-pf-%d", family);
1283 pf = rcu_dereference(net_families[family]);
1284 err = -EAFNOSUPPORT;
1289 * We will call the ->create function, that possibly is in a loadable
1290 * module, so we have to bump that loadable module refcnt first.
1292 if (!try_module_get(pf->owner))
1295 /* Now protected by module ref count */
1298 err = pf->create(net, sock, protocol, kern);
1300 goto out_module_put;
1303 * Now to bump the refcnt of the [loadable] module that owns this
1304 * socket at sock_release time we decrement its refcnt.
1306 if (!try_module_get(sock->ops->owner))
1307 goto out_module_busy;
1310 * Now that we're done with the ->create function, the [loadable]
1311 * module can have its refcnt decremented
1313 module_put(pf->owner);
1314 err = security_socket_post_create(sock, family, type, protocol, kern);
1316 goto out_sock_release;
1322 err = -EAFNOSUPPORT;
1325 module_put(pf->owner);
1332 goto out_sock_release;
1334 EXPORT_SYMBOL(__sock_create);
1336 int sock_create(int family, int type, int protocol, struct socket **res)
1338 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1340 EXPORT_SYMBOL(sock_create);
1342 int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1344 return __sock_create(net, family, type, protocol, res, 1);
1346 EXPORT_SYMBOL(sock_create_kern);
1348 int __sys_socket(int family, int type, int protocol)
1351 struct socket *sock;
1354 /* Check the SOCK_* constants for consistency. */
1355 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1356 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1357 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1358 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1360 flags = type & ~SOCK_TYPE_MASK;
1361 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1363 type &= SOCK_TYPE_MASK;
1365 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1366 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1368 retval = sock_create(family, type, protocol, &sock);
1372 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1375 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1377 return __sys_socket(family, type, protocol);
1381 * Create a pair of connected sockets.
1384 int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1386 struct socket *sock1, *sock2;
1388 struct file *newfile1, *newfile2;
1391 flags = type & ~SOCK_TYPE_MASK;
1392 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1394 type &= SOCK_TYPE_MASK;
1396 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1397 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1400 * reserve descriptors and make sure we won't fail
1401 * to return them to userland.
1403 fd1 = get_unused_fd_flags(flags);
1404 if (unlikely(fd1 < 0))
1407 fd2 = get_unused_fd_flags(flags);
1408 if (unlikely(fd2 < 0)) {
1413 err = put_user(fd1, &usockvec[0]);
1417 err = put_user(fd2, &usockvec[1]);
1422 * Obtain the first socket and check if the underlying protocol
1423 * supports the socketpair call.
1426 err = sock_create(family, type, protocol, &sock1);
1427 if (unlikely(err < 0))
1430 err = sock_create(family, type, protocol, &sock2);
1431 if (unlikely(err < 0)) {
1432 sock_release(sock1);
1436 err = security_socket_socketpair(sock1, sock2);
1437 if (unlikely(err)) {
1438 sock_release(sock2);
1439 sock_release(sock1);
1443 err = sock1->ops->socketpair(sock1, sock2);
1444 if (unlikely(err < 0)) {
1445 sock_release(sock2);
1446 sock_release(sock1);
1450 newfile1 = sock_alloc_file(sock1, flags, NULL);
1451 if (IS_ERR(newfile1)) {
1452 err = PTR_ERR(newfile1);
1453 sock_release(sock2);
1457 newfile2 = sock_alloc_file(sock2, flags, NULL);
1458 if (IS_ERR(newfile2)) {
1459 err = PTR_ERR(newfile2);
1464 audit_fd_pair(fd1, fd2);
1466 fd_install(fd1, newfile1);
1467 fd_install(fd2, newfile2);
1476 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1477 int __user *, usockvec)
1479 return __sys_socketpair(family, type, protocol, usockvec);
1483 * Bind a name to a socket. Nothing much to do here since it's
1484 * the protocol's responsibility to handle the local address.
1486 * We move the socket address to kernel space before we call
1487 * the protocol layer (having also checked the address is ok).
1490 int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1492 struct socket *sock;
1493 struct sockaddr_storage address;
1494 int err, fput_needed;
1496 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1498 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1500 err = security_socket_bind(sock,
1501 (struct sockaddr *)&address,
1504 err = sock->ops->bind(sock,
1508 fput_light(sock->file, fput_needed);
1513 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1515 return __sys_bind(fd, umyaddr, addrlen);
1519 * Perform a listen. Basically, we allow the protocol to do anything
1520 * necessary for a listen, and if that works, we mark the socket as
1521 * ready for listening.
1524 int __sys_listen(int fd, int backlog)
1526 struct socket *sock;
1527 int err, fput_needed;
1530 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1532 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1533 if ((unsigned int)backlog > somaxconn)
1534 backlog = somaxconn;
1536 err = security_socket_listen(sock, backlog);
1538 err = sock->ops->listen(sock, backlog);
1540 fput_light(sock->file, fput_needed);
1545 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1547 return __sys_listen(fd, backlog);
1551 * For accept, we attempt to create a new socket, set up the link
1552 * with the client, wake up the client, then return the new
1553 * connected fd. We collect the address of the connector in kernel
1554 * space and move it to user at the very end. This is unclean because
1555 * we open the socket then return an error.
1557 * 1003.1g adds the ability to recvmsg() to query connection pending
1558 * status to recvmsg. We need to add that support in a way thats
1559 * clean when we restructure accept also.
1562 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1563 int __user *upeer_addrlen, int flags)
1565 struct socket *sock, *newsock;
1566 struct file *newfile;
1567 int err, len, newfd, fput_needed;
1568 struct sockaddr_storage address;
1570 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1573 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1574 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1576 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1581 newsock = sock_alloc();
1585 newsock->type = sock->type;
1586 newsock->ops = sock->ops;
1589 * We don't need try_module_get here, as the listening socket (sock)
1590 * has the protocol module (sock->ops->owner) held.
1592 __module_get(newsock->ops->owner);
1594 newfd = get_unused_fd_flags(flags);
1595 if (unlikely(newfd < 0)) {
1597 sock_release(newsock);
1600 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1601 if (IS_ERR(newfile)) {
1602 err = PTR_ERR(newfile);
1603 put_unused_fd(newfd);
1607 err = security_socket_accept(sock, newsock);
1611 err = sock->ops->accept(sock, newsock, sock->file->f_flags, false);
1615 if (upeer_sockaddr) {
1616 len = newsock->ops->getname(newsock,
1617 (struct sockaddr *)&address, 2);
1619 err = -ECONNABORTED;
1622 err = move_addr_to_user(&address,
1623 len, upeer_sockaddr, upeer_addrlen);
1628 /* File flags are not inherited via accept() unlike another OSes. */
1630 fd_install(newfd, newfile);
1634 fput_light(sock->file, fput_needed);
1639 put_unused_fd(newfd);
1643 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1644 int __user *, upeer_addrlen, int, flags)
1646 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1649 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1650 int __user *, upeer_addrlen)
1652 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1656 * Attempt to connect to a socket with the server address. The address
1657 * is in user space so we verify it is OK and move it to kernel space.
1659 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1662 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1663 * other SEQPACKET protocols that take time to connect() as it doesn't
1664 * include the -EINPROGRESS status for such sockets.
1667 int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1669 struct socket *sock;
1670 struct sockaddr_storage address;
1671 int err, fput_needed;
1673 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1676 err = move_addr_to_kernel(uservaddr, addrlen, &address);
1681 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1685 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1686 sock->file->f_flags);
1688 fput_light(sock->file, fput_needed);
1693 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1696 return __sys_connect(fd, uservaddr, addrlen);
1700 * Get the local address ('name') of a socket object. Move the obtained
1701 * name to user space.
1704 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1705 int __user *usockaddr_len)
1707 struct socket *sock;
1708 struct sockaddr_storage address;
1709 int err, fput_needed;
1711 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1715 err = security_socket_getsockname(sock);
1719 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1722 /* "err" is actually length in this case */
1723 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1726 fput_light(sock->file, fput_needed);
1731 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1732 int __user *, usockaddr_len)
1734 return __sys_getsockname(fd, usockaddr, usockaddr_len);
1738 * Get the remote address ('name') of a socket object. Move the obtained
1739 * name to user space.
1742 int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1743 int __user *usockaddr_len)
1745 struct socket *sock;
1746 struct sockaddr_storage address;
1747 int err, fput_needed;
1749 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1751 err = security_socket_getpeername(sock);
1753 fput_light(sock->file, fput_needed);
1757 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
1759 /* "err" is actually length in this case */
1760 err = move_addr_to_user(&address, err, usockaddr,
1762 fput_light(sock->file, fput_needed);
1767 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1768 int __user *, usockaddr_len)
1770 return __sys_getpeername(fd, usockaddr, usockaddr_len);
1774 * Send a datagram to a given address. We move the address into kernel
1775 * space and check the user space data area is readable before invoking
1778 int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
1779 struct sockaddr __user *addr, int addr_len)
1781 struct socket *sock;
1782 struct sockaddr_storage address;
1788 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
1791 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1795 msg.msg_name = NULL;
1796 msg.msg_control = NULL;
1797 msg.msg_controllen = 0;
1798 msg.msg_namelen = 0;
1800 err = move_addr_to_kernel(addr, addr_len, &address);
1803 msg.msg_name = (struct sockaddr *)&address;
1804 msg.msg_namelen = addr_len;
1806 if (sock->file->f_flags & O_NONBLOCK)
1807 flags |= MSG_DONTWAIT;
1808 msg.msg_flags = flags;
1809 err = sock_sendmsg(sock, &msg);
1812 fput_light(sock->file, fput_needed);
1817 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1818 unsigned int, flags, struct sockaddr __user *, addr,
1821 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
1825 * Send a datagram down a socket.
1828 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1829 unsigned int, flags)
1831 return __sys_sendto(fd, buff, len, flags, NULL, 0);
1835 * Receive a frame from the socket and optionally record the address of the
1836 * sender. We verify the buffers are writable and if needed move the
1837 * sender address from kernel to user space.
1839 int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
1840 struct sockaddr __user *addr, int __user *addr_len)
1842 struct socket *sock;
1845 struct sockaddr_storage address;
1849 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
1852 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1856 msg.msg_control = NULL;
1857 msg.msg_controllen = 0;
1858 /* Save some cycles and don't copy the address if not needed */
1859 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
1860 /* We assume all kernel code knows the size of sockaddr_storage */
1861 msg.msg_namelen = 0;
1862 msg.msg_iocb = NULL;
1864 if (sock->file->f_flags & O_NONBLOCK)
1865 flags |= MSG_DONTWAIT;
1866 err = sock_recvmsg(sock, &msg, flags);
1868 if (err >= 0 && addr != NULL) {
1869 err2 = move_addr_to_user(&address,
1870 msg.msg_namelen, addr, addr_len);
1875 fput_light(sock->file, fput_needed);
1880 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1881 unsigned int, flags, struct sockaddr __user *, addr,
1882 int __user *, addr_len)
1884 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
1888 * Receive a datagram from a socket.
1891 SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
1892 unsigned int, flags)
1894 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1898 * Set a socket option. Because we don't know the option lengths we have
1899 * to pass the user mode parameter for the protocols to sort out.
1902 static int __sys_setsockopt(int fd, int level, int optname,
1903 char __user *optval, int optlen)
1905 int err, fput_needed;
1906 struct socket *sock;
1911 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1913 err = security_socket_setsockopt(sock, level, optname);
1917 if (level == SOL_SOCKET)
1919 sock_setsockopt(sock, level, optname, optval,
1923 sock->ops->setsockopt(sock, level, optname, optval,
1926 fput_light(sock->file, fput_needed);
1931 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1932 char __user *, optval, int, optlen)
1934 return __sys_setsockopt(fd, level, optname, optval, optlen);
1938 * Get a socket option. Because we don't know the option lengths we have
1939 * to pass a user mode parameter for the protocols to sort out.
1942 static int __sys_getsockopt(int fd, int level, int optname,
1943 char __user *optval, int __user *optlen)
1945 int err, fput_needed;
1946 struct socket *sock;
1948 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1950 err = security_socket_getsockopt(sock, level, optname);
1954 if (level == SOL_SOCKET)
1956 sock_getsockopt(sock, level, optname, optval,
1960 sock->ops->getsockopt(sock, level, optname, optval,
1963 fput_light(sock->file, fput_needed);
1968 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1969 char __user *, optval, int __user *, optlen)
1971 return __sys_getsockopt(fd, level, optname, optval, optlen);
1975 * Shutdown a socket.
1978 int __sys_shutdown(int fd, int how)
1980 int err, fput_needed;
1981 struct socket *sock;
1983 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1985 err = security_socket_shutdown(sock, how);
1987 err = sock->ops->shutdown(sock, how);
1988 fput_light(sock->file, fput_needed);
1993 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1995 return __sys_shutdown(fd, how);
1998 /* A couple of helpful macros for getting the address of the 32/64 bit
1999 * fields which are the same type (int / unsigned) on our platforms.
2001 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2002 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2003 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2005 struct used_address {
2006 struct sockaddr_storage name;
2007 unsigned int name_len;
2010 static int copy_msghdr_from_user(struct msghdr *kmsg,
2011 struct user_msghdr __user *umsg,
2012 struct sockaddr __user **save_addr,
2015 struct user_msghdr msg;
2018 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2021 kmsg->msg_control = (void __force *)msg.msg_control;
2022 kmsg->msg_controllen = msg.msg_controllen;
2023 kmsg->msg_flags = msg.msg_flags;
2025 kmsg->msg_namelen = msg.msg_namelen;
2027 kmsg->msg_namelen = 0;
2029 if (kmsg->msg_namelen < 0)
2032 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2033 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2036 *save_addr = msg.msg_name;
2038 if (msg.msg_name && kmsg->msg_namelen) {
2040 err = move_addr_to_kernel(msg.msg_name,
2047 kmsg->msg_name = NULL;
2048 kmsg->msg_namelen = 0;
2051 if (msg.msg_iovlen > UIO_MAXIOV)
2054 kmsg->msg_iocb = NULL;
2056 return import_iovec(save_addr ? READ : WRITE,
2057 msg.msg_iov, msg.msg_iovlen,
2058 UIO_FASTIOV, iov, &kmsg->msg_iter);
2061 static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2062 struct msghdr *msg_sys, unsigned int flags,
2063 struct used_address *used_address,
2064 unsigned int allowed_msghdr_flags)
2066 struct compat_msghdr __user *msg_compat =
2067 (struct compat_msghdr __user *)msg;
2068 struct sockaddr_storage address;
2069 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2070 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2071 __aligned(sizeof(__kernel_size_t));
2072 /* 20 is size of ipv6_pktinfo */
2073 unsigned char *ctl_buf = ctl;
2077 msg_sys->msg_name = &address;
2079 if (MSG_CMSG_COMPAT & flags)
2080 err = get_compat_msghdr(msg_sys, msg_compat, NULL, &iov);
2082 err = copy_msghdr_from_user(msg_sys, msg, NULL, &iov);
2088 if (msg_sys->msg_controllen > INT_MAX)
2090 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2091 ctl_len = msg_sys->msg_controllen;
2092 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2094 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2098 ctl_buf = msg_sys->msg_control;
2099 ctl_len = msg_sys->msg_controllen;
2100 } else if (ctl_len) {
2101 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2102 CMSG_ALIGN(sizeof(struct cmsghdr)));
2103 if (ctl_len > sizeof(ctl)) {
2104 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2105 if (ctl_buf == NULL)
2110 * Careful! Before this, msg_sys->msg_control contains a user pointer.
2111 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
2112 * checking falls down on this.
2114 if (copy_from_user(ctl_buf,
2115 (void __user __force *)msg_sys->msg_control,
2118 msg_sys->msg_control = ctl_buf;
2120 msg_sys->msg_flags = flags;
2122 if (sock->file->f_flags & O_NONBLOCK)
2123 msg_sys->msg_flags |= MSG_DONTWAIT;
2125 * If this is sendmmsg() and current destination address is same as
2126 * previously succeeded address, omit asking LSM's decision.
2127 * used_address->name_len is initialized to UINT_MAX so that the first
2128 * destination address never matches.
2130 if (used_address && msg_sys->msg_name &&
2131 used_address->name_len == msg_sys->msg_namelen &&
2132 !memcmp(&used_address->name, msg_sys->msg_name,
2133 used_address->name_len)) {
2134 err = sock_sendmsg_nosec(sock, msg_sys);
2137 err = sock_sendmsg(sock, msg_sys);
2139 * If this is sendmmsg() and sending to current destination address was
2140 * successful, remember it.
2142 if (used_address && err >= 0) {
2143 used_address->name_len = msg_sys->msg_namelen;
2144 if (msg_sys->msg_name)
2145 memcpy(&used_address->name, msg_sys->msg_name,
2146 used_address->name_len);
2151 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2158 * BSD sendmsg interface
2161 long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2162 bool forbid_cmsg_compat)
2164 int fput_needed, err;
2165 struct msghdr msg_sys;
2166 struct socket *sock;
2168 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2171 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2175 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2177 fput_light(sock->file, fput_needed);
2182 SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2184 return __sys_sendmsg(fd, msg, flags, true);
2188 * Linux sendmmsg interface
2191 int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2192 unsigned int flags, bool forbid_cmsg_compat)
2194 int fput_needed, err, datagrams;
2195 struct socket *sock;
2196 struct mmsghdr __user *entry;
2197 struct compat_mmsghdr __user *compat_entry;
2198 struct msghdr msg_sys;
2199 struct used_address used_address;
2200 unsigned int oflags = flags;
2202 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2205 if (vlen > UIO_MAXIOV)
2210 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2214 used_address.name_len = UINT_MAX;
2216 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2220 while (datagrams < vlen) {
2221 if (datagrams == vlen - 1)
2224 if (MSG_CMSG_COMPAT & flags) {
2225 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2226 &msg_sys, flags, &used_address, MSG_EOR);
2229 err = __put_user(err, &compat_entry->msg_len);
2232 err = ___sys_sendmsg(sock,
2233 (struct user_msghdr __user *)entry,
2234 &msg_sys, flags, &used_address, MSG_EOR);
2237 err = put_user(err, &entry->msg_len);
2244 if (msg_data_left(&msg_sys))
2249 fput_light(sock->file, fput_needed);
2251 /* We only return an error if no datagrams were able to be sent */
2258 SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2259 unsigned int, vlen, unsigned int, flags)
2261 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2264 static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2265 struct msghdr *msg_sys, unsigned int flags, int nosec)
2267 struct compat_msghdr __user *msg_compat =
2268 (struct compat_msghdr __user *)msg;
2269 struct iovec iovstack[UIO_FASTIOV];
2270 struct iovec *iov = iovstack;
2271 unsigned long cmsg_ptr;
2275 /* kernel mode address */
2276 struct sockaddr_storage addr;
2278 /* user mode address pointers */
2279 struct sockaddr __user *uaddr;
2280 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2282 msg_sys->msg_name = &addr;
2284 if (MSG_CMSG_COMPAT & flags)
2285 err = get_compat_msghdr(msg_sys, msg_compat, &uaddr, &iov);
2287 err = copy_msghdr_from_user(msg_sys, msg, &uaddr, &iov);
2291 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2292 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2294 /* We assume all kernel code knows the size of sockaddr_storage */
2295 msg_sys->msg_namelen = 0;
2297 if (sock->file->f_flags & O_NONBLOCK)
2298 flags |= MSG_DONTWAIT;
2299 err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys, flags);
2304 if (uaddr != NULL) {
2305 err = move_addr_to_user(&addr,
2306 msg_sys->msg_namelen, uaddr,
2311 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2315 if (MSG_CMSG_COMPAT & flags)
2316 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2317 &msg_compat->msg_controllen);
2319 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2320 &msg->msg_controllen);
2331 * BSD recvmsg interface
2334 long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2335 bool forbid_cmsg_compat)
2337 int fput_needed, err;
2338 struct msghdr msg_sys;
2339 struct socket *sock;
2341 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2344 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2348 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2350 fput_light(sock->file, fput_needed);
2355 SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2356 unsigned int, flags)
2358 return __sys_recvmsg(fd, msg, flags, true);
2362 * Linux recvmmsg interface
2365 static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2366 unsigned int vlen, unsigned int flags,
2367 struct timespec64 *timeout)
2369 int fput_needed, err, datagrams;
2370 struct socket *sock;
2371 struct mmsghdr __user *entry;
2372 struct compat_mmsghdr __user *compat_entry;
2373 struct msghdr msg_sys;
2374 struct timespec64 end_time;
2375 struct timespec64 timeout64;
2378 poll_select_set_timeout(&end_time, timeout->tv_sec,
2384 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2388 if (likely(!(flags & MSG_ERRQUEUE))) {
2389 err = sock_error(sock->sk);
2397 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2399 while (datagrams < vlen) {
2401 * No need to ask LSM for more than the first datagram.
2403 if (MSG_CMSG_COMPAT & flags) {
2404 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2405 &msg_sys, flags & ~MSG_WAITFORONE,
2409 err = __put_user(err, &compat_entry->msg_len);
2412 err = ___sys_recvmsg(sock,
2413 (struct user_msghdr __user *)entry,
2414 &msg_sys, flags & ~MSG_WAITFORONE,
2418 err = put_user(err, &entry->msg_len);
2426 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2427 if (flags & MSG_WAITFORONE)
2428 flags |= MSG_DONTWAIT;
2431 ktime_get_ts64(&timeout64);
2432 *timeout = timespec64_sub(end_time, timeout64);
2433 if (timeout->tv_sec < 0) {
2434 timeout->tv_sec = timeout->tv_nsec = 0;
2438 /* Timeout, return less than vlen datagrams */
2439 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2443 /* Out of band data, return right away */
2444 if (msg_sys.msg_flags & MSG_OOB)
2452 if (datagrams == 0) {
2458 * We may return less entries than requested (vlen) if the
2459 * sock is non block and there aren't enough datagrams...
2461 if (err != -EAGAIN) {
2463 * ... or if recvmsg returns an error after we
2464 * received some datagrams, where we record the
2465 * error to return on the next call or if the
2466 * app asks about it using getsockopt(SO_ERROR).
2468 sock->sk->sk_err = -err;
2471 fput_light(sock->file, fput_needed);
2476 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2477 unsigned int vlen, unsigned int flags,
2478 struct __kernel_timespec __user *timeout,
2479 struct old_timespec32 __user *timeout32)
2482 struct timespec64 timeout_sys;
2484 if (timeout && get_timespec64(&timeout_sys, timeout))
2487 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2490 if (!timeout && !timeout32)
2491 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2493 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2498 if (timeout && put_timespec64(&timeout_sys, timeout))
2499 datagrams = -EFAULT;
2501 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2502 datagrams = -EFAULT;
2507 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2508 unsigned int, vlen, unsigned int, flags,
2509 struct __kernel_timespec __user *, timeout)
2511 if (flags & MSG_CMSG_COMPAT)
2514 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2517 #ifdef CONFIG_COMPAT_32BIT_TIME
2518 SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2519 unsigned int, vlen, unsigned int, flags,
2520 struct old_timespec32 __user *, timeout)
2522 if (flags & MSG_CMSG_COMPAT)
2525 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2529 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2530 /* Argument list sizes for sys_socketcall */
2531 #define AL(x) ((x) * sizeof(unsigned long))
2532 static const unsigned char nargs[21] = {
2533 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2534 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2535 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2542 * System call vectors.
2544 * Argument checking cleaned up. Saved 20% in size.
2545 * This function doesn't need to set the kernel lock because
2546 * it is set by the callees.
2549 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2551 unsigned long a[AUDITSC_ARGS];
2552 unsigned long a0, a1;
2556 if (call < 1 || call > SYS_SENDMMSG)
2558 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2561 if (len > sizeof(a))
2564 /* copy_from_user should be SMP safe. */
2565 if (copy_from_user(a, args, len))
2568 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2577 err = __sys_socket(a0, a1, a[2]);
2580 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2583 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2586 err = __sys_listen(a0, a1);
2589 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2590 (int __user *)a[2], 0);
2592 case SYS_GETSOCKNAME:
2594 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2595 (int __user *)a[2]);
2597 case SYS_GETPEERNAME:
2599 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2600 (int __user *)a[2]);
2602 case SYS_SOCKETPAIR:
2603 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2606 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2610 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2611 (struct sockaddr __user *)a[4], a[5]);
2614 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2618 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2619 (struct sockaddr __user *)a[4],
2620 (int __user *)a[5]);
2623 err = __sys_shutdown(a0, a1);
2625 case SYS_SETSOCKOPT:
2626 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
2629 case SYS_GETSOCKOPT:
2631 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2632 (int __user *)a[4]);
2635 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
2639 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
2643 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
2647 if (IS_ENABLED(CONFIG_64BIT) || !IS_ENABLED(CONFIG_64BIT_TIME))
2648 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2650 (struct __kernel_timespec __user *)a[4],
2653 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2655 (struct old_timespec32 __user *)a[4]);
2658 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2659 (int __user *)a[2], a[3]);
2668 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
2671 * sock_register - add a socket protocol handler
2672 * @ops: description of protocol
2674 * This function is called by a protocol handler that wants to
2675 * advertise its address family, and have it linked into the
2676 * socket interface. The value ops->family corresponds to the
2677 * socket system call protocol family.
2679 int sock_register(const struct net_proto_family *ops)
2683 if (ops->family >= NPROTO) {
2684 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
2688 spin_lock(&net_family_lock);
2689 if (rcu_dereference_protected(net_families[ops->family],
2690 lockdep_is_held(&net_family_lock)))
2693 rcu_assign_pointer(net_families[ops->family], ops);
2696 spin_unlock(&net_family_lock);
2698 pr_info("NET: Registered protocol family %d\n", ops->family);
2701 EXPORT_SYMBOL(sock_register);
2704 * sock_unregister - remove a protocol handler
2705 * @family: protocol family to remove
2707 * This function is called by a protocol handler that wants to
2708 * remove its address family, and have it unlinked from the
2709 * new socket creation.
2711 * If protocol handler is a module, then it can use module reference
2712 * counts to protect against new references. If protocol handler is not
2713 * a module then it needs to provide its own protection in
2714 * the ops->create routine.
2716 void sock_unregister(int family)
2718 BUG_ON(family < 0 || family >= NPROTO);
2720 spin_lock(&net_family_lock);
2721 RCU_INIT_POINTER(net_families[family], NULL);
2722 spin_unlock(&net_family_lock);
2726 pr_info("NET: Unregistered protocol family %d\n", family);
2728 EXPORT_SYMBOL(sock_unregister);
2730 bool sock_is_registered(int family)
2732 return family < NPROTO && rcu_access_pointer(net_families[family]);
2735 static int __init sock_init(void)
2739 * Initialize the network sysctl infrastructure.
2741 err = net_sysctl_init();
2746 * Initialize skbuff SLAB cache
2751 * Initialize the protocols module.
2756 err = register_filesystem(&sock_fs_type);
2759 sock_mnt = kern_mount(&sock_fs_type);
2760 if (IS_ERR(sock_mnt)) {
2761 err = PTR_ERR(sock_mnt);
2765 /* The real protocol initialization is performed in later initcalls.
2768 #ifdef CONFIG_NETFILTER
2769 err = netfilter_init();
2774 ptp_classifier_init();
2780 unregister_filesystem(&sock_fs_type);
2785 core_initcall(sock_init); /* early initcall */
2787 #ifdef CONFIG_PROC_FS
2788 void socket_seq_show(struct seq_file *seq)
2790 seq_printf(seq, "sockets: used %d\n",
2791 sock_inuse_get(seq->private));
2793 #endif /* CONFIG_PROC_FS */
2795 #ifdef CONFIG_COMPAT
2796 static int do_siocgstamp(struct net *net, struct socket *sock,
2797 unsigned int cmd, void __user *up)
2799 mm_segment_t old_fs = get_fs();
2804 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&ktv);
2807 err = compat_put_timeval(&ktv, up);
2812 static int do_siocgstampns(struct net *net, struct socket *sock,
2813 unsigned int cmd, void __user *up)
2815 mm_segment_t old_fs = get_fs();
2816 struct timespec kts;
2820 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&kts);
2823 err = compat_put_timespec(&kts, up);
2828 static int compat_dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
2830 struct compat_ifconf ifc32;
2834 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
2837 ifc.ifc_len = ifc32.ifc_len;
2838 ifc.ifc_req = compat_ptr(ifc32.ifcbuf);
2841 err = dev_ifconf(net, &ifc, sizeof(struct compat_ifreq));
2846 ifc32.ifc_len = ifc.ifc_len;
2847 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
2853 static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
2855 struct compat_ethtool_rxnfc __user *compat_rxnfc;
2856 bool convert_in = false, convert_out = false;
2857 size_t buf_size = 0;
2858 struct ethtool_rxnfc __user *rxnfc = NULL;
2860 u32 rule_cnt = 0, actual_rule_cnt;
2865 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2868 compat_rxnfc = compat_ptr(data);
2870 if (get_user(ethcmd, &compat_rxnfc->cmd))
2873 /* Most ethtool structures are defined without padding.
2874 * Unfortunately struct ethtool_rxnfc is an exception.
2879 case ETHTOOL_GRXCLSRLALL:
2880 /* Buffer size is variable */
2881 if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
2883 if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
2885 buf_size += rule_cnt * sizeof(u32);
2887 case ETHTOOL_GRXRINGS:
2888 case ETHTOOL_GRXCLSRLCNT:
2889 case ETHTOOL_GRXCLSRULE:
2890 case ETHTOOL_SRXCLSRLINS:
2893 case ETHTOOL_SRXCLSRLDEL:
2894 buf_size += sizeof(struct ethtool_rxnfc);
2896 rxnfc = compat_alloc_user_space(buf_size);
2900 if (copy_from_user(&ifr.ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2903 ifr.ifr_data = convert_in ? rxnfc : (void __user *)compat_rxnfc;
2906 /* We expect there to be holes between fs.m_ext and
2907 * fs.ring_cookie and at the end of fs, but nowhere else.
2909 BUILD_BUG_ON(offsetof(struct compat_ethtool_rxnfc, fs.m_ext) +
2910 sizeof(compat_rxnfc->fs.m_ext) !=
2911 offsetof(struct ethtool_rxnfc, fs.m_ext) +
2912 sizeof(rxnfc->fs.m_ext));
2914 offsetof(struct compat_ethtool_rxnfc, fs.location) -
2915 offsetof(struct compat_ethtool_rxnfc, fs.ring_cookie) !=
2916 offsetof(struct ethtool_rxnfc, fs.location) -
2917 offsetof(struct ethtool_rxnfc, fs.ring_cookie));
2919 if (copy_in_user(rxnfc, compat_rxnfc,
2920 (void __user *)(&rxnfc->fs.m_ext + 1) -
2921 (void __user *)rxnfc) ||
2922 copy_in_user(&rxnfc->fs.ring_cookie,
2923 &compat_rxnfc->fs.ring_cookie,
2924 (void __user *)(&rxnfc->fs.location + 1) -
2925 (void __user *)&rxnfc->fs.ring_cookie))
2927 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
2928 if (put_user(rule_cnt, &rxnfc->rule_cnt))
2930 } else if (copy_in_user(&rxnfc->rule_cnt,
2931 &compat_rxnfc->rule_cnt,
2932 sizeof(rxnfc->rule_cnt)))
2936 ret = dev_ioctl(net, SIOCETHTOOL, &ifr, NULL);
2941 if (copy_in_user(compat_rxnfc, rxnfc,
2942 (const void __user *)(&rxnfc->fs.m_ext + 1) -
2943 (const void __user *)rxnfc) ||
2944 copy_in_user(&compat_rxnfc->fs.ring_cookie,
2945 &rxnfc->fs.ring_cookie,
2946 (const void __user *)(&rxnfc->fs.location + 1) -
2947 (const void __user *)&rxnfc->fs.ring_cookie) ||
2948 copy_in_user(&compat_rxnfc->rule_cnt, &rxnfc->rule_cnt,
2949 sizeof(rxnfc->rule_cnt)))
2952 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
2953 /* As an optimisation, we only copy the actual
2954 * number of rules that the underlying
2955 * function returned. Since Mallory might
2956 * change the rule count in user memory, we
2957 * check that it is less than the rule count
2958 * originally given (as the user buffer size),
2959 * which has been range-checked.
2961 if (get_user(actual_rule_cnt, &rxnfc->rule_cnt))
2963 if (actual_rule_cnt < rule_cnt)
2964 rule_cnt = actual_rule_cnt;
2965 if (copy_in_user(&compat_rxnfc->rule_locs[0],
2966 &rxnfc->rule_locs[0],
2967 rule_cnt * sizeof(u32)))
2975 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
2977 compat_uptr_t uptr32;
2982 if (copy_from_user(&ifr, uifr32, sizeof(struct compat_ifreq)))
2985 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
2988 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
2989 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
2991 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL);
2993 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
2994 if (copy_to_user(uifr32, &ifr, sizeof(struct compat_ifreq)))
3000 /* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3001 static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3002 struct compat_ifreq __user *u_ifreq32)
3007 if (copy_from_user(ifreq.ifr_name, u_ifreq32->ifr_name, IFNAMSIZ))
3009 if (get_user(data32, &u_ifreq32->ifr_data))
3011 ifreq.ifr_data = compat_ptr(data32);
3013 return dev_ioctl(net, cmd, &ifreq, NULL);
3016 static int compat_ifreq_ioctl(struct net *net, struct socket *sock,
3018 struct compat_ifreq __user *uifr32)
3020 struct ifreq __user *uifr;
3023 /* Handle the fact that while struct ifreq has the same *layout* on
3024 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3025 * which are handled elsewhere, it still has different *size* due to
3026 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3027 * resulting in struct ifreq being 32 and 40 bytes respectively).
3028 * As a result, if the struct happens to be at the end of a page and
3029 * the next page isn't readable/writable, we get a fault. To prevent
3030 * that, copy back and forth to the full size.
3033 uifr = compat_alloc_user_space(sizeof(*uifr));
3034 if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
3037 err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
3048 case SIOCGIFBRDADDR:
3049 case SIOCGIFDSTADDR:
3050 case SIOCGIFNETMASK:
3056 if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
3064 static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
3065 struct compat_ifreq __user *uifr32)
3068 struct compat_ifmap __user *uifmap32;
3071 uifmap32 = &uifr32->ifr_ifru.ifru_map;
3072 err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
3073 err |= get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3074 err |= get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3075 err |= get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3076 err |= get_user(ifr.ifr_map.irq, &uifmap32->irq);
3077 err |= get_user(ifr.ifr_map.dma, &uifmap32->dma);
3078 err |= get_user(ifr.ifr_map.port, &uifmap32->port);
3082 err = dev_ioctl(net, cmd, &ifr, NULL);
3084 if (cmd == SIOCGIFMAP && !err) {
3085 err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
3086 err |= put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3087 err |= put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3088 err |= put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3089 err |= put_user(ifr.ifr_map.irq, &uifmap32->irq);
3090 err |= put_user(ifr.ifr_map.dma, &uifmap32->dma);
3091 err |= put_user(ifr.ifr_map.port, &uifmap32->port);
3100 struct sockaddr rt_dst; /* target address */
3101 struct sockaddr rt_gateway; /* gateway addr (RTF_GATEWAY) */
3102 struct sockaddr rt_genmask; /* target network mask (IP) */
3103 unsigned short rt_flags;
3106 unsigned char rt_tos;
3107 unsigned char rt_class;
3109 short rt_metric; /* +1 for binary compatibility! */
3110 /* char * */ u32 rt_dev; /* forcing the device at add */
3111 u32 rt_mtu; /* per route MTU/Window */
3112 u32 rt_window; /* Window clamping */
3113 unsigned short rt_irtt; /* Initial RTT */
3116 struct in6_rtmsg32 {
3117 struct in6_addr rtmsg_dst;
3118 struct in6_addr rtmsg_src;
3119 struct in6_addr rtmsg_gateway;
3129 static int routing_ioctl(struct net *net, struct socket *sock,
3130 unsigned int cmd, void __user *argp)
3134 struct in6_rtmsg r6;
3138 mm_segment_t old_fs = get_fs();
3140 if (sock && sock->sk && sock->sk->sk_family == AF_INET6) { /* ipv6 */
3141 struct in6_rtmsg32 __user *ur6 = argp;
3142 ret = copy_from_user(&r6.rtmsg_dst, &(ur6->rtmsg_dst),
3143 3 * sizeof(struct in6_addr));
3144 ret |= get_user(r6.rtmsg_type, &(ur6->rtmsg_type));
3145 ret |= get_user(r6.rtmsg_dst_len, &(ur6->rtmsg_dst_len));
3146 ret |= get_user(r6.rtmsg_src_len, &(ur6->rtmsg_src_len));
3147 ret |= get_user(r6.rtmsg_metric, &(ur6->rtmsg_metric));
3148 ret |= get_user(r6.rtmsg_info, &(ur6->rtmsg_info));
3149 ret |= get_user(r6.rtmsg_flags, &(ur6->rtmsg_flags));
3150 ret |= get_user(r6.rtmsg_ifindex, &(ur6->rtmsg_ifindex));
3154 struct rtentry32 __user *ur4 = argp;
3155 ret = copy_from_user(&r4.rt_dst, &(ur4->rt_dst),
3156 3 * sizeof(struct sockaddr));
3157 ret |= get_user(r4.rt_flags, &(ur4->rt_flags));
3158 ret |= get_user(r4.rt_metric, &(ur4->rt_metric));
3159 ret |= get_user(r4.rt_mtu, &(ur4->rt_mtu));
3160 ret |= get_user(r4.rt_window, &(ur4->rt_window));
3161 ret |= get_user(r4.rt_irtt, &(ur4->rt_irtt));
3162 ret |= get_user(rtdev, &(ur4->rt_dev));
3164 ret |= copy_from_user(devname, compat_ptr(rtdev), 15);
3165 r4.rt_dev = (char __user __force *)devname;
3179 ret = sock_do_ioctl(net, sock, cmd, (unsigned long) r);
3186 /* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3187 * for some operations; this forces use of the newer bridge-utils that
3188 * use compatible ioctls
3190 static int old_bridge_ioctl(compat_ulong_t __user *argp)
3194 if (get_user(tmp, argp))
3196 if (tmp == BRCTL_GET_VERSION)
3197 return BRCTL_VERSION + 1;
3201 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3202 unsigned int cmd, unsigned long arg)
3204 void __user *argp = compat_ptr(arg);
3205 struct sock *sk = sock->sk;
3206 struct net *net = sock_net(sk);
3208 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3209 return compat_ifr_data_ioctl(net, cmd, argp);
3214 return old_bridge_ioctl(argp);
3216 return compat_dev_ifconf(net, argp);
3218 return ethtool_ioctl(net, argp);
3220 return compat_siocwandev(net, argp);
3223 return compat_sioc_ifmap(net, cmd, argp);
3226 return routing_ioctl(net, sock, cmd, argp);
3228 return do_siocgstamp(net, sock, cmd, argp);
3230 return do_siocgstampns(net, sock, cmd, argp);
3231 case SIOCBONDSLAVEINFOQUERY:
3232 case SIOCBONDINFOQUERY:
3235 return compat_ifr_data_ioctl(net, cmd, argp);
3248 return sock_ioctl(file, cmd, arg);
3265 case SIOCSIFHWBROADCAST:
3267 case SIOCGIFBRDADDR:
3268 case SIOCSIFBRDADDR:
3269 case SIOCGIFDSTADDR:
3270 case SIOCSIFDSTADDR:
3271 case SIOCGIFNETMASK:
3272 case SIOCSIFNETMASK:
3284 case SIOCBONDENSLAVE:
3285 case SIOCBONDRELEASE:
3286 case SIOCBONDSETHWADDR:
3287 case SIOCBONDCHANGEACTIVE:
3288 return compat_ifreq_ioctl(net, sock, cmd, argp);
3294 return sock_do_ioctl(net, sock, cmd, arg);
3297 return -ENOIOCTLCMD;
3300 static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3303 struct socket *sock = file->private_data;
3304 int ret = -ENOIOCTLCMD;
3311 if (sock->ops->compat_ioctl)
3312 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3314 if (ret == -ENOIOCTLCMD &&
3315 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3316 ret = compat_wext_handle_ioctl(net, cmd, arg);
3318 if (ret == -ENOIOCTLCMD)
3319 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3325 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3327 return sock->ops->bind(sock, addr, addrlen);
3329 EXPORT_SYMBOL(kernel_bind);
3331 int kernel_listen(struct socket *sock, int backlog)
3333 return sock->ops->listen(sock, backlog);
3335 EXPORT_SYMBOL(kernel_listen);
3337 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3339 struct sock *sk = sock->sk;
3342 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3347 err = sock->ops->accept(sock, *newsock, flags, true);
3349 sock_release(*newsock);
3354 (*newsock)->ops = sock->ops;
3355 __module_get((*newsock)->ops->owner);
3360 EXPORT_SYMBOL(kernel_accept);
3362 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3365 return sock->ops->connect(sock, addr, addrlen, flags);
3367 EXPORT_SYMBOL(kernel_connect);
3369 int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3371 return sock->ops->getname(sock, addr, 0);
3373 EXPORT_SYMBOL(kernel_getsockname);
3375 int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3377 return sock->ops->getname(sock, addr, 1);
3379 EXPORT_SYMBOL(kernel_getpeername);
3381 int kernel_getsockopt(struct socket *sock, int level, int optname,
3382 char *optval, int *optlen)
3384 mm_segment_t oldfs = get_fs();
3385 char __user *uoptval;
3386 int __user *uoptlen;
3389 uoptval = (char __user __force *) optval;
3390 uoptlen = (int __user __force *) optlen;
3393 if (level == SOL_SOCKET)
3394 err = sock_getsockopt(sock, level, optname, uoptval, uoptlen);
3396 err = sock->ops->getsockopt(sock, level, optname, uoptval,
3401 EXPORT_SYMBOL(kernel_getsockopt);
3403 int kernel_setsockopt(struct socket *sock, int level, int optname,
3404 char *optval, unsigned int optlen)
3406 mm_segment_t oldfs = get_fs();
3407 char __user *uoptval;
3410 uoptval = (char __user __force *) optval;
3413 if (level == SOL_SOCKET)
3414 err = sock_setsockopt(sock, level, optname, uoptval, optlen);
3416 err = sock->ops->setsockopt(sock, level, optname, uoptval,
3421 EXPORT_SYMBOL(kernel_setsockopt);
3423 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3424 size_t size, int flags)
3426 if (sock->ops->sendpage)
3427 return sock->ops->sendpage(sock, page, offset, size, flags);
3429 return sock_no_sendpage(sock, page, offset, size, flags);
3431 EXPORT_SYMBOL(kernel_sendpage);
3433 int kernel_sendpage_locked(struct sock *sk, struct page *page, int offset,
3434 size_t size, int flags)
3436 struct socket *sock = sk->sk_socket;
3438 if (sock->ops->sendpage_locked)
3439 return sock->ops->sendpage_locked(sk, page, offset, size,
3442 return sock_no_sendpage_locked(sk, page, offset, size, flags);
3444 EXPORT_SYMBOL(kernel_sendpage_locked);
3446 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3448 return sock->ops->shutdown(sock, how);
3450 EXPORT_SYMBOL(kernel_sock_shutdown);
3452 /* This routine returns the IP overhead imposed by a socket i.e.
3453 * the length of the underlying IP header, depending on whether
3454 * this is an IPv4 or IPv6 socket and the length from IP options turned
3455 * on at the socket. Assumes that the caller has a lock on the socket.
3457 u32 kernel_sock_ip_overhead(struct sock *sk)
3459 struct inet_sock *inet;
3460 struct ip_options_rcu *opt;
3462 #if IS_ENABLED(CONFIG_IPV6)
3463 struct ipv6_pinfo *np;
3464 struct ipv6_txoptions *optv6 = NULL;
3465 #endif /* IS_ENABLED(CONFIG_IPV6) */
3470 switch (sk->sk_family) {
3473 overhead += sizeof(struct iphdr);
3474 opt = rcu_dereference_protected(inet->inet_opt,
3475 sock_owned_by_user(sk));
3477 overhead += opt->opt.optlen;
3479 #if IS_ENABLED(CONFIG_IPV6)
3482 overhead += sizeof(struct ipv6hdr);
3484 optv6 = rcu_dereference_protected(np->opt,
3485 sock_owned_by_user(sk));
3487 overhead += (optv6->opt_flen + optv6->opt_nflen);
3489 #endif /* IS_ENABLED(CONFIG_IPV6) */
3490 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3494 EXPORT_SYMBOL(kernel_sock_ip_overhead);