1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET An implementation of the SOCKET network access protocol.
5 * Version: @(#)socket.c 1.1.93 18/02/95
7 * Authors: Orest Zborowski, <obz@Kodak.COM>
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
14 * Alan Cox : verify_area() fixes
15 * Alan Cox : Removed DDI
16 * Jonathan Kamens : SOCK_DGRAM reconnect bug
17 * Alan Cox : Moved a load of checks to the very
19 * Alan Cox : Move address structures to/from user
20 * mode above the protocol layers.
21 * Rob Janssen : Allow 0 length sends.
22 * Alan Cox : Asynchronous I/O support (cribbed from the
24 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
25 * Jeff Uphoff : Made max number of sockets command-line
27 * Matti Aarnio : Made the number of sockets dynamic,
28 * to be allocated when needed, and mr.
29 * Uphoff's max is used as max to be
30 * allowed to allocate.
31 * Linus : Argh. removed all the socket allocation
32 * altogether: it's in the inode now.
33 * Alan Cox : Made sock_alloc()/sock_release() public
34 * for NetROM and future kernel nfsd type
36 * Alan Cox : sendmsg/recvmsg basics.
37 * Tom Dyas : Export net symbols.
38 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
39 * Alan Cox : Added thread locking to sys_* calls
40 * for sockets. May have errors at the
42 * Kevin Buhr : Fixed the dumb errors in the above.
43 * Andi Kleen : Some small cleanups, optimizations,
44 * and fixed a copy_from_user() bug.
45 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
46 * Tigran Aivazian : Made listen(2) backlog sanity checks
47 * protocol-independent
49 * This module is effectively the top level interface to the BSD socket
52 * Based upon Swansea University Computer Society NET3.039
55 #include <linux/ethtool.h>
57 #include <linux/socket.h>
58 #include <linux/file.h>
59 #include <linux/net.h>
60 #include <linux/interrupt.h>
61 #include <linux/thread_info.h>
62 #include <linux/rcupdate.h>
63 #include <linux/netdevice.h>
64 #include <linux/proc_fs.h>
65 #include <linux/seq_file.h>
66 #include <linux/mutex.h>
67 #include <linux/if_bridge.h>
68 #include <linux/if_vlan.h>
69 #include <linux/ptp_classify.h>
70 #include <linux/init.h>
71 #include <linux/poll.h>
72 #include <linux/cache.h>
73 #include <linux/module.h>
74 #include <linux/highmem.h>
75 #include <linux/mount.h>
76 #include <linux/pseudo_fs.h>
77 #include <linux/security.h>
78 #include <linux/syscalls.h>
79 #include <linux/compat.h>
80 #include <linux/kmod.h>
81 #include <linux/audit.h>
82 #include <linux/wireless.h>
83 #include <linux/nsproxy.h>
84 #include <linux/magic.h>
85 #include <linux/slab.h>
86 #include <linux/xattr.h>
87 #include <linux/nospec.h>
88 #include <linux/indirect_call_wrapper.h>
90 #include <linux/uaccess.h>
91 #include <asm/unistd.h>
93 #include <net/compat.h>
95 #include <net/cls_cgroup.h>
98 #include <linux/netfilter.h>
100 #include <linux/if_tun.h>
101 #include <linux/ipv6_route.h>
102 #include <linux/route.h>
103 #include <linux/termios.h>
104 #include <linux/sockios.h>
105 #include <net/busy_poll.h>
106 #include <linux/errqueue.h>
107 #include <linux/ptp_clock_kernel.h>
109 #ifdef CONFIG_NET_RX_BUSY_POLL
110 unsigned int sysctl_net_busy_read __read_mostly;
111 unsigned int sysctl_net_busy_poll __read_mostly;
114 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
115 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
116 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
118 static int sock_close(struct inode *inode, struct file *file);
119 static __poll_t sock_poll(struct file *file,
120 struct poll_table_struct *wait);
121 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
123 static long compat_sock_ioctl(struct file *file,
124 unsigned int cmd, unsigned long arg);
126 static int sock_fasync(int fd, struct file *filp, int on);
127 static ssize_t sock_sendpage(struct file *file, struct page *page,
128 int offset, size_t size, loff_t *ppos, int more);
129 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
130 struct pipe_inode_info *pipe, size_t len,
133 #ifdef CONFIG_PROC_FS
134 static void sock_show_fdinfo(struct seq_file *m, struct file *f)
136 struct socket *sock = f->private_data;
138 if (sock->ops->show_fdinfo)
139 sock->ops->show_fdinfo(m, sock);
142 #define sock_show_fdinfo NULL
146 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
147 * in the operation structures but are done directly via the socketcall() multiplexor.
150 static const struct file_operations socket_file_ops = {
151 .owner = THIS_MODULE,
153 .read_iter = sock_read_iter,
154 .write_iter = sock_write_iter,
156 .unlocked_ioctl = sock_ioctl,
158 .compat_ioctl = compat_sock_ioctl,
161 .release = sock_close,
162 .fasync = sock_fasync,
163 .sendpage = sock_sendpage,
164 .splice_write = generic_splice_sendpage,
165 .splice_read = sock_splice_read,
166 .show_fdinfo = sock_show_fdinfo,
169 static const char * const pf_family_names[] = {
170 [PF_UNSPEC] = "PF_UNSPEC",
171 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
172 [PF_INET] = "PF_INET",
173 [PF_AX25] = "PF_AX25",
175 [PF_APPLETALK] = "PF_APPLETALK",
176 [PF_NETROM] = "PF_NETROM",
177 [PF_BRIDGE] = "PF_BRIDGE",
178 [PF_ATMPVC] = "PF_ATMPVC",
180 [PF_INET6] = "PF_INET6",
181 [PF_ROSE] = "PF_ROSE",
182 [PF_DECnet] = "PF_DECnet",
183 [PF_NETBEUI] = "PF_NETBEUI",
184 [PF_SECURITY] = "PF_SECURITY",
186 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
187 [PF_PACKET] = "PF_PACKET",
189 [PF_ECONET] = "PF_ECONET",
190 [PF_ATMSVC] = "PF_ATMSVC",
193 [PF_IRDA] = "PF_IRDA",
194 [PF_PPPOX] = "PF_PPPOX",
195 [PF_WANPIPE] = "PF_WANPIPE",
198 [PF_MPLS] = "PF_MPLS",
200 [PF_TIPC] = "PF_TIPC",
201 [PF_BLUETOOTH] = "PF_BLUETOOTH",
202 [PF_IUCV] = "PF_IUCV",
203 [PF_RXRPC] = "PF_RXRPC",
204 [PF_ISDN] = "PF_ISDN",
205 [PF_PHONET] = "PF_PHONET",
206 [PF_IEEE802154] = "PF_IEEE802154",
207 [PF_CAIF] = "PF_CAIF",
210 [PF_VSOCK] = "PF_VSOCK",
212 [PF_QIPCRTR] = "PF_QIPCRTR",
215 [PF_MCTP] = "PF_MCTP",
219 * The protocol list. Each protocol is registered in here.
222 static DEFINE_SPINLOCK(net_family_lock);
223 static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
227 * Move socket addresses back and forth across the kernel/user
228 * divide and look after the messy bits.
232 * move_addr_to_kernel - copy a socket address into kernel space
233 * @uaddr: Address in user space
234 * @kaddr: Address in kernel space
235 * @ulen: Length in user space
237 * The address is copied into kernel space. If the provided address is
238 * too long an error code of -EINVAL is returned. If the copy gives
239 * invalid addresses -EFAULT is returned. On a success 0 is returned.
242 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
244 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
248 if (copy_from_user(kaddr, uaddr, ulen))
250 return audit_sockaddr(ulen, kaddr);
254 * move_addr_to_user - copy an address to user space
255 * @kaddr: kernel space address
256 * @klen: length of address in kernel
257 * @uaddr: user space address
258 * @ulen: pointer to user length field
260 * The value pointed to by ulen on entry is the buffer length available.
261 * This is overwritten with the buffer space used. -EINVAL is returned
262 * if an overlong buffer is specified or a negative buffer size. -EFAULT
263 * is returned if either the buffer or the length field are not
265 * After copying the data up to the limit the user specifies, the true
266 * length of the data is written over the length limit the user
267 * specified. Zero is returned for a success.
270 static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
271 void __user *uaddr, int __user *ulen)
276 BUG_ON(klen > sizeof(struct sockaddr_storage));
277 err = get_user(len, ulen);
285 if (audit_sockaddr(klen, kaddr))
287 if (copy_to_user(uaddr, kaddr, len))
291 * "fromlen shall refer to the value before truncation.."
294 return __put_user(klen, ulen);
297 static struct kmem_cache *sock_inode_cachep __ro_after_init;
299 static struct inode *sock_alloc_inode(struct super_block *sb)
301 struct socket_alloc *ei;
303 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
306 init_waitqueue_head(&ei->socket.wq.wait);
307 ei->socket.wq.fasync_list = NULL;
308 ei->socket.wq.flags = 0;
310 ei->socket.state = SS_UNCONNECTED;
311 ei->socket.flags = 0;
312 ei->socket.ops = NULL;
313 ei->socket.sk = NULL;
314 ei->socket.file = NULL;
316 return &ei->vfs_inode;
319 static void sock_free_inode(struct inode *inode)
321 struct socket_alloc *ei;
323 ei = container_of(inode, struct socket_alloc, vfs_inode);
324 kmem_cache_free(sock_inode_cachep, ei);
327 static void init_once(void *foo)
329 struct socket_alloc *ei = (struct socket_alloc *)foo;
331 inode_init_once(&ei->vfs_inode);
334 static void init_inodecache(void)
336 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
337 sizeof(struct socket_alloc),
339 (SLAB_HWCACHE_ALIGN |
340 SLAB_RECLAIM_ACCOUNT |
341 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
343 BUG_ON(sock_inode_cachep == NULL);
346 static const struct super_operations sockfs_ops = {
347 .alloc_inode = sock_alloc_inode,
348 .free_inode = sock_free_inode,
349 .statfs = simple_statfs,
353 * sockfs_dname() is called from d_path().
355 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
357 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
358 d_inode(dentry)->i_ino);
361 static const struct dentry_operations sockfs_dentry_operations = {
362 .d_dname = sockfs_dname,
365 static int sockfs_xattr_get(const struct xattr_handler *handler,
366 struct dentry *dentry, struct inode *inode,
367 const char *suffix, void *value, size_t size)
370 if (dentry->d_name.len + 1 > size)
372 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
374 return dentry->d_name.len + 1;
377 #define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
378 #define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
379 #define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
381 static const struct xattr_handler sockfs_xattr_handler = {
382 .name = XATTR_NAME_SOCKPROTONAME,
383 .get = sockfs_xattr_get,
386 static int sockfs_security_xattr_set(const struct xattr_handler *handler,
387 struct user_namespace *mnt_userns,
388 struct dentry *dentry, struct inode *inode,
389 const char *suffix, const void *value,
390 size_t size, int flags)
392 /* Handled by LSM. */
396 static const struct xattr_handler sockfs_security_xattr_handler = {
397 .prefix = XATTR_SECURITY_PREFIX,
398 .set = sockfs_security_xattr_set,
401 static const struct xattr_handler *sockfs_xattr_handlers[] = {
402 &sockfs_xattr_handler,
403 &sockfs_security_xattr_handler,
407 static int sockfs_init_fs_context(struct fs_context *fc)
409 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
412 ctx->ops = &sockfs_ops;
413 ctx->dops = &sockfs_dentry_operations;
414 ctx->xattr = sockfs_xattr_handlers;
418 static struct vfsmount *sock_mnt __read_mostly;
420 static struct file_system_type sock_fs_type = {
422 .init_fs_context = sockfs_init_fs_context,
423 .kill_sb = kill_anon_super,
427 * Obtains the first available file descriptor and sets it up for use.
429 * These functions create file structures and maps them to fd space
430 * of the current process. On success it returns file descriptor
431 * and file struct implicitly stored in sock->file.
432 * Note that another thread may close file descriptor before we return
433 * from this function. We use the fact that now we do not refer
434 * to socket after mapping. If one day we will need it, this
435 * function will increment ref. count on file by 1.
437 * In any case returned fd MAY BE not valid!
438 * This race condition is unavoidable
439 * with shared fd spaces, we cannot solve it inside kernel,
440 * but we take care of internal coherence yet.
444 * sock_alloc_file - Bind a &socket to a &file
446 * @flags: file status flags
447 * @dname: protocol name
449 * Returns the &file bound with @sock, implicitly storing it
450 * in sock->file. If dname is %NULL, sets to "".
451 * On failure the return is a ERR pointer (see linux/err.h).
452 * This function uses GFP_KERNEL internally.
455 struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
460 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
462 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
463 O_RDWR | (flags & O_NONBLOCK),
471 file->private_data = sock;
472 stream_open(SOCK_INODE(sock), file);
475 EXPORT_SYMBOL(sock_alloc_file);
477 static int sock_map_fd(struct socket *sock, int flags)
479 struct file *newfile;
480 int fd = get_unused_fd_flags(flags);
481 if (unlikely(fd < 0)) {
486 newfile = sock_alloc_file(sock, flags, NULL);
487 if (!IS_ERR(newfile)) {
488 fd_install(fd, newfile);
493 return PTR_ERR(newfile);
497 * sock_from_file - Return the &socket bounded to @file.
500 * On failure returns %NULL.
503 struct socket *sock_from_file(struct file *file)
505 if (file->f_op == &socket_file_ops)
506 return file->private_data; /* set in sock_map_fd */
510 EXPORT_SYMBOL(sock_from_file);
513 * sockfd_lookup - Go from a file number to its socket slot
515 * @err: pointer to an error code return
517 * The file handle passed in is locked and the socket it is bound
518 * to is returned. If an error occurs the err pointer is overwritten
519 * with a negative errno code and NULL is returned. The function checks
520 * for both invalid handles and passing a handle which is not a socket.
522 * On a success the socket object pointer is returned.
525 struct socket *sockfd_lookup(int fd, int *err)
536 sock = sock_from_file(file);
543 EXPORT_SYMBOL(sockfd_lookup);
545 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
547 struct fd f = fdget(fd);
552 sock = sock_from_file(f.file);
554 *fput_needed = f.flags & FDPUT_FPUT;
563 static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
569 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
579 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
584 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
591 static int sockfs_setattr(struct user_namespace *mnt_userns,
592 struct dentry *dentry, struct iattr *iattr)
594 int err = simple_setattr(&init_user_ns, dentry, iattr);
596 if (!err && (iattr->ia_valid & ATTR_UID)) {
597 struct socket *sock = SOCKET_I(d_inode(dentry));
600 sock->sk->sk_uid = iattr->ia_uid;
608 static const struct inode_operations sockfs_inode_ops = {
609 .listxattr = sockfs_listxattr,
610 .setattr = sockfs_setattr,
614 * sock_alloc - allocate a socket
616 * Allocate a new inode and socket object. The two are bound together
617 * and initialised. The socket is then returned. If we are out of inodes
618 * NULL is returned. This functions uses GFP_KERNEL internally.
621 struct socket *sock_alloc(void)
626 inode = new_inode_pseudo(sock_mnt->mnt_sb);
630 sock = SOCKET_I(inode);
632 inode->i_ino = get_next_ino();
633 inode->i_mode = S_IFSOCK | S_IRWXUGO;
634 inode->i_uid = current_fsuid();
635 inode->i_gid = current_fsgid();
636 inode->i_op = &sockfs_inode_ops;
640 EXPORT_SYMBOL(sock_alloc);
642 static void __sock_release(struct socket *sock, struct inode *inode)
645 struct module *owner = sock->ops->owner;
649 sock->ops->release(sock);
657 if (sock->wq.fasync_list)
658 pr_err("%s: fasync list not empty!\n", __func__);
661 iput(SOCK_INODE(sock));
668 * sock_release - close a socket
669 * @sock: socket to close
671 * The socket is released from the protocol stack if it has a release
672 * callback, and the inode is then released if the socket is bound to
673 * an inode not a file.
675 void sock_release(struct socket *sock)
677 __sock_release(sock, NULL);
679 EXPORT_SYMBOL(sock_release);
681 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
683 u8 flags = *tx_flags;
685 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
686 flags |= SKBTX_HW_TSTAMP;
688 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
689 flags |= SKBTX_SW_TSTAMP;
691 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
692 flags |= SKBTX_SCHED_TSTAMP;
696 EXPORT_SYMBOL(__sock_tx_timestamp);
698 INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
700 INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
702 static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
704 int ret = INDIRECT_CALL_INET(sock->ops->sendmsg, inet6_sendmsg,
705 inet_sendmsg, sock, msg,
707 BUG_ON(ret == -EIOCBQUEUED);
711 static int __sock_sendmsg(struct socket *sock, struct msghdr *msg)
713 int err = security_socket_sendmsg(sock, msg,
716 return err ?: sock_sendmsg_nosec(sock, msg);
720 * sock_sendmsg - send a message through @sock
722 * @msg: message to send
724 * Sends @msg through @sock, passing through LSM.
725 * Returns the number of bytes sent, or an error code.
727 int sock_sendmsg(struct socket *sock, struct msghdr *msg)
729 struct sockaddr_storage *save_addr = (struct sockaddr_storage *)msg->msg_name;
730 struct sockaddr_storage address;
734 memcpy(&address, msg->msg_name, msg->msg_namelen);
735 msg->msg_name = &address;
738 ret = __sock_sendmsg(sock, msg);
739 msg->msg_name = save_addr;
743 EXPORT_SYMBOL(sock_sendmsg);
746 * kernel_sendmsg - send a message through @sock (kernel-space)
748 * @msg: message header
750 * @num: vec array length
751 * @size: total message data size
753 * Builds the message data with @vec and sends it through @sock.
754 * Returns the number of bytes sent, or an error code.
757 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
758 struct kvec *vec, size_t num, size_t size)
760 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
761 return sock_sendmsg(sock, msg);
763 EXPORT_SYMBOL(kernel_sendmsg);
766 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
768 * @msg: message header
769 * @vec: output s/g array
770 * @num: output s/g array length
771 * @size: total message data size
773 * Builds the message data with @vec and sends it through @sock.
774 * Returns the number of bytes sent, or an error code.
775 * Caller must hold @sk.
778 int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
779 struct kvec *vec, size_t num, size_t size)
781 struct socket *sock = sk->sk_socket;
783 if (!sock->ops->sendmsg_locked)
784 return sock_no_sendmsg_locked(sk, msg, size);
786 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
788 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
790 EXPORT_SYMBOL(kernel_sendmsg_locked);
792 static bool skb_is_err_queue(const struct sk_buff *skb)
794 /* pkt_type of skbs enqueued on the error queue are set to
795 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
796 * in recvmsg, since skbs received on a local socket will never
797 * have a pkt_type of PACKET_OUTGOING.
799 return skb->pkt_type == PACKET_OUTGOING;
802 /* On transmit, software and hardware timestamps are returned independently.
803 * As the two skb clones share the hardware timestamp, which may be updated
804 * before the software timestamp is received, a hardware TX timestamp may be
805 * returned only if there is no software TX timestamp. Ignore false software
806 * timestamps, which may be made in the __sock_recv_timestamp() call when the
807 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
808 * hardware timestamp.
810 static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
812 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
815 static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
817 struct scm_ts_pktinfo ts_pktinfo;
818 struct net_device *orig_dev;
820 if (!skb_mac_header_was_set(skb))
823 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
826 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
828 ts_pktinfo.if_index = orig_dev->ifindex;
831 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
832 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
833 sizeof(ts_pktinfo), &ts_pktinfo);
837 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
839 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
842 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
843 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
844 struct scm_timestamping_internal tss;
846 int empty = 1, false_tstamp = 0;
847 struct skb_shared_hwtstamps *shhwtstamps =
851 /* Race occurred between timestamp enabling and packet
852 receiving. Fill in the current time for now. */
853 if (need_software_tstamp && skb->tstamp == 0) {
854 __net_timestamp(skb);
858 if (need_software_tstamp) {
859 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
861 struct __kernel_sock_timeval tv;
863 skb_get_new_timestamp(skb, &tv);
864 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
867 struct __kernel_old_timeval tv;
869 skb_get_timestamp(skb, &tv);
870 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
875 struct __kernel_timespec ts;
877 skb_get_new_timestampns(skb, &ts);
878 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
881 struct __kernel_old_timespec ts;
883 skb_get_timestampns(skb, &ts);
884 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
890 memset(&tss, 0, sizeof(tss));
891 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
892 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
895 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
896 !skb_is_swtx_tstamp(skb, false_tstamp)) {
897 if (sk->sk_tsflags & SOF_TIMESTAMPING_BIND_PHC)
898 hwtstamp = ptp_convert_timestamp(shhwtstamps,
901 hwtstamp = shhwtstamps->hwtstamp;
903 if (ktime_to_timespec64_cond(hwtstamp, tss.ts + 2)) {
906 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
907 !skb_is_err_queue(skb))
908 put_ts_pktinfo(msg, skb);
912 if (sock_flag(sk, SOCK_TSTAMP_NEW))
913 put_cmsg_scm_timestamping64(msg, &tss);
915 put_cmsg_scm_timestamping(msg, &tss);
917 if (skb_is_err_queue(skb) && skb->len &&
918 SKB_EXT_ERR(skb)->opt_stats)
919 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
920 skb->len, skb->data);
923 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
925 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
930 if (!sock_flag(sk, SOCK_WIFI_STATUS))
932 if (!skb->wifi_acked_valid)
935 ack = skb->wifi_acked;
937 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
939 EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
941 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
944 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
945 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
946 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
949 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
952 sock_recv_timestamp(msg, sk, skb);
953 sock_recv_drops(msg, sk, skb);
955 EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
957 INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
959 INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
961 static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
964 return INDIRECT_CALL_INET(sock->ops->recvmsg, inet6_recvmsg,
965 inet_recvmsg, sock, msg, msg_data_left(msg),
970 * sock_recvmsg - receive a message from @sock
972 * @msg: message to receive
973 * @flags: message flags
975 * Receives @msg from @sock, passing through LSM. Returns the total number
976 * of bytes received, or an error.
978 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
980 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
982 return err ?: sock_recvmsg_nosec(sock, msg, flags);
984 EXPORT_SYMBOL(sock_recvmsg);
987 * kernel_recvmsg - Receive a message from a socket (kernel space)
988 * @sock: The socket to receive the message from
989 * @msg: Received message
990 * @vec: Input s/g array for message data
991 * @num: Size of input s/g array
992 * @size: Number of bytes to read
993 * @flags: Message flags (MSG_DONTWAIT, etc...)
995 * On return the msg structure contains the scatter/gather array passed in the
996 * vec argument. The array is modified so that it consists of the unfilled
997 * portion of the original array.
999 * The returned value is the total number of bytes received, or an error.
1002 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
1003 struct kvec *vec, size_t num, size_t size, int flags)
1005 msg->msg_control_is_user = false;
1006 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
1007 return sock_recvmsg(sock, msg, flags);
1009 EXPORT_SYMBOL(kernel_recvmsg);
1011 static ssize_t sock_sendpage(struct file *file, struct page *page,
1012 int offset, size_t size, loff_t *ppos, int more)
1014 struct socket *sock;
1017 sock = file->private_data;
1019 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
1020 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
1023 return kernel_sendpage(sock, page, offset, size, flags);
1026 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1027 struct pipe_inode_info *pipe, size_t len,
1030 struct socket *sock = file->private_data;
1032 if (unlikely(!sock->ops->splice_read))
1033 return generic_file_splice_read(file, ppos, pipe, len, flags);
1035 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
1038 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1040 struct file *file = iocb->ki_filp;
1041 struct socket *sock = file->private_data;
1042 struct msghdr msg = {.msg_iter = *to,
1046 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1047 msg.msg_flags = MSG_DONTWAIT;
1049 if (iocb->ki_pos != 0)
1052 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1055 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1060 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1062 struct file *file = iocb->ki_filp;
1063 struct socket *sock = file->private_data;
1064 struct msghdr msg = {.msg_iter = *from,
1068 if (iocb->ki_pos != 0)
1071 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1072 msg.msg_flags = MSG_DONTWAIT;
1074 if (sock->type == SOCK_SEQPACKET)
1075 msg.msg_flags |= MSG_EOR;
1077 res = __sock_sendmsg(sock, &msg);
1078 *from = msg.msg_iter;
1083 * Atomic setting of ioctl hooks to avoid race
1084 * with module unload.
1087 static DEFINE_MUTEX(br_ioctl_mutex);
1088 static int (*br_ioctl_hook)(struct net *net, struct net_bridge *br,
1089 unsigned int cmd, struct ifreq *ifr,
1092 void brioctl_set(int (*hook)(struct net *net, struct net_bridge *br,
1093 unsigned int cmd, struct ifreq *ifr,
1096 mutex_lock(&br_ioctl_mutex);
1097 br_ioctl_hook = hook;
1098 mutex_unlock(&br_ioctl_mutex);
1100 EXPORT_SYMBOL(brioctl_set);
1102 int br_ioctl_call(struct net *net, struct net_bridge *br, unsigned int cmd,
1103 struct ifreq *ifr, void __user *uarg)
1108 request_module("bridge");
1110 mutex_lock(&br_ioctl_mutex);
1112 err = br_ioctl_hook(net, br, cmd, ifr, uarg);
1113 mutex_unlock(&br_ioctl_mutex);
1118 static DEFINE_MUTEX(vlan_ioctl_mutex);
1119 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1121 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1123 mutex_lock(&vlan_ioctl_mutex);
1124 vlan_ioctl_hook = hook;
1125 mutex_unlock(&vlan_ioctl_mutex);
1127 EXPORT_SYMBOL(vlan_ioctl_set);
1129 static long sock_do_ioctl(struct net *net, struct socket *sock,
1130 unsigned int cmd, unsigned long arg)
1135 void __user *argp = (void __user *)arg;
1138 err = sock->ops->ioctl(sock, cmd, arg);
1141 * If this ioctl is unknown try to hand it down
1142 * to the NIC driver.
1144 if (err != -ENOIOCTLCMD)
1147 if (!is_socket_ioctl_cmd(cmd))
1150 if (get_user_ifreq(&ifr, &data, argp))
1152 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1153 if (!err && need_copyout)
1154 if (put_user_ifreq(&ifr, argp))
1161 * With an ioctl, arg may well be a user mode pointer, but we don't know
1162 * what to do with it - that's up to the protocol still.
1165 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1167 struct socket *sock;
1169 void __user *argp = (void __user *)arg;
1173 sock = file->private_data;
1176 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1180 if (get_user_ifreq(&ifr, &data, argp))
1182 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1183 if (!err && need_copyout)
1184 if (put_user_ifreq(&ifr, argp))
1187 #ifdef CONFIG_WEXT_CORE
1188 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1189 err = wext_handle_ioctl(net, cmd, argp);
1196 if (get_user(pid, (int __user *)argp))
1198 err = f_setown(sock->file, pid, 1);
1202 err = put_user(f_getown(sock->file),
1203 (int __user *)argp);
1209 err = br_ioctl_call(net, NULL, cmd, NULL, argp);
1214 if (!vlan_ioctl_hook)
1215 request_module("8021q");
1217 mutex_lock(&vlan_ioctl_mutex);
1218 if (vlan_ioctl_hook)
1219 err = vlan_ioctl_hook(net, argp);
1220 mutex_unlock(&vlan_ioctl_mutex);
1224 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1227 err = open_related_ns(&net->ns, get_net_ns);
1229 case SIOCGSTAMP_OLD:
1230 case SIOCGSTAMPNS_OLD:
1231 if (!sock->ops->gettstamp) {
1235 err = sock->ops->gettstamp(sock, argp,
1236 cmd == SIOCGSTAMP_OLD,
1237 !IS_ENABLED(CONFIG_64BIT));
1239 case SIOCGSTAMP_NEW:
1240 case SIOCGSTAMPNS_NEW:
1241 if (!sock->ops->gettstamp) {
1245 err = sock->ops->gettstamp(sock, argp,
1246 cmd == SIOCGSTAMP_NEW,
1251 err = dev_ifconf(net, argp);
1255 err = sock_do_ioctl(net, sock, cmd, arg);
1262 * sock_create_lite - creates a socket
1263 * @family: protocol family (AF_INET, ...)
1264 * @type: communication type (SOCK_STREAM, ...)
1265 * @protocol: protocol (0, ...)
1268 * Creates a new socket and assigns it to @res, passing through LSM.
1269 * The new socket initialization is not complete, see kernel_accept().
1270 * Returns 0 or an error. On failure @res is set to %NULL.
1271 * This function internally uses GFP_KERNEL.
1274 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1277 struct socket *sock = NULL;
1279 err = security_socket_create(family, type, protocol, 1);
1283 sock = sock_alloc();
1290 err = security_socket_post_create(sock, family, type, protocol, 1);
1302 EXPORT_SYMBOL(sock_create_lite);
1304 /* No kernel lock held - perfect */
1305 static __poll_t sock_poll(struct file *file, poll_table *wait)
1307 struct socket *sock = file->private_data;
1308 __poll_t events = poll_requested_events(wait), flag = 0;
1310 if (!sock->ops->poll)
1313 if (sk_can_busy_loop(sock->sk)) {
1314 /* poll once if requested by the syscall */
1315 if (events & POLL_BUSY_LOOP)
1316 sk_busy_loop(sock->sk, 1);
1318 /* if this socket can poll_ll, tell the system call */
1319 flag = POLL_BUSY_LOOP;
1322 return sock->ops->poll(file, sock, wait) | flag;
1325 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1327 struct socket *sock = file->private_data;
1329 return sock->ops->mmap(file, sock, vma);
1332 static int sock_close(struct inode *inode, struct file *filp)
1334 __sock_release(SOCKET_I(inode), inode);
1339 * Update the socket async list
1341 * Fasync_list locking strategy.
1343 * 1. fasync_list is modified only under process context socket lock
1344 * i.e. under semaphore.
1345 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1346 * or under socket lock
1349 static int sock_fasync(int fd, struct file *filp, int on)
1351 struct socket *sock = filp->private_data;
1352 struct sock *sk = sock->sk;
1353 struct socket_wq *wq = &sock->wq;
1359 fasync_helper(fd, filp, on, &wq->fasync_list);
1361 if (!wq->fasync_list)
1362 sock_reset_flag(sk, SOCK_FASYNC);
1364 sock_set_flag(sk, SOCK_FASYNC);
1370 /* This function may be called only under rcu_lock */
1372 int sock_wake_async(struct socket_wq *wq, int how, int band)
1374 if (!wq || !wq->fasync_list)
1378 case SOCK_WAKE_WAITD:
1379 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1382 case SOCK_WAKE_SPACE:
1383 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1388 kill_fasync(&wq->fasync_list, SIGIO, band);
1391 kill_fasync(&wq->fasync_list, SIGURG, band);
1396 EXPORT_SYMBOL(sock_wake_async);
1399 * __sock_create - creates a socket
1400 * @net: net namespace
1401 * @family: protocol family (AF_INET, ...)
1402 * @type: communication type (SOCK_STREAM, ...)
1403 * @protocol: protocol (0, ...)
1405 * @kern: boolean for kernel space sockets
1407 * Creates a new socket and assigns it to @res, passing through LSM.
1408 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1409 * be set to true if the socket resides in kernel space.
1410 * This function internally uses GFP_KERNEL.
1413 int __sock_create(struct net *net, int family, int type, int protocol,
1414 struct socket **res, int kern)
1417 struct socket *sock;
1418 const struct net_proto_family *pf;
1421 * Check protocol is in range
1423 if (family < 0 || family >= NPROTO)
1424 return -EAFNOSUPPORT;
1425 if (type < 0 || type >= SOCK_MAX)
1430 This uglymoron is moved from INET layer to here to avoid
1431 deadlock in module load.
1433 if (family == PF_INET && type == SOCK_PACKET) {
1434 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1439 err = security_socket_create(family, type, protocol, kern);
1444 * Allocate the socket and allow the family to set things up. if
1445 * the protocol is 0, the family is instructed to select an appropriate
1448 sock = sock_alloc();
1450 net_warn_ratelimited("socket: no more sockets\n");
1451 return -ENFILE; /* Not exactly a match, but its the
1452 closest posix thing */
1457 #ifdef CONFIG_MODULES
1458 /* Attempt to load a protocol module if the find failed.
1460 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1461 * requested real, full-featured networking support upon configuration.
1462 * Otherwise module support will break!
1464 if (rcu_access_pointer(net_families[family]) == NULL)
1465 request_module("net-pf-%d", family);
1469 pf = rcu_dereference(net_families[family]);
1470 err = -EAFNOSUPPORT;
1475 * We will call the ->create function, that possibly is in a loadable
1476 * module, so we have to bump that loadable module refcnt first.
1478 if (!try_module_get(pf->owner))
1481 /* Now protected by module ref count */
1484 err = pf->create(net, sock, protocol, kern);
1486 goto out_module_put;
1489 * Now to bump the refcnt of the [loadable] module that owns this
1490 * socket at sock_release time we decrement its refcnt.
1492 if (!try_module_get(sock->ops->owner))
1493 goto out_module_busy;
1496 * Now that we're done with the ->create function, the [loadable]
1497 * module can have its refcnt decremented
1499 module_put(pf->owner);
1500 err = security_socket_post_create(sock, family, type, protocol, kern);
1502 goto out_sock_release;
1508 err = -EAFNOSUPPORT;
1511 module_put(pf->owner);
1518 goto out_sock_release;
1520 EXPORT_SYMBOL(__sock_create);
1523 * sock_create - creates a socket
1524 * @family: protocol family (AF_INET, ...)
1525 * @type: communication type (SOCK_STREAM, ...)
1526 * @protocol: protocol (0, ...)
1529 * A wrapper around __sock_create().
1530 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1533 int sock_create(int family, int type, int protocol, struct socket **res)
1535 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1537 EXPORT_SYMBOL(sock_create);
1540 * sock_create_kern - creates a socket (kernel space)
1541 * @net: net namespace
1542 * @family: protocol family (AF_INET, ...)
1543 * @type: communication type (SOCK_STREAM, ...)
1544 * @protocol: protocol (0, ...)
1547 * A wrapper around __sock_create().
1548 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1551 int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1553 return __sock_create(net, family, type, protocol, res, 1);
1555 EXPORT_SYMBOL(sock_create_kern);
1557 int __sys_socket(int family, int type, int protocol)
1560 struct socket *sock;
1563 /* Check the SOCK_* constants for consistency. */
1564 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1565 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1566 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1567 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1569 flags = type & ~SOCK_TYPE_MASK;
1570 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1572 type &= SOCK_TYPE_MASK;
1574 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1575 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1577 retval = sock_create(family, type, protocol, &sock);
1581 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1584 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1586 return __sys_socket(family, type, protocol);
1590 * Create a pair of connected sockets.
1593 int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1595 struct socket *sock1, *sock2;
1597 struct file *newfile1, *newfile2;
1600 flags = type & ~SOCK_TYPE_MASK;
1601 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1603 type &= SOCK_TYPE_MASK;
1605 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1606 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1609 * reserve descriptors and make sure we won't fail
1610 * to return them to userland.
1612 fd1 = get_unused_fd_flags(flags);
1613 if (unlikely(fd1 < 0))
1616 fd2 = get_unused_fd_flags(flags);
1617 if (unlikely(fd2 < 0)) {
1622 err = put_user(fd1, &usockvec[0]);
1626 err = put_user(fd2, &usockvec[1]);
1631 * Obtain the first socket and check if the underlying protocol
1632 * supports the socketpair call.
1635 err = sock_create(family, type, protocol, &sock1);
1636 if (unlikely(err < 0))
1639 err = sock_create(family, type, protocol, &sock2);
1640 if (unlikely(err < 0)) {
1641 sock_release(sock1);
1645 err = security_socket_socketpair(sock1, sock2);
1646 if (unlikely(err)) {
1647 sock_release(sock2);
1648 sock_release(sock1);
1652 err = sock1->ops->socketpair(sock1, sock2);
1653 if (unlikely(err < 0)) {
1654 sock_release(sock2);
1655 sock_release(sock1);
1659 newfile1 = sock_alloc_file(sock1, flags, NULL);
1660 if (IS_ERR(newfile1)) {
1661 err = PTR_ERR(newfile1);
1662 sock_release(sock2);
1666 newfile2 = sock_alloc_file(sock2, flags, NULL);
1667 if (IS_ERR(newfile2)) {
1668 err = PTR_ERR(newfile2);
1673 audit_fd_pair(fd1, fd2);
1675 fd_install(fd1, newfile1);
1676 fd_install(fd2, newfile2);
1685 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1686 int __user *, usockvec)
1688 return __sys_socketpair(family, type, protocol, usockvec);
1692 * Bind a name to a socket. Nothing much to do here since it's
1693 * the protocol's responsibility to handle the local address.
1695 * We move the socket address to kernel space before we call
1696 * the protocol layer (having also checked the address is ok).
1699 int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1701 struct socket *sock;
1702 struct sockaddr_storage address;
1703 int err, fput_needed;
1705 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1707 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1709 err = security_socket_bind(sock,
1710 (struct sockaddr *)&address,
1713 err = sock->ops->bind(sock,
1717 fput_light(sock->file, fput_needed);
1722 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1724 return __sys_bind(fd, umyaddr, addrlen);
1728 * Perform a listen. Basically, we allow the protocol to do anything
1729 * necessary for a listen, and if that works, we mark the socket as
1730 * ready for listening.
1733 int __sys_listen(int fd, int backlog)
1735 struct socket *sock;
1736 int err, fput_needed;
1739 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1741 somaxconn = READ_ONCE(sock_net(sock->sk)->core.sysctl_somaxconn);
1742 if ((unsigned int)backlog > somaxconn)
1743 backlog = somaxconn;
1745 err = security_socket_listen(sock, backlog);
1747 err = sock->ops->listen(sock, backlog);
1749 fput_light(sock->file, fput_needed);
1754 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1756 return __sys_listen(fd, backlog);
1759 struct file *do_accept(struct file *file, unsigned file_flags,
1760 struct sockaddr __user *upeer_sockaddr,
1761 int __user *upeer_addrlen, int flags)
1763 struct socket *sock, *newsock;
1764 struct file *newfile;
1766 struct sockaddr_storage address;
1768 sock = sock_from_file(file);
1770 return ERR_PTR(-ENOTSOCK);
1772 newsock = sock_alloc();
1774 return ERR_PTR(-ENFILE);
1776 newsock->type = sock->type;
1777 newsock->ops = sock->ops;
1780 * We don't need try_module_get here, as the listening socket (sock)
1781 * has the protocol module (sock->ops->owner) held.
1783 __module_get(newsock->ops->owner);
1785 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1786 if (IS_ERR(newfile))
1789 err = security_socket_accept(sock, newsock);
1793 err = sock->ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1798 if (upeer_sockaddr) {
1799 len = newsock->ops->getname(newsock,
1800 (struct sockaddr *)&address, 2);
1802 err = -ECONNABORTED;
1805 err = move_addr_to_user(&address,
1806 len, upeer_sockaddr, upeer_addrlen);
1811 /* File flags are not inherited via accept() unlike another OSes. */
1815 return ERR_PTR(err);
1818 int __sys_accept4_file(struct file *file, unsigned file_flags,
1819 struct sockaddr __user *upeer_sockaddr,
1820 int __user *upeer_addrlen, int flags,
1821 unsigned long nofile)
1823 struct file *newfile;
1826 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1829 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1830 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1832 newfd = __get_unused_fd_flags(flags, nofile);
1833 if (unlikely(newfd < 0))
1836 newfile = do_accept(file, file_flags, upeer_sockaddr, upeer_addrlen,
1838 if (IS_ERR(newfile)) {
1839 put_unused_fd(newfd);
1840 return PTR_ERR(newfile);
1842 fd_install(newfd, newfile);
1847 * For accept, we attempt to create a new socket, set up the link
1848 * with the client, wake up the client, then return the new
1849 * connected fd. We collect the address of the connector in kernel
1850 * space and move it to user at the very end. This is unclean because
1851 * we open the socket then return an error.
1853 * 1003.1g adds the ability to recvmsg() to query connection pending
1854 * status to recvmsg. We need to add that support in a way thats
1855 * clean when we restructure accept also.
1858 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1859 int __user *upeer_addrlen, int flags)
1866 ret = __sys_accept4_file(f.file, 0, upeer_sockaddr,
1867 upeer_addrlen, flags,
1868 rlimit(RLIMIT_NOFILE));
1875 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1876 int __user *, upeer_addrlen, int, flags)
1878 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1881 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1882 int __user *, upeer_addrlen)
1884 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1888 * Attempt to connect to a socket with the server address. The address
1889 * is in user space so we verify it is OK and move it to kernel space.
1891 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1894 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1895 * other SEQPACKET protocols that take time to connect() as it doesn't
1896 * include the -EINPROGRESS status for such sockets.
1899 int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
1900 int addrlen, int file_flags)
1902 struct socket *sock;
1905 sock = sock_from_file(file);
1912 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1916 err = sock->ops->connect(sock, (struct sockaddr *)address, addrlen,
1917 sock->file->f_flags | file_flags);
1922 int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1929 struct sockaddr_storage address;
1931 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
1933 ret = __sys_connect_file(f.file, &address, addrlen, 0);
1940 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1943 return __sys_connect(fd, uservaddr, addrlen);
1947 * Get the local address ('name') of a socket object. Move the obtained
1948 * name to user space.
1951 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1952 int __user *usockaddr_len)
1954 struct socket *sock;
1955 struct sockaddr_storage address;
1956 int err, fput_needed;
1958 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1962 err = security_socket_getsockname(sock);
1966 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1969 /* "err" is actually length in this case */
1970 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1973 fput_light(sock->file, fput_needed);
1978 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1979 int __user *, usockaddr_len)
1981 return __sys_getsockname(fd, usockaddr, usockaddr_len);
1985 * Get the remote address ('name') of a socket object. Move the obtained
1986 * name to user space.
1989 int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1990 int __user *usockaddr_len)
1992 struct socket *sock;
1993 struct sockaddr_storage address;
1994 int err, fput_needed;
1996 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1998 err = security_socket_getpeername(sock);
2000 fput_light(sock->file, fput_needed);
2004 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
2006 /* "err" is actually length in this case */
2007 err = move_addr_to_user(&address, err, usockaddr,
2009 fput_light(sock->file, fput_needed);
2014 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
2015 int __user *, usockaddr_len)
2017 return __sys_getpeername(fd, usockaddr, usockaddr_len);
2021 * Send a datagram to a given address. We move the address into kernel
2022 * space and check the user space data area is readable before invoking
2025 int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
2026 struct sockaddr __user *addr, int addr_len)
2028 struct socket *sock;
2029 struct sockaddr_storage address;
2035 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
2038 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2042 msg.msg_name = NULL;
2043 msg.msg_control = NULL;
2044 msg.msg_controllen = 0;
2045 msg.msg_namelen = 0;
2047 err = move_addr_to_kernel(addr, addr_len, &address);
2050 msg.msg_name = (struct sockaddr *)&address;
2051 msg.msg_namelen = addr_len;
2053 if (sock->file->f_flags & O_NONBLOCK)
2054 flags |= MSG_DONTWAIT;
2055 msg.msg_flags = flags;
2056 err = __sock_sendmsg(sock, &msg);
2059 fput_light(sock->file, fput_needed);
2064 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2065 unsigned int, flags, struct sockaddr __user *, addr,
2068 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2072 * Send a datagram down a socket.
2075 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2076 unsigned int, flags)
2078 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2082 * Receive a frame from the socket and optionally record the address of the
2083 * sender. We verify the buffers are writable and if needed move the
2084 * sender address from kernel to user space.
2086 int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2087 struct sockaddr __user *addr, int __user *addr_len)
2089 struct socket *sock;
2092 struct sockaddr_storage address;
2096 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
2099 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2103 msg.msg_control = NULL;
2104 msg.msg_controllen = 0;
2105 /* Save some cycles and don't copy the address if not needed */
2106 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
2107 /* We assume all kernel code knows the size of sockaddr_storage */
2108 msg.msg_namelen = 0;
2109 msg.msg_iocb = NULL;
2111 if (sock->file->f_flags & O_NONBLOCK)
2112 flags |= MSG_DONTWAIT;
2113 err = sock_recvmsg(sock, &msg, flags);
2115 if (err >= 0 && addr != NULL) {
2116 err2 = move_addr_to_user(&address,
2117 msg.msg_namelen, addr, addr_len);
2122 fput_light(sock->file, fput_needed);
2127 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2128 unsigned int, flags, struct sockaddr __user *, addr,
2129 int __user *, addr_len)
2131 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2135 * Receive a datagram from a socket.
2138 SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2139 unsigned int, flags)
2141 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2144 static bool sock_use_custom_sol_socket(const struct socket *sock)
2146 const struct sock *sk = sock->sk;
2148 /* Use sock->ops->setsockopt() for MPTCP */
2149 return IS_ENABLED(CONFIG_MPTCP) &&
2150 sk->sk_protocol == IPPROTO_MPTCP &&
2151 sk->sk_type == SOCK_STREAM &&
2152 (sk->sk_family == AF_INET || sk->sk_family == AF_INET6);
2156 * Set a socket option. Because we don't know the option lengths we have
2157 * to pass the user mode parameter for the protocols to sort out.
2159 int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2162 sockptr_t optval = USER_SOCKPTR(user_optval);
2163 char *kernel_optval = NULL;
2164 int err, fput_needed;
2165 struct socket *sock;
2170 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2174 err = security_socket_setsockopt(sock, level, optname);
2178 if (!in_compat_syscall())
2179 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2180 user_optval, &optlen,
2190 optval = KERNEL_SOCKPTR(kernel_optval);
2191 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2192 err = sock_setsockopt(sock, level, optname, optval, optlen);
2193 else if (unlikely(!sock->ops->setsockopt))
2196 err = sock->ops->setsockopt(sock, level, optname, optval,
2198 kfree(kernel_optval);
2200 fput_light(sock->file, fput_needed);
2204 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2205 char __user *, optval, int, optlen)
2207 return __sys_setsockopt(fd, level, optname, optval, optlen);
2210 INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2214 * Get a socket option. Because we don't know the option lengths we have
2215 * to pass a user mode parameter for the protocols to sort out.
2217 int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2220 int err, fput_needed;
2221 struct socket *sock;
2224 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2228 err = security_socket_getsockopt(sock, level, optname);
2232 if (!in_compat_syscall())
2233 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2235 if (level == SOL_SOCKET)
2236 err = sock_getsockopt(sock, level, optname, optval, optlen);
2237 else if (unlikely(!sock->ops->getsockopt))
2240 err = sock->ops->getsockopt(sock, level, optname, optval,
2243 if (!in_compat_syscall())
2244 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2245 optval, optlen, max_optlen,
2248 fput_light(sock->file, fput_needed);
2252 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2253 char __user *, optval, int __user *, optlen)
2255 return __sys_getsockopt(fd, level, optname, optval, optlen);
2259 * Shutdown a socket.
2262 int __sys_shutdown_sock(struct socket *sock, int how)
2266 err = security_socket_shutdown(sock, how);
2268 err = sock->ops->shutdown(sock, how);
2273 int __sys_shutdown(int fd, int how)
2275 int err, fput_needed;
2276 struct socket *sock;
2278 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2280 err = __sys_shutdown_sock(sock, how);
2281 fput_light(sock->file, fput_needed);
2286 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2288 return __sys_shutdown(fd, how);
2291 /* A couple of helpful macros for getting the address of the 32/64 bit
2292 * fields which are the same type (int / unsigned) on our platforms.
2294 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2295 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2296 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2298 struct used_address {
2299 struct sockaddr_storage name;
2300 unsigned int name_len;
2303 int __copy_msghdr_from_user(struct msghdr *kmsg,
2304 struct user_msghdr __user *umsg,
2305 struct sockaddr __user **save_addr,
2306 struct iovec __user **uiov, size_t *nsegs)
2308 struct user_msghdr msg;
2311 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2314 kmsg->msg_control_is_user = true;
2315 kmsg->msg_control_user = msg.msg_control;
2316 kmsg->msg_controllen = msg.msg_controllen;
2317 kmsg->msg_flags = msg.msg_flags;
2319 kmsg->msg_namelen = msg.msg_namelen;
2321 kmsg->msg_namelen = 0;
2323 if (kmsg->msg_namelen < 0)
2326 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2327 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2330 *save_addr = msg.msg_name;
2332 if (msg.msg_name && kmsg->msg_namelen) {
2334 err = move_addr_to_kernel(msg.msg_name,
2341 kmsg->msg_name = NULL;
2342 kmsg->msg_namelen = 0;
2345 if (msg.msg_iovlen > UIO_MAXIOV)
2348 kmsg->msg_iocb = NULL;
2349 *uiov = msg.msg_iov;
2350 *nsegs = msg.msg_iovlen;
2354 static int copy_msghdr_from_user(struct msghdr *kmsg,
2355 struct user_msghdr __user *umsg,
2356 struct sockaddr __user **save_addr,
2359 struct user_msghdr msg;
2362 err = __copy_msghdr_from_user(kmsg, umsg, save_addr, &msg.msg_iov,
2367 err = import_iovec(save_addr ? READ : WRITE,
2368 msg.msg_iov, msg.msg_iovlen,
2369 UIO_FASTIOV, iov, &kmsg->msg_iter);
2370 return err < 0 ? err : 0;
2373 static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2374 unsigned int flags, struct used_address *used_address,
2375 unsigned int allowed_msghdr_flags)
2377 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2378 __aligned(sizeof(__kernel_size_t));
2379 /* 20 is size of ipv6_pktinfo */
2380 unsigned char *ctl_buf = ctl;
2386 if (msg_sys->msg_controllen > INT_MAX)
2388 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2389 ctl_len = msg_sys->msg_controllen;
2390 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2392 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2396 ctl_buf = msg_sys->msg_control;
2397 ctl_len = msg_sys->msg_controllen;
2398 } else if (ctl_len) {
2399 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2400 CMSG_ALIGN(sizeof(struct cmsghdr)));
2401 if (ctl_len > sizeof(ctl)) {
2402 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2403 if (ctl_buf == NULL)
2407 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2409 msg_sys->msg_control = ctl_buf;
2410 msg_sys->msg_control_is_user = false;
2412 msg_sys->msg_flags = flags;
2414 if (sock->file->f_flags & O_NONBLOCK)
2415 msg_sys->msg_flags |= MSG_DONTWAIT;
2417 * If this is sendmmsg() and current destination address is same as
2418 * previously succeeded address, omit asking LSM's decision.
2419 * used_address->name_len is initialized to UINT_MAX so that the first
2420 * destination address never matches.
2422 if (used_address && msg_sys->msg_name &&
2423 used_address->name_len == msg_sys->msg_namelen &&
2424 !memcmp(&used_address->name, msg_sys->msg_name,
2425 used_address->name_len)) {
2426 err = sock_sendmsg_nosec(sock, msg_sys);
2429 err = __sock_sendmsg(sock, msg_sys);
2431 * If this is sendmmsg() and sending to current destination address was
2432 * successful, remember it.
2434 if (used_address && err >= 0) {
2435 used_address->name_len = msg_sys->msg_namelen;
2436 if (msg_sys->msg_name)
2437 memcpy(&used_address->name, msg_sys->msg_name,
2438 used_address->name_len);
2443 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2448 int sendmsg_copy_msghdr(struct msghdr *msg,
2449 struct user_msghdr __user *umsg, unsigned flags,
2454 if (flags & MSG_CMSG_COMPAT) {
2455 struct compat_msghdr __user *msg_compat;
2457 msg_compat = (struct compat_msghdr __user *) umsg;
2458 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2460 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2468 static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2469 struct msghdr *msg_sys, unsigned int flags,
2470 struct used_address *used_address,
2471 unsigned int allowed_msghdr_flags)
2473 struct sockaddr_storage address;
2474 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2477 msg_sys->msg_name = &address;
2479 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2483 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2484 allowed_msghdr_flags);
2490 * BSD sendmsg interface
2492 long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2495 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2498 long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2499 bool forbid_cmsg_compat)
2501 int fput_needed, err;
2502 struct msghdr msg_sys;
2503 struct socket *sock;
2505 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2508 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2512 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2514 fput_light(sock->file, fput_needed);
2519 SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2521 return __sys_sendmsg(fd, msg, flags, true);
2525 * Linux sendmmsg interface
2528 int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2529 unsigned int flags, bool forbid_cmsg_compat)
2531 int fput_needed, err, datagrams;
2532 struct socket *sock;
2533 struct mmsghdr __user *entry;
2534 struct compat_mmsghdr __user *compat_entry;
2535 struct msghdr msg_sys;
2536 struct used_address used_address;
2537 unsigned int oflags = flags;
2539 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2542 if (vlen > UIO_MAXIOV)
2547 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2551 used_address.name_len = UINT_MAX;
2553 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2557 while (datagrams < vlen) {
2558 if (datagrams == vlen - 1)
2561 if (MSG_CMSG_COMPAT & flags) {
2562 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2563 &msg_sys, flags, &used_address, MSG_EOR);
2566 err = __put_user(err, &compat_entry->msg_len);
2569 err = ___sys_sendmsg(sock,
2570 (struct user_msghdr __user *)entry,
2571 &msg_sys, flags, &used_address, MSG_EOR);
2574 err = put_user(err, &entry->msg_len);
2581 if (msg_data_left(&msg_sys))
2586 fput_light(sock->file, fput_needed);
2588 /* We only return an error if no datagrams were able to be sent */
2595 SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2596 unsigned int, vlen, unsigned int, flags)
2598 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2601 int recvmsg_copy_msghdr(struct msghdr *msg,
2602 struct user_msghdr __user *umsg, unsigned flags,
2603 struct sockaddr __user **uaddr,
2608 if (MSG_CMSG_COMPAT & flags) {
2609 struct compat_msghdr __user *msg_compat;
2611 msg_compat = (struct compat_msghdr __user *) umsg;
2612 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2614 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2622 static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2623 struct user_msghdr __user *msg,
2624 struct sockaddr __user *uaddr,
2625 unsigned int flags, int nosec)
2627 struct compat_msghdr __user *msg_compat =
2628 (struct compat_msghdr __user *) msg;
2629 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2630 struct sockaddr_storage addr;
2631 unsigned long cmsg_ptr;
2635 msg_sys->msg_name = &addr;
2636 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2637 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2639 /* We assume all kernel code knows the size of sockaddr_storage */
2640 msg_sys->msg_namelen = 0;
2642 if (sock->file->f_flags & O_NONBLOCK)
2643 flags |= MSG_DONTWAIT;
2645 if (unlikely(nosec))
2646 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2648 err = sock_recvmsg(sock, msg_sys, flags);
2654 if (uaddr != NULL) {
2655 err = move_addr_to_user(&addr,
2656 msg_sys->msg_namelen, uaddr,
2661 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2665 if (MSG_CMSG_COMPAT & flags)
2666 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2667 &msg_compat->msg_controllen);
2669 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2670 &msg->msg_controllen);
2678 static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2679 struct msghdr *msg_sys, unsigned int flags, int nosec)
2681 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2682 /* user mode address pointers */
2683 struct sockaddr __user *uaddr;
2686 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2690 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2696 * BSD recvmsg interface
2699 long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2700 struct user_msghdr __user *umsg,
2701 struct sockaddr __user *uaddr, unsigned int flags)
2703 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2706 long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2707 bool forbid_cmsg_compat)
2709 int fput_needed, err;
2710 struct msghdr msg_sys;
2711 struct socket *sock;
2713 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2716 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2720 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2722 fput_light(sock->file, fput_needed);
2727 SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2728 unsigned int, flags)
2730 return __sys_recvmsg(fd, msg, flags, true);
2734 * Linux recvmmsg interface
2737 static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2738 unsigned int vlen, unsigned int flags,
2739 struct timespec64 *timeout)
2741 int fput_needed, err, datagrams;
2742 struct socket *sock;
2743 struct mmsghdr __user *entry;
2744 struct compat_mmsghdr __user *compat_entry;
2745 struct msghdr msg_sys;
2746 struct timespec64 end_time;
2747 struct timespec64 timeout64;
2750 poll_select_set_timeout(&end_time, timeout->tv_sec,
2756 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2760 if (likely(!(flags & MSG_ERRQUEUE))) {
2761 err = sock_error(sock->sk);
2769 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2771 while (datagrams < vlen) {
2773 * No need to ask LSM for more than the first datagram.
2775 if (MSG_CMSG_COMPAT & flags) {
2776 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2777 &msg_sys, flags & ~MSG_WAITFORONE,
2781 err = __put_user(err, &compat_entry->msg_len);
2784 err = ___sys_recvmsg(sock,
2785 (struct user_msghdr __user *)entry,
2786 &msg_sys, flags & ~MSG_WAITFORONE,
2790 err = put_user(err, &entry->msg_len);
2798 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2799 if (flags & MSG_WAITFORONE)
2800 flags |= MSG_DONTWAIT;
2803 ktime_get_ts64(&timeout64);
2804 *timeout = timespec64_sub(end_time, timeout64);
2805 if (timeout->tv_sec < 0) {
2806 timeout->tv_sec = timeout->tv_nsec = 0;
2810 /* Timeout, return less than vlen datagrams */
2811 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2815 /* Out of band data, return right away */
2816 if (msg_sys.msg_flags & MSG_OOB)
2824 if (datagrams == 0) {
2830 * We may return less entries than requested (vlen) if the
2831 * sock is non block and there aren't enough datagrams...
2833 if (err != -EAGAIN) {
2835 * ... or if recvmsg returns an error after we
2836 * received some datagrams, where we record the
2837 * error to return on the next call or if the
2838 * app asks about it using getsockopt(SO_ERROR).
2840 WRITE_ONCE(sock->sk->sk_err, -err);
2843 fput_light(sock->file, fput_needed);
2848 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2849 unsigned int vlen, unsigned int flags,
2850 struct __kernel_timespec __user *timeout,
2851 struct old_timespec32 __user *timeout32)
2854 struct timespec64 timeout_sys;
2856 if (timeout && get_timespec64(&timeout_sys, timeout))
2859 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2862 if (!timeout && !timeout32)
2863 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2865 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2870 if (timeout && put_timespec64(&timeout_sys, timeout))
2871 datagrams = -EFAULT;
2873 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2874 datagrams = -EFAULT;
2879 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2880 unsigned int, vlen, unsigned int, flags,
2881 struct __kernel_timespec __user *, timeout)
2883 if (flags & MSG_CMSG_COMPAT)
2886 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2889 #ifdef CONFIG_COMPAT_32BIT_TIME
2890 SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2891 unsigned int, vlen, unsigned int, flags,
2892 struct old_timespec32 __user *, timeout)
2894 if (flags & MSG_CMSG_COMPAT)
2897 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2901 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2902 /* Argument list sizes for sys_socketcall */
2903 #define AL(x) ((x) * sizeof(unsigned long))
2904 static const unsigned char nargs[21] = {
2905 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2906 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2907 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2914 * System call vectors.
2916 * Argument checking cleaned up. Saved 20% in size.
2917 * This function doesn't need to set the kernel lock because
2918 * it is set by the callees.
2921 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2923 unsigned long a[AUDITSC_ARGS];
2924 unsigned long a0, a1;
2928 if (call < 1 || call > SYS_SENDMMSG)
2930 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2933 if (len > sizeof(a))
2936 /* copy_from_user should be SMP safe. */
2937 if (copy_from_user(a, args, len))
2940 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2949 err = __sys_socket(a0, a1, a[2]);
2952 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2955 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2958 err = __sys_listen(a0, a1);
2961 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2962 (int __user *)a[2], 0);
2964 case SYS_GETSOCKNAME:
2966 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2967 (int __user *)a[2]);
2969 case SYS_GETPEERNAME:
2971 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2972 (int __user *)a[2]);
2974 case SYS_SOCKETPAIR:
2975 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2978 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2982 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2983 (struct sockaddr __user *)a[4], a[5]);
2986 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2990 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2991 (struct sockaddr __user *)a[4],
2992 (int __user *)a[5]);
2995 err = __sys_shutdown(a0, a1);
2997 case SYS_SETSOCKOPT:
2998 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
3001 case SYS_GETSOCKOPT:
3003 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
3004 (int __user *)a[4]);
3007 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
3011 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
3015 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
3019 if (IS_ENABLED(CONFIG_64BIT))
3020 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3022 (struct __kernel_timespec __user *)a[4],
3025 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3027 (struct old_timespec32 __user *)a[4]);
3030 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3031 (int __user *)a[2], a[3]);
3040 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
3043 * sock_register - add a socket protocol handler
3044 * @ops: description of protocol
3046 * This function is called by a protocol handler that wants to
3047 * advertise its address family, and have it linked into the
3048 * socket interface. The value ops->family corresponds to the
3049 * socket system call protocol family.
3051 int sock_register(const struct net_proto_family *ops)
3055 if (ops->family >= NPROTO) {
3056 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3060 spin_lock(&net_family_lock);
3061 if (rcu_dereference_protected(net_families[ops->family],
3062 lockdep_is_held(&net_family_lock)))
3065 rcu_assign_pointer(net_families[ops->family], ops);
3068 spin_unlock(&net_family_lock);
3070 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3073 EXPORT_SYMBOL(sock_register);
3076 * sock_unregister - remove a protocol handler
3077 * @family: protocol family to remove
3079 * This function is called by a protocol handler that wants to
3080 * remove its address family, and have it unlinked from the
3081 * new socket creation.
3083 * If protocol handler is a module, then it can use module reference
3084 * counts to protect against new references. If protocol handler is not
3085 * a module then it needs to provide its own protection in
3086 * the ops->create routine.
3088 void sock_unregister(int family)
3090 BUG_ON(family < 0 || family >= NPROTO);
3092 spin_lock(&net_family_lock);
3093 RCU_INIT_POINTER(net_families[family], NULL);
3094 spin_unlock(&net_family_lock);
3098 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3100 EXPORT_SYMBOL(sock_unregister);
3102 bool sock_is_registered(int family)
3104 return family < NPROTO && rcu_access_pointer(net_families[family]);
3107 static int __init sock_init(void)
3111 * Initialize the network sysctl infrastructure.
3113 err = net_sysctl_init();
3118 * Initialize skbuff SLAB cache
3123 * Initialize the protocols module.
3128 err = register_filesystem(&sock_fs_type);
3131 sock_mnt = kern_mount(&sock_fs_type);
3132 if (IS_ERR(sock_mnt)) {
3133 err = PTR_ERR(sock_mnt);
3137 /* The real protocol initialization is performed in later initcalls.
3140 #ifdef CONFIG_NETFILTER
3141 err = netfilter_init();
3146 ptp_classifier_init();
3152 unregister_filesystem(&sock_fs_type);
3156 core_initcall(sock_init); /* early initcall */
3158 #ifdef CONFIG_PROC_FS
3159 void socket_seq_show(struct seq_file *seq)
3161 seq_printf(seq, "sockets: used %d\n",
3162 sock_inuse_get(seq->private));
3164 #endif /* CONFIG_PROC_FS */
3166 /* Handle the fact that while struct ifreq has the same *layout* on
3167 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3168 * which are handled elsewhere, it still has different *size* due to
3169 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3170 * resulting in struct ifreq being 32 and 40 bytes respectively).
3171 * As a result, if the struct happens to be at the end of a page and
3172 * the next page isn't readable/writable, we get a fault. To prevent
3173 * that, copy back and forth to the full size.
3175 int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3177 if (in_compat_syscall()) {
3178 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3180 memset(ifr, 0, sizeof(*ifr));
3181 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3185 *ifrdata = compat_ptr(ifr32->ifr_data);
3190 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3194 *ifrdata = ifr->ifr_data;
3198 EXPORT_SYMBOL(get_user_ifreq);
3200 int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3202 size_t size = sizeof(*ifr);
3204 if (in_compat_syscall())
3205 size = sizeof(struct compat_ifreq);
3207 if (copy_to_user(arg, ifr, size))
3212 EXPORT_SYMBOL(put_user_ifreq);
3214 #ifdef CONFIG_COMPAT
3215 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3217 compat_uptr_t uptr32;
3222 if (get_user_ifreq(&ifr, NULL, uifr32))
3225 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3228 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3229 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3231 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3233 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3234 if (put_user_ifreq(&ifr, uifr32))
3240 /* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3241 static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3242 struct compat_ifreq __user *u_ifreq32)
3247 if (!is_socket_ioctl_cmd(cmd))
3249 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3251 ifreq.ifr_data = data;
3253 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3256 /* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3257 * for some operations; this forces use of the newer bridge-utils that
3258 * use compatible ioctls
3260 static int old_bridge_ioctl(compat_ulong_t __user *argp)
3264 if (get_user(tmp, argp))
3266 if (tmp == BRCTL_GET_VERSION)
3267 return BRCTL_VERSION + 1;
3271 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3272 unsigned int cmd, unsigned long arg)
3274 void __user *argp = compat_ptr(arg);
3275 struct sock *sk = sock->sk;
3276 struct net *net = sock_net(sk);
3278 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3279 return sock_ioctl(file, cmd, (unsigned long)argp);
3284 return old_bridge_ioctl(argp);
3286 return compat_siocwandev(net, argp);
3287 case SIOCGSTAMP_OLD:
3288 case SIOCGSTAMPNS_OLD:
3289 if (!sock->ops->gettstamp)
3290 return -ENOIOCTLCMD;
3291 return sock->ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3292 !COMPAT_USE_64BIT_TIME);
3295 case SIOCBONDSLAVEINFOQUERY:
3296 case SIOCBONDINFOQUERY:
3299 return compat_ifr_data_ioctl(net, cmd, argp);
3310 case SIOCGSTAMP_NEW:
3311 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 struct sockaddr_storage address;
3405 memcpy(&address, addr, addrlen);
3407 return sock->ops->bind(sock, (struct sockaddr *)&address, addrlen);
3409 EXPORT_SYMBOL(kernel_bind);
3412 * kernel_listen - move socket to listening state (kernel space)
3414 * @backlog: pending connections queue size
3416 * Returns 0 or an error.
3419 int kernel_listen(struct socket *sock, int backlog)
3421 return sock->ops->listen(sock, backlog);
3423 EXPORT_SYMBOL(kernel_listen);
3426 * kernel_accept - accept a connection (kernel space)
3427 * @sock: listening socket
3428 * @newsock: new connected socket
3431 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3432 * If it fails, @newsock is guaranteed to be %NULL.
3433 * Returns 0 or an error.
3436 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3438 struct sock *sk = sock->sk;
3441 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3446 err = sock->ops->accept(sock, *newsock, flags, true);
3448 sock_release(*newsock);
3453 (*newsock)->ops = sock->ops;
3454 __module_get((*newsock)->ops->owner);
3459 EXPORT_SYMBOL(kernel_accept);
3462 * kernel_connect - connect a socket (kernel space)
3465 * @addrlen: address length
3466 * @flags: flags (O_NONBLOCK, ...)
3468 * For datagram sockets, @addr is the address to which datagrams are sent
3469 * by default, and the only address from which datagrams are received.
3470 * For stream sockets, attempts to connect to @addr.
3471 * Returns 0 or an error code.
3474 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3477 struct sockaddr_storage address;
3479 memcpy(&address, addr, addrlen);
3481 return sock->ops->connect(sock, (struct sockaddr *)&address, addrlen, flags);
3483 EXPORT_SYMBOL(kernel_connect);
3486 * kernel_getsockname - get the address which the socket is bound (kernel space)
3488 * @addr: address holder
3490 * Fills the @addr pointer with the address which the socket is bound.
3491 * Returns 0 or an error code.
3494 int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3496 return sock->ops->getname(sock, addr, 0);
3498 EXPORT_SYMBOL(kernel_getsockname);
3501 * kernel_getpeername - get the address which the socket is connected (kernel space)
3503 * @addr: address holder
3505 * Fills the @addr pointer with the address which the socket is connected.
3506 * Returns 0 or an error code.
3509 int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3511 return sock->ops->getname(sock, addr, 1);
3513 EXPORT_SYMBOL(kernel_getpeername);
3516 * kernel_sendpage - send a &page through a socket (kernel space)
3519 * @offset: page offset
3520 * @size: total size in bytes
3521 * @flags: flags (MSG_DONTWAIT, ...)
3523 * Returns the total amount sent in bytes or an error.
3526 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3527 size_t size, int flags)
3529 if (sock->ops->sendpage) {
3530 /* Warn in case the improper page to zero-copy send */
3531 WARN_ONCE(!sendpage_ok(page), "improper page for zero-copy send");
3532 return sock->ops->sendpage(sock, page, offset, size, flags);
3534 return sock_no_sendpage(sock, page, offset, size, flags);
3536 EXPORT_SYMBOL(kernel_sendpage);
3539 * kernel_sendpage_locked - send a &page through the locked sock (kernel space)
3542 * @offset: page offset
3543 * @size: total size in bytes
3544 * @flags: flags (MSG_DONTWAIT, ...)
3546 * Returns the total amount sent in bytes or an error.
3547 * Caller must hold @sk.
3550 int kernel_sendpage_locked(struct sock *sk, struct page *page, int offset,
3551 size_t size, int flags)
3553 struct socket *sock = sk->sk_socket;
3555 if (sock->ops->sendpage_locked)
3556 return sock->ops->sendpage_locked(sk, page, offset, size,
3559 return sock_no_sendpage_locked(sk, page, offset, size, flags);
3561 EXPORT_SYMBOL(kernel_sendpage_locked);
3564 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3566 * @how: connection part
3568 * Returns 0 or an error.
3571 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3573 return sock->ops->shutdown(sock, how);
3575 EXPORT_SYMBOL(kernel_sock_shutdown);
3578 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3581 * This routine returns the IP overhead imposed by a socket i.e.
3582 * the length of the underlying IP header, depending on whether
3583 * this is an IPv4 or IPv6 socket and the length from IP options turned
3584 * on at the socket. Assumes that the caller has a lock on the socket.
3587 u32 kernel_sock_ip_overhead(struct sock *sk)
3589 struct inet_sock *inet;
3590 struct ip_options_rcu *opt;
3592 #if IS_ENABLED(CONFIG_IPV6)
3593 struct ipv6_pinfo *np;
3594 struct ipv6_txoptions *optv6 = NULL;
3595 #endif /* IS_ENABLED(CONFIG_IPV6) */
3600 switch (sk->sk_family) {
3603 overhead += sizeof(struct iphdr);
3604 opt = rcu_dereference_protected(inet->inet_opt,
3605 sock_owned_by_user(sk));
3607 overhead += opt->opt.optlen;
3609 #if IS_ENABLED(CONFIG_IPV6)
3612 overhead += sizeof(struct ipv6hdr);
3614 optv6 = rcu_dereference_protected(np->opt,
3615 sock_owned_by_user(sk));
3617 overhead += (optv6->opt_flen + optv6->opt_nflen);
3619 #endif /* IS_ENABLED(CONFIG_IPV6) */
3620 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3624 EXPORT_SYMBOL(kernel_sock_ip_overhead);