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);
712 * sock_sendmsg - send a message through @sock
714 * @msg: message to send
716 * Sends @msg through @sock, passing through LSM.
717 * Returns the number of bytes sent, or an error code.
719 int sock_sendmsg(struct socket *sock, struct msghdr *msg)
721 int err = security_socket_sendmsg(sock, msg,
724 return err ?: sock_sendmsg_nosec(sock, msg);
726 EXPORT_SYMBOL(sock_sendmsg);
729 * kernel_sendmsg - send a message through @sock (kernel-space)
731 * @msg: message header
733 * @num: vec array length
734 * @size: total message data size
736 * Builds the message data with @vec and sends it through @sock.
737 * Returns the number of bytes sent, or an error code.
740 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
741 struct kvec *vec, size_t num, size_t size)
743 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
744 return sock_sendmsg(sock, msg);
746 EXPORT_SYMBOL(kernel_sendmsg);
749 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
751 * @msg: message header
752 * @vec: output s/g array
753 * @num: output s/g array length
754 * @size: total message data size
756 * Builds the message data with @vec and sends it through @sock.
757 * Returns the number of bytes sent, or an error code.
758 * Caller must hold @sk.
761 int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
762 struct kvec *vec, size_t num, size_t size)
764 struct socket *sock = sk->sk_socket;
766 if (!sock->ops->sendmsg_locked)
767 return sock_no_sendmsg_locked(sk, msg, size);
769 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
771 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
773 EXPORT_SYMBOL(kernel_sendmsg_locked);
775 static bool skb_is_err_queue(const struct sk_buff *skb)
777 /* pkt_type of skbs enqueued on the error queue are set to
778 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
779 * in recvmsg, since skbs received on a local socket will never
780 * have a pkt_type of PACKET_OUTGOING.
782 return skb->pkt_type == PACKET_OUTGOING;
785 /* On transmit, software and hardware timestamps are returned independently.
786 * As the two skb clones share the hardware timestamp, which may be updated
787 * before the software timestamp is received, a hardware TX timestamp may be
788 * returned only if there is no software TX timestamp. Ignore false software
789 * timestamps, which may be made in the __sock_recv_timestamp() call when the
790 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
791 * hardware timestamp.
793 static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
795 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
798 static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
800 struct scm_ts_pktinfo ts_pktinfo;
801 struct net_device *orig_dev;
803 if (!skb_mac_header_was_set(skb))
806 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
809 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
811 ts_pktinfo.if_index = orig_dev->ifindex;
814 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
815 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
816 sizeof(ts_pktinfo), &ts_pktinfo);
820 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
822 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
825 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
826 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
827 struct scm_timestamping_internal tss;
829 int empty = 1, false_tstamp = 0;
830 struct skb_shared_hwtstamps *shhwtstamps =
833 /* Race occurred between timestamp enabling and packet
834 receiving. Fill in the current time for now. */
835 if (need_software_tstamp && skb->tstamp == 0) {
836 __net_timestamp(skb);
840 if (need_software_tstamp) {
841 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
843 struct __kernel_sock_timeval tv;
845 skb_get_new_timestamp(skb, &tv);
846 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
849 struct __kernel_old_timeval tv;
851 skb_get_timestamp(skb, &tv);
852 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
857 struct __kernel_timespec ts;
859 skb_get_new_timestampns(skb, &ts);
860 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
863 struct __kernel_old_timespec ts;
865 skb_get_timestampns(skb, &ts);
866 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
872 memset(&tss, 0, sizeof(tss));
873 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
874 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
877 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
878 !skb_is_swtx_tstamp(skb, false_tstamp)) {
879 if (sk->sk_tsflags & SOF_TIMESTAMPING_BIND_PHC)
880 ptp_convert_timestamp(shhwtstamps, sk->sk_bind_phc);
882 if (ktime_to_timespec64_cond(shhwtstamps->hwtstamp,
886 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
887 !skb_is_err_queue(skb))
888 put_ts_pktinfo(msg, skb);
892 if (sock_flag(sk, SOCK_TSTAMP_NEW))
893 put_cmsg_scm_timestamping64(msg, &tss);
895 put_cmsg_scm_timestamping(msg, &tss);
897 if (skb_is_err_queue(skb) && skb->len &&
898 SKB_EXT_ERR(skb)->opt_stats)
899 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
900 skb->len, skb->data);
903 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
905 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
910 if (!sock_flag(sk, SOCK_WIFI_STATUS))
912 if (!skb->wifi_acked_valid)
915 ack = skb->wifi_acked;
917 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
919 EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
921 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
924 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
925 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
926 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
929 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
932 sock_recv_timestamp(msg, sk, skb);
933 sock_recv_drops(msg, sk, skb);
935 EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
937 INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
939 INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
941 static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
944 return INDIRECT_CALL_INET(sock->ops->recvmsg, inet6_recvmsg,
945 inet_recvmsg, sock, msg, msg_data_left(msg),
950 * sock_recvmsg - receive a message from @sock
952 * @msg: message to receive
953 * @flags: message flags
955 * Receives @msg from @sock, passing through LSM. Returns the total number
956 * of bytes received, or an error.
958 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
960 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
962 return err ?: sock_recvmsg_nosec(sock, msg, flags);
964 EXPORT_SYMBOL(sock_recvmsg);
967 * kernel_recvmsg - Receive a message from a socket (kernel space)
968 * @sock: The socket to receive the message from
969 * @msg: Received message
970 * @vec: Input s/g array for message data
971 * @num: Size of input s/g array
972 * @size: Number of bytes to read
973 * @flags: Message flags (MSG_DONTWAIT, etc...)
975 * On return the msg structure contains the scatter/gather array passed in the
976 * vec argument. The array is modified so that it consists of the unfilled
977 * portion of the original array.
979 * The returned value is the total number of bytes received, or an error.
982 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
983 struct kvec *vec, size_t num, size_t size, int flags)
985 msg->msg_control_is_user = false;
986 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
987 return sock_recvmsg(sock, msg, flags);
989 EXPORT_SYMBOL(kernel_recvmsg);
991 static ssize_t sock_sendpage(struct file *file, struct page *page,
992 int offset, size_t size, loff_t *ppos, int more)
997 sock = file->private_data;
999 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
1000 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
1003 return kernel_sendpage(sock, page, offset, size, flags);
1006 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1007 struct pipe_inode_info *pipe, size_t len,
1010 struct socket *sock = file->private_data;
1012 if (unlikely(!sock->ops->splice_read))
1013 return generic_file_splice_read(file, ppos, pipe, len, flags);
1015 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
1018 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1020 struct file *file = iocb->ki_filp;
1021 struct socket *sock = file->private_data;
1022 struct msghdr msg = {.msg_iter = *to,
1026 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1027 msg.msg_flags = MSG_DONTWAIT;
1029 if (iocb->ki_pos != 0)
1032 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1035 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1040 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1042 struct file *file = iocb->ki_filp;
1043 struct socket *sock = file->private_data;
1044 struct msghdr msg = {.msg_iter = *from,
1048 if (iocb->ki_pos != 0)
1051 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1052 msg.msg_flags = MSG_DONTWAIT;
1054 if (sock->type == SOCK_SEQPACKET)
1055 msg.msg_flags |= MSG_EOR;
1057 res = sock_sendmsg(sock, &msg);
1058 *from = msg.msg_iter;
1063 * Atomic setting of ioctl hooks to avoid race
1064 * with module unload.
1067 static DEFINE_MUTEX(br_ioctl_mutex);
1068 static int (*br_ioctl_hook)(struct net *net, struct net_bridge *br,
1069 unsigned int cmd, struct ifreq *ifr,
1072 void brioctl_set(int (*hook)(struct net *net, struct net_bridge *br,
1073 unsigned int cmd, struct ifreq *ifr,
1076 mutex_lock(&br_ioctl_mutex);
1077 br_ioctl_hook = hook;
1078 mutex_unlock(&br_ioctl_mutex);
1080 EXPORT_SYMBOL(brioctl_set);
1082 int br_ioctl_call(struct net *net, struct net_bridge *br, unsigned int cmd,
1083 struct ifreq *ifr, void __user *uarg)
1088 request_module("bridge");
1090 mutex_lock(&br_ioctl_mutex);
1092 err = br_ioctl_hook(net, br, cmd, ifr, uarg);
1093 mutex_unlock(&br_ioctl_mutex);
1098 static DEFINE_MUTEX(vlan_ioctl_mutex);
1099 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1101 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1103 mutex_lock(&vlan_ioctl_mutex);
1104 vlan_ioctl_hook = hook;
1105 mutex_unlock(&vlan_ioctl_mutex);
1107 EXPORT_SYMBOL(vlan_ioctl_set);
1109 static long sock_do_ioctl(struct net *net, struct socket *sock,
1110 unsigned int cmd, unsigned long arg)
1115 void __user *argp = (void __user *)arg;
1118 err = sock->ops->ioctl(sock, cmd, arg);
1121 * If this ioctl is unknown try to hand it down
1122 * to the NIC driver.
1124 if (err != -ENOIOCTLCMD)
1127 if (!is_socket_ioctl_cmd(cmd))
1130 if (get_user_ifreq(&ifr, &data, argp))
1132 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1133 if (!err && need_copyout)
1134 if (put_user_ifreq(&ifr, argp))
1141 * With an ioctl, arg may well be a user mode pointer, but we don't know
1142 * what to do with it - that's up to the protocol still.
1145 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1147 struct socket *sock;
1149 void __user *argp = (void __user *)arg;
1153 sock = file->private_data;
1156 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1160 if (get_user_ifreq(&ifr, &data, argp))
1162 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1163 if (!err && need_copyout)
1164 if (put_user_ifreq(&ifr, argp))
1167 #ifdef CONFIG_WEXT_CORE
1168 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1169 err = wext_handle_ioctl(net, cmd, argp);
1176 if (get_user(pid, (int __user *)argp))
1178 err = f_setown(sock->file, pid, 1);
1182 err = put_user(f_getown(sock->file),
1183 (int __user *)argp);
1189 err = br_ioctl_call(net, NULL, cmd, NULL, argp);
1194 if (!vlan_ioctl_hook)
1195 request_module("8021q");
1197 mutex_lock(&vlan_ioctl_mutex);
1198 if (vlan_ioctl_hook)
1199 err = vlan_ioctl_hook(net, argp);
1200 mutex_unlock(&vlan_ioctl_mutex);
1204 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1207 err = open_related_ns(&net->ns, get_net_ns);
1209 case SIOCGSTAMP_OLD:
1210 case SIOCGSTAMPNS_OLD:
1211 if (!sock->ops->gettstamp) {
1215 err = sock->ops->gettstamp(sock, argp,
1216 cmd == SIOCGSTAMP_OLD,
1217 !IS_ENABLED(CONFIG_64BIT));
1219 case SIOCGSTAMP_NEW:
1220 case SIOCGSTAMPNS_NEW:
1221 if (!sock->ops->gettstamp) {
1225 err = sock->ops->gettstamp(sock, argp,
1226 cmd == SIOCGSTAMP_NEW,
1231 err = dev_ifconf(net, argp);
1235 err = sock_do_ioctl(net, sock, cmd, arg);
1242 * sock_create_lite - creates a socket
1243 * @family: protocol family (AF_INET, ...)
1244 * @type: communication type (SOCK_STREAM, ...)
1245 * @protocol: protocol (0, ...)
1248 * Creates a new socket and assigns it to @res, passing through LSM.
1249 * The new socket initialization is not complete, see kernel_accept().
1250 * Returns 0 or an error. On failure @res is set to %NULL.
1251 * This function internally uses GFP_KERNEL.
1254 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1257 struct socket *sock = NULL;
1259 err = security_socket_create(family, type, protocol, 1);
1263 sock = sock_alloc();
1270 err = security_socket_post_create(sock, family, type, protocol, 1);
1282 EXPORT_SYMBOL(sock_create_lite);
1284 /* No kernel lock held - perfect */
1285 static __poll_t sock_poll(struct file *file, poll_table *wait)
1287 struct socket *sock = file->private_data;
1288 __poll_t events = poll_requested_events(wait), flag = 0;
1290 if (!sock->ops->poll)
1293 if (sk_can_busy_loop(sock->sk)) {
1294 /* poll once if requested by the syscall */
1295 if (events & POLL_BUSY_LOOP)
1296 sk_busy_loop(sock->sk, 1);
1298 /* if this socket can poll_ll, tell the system call */
1299 flag = POLL_BUSY_LOOP;
1302 return sock->ops->poll(file, sock, wait) | flag;
1305 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1307 struct socket *sock = file->private_data;
1309 return sock->ops->mmap(file, sock, vma);
1312 static int sock_close(struct inode *inode, struct file *filp)
1314 __sock_release(SOCKET_I(inode), inode);
1319 * Update the socket async list
1321 * Fasync_list locking strategy.
1323 * 1. fasync_list is modified only under process context socket lock
1324 * i.e. under semaphore.
1325 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1326 * or under socket lock
1329 static int sock_fasync(int fd, struct file *filp, int on)
1331 struct socket *sock = filp->private_data;
1332 struct sock *sk = sock->sk;
1333 struct socket_wq *wq = &sock->wq;
1339 fasync_helper(fd, filp, on, &wq->fasync_list);
1341 if (!wq->fasync_list)
1342 sock_reset_flag(sk, SOCK_FASYNC);
1344 sock_set_flag(sk, SOCK_FASYNC);
1350 /* This function may be called only under rcu_lock */
1352 int sock_wake_async(struct socket_wq *wq, int how, int band)
1354 if (!wq || !wq->fasync_list)
1358 case SOCK_WAKE_WAITD:
1359 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1362 case SOCK_WAKE_SPACE:
1363 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1368 kill_fasync(&wq->fasync_list, SIGIO, band);
1371 kill_fasync(&wq->fasync_list, SIGURG, band);
1376 EXPORT_SYMBOL(sock_wake_async);
1379 * __sock_create - creates a socket
1380 * @net: net namespace
1381 * @family: protocol family (AF_INET, ...)
1382 * @type: communication type (SOCK_STREAM, ...)
1383 * @protocol: protocol (0, ...)
1385 * @kern: boolean for kernel space sockets
1387 * Creates a new socket and assigns it to @res, passing through LSM.
1388 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1389 * be set to true if the socket resides in kernel space.
1390 * This function internally uses GFP_KERNEL.
1393 int __sock_create(struct net *net, int family, int type, int protocol,
1394 struct socket **res, int kern)
1397 struct socket *sock;
1398 const struct net_proto_family *pf;
1401 * Check protocol is in range
1403 if (family < 0 || family >= NPROTO)
1404 return -EAFNOSUPPORT;
1405 if (type < 0 || type >= SOCK_MAX)
1410 This uglymoron is moved from INET layer to here to avoid
1411 deadlock in module load.
1413 if (family == PF_INET && type == SOCK_PACKET) {
1414 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1419 err = security_socket_create(family, type, protocol, kern);
1424 * Allocate the socket and allow the family to set things up. if
1425 * the protocol is 0, the family is instructed to select an appropriate
1428 sock = sock_alloc();
1430 net_warn_ratelimited("socket: no more sockets\n");
1431 return -ENFILE; /* Not exactly a match, but its the
1432 closest posix thing */
1437 #ifdef CONFIG_MODULES
1438 /* Attempt to load a protocol module if the find failed.
1440 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1441 * requested real, full-featured networking support upon configuration.
1442 * Otherwise module support will break!
1444 if (rcu_access_pointer(net_families[family]) == NULL)
1445 request_module("net-pf-%d", family);
1449 pf = rcu_dereference(net_families[family]);
1450 err = -EAFNOSUPPORT;
1455 * We will call the ->create function, that possibly is in a loadable
1456 * module, so we have to bump that loadable module refcnt first.
1458 if (!try_module_get(pf->owner))
1461 /* Now protected by module ref count */
1464 err = pf->create(net, sock, protocol, kern);
1466 goto out_module_put;
1469 * Now to bump the refcnt of the [loadable] module that owns this
1470 * socket at sock_release time we decrement its refcnt.
1472 if (!try_module_get(sock->ops->owner))
1473 goto out_module_busy;
1476 * Now that we're done with the ->create function, the [loadable]
1477 * module can have its refcnt decremented
1479 module_put(pf->owner);
1480 err = security_socket_post_create(sock, family, type, protocol, kern);
1482 goto out_sock_release;
1488 err = -EAFNOSUPPORT;
1491 module_put(pf->owner);
1498 goto out_sock_release;
1500 EXPORT_SYMBOL(__sock_create);
1503 * sock_create - creates a socket
1504 * @family: protocol family (AF_INET, ...)
1505 * @type: communication type (SOCK_STREAM, ...)
1506 * @protocol: protocol (0, ...)
1509 * A wrapper around __sock_create().
1510 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1513 int sock_create(int family, int type, int protocol, struct socket **res)
1515 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1517 EXPORT_SYMBOL(sock_create);
1520 * sock_create_kern - creates a socket (kernel space)
1521 * @net: net namespace
1522 * @family: protocol family (AF_INET, ...)
1523 * @type: communication type (SOCK_STREAM, ...)
1524 * @protocol: protocol (0, ...)
1527 * A wrapper around __sock_create().
1528 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1531 int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1533 return __sock_create(net, family, type, protocol, res, 1);
1535 EXPORT_SYMBOL(sock_create_kern);
1537 int __sys_socket(int family, int type, int protocol)
1540 struct socket *sock;
1543 /* Check the SOCK_* constants for consistency. */
1544 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1545 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1546 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1547 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1549 flags = type & ~SOCK_TYPE_MASK;
1550 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1552 type &= SOCK_TYPE_MASK;
1554 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1555 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1557 retval = sock_create(family, type, protocol, &sock);
1561 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1564 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1566 return __sys_socket(family, type, protocol);
1570 * Create a pair of connected sockets.
1573 int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1575 struct socket *sock1, *sock2;
1577 struct file *newfile1, *newfile2;
1580 flags = type & ~SOCK_TYPE_MASK;
1581 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1583 type &= SOCK_TYPE_MASK;
1585 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1586 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1589 * reserve descriptors and make sure we won't fail
1590 * to return them to userland.
1592 fd1 = get_unused_fd_flags(flags);
1593 if (unlikely(fd1 < 0))
1596 fd2 = get_unused_fd_flags(flags);
1597 if (unlikely(fd2 < 0)) {
1602 err = put_user(fd1, &usockvec[0]);
1606 err = put_user(fd2, &usockvec[1]);
1611 * Obtain the first socket and check if the underlying protocol
1612 * supports the socketpair call.
1615 err = sock_create(family, type, protocol, &sock1);
1616 if (unlikely(err < 0))
1619 err = sock_create(family, type, protocol, &sock2);
1620 if (unlikely(err < 0)) {
1621 sock_release(sock1);
1625 err = security_socket_socketpair(sock1, sock2);
1626 if (unlikely(err)) {
1627 sock_release(sock2);
1628 sock_release(sock1);
1632 err = sock1->ops->socketpair(sock1, sock2);
1633 if (unlikely(err < 0)) {
1634 sock_release(sock2);
1635 sock_release(sock1);
1639 newfile1 = sock_alloc_file(sock1, flags, NULL);
1640 if (IS_ERR(newfile1)) {
1641 err = PTR_ERR(newfile1);
1642 sock_release(sock2);
1646 newfile2 = sock_alloc_file(sock2, flags, NULL);
1647 if (IS_ERR(newfile2)) {
1648 err = PTR_ERR(newfile2);
1653 audit_fd_pair(fd1, fd2);
1655 fd_install(fd1, newfile1);
1656 fd_install(fd2, newfile2);
1665 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1666 int __user *, usockvec)
1668 return __sys_socketpair(family, type, protocol, usockvec);
1672 * Bind a name to a socket. Nothing much to do here since it's
1673 * the protocol's responsibility to handle the local address.
1675 * We move the socket address to kernel space before we call
1676 * the protocol layer (having also checked the address is ok).
1679 int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1681 struct socket *sock;
1682 struct sockaddr_storage address;
1683 int err, fput_needed;
1685 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1687 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1689 err = security_socket_bind(sock,
1690 (struct sockaddr *)&address,
1693 err = sock->ops->bind(sock,
1697 fput_light(sock->file, fput_needed);
1702 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1704 return __sys_bind(fd, umyaddr, addrlen);
1708 * Perform a listen. Basically, we allow the protocol to do anything
1709 * necessary for a listen, and if that works, we mark the socket as
1710 * ready for listening.
1713 int __sys_listen(int fd, int backlog)
1715 struct socket *sock;
1716 int err, fput_needed;
1719 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1721 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1722 if ((unsigned int)backlog > somaxconn)
1723 backlog = somaxconn;
1725 err = security_socket_listen(sock, backlog);
1727 err = sock->ops->listen(sock, backlog);
1729 fput_light(sock->file, fput_needed);
1734 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1736 return __sys_listen(fd, backlog);
1739 int __sys_accept4_file(struct file *file, unsigned file_flags,
1740 struct sockaddr __user *upeer_sockaddr,
1741 int __user *upeer_addrlen, int flags,
1742 unsigned long nofile)
1744 struct socket *sock, *newsock;
1745 struct file *newfile;
1746 int err, len, newfd;
1747 struct sockaddr_storage address;
1749 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1752 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1753 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1755 sock = sock_from_file(file);
1762 newsock = sock_alloc();
1766 newsock->type = sock->type;
1767 newsock->ops = sock->ops;
1770 * We don't need try_module_get here, as the listening socket (sock)
1771 * has the protocol module (sock->ops->owner) held.
1773 __module_get(newsock->ops->owner);
1775 newfd = __get_unused_fd_flags(flags, nofile);
1776 if (unlikely(newfd < 0)) {
1778 sock_release(newsock);
1781 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1782 if (IS_ERR(newfile)) {
1783 err = PTR_ERR(newfile);
1784 put_unused_fd(newfd);
1788 err = security_socket_accept(sock, newsock);
1792 err = sock->ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1797 if (upeer_sockaddr) {
1798 len = newsock->ops->getname(newsock,
1799 (struct sockaddr *)&address, 2);
1801 err = -ECONNABORTED;
1804 err = move_addr_to_user(&address,
1805 len, upeer_sockaddr, upeer_addrlen);
1810 /* File flags are not inherited via accept() unlike another OSes. */
1812 fd_install(newfd, newfile);
1818 put_unused_fd(newfd);
1824 * For accept, we attempt to create a new socket, set up the link
1825 * with the client, wake up the client, then return the new
1826 * connected fd. We collect the address of the connector in kernel
1827 * space and move it to user at the very end. This is unclean because
1828 * we open the socket then return an error.
1830 * 1003.1g adds the ability to recvmsg() to query connection pending
1831 * status to recvmsg. We need to add that support in a way thats
1832 * clean when we restructure accept also.
1835 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1836 int __user *upeer_addrlen, int flags)
1843 ret = __sys_accept4_file(f.file, 0, upeer_sockaddr,
1844 upeer_addrlen, flags,
1845 rlimit(RLIMIT_NOFILE));
1852 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1853 int __user *, upeer_addrlen, int, flags)
1855 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1858 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1859 int __user *, upeer_addrlen)
1861 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1865 * Attempt to connect to a socket with the server address. The address
1866 * is in user space so we verify it is OK and move it to kernel space.
1868 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1871 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1872 * other SEQPACKET protocols that take time to connect() as it doesn't
1873 * include the -EINPROGRESS status for such sockets.
1876 int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
1877 int addrlen, int file_flags)
1879 struct socket *sock;
1882 sock = sock_from_file(file);
1889 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1893 err = sock->ops->connect(sock, (struct sockaddr *)address, addrlen,
1894 sock->file->f_flags | file_flags);
1899 int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1906 struct sockaddr_storage address;
1908 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
1910 ret = __sys_connect_file(f.file, &address, addrlen, 0);
1917 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1920 return __sys_connect(fd, uservaddr, addrlen);
1924 * Get the local address ('name') of a socket object. Move the obtained
1925 * name to user space.
1928 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1929 int __user *usockaddr_len)
1931 struct socket *sock;
1932 struct sockaddr_storage address;
1933 int err, fput_needed;
1935 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1939 err = security_socket_getsockname(sock);
1943 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1946 /* "err" is actually length in this case */
1947 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1950 fput_light(sock->file, fput_needed);
1955 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1956 int __user *, usockaddr_len)
1958 return __sys_getsockname(fd, usockaddr, usockaddr_len);
1962 * Get the remote address ('name') of a socket object. Move the obtained
1963 * name to user space.
1966 int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1967 int __user *usockaddr_len)
1969 struct socket *sock;
1970 struct sockaddr_storage address;
1971 int err, fput_needed;
1973 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1975 err = security_socket_getpeername(sock);
1977 fput_light(sock->file, fput_needed);
1981 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
1983 /* "err" is actually length in this case */
1984 err = move_addr_to_user(&address, err, usockaddr,
1986 fput_light(sock->file, fput_needed);
1991 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1992 int __user *, usockaddr_len)
1994 return __sys_getpeername(fd, usockaddr, usockaddr_len);
1998 * Send a datagram to a given address. We move the address into kernel
1999 * space and check the user space data area is readable before invoking
2002 int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
2003 struct sockaddr __user *addr, int addr_len)
2005 struct socket *sock;
2006 struct sockaddr_storage address;
2012 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
2015 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2019 msg.msg_name = NULL;
2020 msg.msg_control = NULL;
2021 msg.msg_controllen = 0;
2022 msg.msg_namelen = 0;
2024 err = move_addr_to_kernel(addr, addr_len, &address);
2027 msg.msg_name = (struct sockaddr *)&address;
2028 msg.msg_namelen = addr_len;
2030 if (sock->file->f_flags & O_NONBLOCK)
2031 flags |= MSG_DONTWAIT;
2032 msg.msg_flags = flags;
2033 err = sock_sendmsg(sock, &msg);
2036 fput_light(sock->file, fput_needed);
2041 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2042 unsigned int, flags, struct sockaddr __user *, addr,
2045 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2049 * Send a datagram down a socket.
2052 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2053 unsigned int, flags)
2055 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2059 * Receive a frame from the socket and optionally record the address of the
2060 * sender. We verify the buffers are writable and if needed move the
2061 * sender address from kernel to user space.
2063 int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2064 struct sockaddr __user *addr, int __user *addr_len)
2066 struct socket *sock;
2069 struct sockaddr_storage address;
2073 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
2076 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2080 msg.msg_control = NULL;
2081 msg.msg_controllen = 0;
2082 /* Save some cycles and don't copy the address if not needed */
2083 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
2084 /* We assume all kernel code knows the size of sockaddr_storage */
2085 msg.msg_namelen = 0;
2086 msg.msg_iocb = NULL;
2088 if (sock->file->f_flags & O_NONBLOCK)
2089 flags |= MSG_DONTWAIT;
2090 err = sock_recvmsg(sock, &msg, flags);
2092 if (err >= 0 && addr != NULL) {
2093 err2 = move_addr_to_user(&address,
2094 msg.msg_namelen, addr, addr_len);
2099 fput_light(sock->file, fput_needed);
2104 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2105 unsigned int, flags, struct sockaddr __user *, addr,
2106 int __user *, addr_len)
2108 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2112 * Receive a datagram from a socket.
2115 SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2116 unsigned int, flags)
2118 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2121 static bool sock_use_custom_sol_socket(const struct socket *sock)
2123 const struct sock *sk = sock->sk;
2125 /* Use sock->ops->setsockopt() for MPTCP */
2126 return IS_ENABLED(CONFIG_MPTCP) &&
2127 sk->sk_protocol == IPPROTO_MPTCP &&
2128 sk->sk_type == SOCK_STREAM &&
2129 (sk->sk_family == AF_INET || sk->sk_family == AF_INET6);
2133 * Set a socket option. Because we don't know the option lengths we have
2134 * to pass the user mode parameter for the protocols to sort out.
2136 int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2139 sockptr_t optval = USER_SOCKPTR(user_optval);
2140 char *kernel_optval = NULL;
2141 int err, fput_needed;
2142 struct socket *sock;
2147 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2151 err = security_socket_setsockopt(sock, level, optname);
2155 if (!in_compat_syscall())
2156 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2157 user_optval, &optlen,
2167 optval = KERNEL_SOCKPTR(kernel_optval);
2168 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2169 err = sock_setsockopt(sock, level, optname, optval, optlen);
2170 else if (unlikely(!sock->ops->setsockopt))
2173 err = sock->ops->setsockopt(sock, level, optname, optval,
2175 kfree(kernel_optval);
2177 fput_light(sock->file, fput_needed);
2181 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2182 char __user *, optval, int, optlen)
2184 return __sys_setsockopt(fd, level, optname, optval, optlen);
2187 INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2191 * Get a socket option. Because we don't know the option lengths we have
2192 * to pass a user mode parameter for the protocols to sort out.
2194 int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2197 int err, fput_needed;
2198 struct socket *sock;
2201 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2205 err = security_socket_getsockopt(sock, level, optname);
2209 if (!in_compat_syscall())
2210 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2212 if (level == SOL_SOCKET)
2213 err = sock_getsockopt(sock, level, optname, optval, optlen);
2214 else if (unlikely(!sock->ops->getsockopt))
2217 err = sock->ops->getsockopt(sock, level, optname, optval,
2220 if (!in_compat_syscall())
2221 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2222 optval, optlen, max_optlen,
2225 fput_light(sock->file, fput_needed);
2229 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2230 char __user *, optval, int __user *, optlen)
2232 return __sys_getsockopt(fd, level, optname, optval, optlen);
2236 * Shutdown a socket.
2239 int __sys_shutdown_sock(struct socket *sock, int how)
2243 err = security_socket_shutdown(sock, how);
2245 err = sock->ops->shutdown(sock, how);
2250 int __sys_shutdown(int fd, int how)
2252 int err, fput_needed;
2253 struct socket *sock;
2255 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2257 err = __sys_shutdown_sock(sock, how);
2258 fput_light(sock->file, fput_needed);
2263 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2265 return __sys_shutdown(fd, how);
2268 /* A couple of helpful macros for getting the address of the 32/64 bit
2269 * fields which are the same type (int / unsigned) on our platforms.
2271 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2272 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2273 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2275 struct used_address {
2276 struct sockaddr_storage name;
2277 unsigned int name_len;
2280 int __copy_msghdr_from_user(struct msghdr *kmsg,
2281 struct user_msghdr __user *umsg,
2282 struct sockaddr __user **save_addr,
2283 struct iovec __user **uiov, size_t *nsegs)
2285 struct user_msghdr msg;
2288 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2291 kmsg->msg_control_is_user = true;
2292 kmsg->msg_control_user = msg.msg_control;
2293 kmsg->msg_controllen = msg.msg_controllen;
2294 kmsg->msg_flags = msg.msg_flags;
2296 kmsg->msg_namelen = msg.msg_namelen;
2298 kmsg->msg_namelen = 0;
2300 if (kmsg->msg_namelen < 0)
2303 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2304 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2307 *save_addr = msg.msg_name;
2309 if (msg.msg_name && kmsg->msg_namelen) {
2311 err = move_addr_to_kernel(msg.msg_name,
2318 kmsg->msg_name = NULL;
2319 kmsg->msg_namelen = 0;
2322 if (msg.msg_iovlen > UIO_MAXIOV)
2325 kmsg->msg_iocb = NULL;
2326 *uiov = msg.msg_iov;
2327 *nsegs = msg.msg_iovlen;
2331 static int copy_msghdr_from_user(struct msghdr *kmsg,
2332 struct user_msghdr __user *umsg,
2333 struct sockaddr __user **save_addr,
2336 struct user_msghdr msg;
2339 err = __copy_msghdr_from_user(kmsg, umsg, save_addr, &msg.msg_iov,
2344 err = import_iovec(save_addr ? READ : WRITE,
2345 msg.msg_iov, msg.msg_iovlen,
2346 UIO_FASTIOV, iov, &kmsg->msg_iter);
2347 return err < 0 ? err : 0;
2350 static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2351 unsigned int flags, struct used_address *used_address,
2352 unsigned int allowed_msghdr_flags)
2354 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2355 __aligned(sizeof(__kernel_size_t));
2356 /* 20 is size of ipv6_pktinfo */
2357 unsigned char *ctl_buf = ctl;
2363 if (msg_sys->msg_controllen > INT_MAX)
2365 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2366 ctl_len = msg_sys->msg_controllen;
2367 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2369 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2373 ctl_buf = msg_sys->msg_control;
2374 ctl_len = msg_sys->msg_controllen;
2375 } else if (ctl_len) {
2376 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2377 CMSG_ALIGN(sizeof(struct cmsghdr)));
2378 if (ctl_len > sizeof(ctl)) {
2379 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2380 if (ctl_buf == NULL)
2384 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2386 msg_sys->msg_control = ctl_buf;
2387 msg_sys->msg_control_is_user = false;
2389 msg_sys->msg_flags = flags;
2391 if (sock->file->f_flags & O_NONBLOCK)
2392 msg_sys->msg_flags |= MSG_DONTWAIT;
2394 * If this is sendmmsg() and current destination address is same as
2395 * previously succeeded address, omit asking LSM's decision.
2396 * used_address->name_len is initialized to UINT_MAX so that the first
2397 * destination address never matches.
2399 if (used_address && msg_sys->msg_name &&
2400 used_address->name_len == msg_sys->msg_namelen &&
2401 !memcmp(&used_address->name, msg_sys->msg_name,
2402 used_address->name_len)) {
2403 err = sock_sendmsg_nosec(sock, msg_sys);
2406 err = sock_sendmsg(sock, msg_sys);
2408 * If this is sendmmsg() and sending to current destination address was
2409 * successful, remember it.
2411 if (used_address && err >= 0) {
2412 used_address->name_len = msg_sys->msg_namelen;
2413 if (msg_sys->msg_name)
2414 memcpy(&used_address->name, msg_sys->msg_name,
2415 used_address->name_len);
2420 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2425 int sendmsg_copy_msghdr(struct msghdr *msg,
2426 struct user_msghdr __user *umsg, unsigned flags,
2431 if (flags & MSG_CMSG_COMPAT) {
2432 struct compat_msghdr __user *msg_compat;
2434 msg_compat = (struct compat_msghdr __user *) umsg;
2435 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2437 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2445 static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2446 struct msghdr *msg_sys, unsigned int flags,
2447 struct used_address *used_address,
2448 unsigned int allowed_msghdr_flags)
2450 struct sockaddr_storage address;
2451 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2454 msg_sys->msg_name = &address;
2456 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2460 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2461 allowed_msghdr_flags);
2467 * BSD sendmsg interface
2469 long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2472 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2475 long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2476 bool forbid_cmsg_compat)
2478 int fput_needed, err;
2479 struct msghdr msg_sys;
2480 struct socket *sock;
2482 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2485 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2489 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2491 fput_light(sock->file, fput_needed);
2496 SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2498 return __sys_sendmsg(fd, msg, flags, true);
2502 * Linux sendmmsg interface
2505 int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2506 unsigned int flags, bool forbid_cmsg_compat)
2508 int fput_needed, err, datagrams;
2509 struct socket *sock;
2510 struct mmsghdr __user *entry;
2511 struct compat_mmsghdr __user *compat_entry;
2512 struct msghdr msg_sys;
2513 struct used_address used_address;
2514 unsigned int oflags = flags;
2516 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2519 if (vlen > UIO_MAXIOV)
2524 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2528 used_address.name_len = UINT_MAX;
2530 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2534 while (datagrams < vlen) {
2535 if (datagrams == vlen - 1)
2538 if (MSG_CMSG_COMPAT & flags) {
2539 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2540 &msg_sys, flags, &used_address, MSG_EOR);
2543 err = __put_user(err, &compat_entry->msg_len);
2546 err = ___sys_sendmsg(sock,
2547 (struct user_msghdr __user *)entry,
2548 &msg_sys, flags, &used_address, MSG_EOR);
2551 err = put_user(err, &entry->msg_len);
2558 if (msg_data_left(&msg_sys))
2563 fput_light(sock->file, fput_needed);
2565 /* We only return an error if no datagrams were able to be sent */
2572 SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2573 unsigned int, vlen, unsigned int, flags)
2575 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2578 int recvmsg_copy_msghdr(struct msghdr *msg,
2579 struct user_msghdr __user *umsg, unsigned flags,
2580 struct sockaddr __user **uaddr,
2585 if (MSG_CMSG_COMPAT & flags) {
2586 struct compat_msghdr __user *msg_compat;
2588 msg_compat = (struct compat_msghdr __user *) umsg;
2589 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2591 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2599 static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2600 struct user_msghdr __user *msg,
2601 struct sockaddr __user *uaddr,
2602 unsigned int flags, int nosec)
2604 struct compat_msghdr __user *msg_compat =
2605 (struct compat_msghdr __user *) msg;
2606 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2607 struct sockaddr_storage addr;
2608 unsigned long cmsg_ptr;
2612 msg_sys->msg_name = &addr;
2613 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2614 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2616 /* We assume all kernel code knows the size of sockaddr_storage */
2617 msg_sys->msg_namelen = 0;
2619 if (sock->file->f_flags & O_NONBLOCK)
2620 flags |= MSG_DONTWAIT;
2622 if (unlikely(nosec))
2623 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2625 err = sock_recvmsg(sock, msg_sys, flags);
2631 if (uaddr != NULL) {
2632 err = move_addr_to_user(&addr,
2633 msg_sys->msg_namelen, uaddr,
2638 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2642 if (MSG_CMSG_COMPAT & flags)
2643 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2644 &msg_compat->msg_controllen);
2646 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2647 &msg->msg_controllen);
2655 static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2656 struct msghdr *msg_sys, unsigned int flags, int nosec)
2658 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2659 /* user mode address pointers */
2660 struct sockaddr __user *uaddr;
2663 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2667 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2673 * BSD recvmsg interface
2676 long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2677 struct user_msghdr __user *umsg,
2678 struct sockaddr __user *uaddr, unsigned int flags)
2680 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2683 long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2684 bool forbid_cmsg_compat)
2686 int fput_needed, err;
2687 struct msghdr msg_sys;
2688 struct socket *sock;
2690 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2693 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2697 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2699 fput_light(sock->file, fput_needed);
2704 SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2705 unsigned int, flags)
2707 return __sys_recvmsg(fd, msg, flags, true);
2711 * Linux recvmmsg interface
2714 static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2715 unsigned int vlen, unsigned int flags,
2716 struct timespec64 *timeout)
2718 int fput_needed, err, datagrams;
2719 struct socket *sock;
2720 struct mmsghdr __user *entry;
2721 struct compat_mmsghdr __user *compat_entry;
2722 struct msghdr msg_sys;
2723 struct timespec64 end_time;
2724 struct timespec64 timeout64;
2727 poll_select_set_timeout(&end_time, timeout->tv_sec,
2733 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2737 if (likely(!(flags & MSG_ERRQUEUE))) {
2738 err = sock_error(sock->sk);
2746 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2748 while (datagrams < vlen) {
2750 * No need to ask LSM for more than the first datagram.
2752 if (MSG_CMSG_COMPAT & flags) {
2753 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2754 &msg_sys, flags & ~MSG_WAITFORONE,
2758 err = __put_user(err, &compat_entry->msg_len);
2761 err = ___sys_recvmsg(sock,
2762 (struct user_msghdr __user *)entry,
2763 &msg_sys, flags & ~MSG_WAITFORONE,
2767 err = put_user(err, &entry->msg_len);
2775 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2776 if (flags & MSG_WAITFORONE)
2777 flags |= MSG_DONTWAIT;
2780 ktime_get_ts64(&timeout64);
2781 *timeout = timespec64_sub(end_time, timeout64);
2782 if (timeout->tv_sec < 0) {
2783 timeout->tv_sec = timeout->tv_nsec = 0;
2787 /* Timeout, return less than vlen datagrams */
2788 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2792 /* Out of band data, return right away */
2793 if (msg_sys.msg_flags & MSG_OOB)
2801 if (datagrams == 0) {
2807 * We may return less entries than requested (vlen) if the
2808 * sock is non block and there aren't enough datagrams...
2810 if (err != -EAGAIN) {
2812 * ... or if recvmsg returns an error after we
2813 * received some datagrams, where we record the
2814 * error to return on the next call or if the
2815 * app asks about it using getsockopt(SO_ERROR).
2817 sock->sk->sk_err = -err;
2820 fput_light(sock->file, fput_needed);
2825 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2826 unsigned int vlen, unsigned int flags,
2827 struct __kernel_timespec __user *timeout,
2828 struct old_timespec32 __user *timeout32)
2831 struct timespec64 timeout_sys;
2833 if (timeout && get_timespec64(&timeout_sys, timeout))
2836 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2839 if (!timeout && !timeout32)
2840 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2842 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2847 if (timeout && put_timespec64(&timeout_sys, timeout))
2848 datagrams = -EFAULT;
2850 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2851 datagrams = -EFAULT;
2856 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2857 unsigned int, vlen, unsigned int, flags,
2858 struct __kernel_timespec __user *, timeout)
2860 if (flags & MSG_CMSG_COMPAT)
2863 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2866 #ifdef CONFIG_COMPAT_32BIT_TIME
2867 SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2868 unsigned int, vlen, unsigned int, flags,
2869 struct old_timespec32 __user *, timeout)
2871 if (flags & MSG_CMSG_COMPAT)
2874 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2878 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2879 /* Argument list sizes for sys_socketcall */
2880 #define AL(x) ((x) * sizeof(unsigned long))
2881 static const unsigned char nargs[21] = {
2882 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2883 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2884 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2891 * System call vectors.
2893 * Argument checking cleaned up. Saved 20% in size.
2894 * This function doesn't need to set the kernel lock because
2895 * it is set by the callees.
2898 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2900 unsigned long a[AUDITSC_ARGS];
2901 unsigned long a0, a1;
2905 if (call < 1 || call > SYS_SENDMMSG)
2907 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2910 if (len > sizeof(a))
2913 /* copy_from_user should be SMP safe. */
2914 if (copy_from_user(a, args, len))
2917 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2926 err = __sys_socket(a0, a1, a[2]);
2929 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2932 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2935 err = __sys_listen(a0, a1);
2938 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2939 (int __user *)a[2], 0);
2941 case SYS_GETSOCKNAME:
2943 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2944 (int __user *)a[2]);
2946 case SYS_GETPEERNAME:
2948 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2949 (int __user *)a[2]);
2951 case SYS_SOCKETPAIR:
2952 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2955 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2959 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2960 (struct sockaddr __user *)a[4], a[5]);
2963 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2967 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2968 (struct sockaddr __user *)a[4],
2969 (int __user *)a[5]);
2972 err = __sys_shutdown(a0, a1);
2974 case SYS_SETSOCKOPT:
2975 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
2978 case SYS_GETSOCKOPT:
2980 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2981 (int __user *)a[4]);
2984 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
2988 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
2992 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
2996 if (IS_ENABLED(CONFIG_64BIT))
2997 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2999 (struct __kernel_timespec __user *)a[4],
3002 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3004 (struct old_timespec32 __user *)a[4]);
3007 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3008 (int __user *)a[2], a[3]);
3017 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
3020 * sock_register - add a socket protocol handler
3021 * @ops: description of protocol
3023 * This function is called by a protocol handler that wants to
3024 * advertise its address family, and have it linked into the
3025 * socket interface. The value ops->family corresponds to the
3026 * socket system call protocol family.
3028 int sock_register(const struct net_proto_family *ops)
3032 if (ops->family >= NPROTO) {
3033 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3037 spin_lock(&net_family_lock);
3038 if (rcu_dereference_protected(net_families[ops->family],
3039 lockdep_is_held(&net_family_lock)))
3042 rcu_assign_pointer(net_families[ops->family], ops);
3045 spin_unlock(&net_family_lock);
3047 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3050 EXPORT_SYMBOL(sock_register);
3053 * sock_unregister - remove a protocol handler
3054 * @family: protocol family to remove
3056 * This function is called by a protocol handler that wants to
3057 * remove its address family, and have it unlinked from the
3058 * new socket creation.
3060 * If protocol handler is a module, then it can use module reference
3061 * counts to protect against new references. If protocol handler is not
3062 * a module then it needs to provide its own protection in
3063 * the ops->create routine.
3065 void sock_unregister(int family)
3067 BUG_ON(family < 0 || family >= NPROTO);
3069 spin_lock(&net_family_lock);
3070 RCU_INIT_POINTER(net_families[family], NULL);
3071 spin_unlock(&net_family_lock);
3075 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3077 EXPORT_SYMBOL(sock_unregister);
3079 bool sock_is_registered(int family)
3081 return family < NPROTO && rcu_access_pointer(net_families[family]);
3084 static int __init sock_init(void)
3088 * Initialize the network sysctl infrastructure.
3090 err = net_sysctl_init();
3095 * Initialize skbuff SLAB cache
3100 * Initialize the protocols module.
3105 err = register_filesystem(&sock_fs_type);
3108 sock_mnt = kern_mount(&sock_fs_type);
3109 if (IS_ERR(sock_mnt)) {
3110 err = PTR_ERR(sock_mnt);
3114 /* The real protocol initialization is performed in later initcalls.
3117 #ifdef CONFIG_NETFILTER
3118 err = netfilter_init();
3123 ptp_classifier_init();
3129 unregister_filesystem(&sock_fs_type);
3133 core_initcall(sock_init); /* early initcall */
3135 #ifdef CONFIG_PROC_FS
3136 void socket_seq_show(struct seq_file *seq)
3138 seq_printf(seq, "sockets: used %d\n",
3139 sock_inuse_get(seq->private));
3141 #endif /* CONFIG_PROC_FS */
3143 /* Handle the fact that while struct ifreq has the same *layout* on
3144 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3145 * which are handled elsewhere, it still has different *size* due to
3146 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3147 * resulting in struct ifreq being 32 and 40 bytes respectively).
3148 * As a result, if the struct happens to be at the end of a page and
3149 * the next page isn't readable/writable, we get a fault. To prevent
3150 * that, copy back and forth to the full size.
3152 int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3154 if (in_compat_syscall()) {
3155 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3157 memset(ifr, 0, sizeof(*ifr));
3158 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3162 *ifrdata = compat_ptr(ifr32->ifr_data);
3167 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3171 *ifrdata = ifr->ifr_data;
3175 EXPORT_SYMBOL(get_user_ifreq);
3177 int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3179 size_t size = sizeof(*ifr);
3181 if (in_compat_syscall())
3182 size = sizeof(struct compat_ifreq);
3184 if (copy_to_user(arg, ifr, size))
3189 EXPORT_SYMBOL(put_user_ifreq);
3191 #ifdef CONFIG_COMPAT
3192 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3194 compat_uptr_t uptr32;
3199 if (get_user_ifreq(&ifr, NULL, uifr32))
3202 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3205 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3206 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3208 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3210 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3211 if (put_user_ifreq(&ifr, uifr32))
3217 /* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3218 static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3219 struct compat_ifreq __user *u_ifreq32)
3224 if (!is_socket_ioctl_cmd(cmd))
3226 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3228 ifreq.ifr_data = data;
3230 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3233 /* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3234 * for some operations; this forces use of the newer bridge-utils that
3235 * use compatible ioctls
3237 static int old_bridge_ioctl(compat_ulong_t __user *argp)
3241 if (get_user(tmp, argp))
3243 if (tmp == BRCTL_GET_VERSION)
3244 return BRCTL_VERSION + 1;
3248 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3249 unsigned int cmd, unsigned long arg)
3251 void __user *argp = compat_ptr(arg);
3252 struct sock *sk = sock->sk;
3253 struct net *net = sock_net(sk);
3255 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3256 return sock_ioctl(file, cmd, (unsigned long)argp);
3261 return old_bridge_ioctl(argp);
3263 return compat_siocwandev(net, argp);
3264 case SIOCGSTAMP_OLD:
3265 case SIOCGSTAMPNS_OLD:
3266 if (!sock->ops->gettstamp)
3267 return -ENOIOCTLCMD;
3268 return sock->ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3269 !COMPAT_USE_64BIT_TIME);
3272 case SIOCBONDSLAVEINFOQUERY:
3273 case SIOCBONDINFOQUERY:
3276 return compat_ifr_data_ioctl(net, cmd, argp);
3287 case SIOCGSTAMP_NEW:
3288 case SIOCGSTAMPNS_NEW:
3290 return sock_ioctl(file, cmd, arg);
3309 case SIOCSIFHWBROADCAST:
3311 case SIOCGIFBRDADDR:
3312 case SIOCSIFBRDADDR:
3313 case SIOCGIFDSTADDR:
3314 case SIOCSIFDSTADDR:
3315 case SIOCGIFNETMASK:
3316 case SIOCSIFNETMASK:
3328 case SIOCBONDENSLAVE:
3329 case SIOCBONDRELEASE:
3330 case SIOCBONDSETHWADDR:
3331 case SIOCBONDCHANGEACTIVE:
3338 return sock_do_ioctl(net, sock, cmd, arg);
3341 return -ENOIOCTLCMD;
3344 static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3347 struct socket *sock = file->private_data;
3348 int ret = -ENOIOCTLCMD;
3355 if (sock->ops->compat_ioctl)
3356 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3358 if (ret == -ENOIOCTLCMD &&
3359 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3360 ret = compat_wext_handle_ioctl(net, cmd, arg);
3362 if (ret == -ENOIOCTLCMD)
3363 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3370 * kernel_bind - bind an address to a socket (kernel space)
3373 * @addrlen: length of address
3375 * Returns 0 or an error.
3378 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3380 return sock->ops->bind(sock, addr, addrlen);
3382 EXPORT_SYMBOL(kernel_bind);
3385 * kernel_listen - move socket to listening state (kernel space)
3387 * @backlog: pending connections queue size
3389 * Returns 0 or an error.
3392 int kernel_listen(struct socket *sock, int backlog)
3394 return sock->ops->listen(sock, backlog);
3396 EXPORT_SYMBOL(kernel_listen);
3399 * kernel_accept - accept a connection (kernel space)
3400 * @sock: listening socket
3401 * @newsock: new connected socket
3404 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3405 * If it fails, @newsock is guaranteed to be %NULL.
3406 * Returns 0 or an error.
3409 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3411 struct sock *sk = sock->sk;
3414 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3419 err = sock->ops->accept(sock, *newsock, flags, true);
3421 sock_release(*newsock);
3426 (*newsock)->ops = sock->ops;
3427 __module_get((*newsock)->ops->owner);
3432 EXPORT_SYMBOL(kernel_accept);
3435 * kernel_connect - connect a socket (kernel space)
3438 * @addrlen: address length
3439 * @flags: flags (O_NONBLOCK, ...)
3441 * For datagram sockets, @addr is the address to which datagrams are sent
3442 * by default, and the only address from which datagrams are received.
3443 * For stream sockets, attempts to connect to @addr.
3444 * Returns 0 or an error code.
3447 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3450 return sock->ops->connect(sock, addr, addrlen, flags);
3452 EXPORT_SYMBOL(kernel_connect);
3455 * kernel_getsockname - get the address which the socket is bound (kernel space)
3457 * @addr: address holder
3459 * Fills the @addr pointer with the address which the socket is bound.
3460 * Returns 0 or an error code.
3463 int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3465 return sock->ops->getname(sock, addr, 0);
3467 EXPORT_SYMBOL(kernel_getsockname);
3470 * kernel_getpeername - get the address which the socket is connected (kernel space)
3472 * @addr: address holder
3474 * Fills the @addr pointer with the address which the socket is connected.
3475 * Returns 0 or an error code.
3478 int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3480 return sock->ops->getname(sock, addr, 1);
3482 EXPORT_SYMBOL(kernel_getpeername);
3485 * kernel_sendpage - send a &page through a socket (kernel space)
3488 * @offset: page offset
3489 * @size: total size in bytes
3490 * @flags: flags (MSG_DONTWAIT, ...)
3492 * Returns the total amount sent in bytes or an error.
3495 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3496 size_t size, int flags)
3498 if (sock->ops->sendpage) {
3499 /* Warn in case the improper page to zero-copy send */
3500 WARN_ONCE(!sendpage_ok(page), "improper page for zero-copy send");
3501 return sock->ops->sendpage(sock, page, offset, size, flags);
3503 return sock_no_sendpage(sock, page, offset, size, flags);
3505 EXPORT_SYMBOL(kernel_sendpage);
3508 * kernel_sendpage_locked - send a &page through the locked sock (kernel space)
3511 * @offset: page offset
3512 * @size: total size in bytes
3513 * @flags: flags (MSG_DONTWAIT, ...)
3515 * Returns the total amount sent in bytes or an error.
3516 * Caller must hold @sk.
3519 int kernel_sendpage_locked(struct sock *sk, struct page *page, int offset,
3520 size_t size, int flags)
3522 struct socket *sock = sk->sk_socket;
3524 if (sock->ops->sendpage_locked)
3525 return sock->ops->sendpage_locked(sk, page, offset, size,
3528 return sock_no_sendpage_locked(sk, page, offset, size, flags);
3530 EXPORT_SYMBOL(kernel_sendpage_locked);
3533 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3535 * @how: connection part
3537 * Returns 0 or an error.
3540 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3542 return sock->ops->shutdown(sock, how);
3544 EXPORT_SYMBOL(kernel_sock_shutdown);
3547 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3550 * This routine returns the IP overhead imposed by a socket i.e.
3551 * the length of the underlying IP header, depending on whether
3552 * this is an IPv4 or IPv6 socket and the length from IP options turned
3553 * on at the socket. Assumes that the caller has a lock on the socket.
3556 u32 kernel_sock_ip_overhead(struct sock *sk)
3558 struct inet_sock *inet;
3559 struct ip_options_rcu *opt;
3561 #if IS_ENABLED(CONFIG_IPV6)
3562 struct ipv6_pinfo *np;
3563 struct ipv6_txoptions *optv6 = NULL;
3564 #endif /* IS_ENABLED(CONFIG_IPV6) */
3569 switch (sk->sk_family) {
3572 overhead += sizeof(struct iphdr);
3573 opt = rcu_dereference_protected(inet->inet_opt,
3574 sock_owned_by_user(sk));
3576 overhead += opt->opt.optlen;
3578 #if IS_ENABLED(CONFIG_IPV6)
3581 overhead += sizeof(struct ipv6hdr);
3583 optv6 = rcu_dereference_protected(np->opt,
3584 sock_owned_by_user(sk));
3586 overhead += (optv6->opt_flen + optv6->opt_nflen);
3588 #endif /* IS_ENABLED(CONFIG_IPV6) */
3589 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3593 EXPORT_SYMBOL(kernel_sock_ip_overhead);