Loading...
1/*
2 * NET An implementation of the SOCKET network access protocol.
3 *
4 * Version: @(#)socket.c 1.1.93 18/02/95
5 *
6 * Authors: Orest Zborowski, <obz@Kodak.COM>
7 * Ross Biro
8 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
9 *
10 * Fixes:
11 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
12 * shutdown()
13 * Alan Cox : verify_area() fixes
14 * Alan Cox : Removed DDI
15 * Jonathan Kamens : SOCK_DGRAM reconnect bug
16 * Alan Cox : Moved a load of checks to the very
17 * top level.
18 * Alan Cox : Move address structures to/from user
19 * mode above the protocol layers.
20 * Rob Janssen : Allow 0 length sends.
21 * Alan Cox : Asynchronous I/O support (cribbed from the
22 * tty drivers).
23 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
24 * Jeff Uphoff : Made max number of sockets command-line
25 * configurable.
26 * Matti Aarnio : Made the number of sockets dynamic,
27 * to be allocated when needed, and mr.
28 * Uphoff's max is used as max to be
29 * allowed to allocate.
30 * Linus : Argh. removed all the socket allocation
31 * altogether: it's in the inode now.
32 * Alan Cox : Made sock_alloc()/sock_release() public
33 * for NetROM and future kernel nfsd type
34 * stuff.
35 * Alan Cox : sendmsg/recvmsg basics.
36 * Tom Dyas : Export net symbols.
37 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
38 * Alan Cox : Added thread locking to sys_* calls
39 * for sockets. May have errors at the
40 * moment.
41 * Kevin Buhr : Fixed the dumb errors in the above.
42 * Andi Kleen : Some small cleanups, optimizations,
43 * and fixed a copy_from_user() bug.
44 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
45 * Tigran Aivazian : Made listen(2) backlog sanity checks
46 * protocol-independent
47 *
48 *
49 * This program is free software; you can redistribute it and/or
50 * modify it under the terms of the GNU General Public License
51 * as published by the Free Software Foundation; either version
52 * 2 of the License, or (at your option) any later version.
53 *
54 *
55 * This module is effectively the top level interface to the BSD socket
56 * paradigm.
57 *
58 * Based upon Swansea University Computer Society NET3.039
59 */
60
61#include <linux/mm.h>
62#include <linux/socket.h>
63#include <linux/file.h>
64#include <linux/net.h>
65#include <linux/interrupt.h>
66#include <linux/thread_info.h>
67#include <linux/rcupdate.h>
68#include <linux/netdevice.h>
69#include <linux/proc_fs.h>
70#include <linux/seq_file.h>
71#include <linux/mutex.h>
72#include <linux/if_bridge.h>
73#include <linux/if_frad.h>
74#include <linux/if_vlan.h>
75#include <linux/ptp_classify.h>
76#include <linux/init.h>
77#include <linux/poll.h>
78#include <linux/cache.h>
79#include <linux/module.h>
80#include <linux/highmem.h>
81#include <linux/mount.h>
82#include <linux/security.h>
83#include <linux/syscalls.h>
84#include <linux/compat.h>
85#include <linux/kmod.h>
86#include <linux/audit.h>
87#include <linux/wireless.h>
88#include <linux/nsproxy.h>
89#include <linux/magic.h>
90#include <linux/slab.h>
91#include <linux/xattr.h>
92
93#include <linux/uaccess.h>
94#include <asm/unistd.h>
95
96#include <net/compat.h>
97#include <net/wext.h>
98#include <net/cls_cgroup.h>
99
100#include <net/sock.h>
101#include <linux/netfilter.h>
102
103#include <linux/if_tun.h>
104#include <linux/ipv6_route.h>
105#include <linux/route.h>
106#include <linux/sockios.h>
107#include <linux/atalk.h>
108#include <net/busy_poll.h>
109#include <linux/errqueue.h>
110
111#ifdef CONFIG_NET_RX_BUSY_POLL
112unsigned int sysctl_net_busy_read __read_mostly;
113unsigned int sysctl_net_busy_poll __read_mostly;
114#endif
115
116static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
117static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
118static int sock_mmap(struct file *file, struct vm_area_struct *vma);
119
120static int sock_close(struct inode *inode, struct file *file);
121static unsigned int sock_poll(struct file *file,
122 struct poll_table_struct *wait);
123static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
124#ifdef CONFIG_COMPAT
125static long compat_sock_ioctl(struct file *file,
126 unsigned int cmd, unsigned long arg);
127#endif
128static int sock_fasync(int fd, struct file *filp, int on);
129static ssize_t sock_sendpage(struct file *file, struct page *page,
130 int offset, size_t size, loff_t *ppos, int more);
131static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
132 struct pipe_inode_info *pipe, size_t len,
133 unsigned int flags);
134
135/*
136 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
137 * in the operation structures but are done directly via the socketcall() multiplexor.
138 */
139
140static const struct file_operations socket_file_ops = {
141 .owner = THIS_MODULE,
142 .llseek = no_llseek,
143 .read_iter = sock_read_iter,
144 .write_iter = sock_write_iter,
145 .poll = sock_poll,
146 .unlocked_ioctl = sock_ioctl,
147#ifdef CONFIG_COMPAT
148 .compat_ioctl = compat_sock_ioctl,
149#endif
150 .mmap = sock_mmap,
151 .release = sock_close,
152 .fasync = sock_fasync,
153 .sendpage = sock_sendpage,
154 .splice_write = generic_splice_sendpage,
155 .splice_read = sock_splice_read,
156};
157
158/*
159 * The protocol list. Each protocol is registered in here.
160 */
161
162static DEFINE_SPINLOCK(net_family_lock);
163static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
164
165/*
166 * Statistics counters of the socket lists
167 */
168
169static DEFINE_PER_CPU(int, sockets_in_use);
170
171/*
172 * Support routines.
173 * Move socket addresses back and forth across the kernel/user
174 * divide and look after the messy bits.
175 */
176
177/**
178 * move_addr_to_kernel - copy a socket address into kernel space
179 * @uaddr: Address in user space
180 * @kaddr: Address in kernel space
181 * @ulen: Length in user space
182 *
183 * The address is copied into kernel space. If the provided address is
184 * too long an error code of -EINVAL is returned. If the copy gives
185 * invalid addresses -EFAULT is returned. On a success 0 is returned.
186 */
187
188int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
189{
190 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
191 return -EINVAL;
192 if (ulen == 0)
193 return 0;
194 if (copy_from_user(kaddr, uaddr, ulen))
195 return -EFAULT;
196 return audit_sockaddr(ulen, kaddr);
197}
198
199/**
200 * move_addr_to_user - copy an address to user space
201 * @kaddr: kernel space address
202 * @klen: length of address in kernel
203 * @uaddr: user space address
204 * @ulen: pointer to user length field
205 *
206 * The value pointed to by ulen on entry is the buffer length available.
207 * This is overwritten with the buffer space used. -EINVAL is returned
208 * if an overlong buffer is specified or a negative buffer size. -EFAULT
209 * is returned if either the buffer or the length field are not
210 * accessible.
211 * After copying the data up to the limit the user specifies, the true
212 * length of the data is written over the length limit the user
213 * specified. Zero is returned for a success.
214 */
215
216static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
217 void __user *uaddr, int __user *ulen)
218{
219 int err;
220 int len;
221
222 BUG_ON(klen > sizeof(struct sockaddr_storage));
223 err = get_user(len, ulen);
224 if (err)
225 return err;
226 if (len > klen)
227 len = klen;
228 if (len < 0)
229 return -EINVAL;
230 if (len) {
231 if (audit_sockaddr(klen, kaddr))
232 return -ENOMEM;
233 if (copy_to_user(uaddr, kaddr, len))
234 return -EFAULT;
235 }
236 /*
237 * "fromlen shall refer to the value before truncation.."
238 * 1003.1g
239 */
240 return __put_user(klen, ulen);
241}
242
243static struct kmem_cache *sock_inode_cachep __read_mostly;
244
245static struct inode *sock_alloc_inode(struct super_block *sb)
246{
247 struct socket_alloc *ei;
248 struct socket_wq *wq;
249
250 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
251 if (!ei)
252 return NULL;
253 wq = kmalloc(sizeof(*wq), GFP_KERNEL);
254 if (!wq) {
255 kmem_cache_free(sock_inode_cachep, ei);
256 return NULL;
257 }
258 init_waitqueue_head(&wq->wait);
259 wq->fasync_list = NULL;
260 wq->flags = 0;
261 RCU_INIT_POINTER(ei->socket.wq, wq);
262
263 ei->socket.state = SS_UNCONNECTED;
264 ei->socket.flags = 0;
265 ei->socket.ops = NULL;
266 ei->socket.sk = NULL;
267 ei->socket.file = NULL;
268
269 return &ei->vfs_inode;
270}
271
272static void sock_destroy_inode(struct inode *inode)
273{
274 struct socket_alloc *ei;
275 struct socket_wq *wq;
276
277 ei = container_of(inode, struct socket_alloc, vfs_inode);
278 wq = rcu_dereference_protected(ei->socket.wq, 1);
279 kfree_rcu(wq, rcu);
280 kmem_cache_free(sock_inode_cachep, ei);
281}
282
283static void init_once(void *foo)
284{
285 struct socket_alloc *ei = (struct socket_alloc *)foo;
286
287 inode_init_once(&ei->vfs_inode);
288}
289
290static int init_inodecache(void)
291{
292 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
293 sizeof(struct socket_alloc),
294 0,
295 (SLAB_HWCACHE_ALIGN |
296 SLAB_RECLAIM_ACCOUNT |
297 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
298 init_once);
299 if (sock_inode_cachep == NULL)
300 return -ENOMEM;
301 return 0;
302}
303
304static const struct super_operations sockfs_ops = {
305 .alloc_inode = sock_alloc_inode,
306 .destroy_inode = sock_destroy_inode,
307 .statfs = simple_statfs,
308};
309
310/*
311 * sockfs_dname() is called from d_path().
312 */
313static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
314{
315 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
316 d_inode(dentry)->i_ino);
317}
318
319static const struct dentry_operations sockfs_dentry_operations = {
320 .d_dname = sockfs_dname,
321};
322
323static int sockfs_xattr_get(const struct xattr_handler *handler,
324 struct dentry *dentry, struct inode *inode,
325 const char *suffix, void *value, size_t size)
326{
327 if (value) {
328 if (dentry->d_name.len + 1 > size)
329 return -ERANGE;
330 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
331 }
332 return dentry->d_name.len + 1;
333}
334
335#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
336#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
337#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
338
339static const struct xattr_handler sockfs_xattr_handler = {
340 .name = XATTR_NAME_SOCKPROTONAME,
341 .get = sockfs_xattr_get,
342};
343
344static int sockfs_security_xattr_set(const struct xattr_handler *handler,
345 struct dentry *dentry, struct inode *inode,
346 const char *suffix, const void *value,
347 size_t size, int flags)
348{
349 /* Handled by LSM. */
350 return -EAGAIN;
351}
352
353static const struct xattr_handler sockfs_security_xattr_handler = {
354 .prefix = XATTR_SECURITY_PREFIX,
355 .set = sockfs_security_xattr_set,
356};
357
358static const struct xattr_handler *sockfs_xattr_handlers[] = {
359 &sockfs_xattr_handler,
360 &sockfs_security_xattr_handler,
361 NULL
362};
363
364static struct dentry *sockfs_mount(struct file_system_type *fs_type,
365 int flags, const char *dev_name, void *data)
366{
367 return mount_pseudo_xattr(fs_type, "socket:", &sockfs_ops,
368 sockfs_xattr_handlers,
369 &sockfs_dentry_operations, SOCKFS_MAGIC);
370}
371
372static struct vfsmount *sock_mnt __read_mostly;
373
374static struct file_system_type sock_fs_type = {
375 .name = "sockfs",
376 .mount = sockfs_mount,
377 .kill_sb = kill_anon_super,
378};
379
380/*
381 * Obtains the first available file descriptor and sets it up for use.
382 *
383 * These functions create file structures and maps them to fd space
384 * of the current process. On success it returns file descriptor
385 * and file struct implicitly stored in sock->file.
386 * Note that another thread may close file descriptor before we return
387 * from this function. We use the fact that now we do not refer
388 * to socket after mapping. If one day we will need it, this
389 * function will increment ref. count on file by 1.
390 *
391 * In any case returned fd MAY BE not valid!
392 * This race condition is unavoidable
393 * with shared fd spaces, we cannot solve it inside kernel,
394 * but we take care of internal coherence yet.
395 */
396
397struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
398{
399 struct qstr name = { .name = "" };
400 struct path path;
401 struct file *file;
402
403 if (dname) {
404 name.name = dname;
405 name.len = strlen(name.name);
406 } else if (sock->sk) {
407 name.name = sock->sk->sk_prot_creator->name;
408 name.len = strlen(name.name);
409 }
410 path.dentry = d_alloc_pseudo(sock_mnt->mnt_sb, &name);
411 if (unlikely(!path.dentry))
412 return ERR_PTR(-ENOMEM);
413 path.mnt = mntget(sock_mnt);
414
415 d_instantiate(path.dentry, SOCK_INODE(sock));
416
417 file = alloc_file(&path, FMODE_READ | FMODE_WRITE,
418 &socket_file_ops);
419 if (IS_ERR(file)) {
420 /* drop dentry, keep inode */
421 ihold(d_inode(path.dentry));
422 path_put(&path);
423 return file;
424 }
425
426 sock->file = file;
427 file->f_flags = O_RDWR | (flags & O_NONBLOCK);
428 file->private_data = sock;
429 return file;
430}
431EXPORT_SYMBOL(sock_alloc_file);
432
433static int sock_map_fd(struct socket *sock, int flags)
434{
435 struct file *newfile;
436 int fd = get_unused_fd_flags(flags);
437 if (unlikely(fd < 0))
438 return fd;
439
440 newfile = sock_alloc_file(sock, flags, NULL);
441 if (likely(!IS_ERR(newfile))) {
442 fd_install(fd, newfile);
443 return fd;
444 }
445
446 put_unused_fd(fd);
447 return PTR_ERR(newfile);
448}
449
450struct socket *sock_from_file(struct file *file, int *err)
451{
452 if (file->f_op == &socket_file_ops)
453 return file->private_data; /* set in sock_map_fd */
454
455 *err = -ENOTSOCK;
456 return NULL;
457}
458EXPORT_SYMBOL(sock_from_file);
459
460/**
461 * sockfd_lookup - Go from a file number to its socket slot
462 * @fd: file handle
463 * @err: pointer to an error code return
464 *
465 * The file handle passed in is locked and the socket it is bound
466 * too is returned. If an error occurs the err pointer is overwritten
467 * with a negative errno code and NULL is returned. The function checks
468 * for both invalid handles and passing a handle which is not a socket.
469 *
470 * On a success the socket object pointer is returned.
471 */
472
473struct socket *sockfd_lookup(int fd, int *err)
474{
475 struct file *file;
476 struct socket *sock;
477
478 file = fget(fd);
479 if (!file) {
480 *err = -EBADF;
481 return NULL;
482 }
483
484 sock = sock_from_file(file, err);
485 if (!sock)
486 fput(file);
487 return sock;
488}
489EXPORT_SYMBOL(sockfd_lookup);
490
491static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
492{
493 struct fd f = fdget(fd);
494 struct socket *sock;
495
496 *err = -EBADF;
497 if (f.file) {
498 sock = sock_from_file(f.file, err);
499 if (likely(sock)) {
500 *fput_needed = f.flags;
501 return sock;
502 }
503 fdput(f);
504 }
505 return NULL;
506}
507
508static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
509 size_t size)
510{
511 ssize_t len;
512 ssize_t used = 0;
513
514 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
515 if (len < 0)
516 return len;
517 used += len;
518 if (buffer) {
519 if (size < used)
520 return -ERANGE;
521 buffer += len;
522 }
523
524 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
525 used += len;
526 if (buffer) {
527 if (size < used)
528 return -ERANGE;
529 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
530 buffer += len;
531 }
532
533 return used;
534}
535
536static int sockfs_setattr(struct dentry *dentry, struct iattr *iattr)
537{
538 int err = simple_setattr(dentry, iattr);
539
540 if (!err && (iattr->ia_valid & ATTR_UID)) {
541 struct socket *sock = SOCKET_I(d_inode(dentry));
542
543 sock->sk->sk_uid = iattr->ia_uid;
544 }
545
546 return err;
547}
548
549static const struct inode_operations sockfs_inode_ops = {
550 .listxattr = sockfs_listxattr,
551 .setattr = sockfs_setattr,
552};
553
554/**
555 * sock_alloc - allocate a socket
556 *
557 * Allocate a new inode and socket object. The two are bound together
558 * and initialised. The socket is then returned. If we are out of inodes
559 * NULL is returned.
560 */
561
562struct socket *sock_alloc(void)
563{
564 struct inode *inode;
565 struct socket *sock;
566
567 inode = new_inode_pseudo(sock_mnt->mnt_sb);
568 if (!inode)
569 return NULL;
570
571 sock = SOCKET_I(inode);
572
573 kmemcheck_annotate_bitfield(sock, type);
574 inode->i_ino = get_next_ino();
575 inode->i_mode = S_IFSOCK | S_IRWXUGO;
576 inode->i_uid = current_fsuid();
577 inode->i_gid = current_fsgid();
578 inode->i_op = &sockfs_inode_ops;
579
580 this_cpu_add(sockets_in_use, 1);
581 return sock;
582}
583EXPORT_SYMBOL(sock_alloc);
584
585/**
586 * sock_release - close a socket
587 * @sock: socket to close
588 *
589 * The socket is released from the protocol stack if it has a release
590 * callback, and the inode is then released if the socket is bound to
591 * an inode not a file.
592 */
593
594void sock_release(struct socket *sock)
595{
596 if (sock->ops) {
597 struct module *owner = sock->ops->owner;
598
599 sock->ops->release(sock);
600 sock->ops = NULL;
601 module_put(owner);
602 }
603
604 if (rcu_dereference_protected(sock->wq, 1)->fasync_list)
605 pr_err("%s: fasync list not empty!\n", __func__);
606
607 this_cpu_sub(sockets_in_use, 1);
608 if (!sock->file) {
609 iput(SOCK_INODE(sock));
610 return;
611 }
612 sock->file = NULL;
613}
614EXPORT_SYMBOL(sock_release);
615
616void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
617{
618 u8 flags = *tx_flags;
619
620 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
621 flags |= SKBTX_HW_TSTAMP;
622
623 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
624 flags |= SKBTX_SW_TSTAMP;
625
626 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
627 flags |= SKBTX_SCHED_TSTAMP;
628
629 *tx_flags = flags;
630}
631EXPORT_SYMBOL(__sock_tx_timestamp);
632
633static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
634{
635 int ret = sock->ops->sendmsg(sock, msg, msg_data_left(msg));
636 BUG_ON(ret == -EIOCBQUEUED);
637 return ret;
638}
639
640int sock_sendmsg(struct socket *sock, struct msghdr *msg)
641{
642 int err = security_socket_sendmsg(sock, msg,
643 msg_data_left(msg));
644
645 return err ?: sock_sendmsg_nosec(sock, msg);
646}
647EXPORT_SYMBOL(sock_sendmsg);
648
649int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
650 struct kvec *vec, size_t num, size_t size)
651{
652 iov_iter_kvec(&msg->msg_iter, WRITE | ITER_KVEC, vec, num, size);
653 return sock_sendmsg(sock, msg);
654}
655EXPORT_SYMBOL(kernel_sendmsg);
656
657/*
658 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
659 */
660void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
661 struct sk_buff *skb)
662{
663 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
664 struct scm_timestamping tss;
665 int empty = 1;
666 struct skb_shared_hwtstamps *shhwtstamps =
667 skb_hwtstamps(skb);
668
669 /* Race occurred between timestamp enabling and packet
670 receiving. Fill in the current time for now. */
671 if (need_software_tstamp && skb->tstamp == 0)
672 __net_timestamp(skb);
673
674 if (need_software_tstamp) {
675 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
676 struct timeval tv;
677 skb_get_timestamp(skb, &tv);
678 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
679 sizeof(tv), &tv);
680 } else {
681 struct timespec ts;
682 skb_get_timestampns(skb, &ts);
683 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
684 sizeof(ts), &ts);
685 }
686 }
687
688 memset(&tss, 0, sizeof(tss));
689 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
690 ktime_to_timespec_cond(skb->tstamp, tss.ts + 0))
691 empty = 0;
692 if (shhwtstamps &&
693 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
694 ktime_to_timespec_cond(shhwtstamps->hwtstamp, tss.ts + 2))
695 empty = 0;
696 if (!empty) {
697 put_cmsg(msg, SOL_SOCKET,
698 SCM_TIMESTAMPING, sizeof(tss), &tss);
699
700 if (skb->len && (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS))
701 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
702 skb->len, skb->data);
703 }
704}
705EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
706
707void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
708 struct sk_buff *skb)
709{
710 int ack;
711
712 if (!sock_flag(sk, SOCK_WIFI_STATUS))
713 return;
714 if (!skb->wifi_acked_valid)
715 return;
716
717 ack = skb->wifi_acked;
718
719 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
720}
721EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
722
723static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
724 struct sk_buff *skb)
725{
726 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
727 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
728 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
729}
730
731void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
732 struct sk_buff *skb)
733{
734 sock_recv_timestamp(msg, sk, skb);
735 sock_recv_drops(msg, sk, skb);
736}
737EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
738
739static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
740 int flags)
741{
742 return sock->ops->recvmsg(sock, msg, msg_data_left(msg), flags);
743}
744
745int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
746{
747 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
748
749 return err ?: sock_recvmsg_nosec(sock, msg, flags);
750}
751EXPORT_SYMBOL(sock_recvmsg);
752
753/**
754 * kernel_recvmsg - Receive a message from a socket (kernel space)
755 * @sock: The socket to receive the message from
756 * @msg: Received message
757 * @vec: Input s/g array for message data
758 * @num: Size of input s/g array
759 * @size: Number of bytes to read
760 * @flags: Message flags (MSG_DONTWAIT, etc...)
761 *
762 * On return the msg structure contains the scatter/gather array passed in the
763 * vec argument. The array is modified so that it consists of the unfilled
764 * portion of the original array.
765 *
766 * The returned value is the total number of bytes received, or an error.
767 */
768int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
769 struct kvec *vec, size_t num, size_t size, int flags)
770{
771 mm_segment_t oldfs = get_fs();
772 int result;
773
774 iov_iter_kvec(&msg->msg_iter, READ | ITER_KVEC, vec, num, size);
775 set_fs(KERNEL_DS);
776 result = sock_recvmsg(sock, msg, flags);
777 set_fs(oldfs);
778 return result;
779}
780EXPORT_SYMBOL(kernel_recvmsg);
781
782static ssize_t sock_sendpage(struct file *file, struct page *page,
783 int offset, size_t size, loff_t *ppos, int more)
784{
785 struct socket *sock;
786 int flags;
787
788 sock = file->private_data;
789
790 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
791 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
792 flags |= more;
793
794 return kernel_sendpage(sock, page, offset, size, flags);
795}
796
797static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
798 struct pipe_inode_info *pipe, size_t len,
799 unsigned int flags)
800{
801 struct socket *sock = file->private_data;
802
803 if (unlikely(!sock->ops->splice_read))
804 return -EINVAL;
805
806 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
807}
808
809static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
810{
811 struct file *file = iocb->ki_filp;
812 struct socket *sock = file->private_data;
813 struct msghdr msg = {.msg_iter = *to,
814 .msg_iocb = iocb};
815 ssize_t res;
816
817 if (file->f_flags & O_NONBLOCK)
818 msg.msg_flags = MSG_DONTWAIT;
819
820 if (iocb->ki_pos != 0)
821 return -ESPIPE;
822
823 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
824 return 0;
825
826 res = sock_recvmsg(sock, &msg, msg.msg_flags);
827 *to = msg.msg_iter;
828 return res;
829}
830
831static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
832{
833 struct file *file = iocb->ki_filp;
834 struct socket *sock = file->private_data;
835 struct msghdr msg = {.msg_iter = *from,
836 .msg_iocb = iocb};
837 ssize_t res;
838
839 if (iocb->ki_pos != 0)
840 return -ESPIPE;
841
842 if (file->f_flags & O_NONBLOCK)
843 msg.msg_flags = MSG_DONTWAIT;
844
845 if (sock->type == SOCK_SEQPACKET)
846 msg.msg_flags |= MSG_EOR;
847
848 res = sock_sendmsg(sock, &msg);
849 *from = msg.msg_iter;
850 return res;
851}
852
853/*
854 * Atomic setting of ioctl hooks to avoid race
855 * with module unload.
856 */
857
858static DEFINE_MUTEX(br_ioctl_mutex);
859static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
860
861void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
862{
863 mutex_lock(&br_ioctl_mutex);
864 br_ioctl_hook = hook;
865 mutex_unlock(&br_ioctl_mutex);
866}
867EXPORT_SYMBOL(brioctl_set);
868
869static DEFINE_MUTEX(vlan_ioctl_mutex);
870static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
871
872void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
873{
874 mutex_lock(&vlan_ioctl_mutex);
875 vlan_ioctl_hook = hook;
876 mutex_unlock(&vlan_ioctl_mutex);
877}
878EXPORT_SYMBOL(vlan_ioctl_set);
879
880static DEFINE_MUTEX(dlci_ioctl_mutex);
881static int (*dlci_ioctl_hook) (unsigned int, void __user *);
882
883void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
884{
885 mutex_lock(&dlci_ioctl_mutex);
886 dlci_ioctl_hook = hook;
887 mutex_unlock(&dlci_ioctl_mutex);
888}
889EXPORT_SYMBOL(dlci_ioctl_set);
890
891static long sock_do_ioctl(struct net *net, struct socket *sock,
892 unsigned int cmd, unsigned long arg)
893{
894 int err;
895 void __user *argp = (void __user *)arg;
896
897 err = sock->ops->ioctl(sock, cmd, arg);
898
899 /*
900 * If this ioctl is unknown try to hand it down
901 * to the NIC driver.
902 */
903 if (err == -ENOIOCTLCMD)
904 err = dev_ioctl(net, cmd, argp);
905
906 return err;
907}
908
909/*
910 * With an ioctl, arg may well be a user mode pointer, but we don't know
911 * what to do with it - that's up to the protocol still.
912 */
913
914static struct ns_common *get_net_ns(struct ns_common *ns)
915{
916 return &get_net(container_of(ns, struct net, ns))->ns;
917}
918
919static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
920{
921 struct socket *sock;
922 struct sock *sk;
923 void __user *argp = (void __user *)arg;
924 int pid, err;
925 struct net *net;
926
927 sock = file->private_data;
928 sk = sock->sk;
929 net = sock_net(sk);
930 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
931 err = dev_ioctl(net, cmd, argp);
932 } else
933#ifdef CONFIG_WEXT_CORE
934 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
935 err = dev_ioctl(net, cmd, argp);
936 } else
937#endif
938 switch (cmd) {
939 case FIOSETOWN:
940 case SIOCSPGRP:
941 err = -EFAULT;
942 if (get_user(pid, (int __user *)argp))
943 break;
944 f_setown(sock->file, pid, 1);
945 err = 0;
946 break;
947 case FIOGETOWN:
948 case SIOCGPGRP:
949 err = put_user(f_getown(sock->file),
950 (int __user *)argp);
951 break;
952 case SIOCGIFBR:
953 case SIOCSIFBR:
954 case SIOCBRADDBR:
955 case SIOCBRDELBR:
956 err = -ENOPKG;
957 if (!br_ioctl_hook)
958 request_module("bridge");
959
960 mutex_lock(&br_ioctl_mutex);
961 if (br_ioctl_hook)
962 err = br_ioctl_hook(net, cmd, argp);
963 mutex_unlock(&br_ioctl_mutex);
964 break;
965 case SIOCGIFVLAN:
966 case SIOCSIFVLAN:
967 err = -ENOPKG;
968 if (!vlan_ioctl_hook)
969 request_module("8021q");
970
971 mutex_lock(&vlan_ioctl_mutex);
972 if (vlan_ioctl_hook)
973 err = vlan_ioctl_hook(net, argp);
974 mutex_unlock(&vlan_ioctl_mutex);
975 break;
976 case SIOCADDDLCI:
977 case SIOCDELDLCI:
978 err = -ENOPKG;
979 if (!dlci_ioctl_hook)
980 request_module("dlci");
981
982 mutex_lock(&dlci_ioctl_mutex);
983 if (dlci_ioctl_hook)
984 err = dlci_ioctl_hook(cmd, argp);
985 mutex_unlock(&dlci_ioctl_mutex);
986 break;
987 case SIOCGSKNS:
988 err = -EPERM;
989 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
990 break;
991
992 err = open_related_ns(&net->ns, get_net_ns);
993 break;
994 default:
995 err = sock_do_ioctl(net, sock, cmd, arg);
996 break;
997 }
998 return err;
999}
1000
1001int sock_create_lite(int family, int type, int protocol, struct socket **res)
1002{
1003 int err;
1004 struct socket *sock = NULL;
1005
1006 err = security_socket_create(family, type, protocol, 1);
1007 if (err)
1008 goto out;
1009
1010 sock = sock_alloc();
1011 if (!sock) {
1012 err = -ENOMEM;
1013 goto out;
1014 }
1015
1016 sock->type = type;
1017 err = security_socket_post_create(sock, family, type, protocol, 1);
1018 if (err)
1019 goto out_release;
1020
1021out:
1022 *res = sock;
1023 return err;
1024out_release:
1025 sock_release(sock);
1026 sock = NULL;
1027 goto out;
1028}
1029EXPORT_SYMBOL(sock_create_lite);
1030
1031/* No kernel lock held - perfect */
1032static unsigned int sock_poll(struct file *file, poll_table *wait)
1033{
1034 unsigned int busy_flag = 0;
1035 struct socket *sock;
1036
1037 /*
1038 * We can't return errors to poll, so it's either yes or no.
1039 */
1040 sock = file->private_data;
1041
1042 if (sk_can_busy_loop(sock->sk)) {
1043 /* this socket can poll_ll so tell the system call */
1044 busy_flag = POLL_BUSY_LOOP;
1045
1046 /* once, only if requested by syscall */
1047 if (wait && (wait->_key & POLL_BUSY_LOOP))
1048 sk_busy_loop(sock->sk, 1);
1049 }
1050
1051 return busy_flag | sock->ops->poll(file, sock, wait);
1052}
1053
1054static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1055{
1056 struct socket *sock = file->private_data;
1057
1058 return sock->ops->mmap(file, sock, vma);
1059}
1060
1061static int sock_close(struct inode *inode, struct file *filp)
1062{
1063 sock_release(SOCKET_I(inode));
1064 return 0;
1065}
1066
1067/*
1068 * Update the socket async list
1069 *
1070 * Fasync_list locking strategy.
1071 *
1072 * 1. fasync_list is modified only under process context socket lock
1073 * i.e. under semaphore.
1074 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1075 * or under socket lock
1076 */
1077
1078static int sock_fasync(int fd, struct file *filp, int on)
1079{
1080 struct socket *sock = filp->private_data;
1081 struct sock *sk = sock->sk;
1082 struct socket_wq *wq;
1083
1084 if (sk == NULL)
1085 return -EINVAL;
1086
1087 lock_sock(sk);
1088 wq = rcu_dereference_protected(sock->wq, lockdep_sock_is_held(sk));
1089 fasync_helper(fd, filp, on, &wq->fasync_list);
1090
1091 if (!wq->fasync_list)
1092 sock_reset_flag(sk, SOCK_FASYNC);
1093 else
1094 sock_set_flag(sk, SOCK_FASYNC);
1095
1096 release_sock(sk);
1097 return 0;
1098}
1099
1100/* This function may be called only under rcu_lock */
1101
1102int sock_wake_async(struct socket_wq *wq, int how, int band)
1103{
1104 if (!wq || !wq->fasync_list)
1105 return -1;
1106
1107 switch (how) {
1108 case SOCK_WAKE_WAITD:
1109 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1110 break;
1111 goto call_kill;
1112 case SOCK_WAKE_SPACE:
1113 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1114 break;
1115 /* fall through */
1116 case SOCK_WAKE_IO:
1117call_kill:
1118 kill_fasync(&wq->fasync_list, SIGIO, band);
1119 break;
1120 case SOCK_WAKE_URG:
1121 kill_fasync(&wq->fasync_list, SIGURG, band);
1122 }
1123
1124 return 0;
1125}
1126EXPORT_SYMBOL(sock_wake_async);
1127
1128int __sock_create(struct net *net, int family, int type, int protocol,
1129 struct socket **res, int kern)
1130{
1131 int err;
1132 struct socket *sock;
1133 const struct net_proto_family *pf;
1134
1135 /*
1136 * Check protocol is in range
1137 */
1138 if (family < 0 || family >= NPROTO)
1139 return -EAFNOSUPPORT;
1140 if (type < 0 || type >= SOCK_MAX)
1141 return -EINVAL;
1142
1143 /* Compatibility.
1144
1145 This uglymoron is moved from INET layer to here to avoid
1146 deadlock in module load.
1147 */
1148 if (family == PF_INET && type == SOCK_PACKET) {
1149 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1150 current->comm);
1151 family = PF_PACKET;
1152 }
1153
1154 err = security_socket_create(family, type, protocol, kern);
1155 if (err)
1156 return err;
1157
1158 /*
1159 * Allocate the socket and allow the family to set things up. if
1160 * the protocol is 0, the family is instructed to select an appropriate
1161 * default.
1162 */
1163 sock = sock_alloc();
1164 if (!sock) {
1165 net_warn_ratelimited("socket: no more sockets\n");
1166 return -ENFILE; /* Not exactly a match, but its the
1167 closest posix thing */
1168 }
1169
1170 sock->type = type;
1171
1172#ifdef CONFIG_MODULES
1173 /* Attempt to load a protocol module if the find failed.
1174 *
1175 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1176 * requested real, full-featured networking support upon configuration.
1177 * Otherwise module support will break!
1178 */
1179 if (rcu_access_pointer(net_families[family]) == NULL)
1180 request_module("net-pf-%d", family);
1181#endif
1182
1183 rcu_read_lock();
1184 pf = rcu_dereference(net_families[family]);
1185 err = -EAFNOSUPPORT;
1186 if (!pf)
1187 goto out_release;
1188
1189 /*
1190 * We will call the ->create function, that possibly is in a loadable
1191 * module, so we have to bump that loadable module refcnt first.
1192 */
1193 if (!try_module_get(pf->owner))
1194 goto out_release;
1195
1196 /* Now protected by module ref count */
1197 rcu_read_unlock();
1198
1199 err = pf->create(net, sock, protocol, kern);
1200 if (err < 0)
1201 goto out_module_put;
1202
1203 /*
1204 * Now to bump the refcnt of the [loadable] module that owns this
1205 * socket at sock_release time we decrement its refcnt.
1206 */
1207 if (!try_module_get(sock->ops->owner))
1208 goto out_module_busy;
1209
1210 /*
1211 * Now that we're done with the ->create function, the [loadable]
1212 * module can have its refcnt decremented
1213 */
1214 module_put(pf->owner);
1215 err = security_socket_post_create(sock, family, type, protocol, kern);
1216 if (err)
1217 goto out_sock_release;
1218 *res = sock;
1219
1220 return 0;
1221
1222out_module_busy:
1223 err = -EAFNOSUPPORT;
1224out_module_put:
1225 sock->ops = NULL;
1226 module_put(pf->owner);
1227out_sock_release:
1228 sock_release(sock);
1229 return err;
1230
1231out_release:
1232 rcu_read_unlock();
1233 goto out_sock_release;
1234}
1235EXPORT_SYMBOL(__sock_create);
1236
1237int sock_create(int family, int type, int protocol, struct socket **res)
1238{
1239 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1240}
1241EXPORT_SYMBOL(sock_create);
1242
1243int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1244{
1245 return __sock_create(net, family, type, protocol, res, 1);
1246}
1247EXPORT_SYMBOL(sock_create_kern);
1248
1249SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1250{
1251 int retval;
1252 struct socket *sock;
1253 int flags;
1254
1255 /* Check the SOCK_* constants for consistency. */
1256 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1257 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1258 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1259 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1260
1261 flags = type & ~SOCK_TYPE_MASK;
1262 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1263 return -EINVAL;
1264 type &= SOCK_TYPE_MASK;
1265
1266 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1267 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1268
1269 retval = sock_create(family, type, protocol, &sock);
1270 if (retval < 0)
1271 goto out;
1272
1273 retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1274 if (retval < 0)
1275 goto out_release;
1276
1277out:
1278 /* It may be already another descriptor 8) Not kernel problem. */
1279 return retval;
1280
1281out_release:
1282 sock_release(sock);
1283 return retval;
1284}
1285
1286/*
1287 * Create a pair of connected sockets.
1288 */
1289
1290SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1291 int __user *, usockvec)
1292{
1293 struct socket *sock1, *sock2;
1294 int fd1, fd2, err;
1295 struct file *newfile1, *newfile2;
1296 int flags;
1297
1298 flags = type & ~SOCK_TYPE_MASK;
1299 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1300 return -EINVAL;
1301 type &= SOCK_TYPE_MASK;
1302
1303 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1304 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1305
1306 /*
1307 * Obtain the first socket and check if the underlying protocol
1308 * supports the socketpair call.
1309 */
1310
1311 err = sock_create(family, type, protocol, &sock1);
1312 if (err < 0)
1313 goto out;
1314
1315 err = sock_create(family, type, protocol, &sock2);
1316 if (err < 0)
1317 goto out_release_1;
1318
1319 err = sock1->ops->socketpair(sock1, sock2);
1320 if (err < 0)
1321 goto out_release_both;
1322
1323 fd1 = get_unused_fd_flags(flags);
1324 if (unlikely(fd1 < 0)) {
1325 err = fd1;
1326 goto out_release_both;
1327 }
1328
1329 fd2 = get_unused_fd_flags(flags);
1330 if (unlikely(fd2 < 0)) {
1331 err = fd2;
1332 goto out_put_unused_1;
1333 }
1334
1335 newfile1 = sock_alloc_file(sock1, flags, NULL);
1336 if (IS_ERR(newfile1)) {
1337 err = PTR_ERR(newfile1);
1338 goto out_put_unused_both;
1339 }
1340
1341 newfile2 = sock_alloc_file(sock2, flags, NULL);
1342 if (IS_ERR(newfile2)) {
1343 err = PTR_ERR(newfile2);
1344 goto out_fput_1;
1345 }
1346
1347 err = put_user(fd1, &usockvec[0]);
1348 if (err)
1349 goto out_fput_both;
1350
1351 err = put_user(fd2, &usockvec[1]);
1352 if (err)
1353 goto out_fput_both;
1354
1355 audit_fd_pair(fd1, fd2);
1356
1357 fd_install(fd1, newfile1);
1358 fd_install(fd2, newfile2);
1359 /* fd1 and fd2 may be already another descriptors.
1360 * Not kernel problem.
1361 */
1362
1363 return 0;
1364
1365out_fput_both:
1366 fput(newfile2);
1367 fput(newfile1);
1368 put_unused_fd(fd2);
1369 put_unused_fd(fd1);
1370 goto out;
1371
1372out_fput_1:
1373 fput(newfile1);
1374 put_unused_fd(fd2);
1375 put_unused_fd(fd1);
1376 sock_release(sock2);
1377 goto out;
1378
1379out_put_unused_both:
1380 put_unused_fd(fd2);
1381out_put_unused_1:
1382 put_unused_fd(fd1);
1383out_release_both:
1384 sock_release(sock2);
1385out_release_1:
1386 sock_release(sock1);
1387out:
1388 return err;
1389}
1390
1391/*
1392 * Bind a name to a socket. Nothing much to do here since it's
1393 * the protocol's responsibility to handle the local address.
1394 *
1395 * We move the socket address to kernel space before we call
1396 * the protocol layer (having also checked the address is ok).
1397 */
1398
1399SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1400{
1401 struct socket *sock;
1402 struct sockaddr_storage address;
1403 int err, fput_needed;
1404
1405 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1406 if (sock) {
1407 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1408 if (err >= 0) {
1409 err = security_socket_bind(sock,
1410 (struct sockaddr *)&address,
1411 addrlen);
1412 if (!err)
1413 err = sock->ops->bind(sock,
1414 (struct sockaddr *)
1415 &address, addrlen);
1416 }
1417 fput_light(sock->file, fput_needed);
1418 }
1419 return err;
1420}
1421
1422/*
1423 * Perform a listen. Basically, we allow the protocol to do anything
1424 * necessary for a listen, and if that works, we mark the socket as
1425 * ready for listening.
1426 */
1427
1428SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1429{
1430 struct socket *sock;
1431 int err, fput_needed;
1432 int somaxconn;
1433
1434 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1435 if (sock) {
1436 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1437 if ((unsigned int)backlog > somaxconn)
1438 backlog = somaxconn;
1439
1440 err = security_socket_listen(sock, backlog);
1441 if (!err)
1442 err = sock->ops->listen(sock, backlog);
1443
1444 fput_light(sock->file, fput_needed);
1445 }
1446 return err;
1447}
1448
1449/*
1450 * For accept, we attempt to create a new socket, set up the link
1451 * with the client, wake up the client, then return the new
1452 * connected fd. We collect the address of the connector in kernel
1453 * space and move it to user at the very end. This is unclean because
1454 * we open the socket then return an error.
1455 *
1456 * 1003.1g adds the ability to recvmsg() to query connection pending
1457 * status to recvmsg. We need to add that support in a way thats
1458 * clean when we restucture accept also.
1459 */
1460
1461SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1462 int __user *, upeer_addrlen, int, flags)
1463{
1464 struct socket *sock, *newsock;
1465 struct file *newfile;
1466 int err, len, newfd, fput_needed;
1467 struct sockaddr_storage address;
1468
1469 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1470 return -EINVAL;
1471
1472 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1473 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1474
1475 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1476 if (!sock)
1477 goto out;
1478
1479 err = -ENFILE;
1480 newsock = sock_alloc();
1481 if (!newsock)
1482 goto out_put;
1483
1484 newsock->type = sock->type;
1485 newsock->ops = sock->ops;
1486
1487 /*
1488 * We don't need try_module_get here, as the listening socket (sock)
1489 * has the protocol module (sock->ops->owner) held.
1490 */
1491 __module_get(newsock->ops->owner);
1492
1493 newfd = get_unused_fd_flags(flags);
1494 if (unlikely(newfd < 0)) {
1495 err = newfd;
1496 sock_release(newsock);
1497 goto out_put;
1498 }
1499 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1500 if (IS_ERR(newfile)) {
1501 err = PTR_ERR(newfile);
1502 put_unused_fd(newfd);
1503 sock_release(newsock);
1504 goto out_put;
1505 }
1506
1507 err = security_socket_accept(sock, newsock);
1508 if (err)
1509 goto out_fd;
1510
1511 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1512 if (err < 0)
1513 goto out_fd;
1514
1515 if (upeer_sockaddr) {
1516 if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1517 &len, 2) < 0) {
1518 err = -ECONNABORTED;
1519 goto out_fd;
1520 }
1521 err = move_addr_to_user(&address,
1522 len, upeer_sockaddr, upeer_addrlen);
1523 if (err < 0)
1524 goto out_fd;
1525 }
1526
1527 /* File flags are not inherited via accept() unlike another OSes. */
1528
1529 fd_install(newfd, newfile);
1530 err = newfd;
1531
1532out_put:
1533 fput_light(sock->file, fput_needed);
1534out:
1535 return err;
1536out_fd:
1537 fput(newfile);
1538 put_unused_fd(newfd);
1539 goto out_put;
1540}
1541
1542SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1543 int __user *, upeer_addrlen)
1544{
1545 return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1546}
1547
1548/*
1549 * Attempt to connect to a socket with the server address. The address
1550 * is in user space so we verify it is OK and move it to kernel space.
1551 *
1552 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1553 * break bindings
1554 *
1555 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1556 * other SEQPACKET protocols that take time to connect() as it doesn't
1557 * include the -EINPROGRESS status for such sockets.
1558 */
1559
1560SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1561 int, addrlen)
1562{
1563 struct socket *sock;
1564 struct sockaddr_storage address;
1565 int err, fput_needed;
1566
1567 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1568 if (!sock)
1569 goto out;
1570 err = move_addr_to_kernel(uservaddr, addrlen, &address);
1571 if (err < 0)
1572 goto out_put;
1573
1574 err =
1575 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1576 if (err)
1577 goto out_put;
1578
1579 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1580 sock->file->f_flags);
1581out_put:
1582 fput_light(sock->file, fput_needed);
1583out:
1584 return err;
1585}
1586
1587/*
1588 * Get the local address ('name') of a socket object. Move the obtained
1589 * name to user space.
1590 */
1591
1592SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1593 int __user *, usockaddr_len)
1594{
1595 struct socket *sock;
1596 struct sockaddr_storage address;
1597 int len, err, fput_needed;
1598
1599 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1600 if (!sock)
1601 goto out;
1602
1603 err = security_socket_getsockname(sock);
1604 if (err)
1605 goto out_put;
1606
1607 err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1608 if (err)
1609 goto out_put;
1610 err = move_addr_to_user(&address, len, usockaddr, usockaddr_len);
1611
1612out_put:
1613 fput_light(sock->file, fput_needed);
1614out:
1615 return err;
1616}
1617
1618/*
1619 * Get the remote address ('name') of a socket object. Move the obtained
1620 * name to user space.
1621 */
1622
1623SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1624 int __user *, usockaddr_len)
1625{
1626 struct socket *sock;
1627 struct sockaddr_storage address;
1628 int len, err, fput_needed;
1629
1630 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1631 if (sock != NULL) {
1632 err = security_socket_getpeername(sock);
1633 if (err) {
1634 fput_light(sock->file, fput_needed);
1635 return err;
1636 }
1637
1638 err =
1639 sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1640 1);
1641 if (!err)
1642 err = move_addr_to_user(&address, len, usockaddr,
1643 usockaddr_len);
1644 fput_light(sock->file, fput_needed);
1645 }
1646 return err;
1647}
1648
1649/*
1650 * Send a datagram to a given address. We move the address into kernel
1651 * space and check the user space data area is readable before invoking
1652 * the protocol.
1653 */
1654
1655SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1656 unsigned int, flags, struct sockaddr __user *, addr,
1657 int, addr_len)
1658{
1659 struct socket *sock;
1660 struct sockaddr_storage address;
1661 int err;
1662 struct msghdr msg;
1663 struct iovec iov;
1664 int fput_needed;
1665
1666 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
1667 if (unlikely(err))
1668 return err;
1669 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1670 if (!sock)
1671 goto out;
1672
1673 msg.msg_name = NULL;
1674 msg.msg_control = NULL;
1675 msg.msg_controllen = 0;
1676 msg.msg_namelen = 0;
1677 if (addr) {
1678 err = move_addr_to_kernel(addr, addr_len, &address);
1679 if (err < 0)
1680 goto out_put;
1681 msg.msg_name = (struct sockaddr *)&address;
1682 msg.msg_namelen = addr_len;
1683 }
1684 if (sock->file->f_flags & O_NONBLOCK)
1685 flags |= MSG_DONTWAIT;
1686 msg.msg_flags = flags;
1687 err = sock_sendmsg(sock, &msg);
1688
1689out_put:
1690 fput_light(sock->file, fput_needed);
1691out:
1692 return err;
1693}
1694
1695/*
1696 * Send a datagram down a socket.
1697 */
1698
1699SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1700 unsigned int, flags)
1701{
1702 return sys_sendto(fd, buff, len, flags, NULL, 0);
1703}
1704
1705/*
1706 * Receive a frame from the socket and optionally record the address of the
1707 * sender. We verify the buffers are writable and if needed move the
1708 * sender address from kernel to user space.
1709 */
1710
1711SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1712 unsigned int, flags, struct sockaddr __user *, addr,
1713 int __user *, addr_len)
1714{
1715 struct socket *sock;
1716 struct iovec iov;
1717 struct msghdr msg;
1718 struct sockaddr_storage address;
1719 int err, err2;
1720 int fput_needed;
1721
1722 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
1723 if (unlikely(err))
1724 return err;
1725 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1726 if (!sock)
1727 goto out;
1728
1729 msg.msg_control = NULL;
1730 msg.msg_controllen = 0;
1731 /* Save some cycles and don't copy the address if not needed */
1732 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
1733 /* We assume all kernel code knows the size of sockaddr_storage */
1734 msg.msg_namelen = 0;
1735 msg.msg_iocb = NULL;
1736 if (sock->file->f_flags & O_NONBLOCK)
1737 flags |= MSG_DONTWAIT;
1738 err = sock_recvmsg(sock, &msg, flags);
1739
1740 if (err >= 0 && addr != NULL) {
1741 err2 = move_addr_to_user(&address,
1742 msg.msg_namelen, addr, addr_len);
1743 if (err2 < 0)
1744 err = err2;
1745 }
1746
1747 fput_light(sock->file, fput_needed);
1748out:
1749 return err;
1750}
1751
1752/*
1753 * Receive a datagram from a socket.
1754 */
1755
1756SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
1757 unsigned int, flags)
1758{
1759 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1760}
1761
1762/*
1763 * Set a socket option. Because we don't know the option lengths we have
1764 * to pass the user mode parameter for the protocols to sort out.
1765 */
1766
1767SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1768 char __user *, optval, int, optlen)
1769{
1770 int err, fput_needed;
1771 struct socket *sock;
1772
1773 if (optlen < 0)
1774 return -EINVAL;
1775
1776 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1777 if (sock != NULL) {
1778 err = security_socket_setsockopt(sock, level, optname);
1779 if (err)
1780 goto out_put;
1781
1782 if (level == SOL_SOCKET)
1783 err =
1784 sock_setsockopt(sock, level, optname, optval,
1785 optlen);
1786 else
1787 err =
1788 sock->ops->setsockopt(sock, level, optname, optval,
1789 optlen);
1790out_put:
1791 fput_light(sock->file, fput_needed);
1792 }
1793 return err;
1794}
1795
1796/*
1797 * Get a socket option. Because we don't know the option lengths we have
1798 * to pass a user mode parameter for the protocols to sort out.
1799 */
1800
1801SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1802 char __user *, optval, int __user *, optlen)
1803{
1804 int err, fput_needed;
1805 struct socket *sock;
1806
1807 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1808 if (sock != NULL) {
1809 err = security_socket_getsockopt(sock, level, optname);
1810 if (err)
1811 goto out_put;
1812
1813 if (level == SOL_SOCKET)
1814 err =
1815 sock_getsockopt(sock, level, optname, optval,
1816 optlen);
1817 else
1818 err =
1819 sock->ops->getsockopt(sock, level, optname, optval,
1820 optlen);
1821out_put:
1822 fput_light(sock->file, fput_needed);
1823 }
1824 return err;
1825}
1826
1827/*
1828 * Shutdown a socket.
1829 */
1830
1831SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1832{
1833 int err, fput_needed;
1834 struct socket *sock;
1835
1836 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1837 if (sock != NULL) {
1838 err = security_socket_shutdown(sock, how);
1839 if (!err)
1840 err = sock->ops->shutdown(sock, how);
1841 fput_light(sock->file, fput_needed);
1842 }
1843 return err;
1844}
1845
1846/* A couple of helpful macros for getting the address of the 32/64 bit
1847 * fields which are the same type (int / unsigned) on our platforms.
1848 */
1849#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1850#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1851#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1852
1853struct used_address {
1854 struct sockaddr_storage name;
1855 unsigned int name_len;
1856};
1857
1858static int copy_msghdr_from_user(struct msghdr *kmsg,
1859 struct user_msghdr __user *umsg,
1860 struct sockaddr __user **save_addr,
1861 struct iovec **iov)
1862{
1863 struct sockaddr __user *uaddr;
1864 struct iovec __user *uiov;
1865 size_t nr_segs;
1866 ssize_t err;
1867
1868 if (!access_ok(VERIFY_READ, umsg, sizeof(*umsg)) ||
1869 __get_user(uaddr, &umsg->msg_name) ||
1870 __get_user(kmsg->msg_namelen, &umsg->msg_namelen) ||
1871 __get_user(uiov, &umsg->msg_iov) ||
1872 __get_user(nr_segs, &umsg->msg_iovlen) ||
1873 __get_user(kmsg->msg_control, &umsg->msg_control) ||
1874 __get_user(kmsg->msg_controllen, &umsg->msg_controllen) ||
1875 __get_user(kmsg->msg_flags, &umsg->msg_flags))
1876 return -EFAULT;
1877
1878 if (!uaddr)
1879 kmsg->msg_namelen = 0;
1880
1881 if (kmsg->msg_namelen < 0)
1882 return -EINVAL;
1883
1884 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
1885 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
1886
1887 if (save_addr)
1888 *save_addr = uaddr;
1889
1890 if (uaddr && kmsg->msg_namelen) {
1891 if (!save_addr) {
1892 err = move_addr_to_kernel(uaddr, kmsg->msg_namelen,
1893 kmsg->msg_name);
1894 if (err < 0)
1895 return err;
1896 }
1897 } else {
1898 kmsg->msg_name = NULL;
1899 kmsg->msg_namelen = 0;
1900 }
1901
1902 if (nr_segs > UIO_MAXIOV)
1903 return -EMSGSIZE;
1904
1905 kmsg->msg_iocb = NULL;
1906
1907 return import_iovec(save_addr ? READ : WRITE, uiov, nr_segs,
1908 UIO_FASTIOV, iov, &kmsg->msg_iter);
1909}
1910
1911static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
1912 struct msghdr *msg_sys, unsigned int flags,
1913 struct used_address *used_address,
1914 unsigned int allowed_msghdr_flags)
1915{
1916 struct compat_msghdr __user *msg_compat =
1917 (struct compat_msghdr __user *)msg;
1918 struct sockaddr_storage address;
1919 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1920 unsigned char ctl[sizeof(struct cmsghdr) + 20]
1921 __aligned(sizeof(__kernel_size_t));
1922 /* 20 is size of ipv6_pktinfo */
1923 unsigned char *ctl_buf = ctl;
1924 int ctl_len;
1925 ssize_t err;
1926
1927 msg_sys->msg_name = &address;
1928
1929 if (MSG_CMSG_COMPAT & flags)
1930 err = get_compat_msghdr(msg_sys, msg_compat, NULL, &iov);
1931 else
1932 err = copy_msghdr_from_user(msg_sys, msg, NULL, &iov);
1933 if (err < 0)
1934 return err;
1935
1936 err = -ENOBUFS;
1937
1938 if (msg_sys->msg_controllen > INT_MAX)
1939 goto out_freeiov;
1940 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
1941 ctl_len = msg_sys->msg_controllen;
1942 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1943 err =
1944 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
1945 sizeof(ctl));
1946 if (err)
1947 goto out_freeiov;
1948 ctl_buf = msg_sys->msg_control;
1949 ctl_len = msg_sys->msg_controllen;
1950 } else if (ctl_len) {
1951 if (ctl_len > sizeof(ctl)) {
1952 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1953 if (ctl_buf == NULL)
1954 goto out_freeiov;
1955 }
1956 err = -EFAULT;
1957 /*
1958 * Careful! Before this, msg_sys->msg_control contains a user pointer.
1959 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1960 * checking falls down on this.
1961 */
1962 if (copy_from_user(ctl_buf,
1963 (void __user __force *)msg_sys->msg_control,
1964 ctl_len))
1965 goto out_freectl;
1966 msg_sys->msg_control = ctl_buf;
1967 }
1968 msg_sys->msg_flags = flags;
1969
1970 if (sock->file->f_flags & O_NONBLOCK)
1971 msg_sys->msg_flags |= MSG_DONTWAIT;
1972 /*
1973 * If this is sendmmsg() and current destination address is same as
1974 * previously succeeded address, omit asking LSM's decision.
1975 * used_address->name_len is initialized to UINT_MAX so that the first
1976 * destination address never matches.
1977 */
1978 if (used_address && msg_sys->msg_name &&
1979 used_address->name_len == msg_sys->msg_namelen &&
1980 !memcmp(&used_address->name, msg_sys->msg_name,
1981 used_address->name_len)) {
1982 err = sock_sendmsg_nosec(sock, msg_sys);
1983 goto out_freectl;
1984 }
1985 err = sock_sendmsg(sock, msg_sys);
1986 /*
1987 * If this is sendmmsg() and sending to current destination address was
1988 * successful, remember it.
1989 */
1990 if (used_address && err >= 0) {
1991 used_address->name_len = msg_sys->msg_namelen;
1992 if (msg_sys->msg_name)
1993 memcpy(&used_address->name, msg_sys->msg_name,
1994 used_address->name_len);
1995 }
1996
1997out_freectl:
1998 if (ctl_buf != ctl)
1999 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2000out_freeiov:
2001 kfree(iov);
2002 return err;
2003}
2004
2005/*
2006 * BSD sendmsg interface
2007 */
2008
2009long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned flags)
2010{
2011 int fput_needed, err;
2012 struct msghdr msg_sys;
2013 struct socket *sock;
2014
2015 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2016 if (!sock)
2017 goto out;
2018
2019 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2020
2021 fput_light(sock->file, fput_needed);
2022out:
2023 return err;
2024}
2025
2026SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2027{
2028 if (flags & MSG_CMSG_COMPAT)
2029 return -EINVAL;
2030 return __sys_sendmsg(fd, msg, flags);
2031}
2032
2033/*
2034 * Linux sendmmsg interface
2035 */
2036
2037int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2038 unsigned int flags)
2039{
2040 int fput_needed, err, datagrams;
2041 struct socket *sock;
2042 struct mmsghdr __user *entry;
2043 struct compat_mmsghdr __user *compat_entry;
2044 struct msghdr msg_sys;
2045 struct used_address used_address;
2046 unsigned int oflags = flags;
2047
2048 if (vlen > UIO_MAXIOV)
2049 vlen = UIO_MAXIOV;
2050
2051 datagrams = 0;
2052
2053 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2054 if (!sock)
2055 return err;
2056
2057 used_address.name_len = UINT_MAX;
2058 entry = mmsg;
2059 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2060 err = 0;
2061 flags |= MSG_BATCH;
2062
2063 while (datagrams < vlen) {
2064 if (datagrams == vlen - 1)
2065 flags = oflags;
2066
2067 if (MSG_CMSG_COMPAT & flags) {
2068 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2069 &msg_sys, flags, &used_address, MSG_EOR);
2070 if (err < 0)
2071 break;
2072 err = __put_user(err, &compat_entry->msg_len);
2073 ++compat_entry;
2074 } else {
2075 err = ___sys_sendmsg(sock,
2076 (struct user_msghdr __user *)entry,
2077 &msg_sys, flags, &used_address, MSG_EOR);
2078 if (err < 0)
2079 break;
2080 err = put_user(err, &entry->msg_len);
2081 ++entry;
2082 }
2083
2084 if (err)
2085 break;
2086 ++datagrams;
2087 if (msg_data_left(&msg_sys))
2088 break;
2089 cond_resched();
2090 }
2091
2092 fput_light(sock->file, fput_needed);
2093
2094 /* We only return an error if no datagrams were able to be sent */
2095 if (datagrams != 0)
2096 return datagrams;
2097
2098 return err;
2099}
2100
2101SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2102 unsigned int, vlen, unsigned int, flags)
2103{
2104 if (flags & MSG_CMSG_COMPAT)
2105 return -EINVAL;
2106 return __sys_sendmmsg(fd, mmsg, vlen, flags);
2107}
2108
2109static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2110 struct msghdr *msg_sys, unsigned int flags, int nosec)
2111{
2112 struct compat_msghdr __user *msg_compat =
2113 (struct compat_msghdr __user *)msg;
2114 struct iovec iovstack[UIO_FASTIOV];
2115 struct iovec *iov = iovstack;
2116 unsigned long cmsg_ptr;
2117 int len;
2118 ssize_t err;
2119
2120 /* kernel mode address */
2121 struct sockaddr_storage addr;
2122
2123 /* user mode address pointers */
2124 struct sockaddr __user *uaddr;
2125 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2126
2127 msg_sys->msg_name = &addr;
2128
2129 if (MSG_CMSG_COMPAT & flags)
2130 err = get_compat_msghdr(msg_sys, msg_compat, &uaddr, &iov);
2131 else
2132 err = copy_msghdr_from_user(msg_sys, msg, &uaddr, &iov);
2133 if (err < 0)
2134 return err;
2135
2136 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2137 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2138
2139 /* We assume all kernel code knows the size of sockaddr_storage */
2140 msg_sys->msg_namelen = 0;
2141
2142 if (sock->file->f_flags & O_NONBLOCK)
2143 flags |= MSG_DONTWAIT;
2144 err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys, flags);
2145 if (err < 0)
2146 goto out_freeiov;
2147 len = err;
2148
2149 if (uaddr != NULL) {
2150 err = move_addr_to_user(&addr,
2151 msg_sys->msg_namelen, uaddr,
2152 uaddr_len);
2153 if (err < 0)
2154 goto out_freeiov;
2155 }
2156 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2157 COMPAT_FLAGS(msg));
2158 if (err)
2159 goto out_freeiov;
2160 if (MSG_CMSG_COMPAT & flags)
2161 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2162 &msg_compat->msg_controllen);
2163 else
2164 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2165 &msg->msg_controllen);
2166 if (err)
2167 goto out_freeiov;
2168 err = len;
2169
2170out_freeiov:
2171 kfree(iov);
2172 return err;
2173}
2174
2175/*
2176 * BSD recvmsg interface
2177 */
2178
2179long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned flags)
2180{
2181 int fput_needed, err;
2182 struct msghdr msg_sys;
2183 struct socket *sock;
2184
2185 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2186 if (!sock)
2187 goto out;
2188
2189 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2190
2191 fput_light(sock->file, fput_needed);
2192out:
2193 return err;
2194}
2195
2196SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2197 unsigned int, flags)
2198{
2199 if (flags & MSG_CMSG_COMPAT)
2200 return -EINVAL;
2201 return __sys_recvmsg(fd, msg, flags);
2202}
2203
2204/*
2205 * Linux recvmmsg interface
2206 */
2207
2208int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2209 unsigned int flags, struct timespec *timeout)
2210{
2211 int fput_needed, err, datagrams;
2212 struct socket *sock;
2213 struct mmsghdr __user *entry;
2214 struct compat_mmsghdr __user *compat_entry;
2215 struct msghdr msg_sys;
2216 struct timespec64 end_time;
2217 struct timespec64 timeout64;
2218
2219 if (timeout &&
2220 poll_select_set_timeout(&end_time, timeout->tv_sec,
2221 timeout->tv_nsec))
2222 return -EINVAL;
2223
2224 datagrams = 0;
2225
2226 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2227 if (!sock)
2228 return err;
2229
2230 err = sock_error(sock->sk);
2231 if (err) {
2232 datagrams = err;
2233 goto out_put;
2234 }
2235
2236 entry = mmsg;
2237 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2238
2239 while (datagrams < vlen) {
2240 /*
2241 * No need to ask LSM for more than the first datagram.
2242 */
2243 if (MSG_CMSG_COMPAT & flags) {
2244 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2245 &msg_sys, flags & ~MSG_WAITFORONE,
2246 datagrams);
2247 if (err < 0)
2248 break;
2249 err = __put_user(err, &compat_entry->msg_len);
2250 ++compat_entry;
2251 } else {
2252 err = ___sys_recvmsg(sock,
2253 (struct user_msghdr __user *)entry,
2254 &msg_sys, flags & ~MSG_WAITFORONE,
2255 datagrams);
2256 if (err < 0)
2257 break;
2258 err = put_user(err, &entry->msg_len);
2259 ++entry;
2260 }
2261
2262 if (err)
2263 break;
2264 ++datagrams;
2265
2266 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2267 if (flags & MSG_WAITFORONE)
2268 flags |= MSG_DONTWAIT;
2269
2270 if (timeout) {
2271 ktime_get_ts64(&timeout64);
2272 *timeout = timespec64_to_timespec(
2273 timespec64_sub(end_time, timeout64));
2274 if (timeout->tv_sec < 0) {
2275 timeout->tv_sec = timeout->tv_nsec = 0;
2276 break;
2277 }
2278
2279 /* Timeout, return less than vlen datagrams */
2280 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2281 break;
2282 }
2283
2284 /* Out of band data, return right away */
2285 if (msg_sys.msg_flags & MSG_OOB)
2286 break;
2287 cond_resched();
2288 }
2289
2290 if (err == 0)
2291 goto out_put;
2292
2293 if (datagrams == 0) {
2294 datagrams = err;
2295 goto out_put;
2296 }
2297
2298 /*
2299 * We may return less entries than requested (vlen) if the
2300 * sock is non block and there aren't enough datagrams...
2301 */
2302 if (err != -EAGAIN) {
2303 /*
2304 * ... or if recvmsg returns an error after we
2305 * received some datagrams, where we record the
2306 * error to return on the next call or if the
2307 * app asks about it using getsockopt(SO_ERROR).
2308 */
2309 sock->sk->sk_err = -err;
2310 }
2311out_put:
2312 fput_light(sock->file, fput_needed);
2313
2314 return datagrams;
2315}
2316
2317SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2318 unsigned int, vlen, unsigned int, flags,
2319 struct timespec __user *, timeout)
2320{
2321 int datagrams;
2322 struct timespec timeout_sys;
2323
2324 if (flags & MSG_CMSG_COMPAT)
2325 return -EINVAL;
2326
2327 if (!timeout)
2328 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL);
2329
2330 if (copy_from_user(&timeout_sys, timeout, sizeof(timeout_sys)))
2331 return -EFAULT;
2332
2333 datagrams = __sys_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2334
2335 if (datagrams > 0 &&
2336 copy_to_user(timeout, &timeout_sys, sizeof(timeout_sys)))
2337 datagrams = -EFAULT;
2338
2339 return datagrams;
2340}
2341
2342#ifdef __ARCH_WANT_SYS_SOCKETCALL
2343/* Argument list sizes for sys_socketcall */
2344#define AL(x) ((x) * sizeof(unsigned long))
2345static const unsigned char nargs[21] = {
2346 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2347 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2348 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2349 AL(4), AL(5), AL(4)
2350};
2351
2352#undef AL
2353
2354/*
2355 * System call vectors.
2356 *
2357 * Argument checking cleaned up. Saved 20% in size.
2358 * This function doesn't need to set the kernel lock because
2359 * it is set by the callees.
2360 */
2361
2362SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2363{
2364 unsigned long a[AUDITSC_ARGS];
2365 unsigned long a0, a1;
2366 int err;
2367 unsigned int len;
2368
2369 if (call < 1 || call > SYS_SENDMMSG)
2370 return -EINVAL;
2371
2372 len = nargs[call];
2373 if (len > sizeof(a))
2374 return -EINVAL;
2375
2376 /* copy_from_user should be SMP safe. */
2377 if (copy_from_user(a, args, len))
2378 return -EFAULT;
2379
2380 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2381 if (err)
2382 return err;
2383
2384 a0 = a[0];
2385 a1 = a[1];
2386
2387 switch (call) {
2388 case SYS_SOCKET:
2389 err = sys_socket(a0, a1, a[2]);
2390 break;
2391 case SYS_BIND:
2392 err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2393 break;
2394 case SYS_CONNECT:
2395 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2396 break;
2397 case SYS_LISTEN:
2398 err = sys_listen(a0, a1);
2399 break;
2400 case SYS_ACCEPT:
2401 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2402 (int __user *)a[2], 0);
2403 break;
2404 case SYS_GETSOCKNAME:
2405 err =
2406 sys_getsockname(a0, (struct sockaddr __user *)a1,
2407 (int __user *)a[2]);
2408 break;
2409 case SYS_GETPEERNAME:
2410 err =
2411 sys_getpeername(a0, (struct sockaddr __user *)a1,
2412 (int __user *)a[2]);
2413 break;
2414 case SYS_SOCKETPAIR:
2415 err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2416 break;
2417 case SYS_SEND:
2418 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2419 break;
2420 case SYS_SENDTO:
2421 err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2422 (struct sockaddr __user *)a[4], a[5]);
2423 break;
2424 case SYS_RECV:
2425 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2426 break;
2427 case SYS_RECVFROM:
2428 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2429 (struct sockaddr __user *)a[4],
2430 (int __user *)a[5]);
2431 break;
2432 case SYS_SHUTDOWN:
2433 err = sys_shutdown(a0, a1);
2434 break;
2435 case SYS_SETSOCKOPT:
2436 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2437 break;
2438 case SYS_GETSOCKOPT:
2439 err =
2440 sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2441 (int __user *)a[4]);
2442 break;
2443 case SYS_SENDMSG:
2444 err = sys_sendmsg(a0, (struct user_msghdr __user *)a1, a[2]);
2445 break;
2446 case SYS_SENDMMSG:
2447 err = sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3]);
2448 break;
2449 case SYS_RECVMSG:
2450 err = sys_recvmsg(a0, (struct user_msghdr __user *)a1, a[2]);
2451 break;
2452 case SYS_RECVMMSG:
2453 err = sys_recvmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3],
2454 (struct timespec __user *)a[4]);
2455 break;
2456 case SYS_ACCEPT4:
2457 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2458 (int __user *)a[2], a[3]);
2459 break;
2460 default:
2461 err = -EINVAL;
2462 break;
2463 }
2464 return err;
2465}
2466
2467#endif /* __ARCH_WANT_SYS_SOCKETCALL */
2468
2469/**
2470 * sock_register - add a socket protocol handler
2471 * @ops: description of protocol
2472 *
2473 * This function is called by a protocol handler that wants to
2474 * advertise its address family, and have it linked into the
2475 * socket interface. The value ops->family corresponds to the
2476 * socket system call protocol family.
2477 */
2478int sock_register(const struct net_proto_family *ops)
2479{
2480 int err;
2481
2482 if (ops->family >= NPROTO) {
2483 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
2484 return -ENOBUFS;
2485 }
2486
2487 spin_lock(&net_family_lock);
2488 if (rcu_dereference_protected(net_families[ops->family],
2489 lockdep_is_held(&net_family_lock)))
2490 err = -EEXIST;
2491 else {
2492 rcu_assign_pointer(net_families[ops->family], ops);
2493 err = 0;
2494 }
2495 spin_unlock(&net_family_lock);
2496
2497 pr_info("NET: Registered protocol family %d\n", ops->family);
2498 return err;
2499}
2500EXPORT_SYMBOL(sock_register);
2501
2502/**
2503 * sock_unregister - remove a protocol handler
2504 * @family: protocol family to remove
2505 *
2506 * This function is called by a protocol handler that wants to
2507 * remove its address family, and have it unlinked from the
2508 * new socket creation.
2509 *
2510 * If protocol handler is a module, then it can use module reference
2511 * counts to protect against new references. If protocol handler is not
2512 * a module then it needs to provide its own protection in
2513 * the ops->create routine.
2514 */
2515void sock_unregister(int family)
2516{
2517 BUG_ON(family < 0 || family >= NPROTO);
2518
2519 spin_lock(&net_family_lock);
2520 RCU_INIT_POINTER(net_families[family], NULL);
2521 spin_unlock(&net_family_lock);
2522
2523 synchronize_rcu();
2524
2525 pr_info("NET: Unregistered protocol family %d\n", family);
2526}
2527EXPORT_SYMBOL(sock_unregister);
2528
2529static int __init sock_init(void)
2530{
2531 int err;
2532 /*
2533 * Initialize the network sysctl infrastructure.
2534 */
2535 err = net_sysctl_init();
2536 if (err)
2537 goto out;
2538
2539 /*
2540 * Initialize skbuff SLAB cache
2541 */
2542 skb_init();
2543
2544 /*
2545 * Initialize the protocols module.
2546 */
2547
2548 init_inodecache();
2549
2550 err = register_filesystem(&sock_fs_type);
2551 if (err)
2552 goto out_fs;
2553 sock_mnt = kern_mount(&sock_fs_type);
2554 if (IS_ERR(sock_mnt)) {
2555 err = PTR_ERR(sock_mnt);
2556 goto out_mount;
2557 }
2558
2559 /* The real protocol initialization is performed in later initcalls.
2560 */
2561
2562#ifdef CONFIG_NETFILTER
2563 err = netfilter_init();
2564 if (err)
2565 goto out;
2566#endif
2567
2568 ptp_classifier_init();
2569
2570out:
2571 return err;
2572
2573out_mount:
2574 unregister_filesystem(&sock_fs_type);
2575out_fs:
2576 goto out;
2577}
2578
2579core_initcall(sock_init); /* early initcall */
2580
2581#ifdef CONFIG_PROC_FS
2582void socket_seq_show(struct seq_file *seq)
2583{
2584 int cpu;
2585 int counter = 0;
2586
2587 for_each_possible_cpu(cpu)
2588 counter += per_cpu(sockets_in_use, cpu);
2589
2590 /* It can be negative, by the way. 8) */
2591 if (counter < 0)
2592 counter = 0;
2593
2594 seq_printf(seq, "sockets: used %d\n", counter);
2595}
2596#endif /* CONFIG_PROC_FS */
2597
2598#ifdef CONFIG_COMPAT
2599static int do_siocgstamp(struct net *net, struct socket *sock,
2600 unsigned int cmd, void __user *up)
2601{
2602 mm_segment_t old_fs = get_fs();
2603 struct timeval ktv;
2604 int err;
2605
2606 set_fs(KERNEL_DS);
2607 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&ktv);
2608 set_fs(old_fs);
2609 if (!err)
2610 err = compat_put_timeval(&ktv, up);
2611
2612 return err;
2613}
2614
2615static int do_siocgstampns(struct net *net, struct socket *sock,
2616 unsigned int cmd, void __user *up)
2617{
2618 mm_segment_t old_fs = get_fs();
2619 struct timespec kts;
2620 int err;
2621
2622 set_fs(KERNEL_DS);
2623 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&kts);
2624 set_fs(old_fs);
2625 if (!err)
2626 err = compat_put_timespec(&kts, up);
2627
2628 return err;
2629}
2630
2631static int dev_ifname32(struct net *net, struct compat_ifreq __user *uifr32)
2632{
2633 struct ifreq __user *uifr;
2634 int err;
2635
2636 uifr = compat_alloc_user_space(sizeof(struct ifreq));
2637 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2638 return -EFAULT;
2639
2640 err = dev_ioctl(net, SIOCGIFNAME, uifr);
2641 if (err)
2642 return err;
2643
2644 if (copy_in_user(uifr32, uifr, sizeof(struct compat_ifreq)))
2645 return -EFAULT;
2646
2647 return 0;
2648}
2649
2650static int dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
2651{
2652 struct compat_ifconf ifc32;
2653 struct ifconf ifc;
2654 struct ifconf __user *uifc;
2655 struct compat_ifreq __user *ifr32;
2656 struct ifreq __user *ifr;
2657 unsigned int i, j;
2658 int err;
2659
2660 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
2661 return -EFAULT;
2662
2663 memset(&ifc, 0, sizeof(ifc));
2664 if (ifc32.ifcbuf == 0) {
2665 ifc32.ifc_len = 0;
2666 ifc.ifc_len = 0;
2667 ifc.ifc_req = NULL;
2668 uifc = compat_alloc_user_space(sizeof(struct ifconf));
2669 } else {
2670 size_t len = ((ifc32.ifc_len / sizeof(struct compat_ifreq)) + 1) *
2671 sizeof(struct ifreq);
2672 uifc = compat_alloc_user_space(sizeof(struct ifconf) + len);
2673 ifc.ifc_len = len;
2674 ifr = ifc.ifc_req = (void __user *)(uifc + 1);
2675 ifr32 = compat_ptr(ifc32.ifcbuf);
2676 for (i = 0; i < ifc32.ifc_len; i += sizeof(struct compat_ifreq)) {
2677 if (copy_in_user(ifr, ifr32, sizeof(struct compat_ifreq)))
2678 return -EFAULT;
2679 ifr++;
2680 ifr32++;
2681 }
2682 }
2683 if (copy_to_user(uifc, &ifc, sizeof(struct ifconf)))
2684 return -EFAULT;
2685
2686 err = dev_ioctl(net, SIOCGIFCONF, uifc);
2687 if (err)
2688 return err;
2689
2690 if (copy_from_user(&ifc, uifc, sizeof(struct ifconf)))
2691 return -EFAULT;
2692
2693 ifr = ifc.ifc_req;
2694 ifr32 = compat_ptr(ifc32.ifcbuf);
2695 for (i = 0, j = 0;
2696 i + sizeof(struct compat_ifreq) <= ifc32.ifc_len && j < ifc.ifc_len;
2697 i += sizeof(struct compat_ifreq), j += sizeof(struct ifreq)) {
2698 if (copy_in_user(ifr32, ifr, sizeof(struct compat_ifreq)))
2699 return -EFAULT;
2700 ifr32++;
2701 ifr++;
2702 }
2703
2704 if (ifc32.ifcbuf == 0) {
2705 /* Translate from 64-bit structure multiple to
2706 * a 32-bit one.
2707 */
2708 i = ifc.ifc_len;
2709 i = ((i / sizeof(struct ifreq)) * sizeof(struct compat_ifreq));
2710 ifc32.ifc_len = i;
2711 } else {
2712 ifc32.ifc_len = i;
2713 }
2714 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
2715 return -EFAULT;
2716
2717 return 0;
2718}
2719
2720static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
2721{
2722 struct compat_ethtool_rxnfc __user *compat_rxnfc;
2723 bool convert_in = false, convert_out = false;
2724 size_t buf_size = ALIGN(sizeof(struct ifreq), 8);
2725 struct ethtool_rxnfc __user *rxnfc;
2726 struct ifreq __user *ifr;
2727 u32 rule_cnt = 0, actual_rule_cnt;
2728 u32 ethcmd;
2729 u32 data;
2730 int ret;
2731
2732 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2733 return -EFAULT;
2734
2735 compat_rxnfc = compat_ptr(data);
2736
2737 if (get_user(ethcmd, &compat_rxnfc->cmd))
2738 return -EFAULT;
2739
2740 /* Most ethtool structures are defined without padding.
2741 * Unfortunately struct ethtool_rxnfc is an exception.
2742 */
2743 switch (ethcmd) {
2744 default:
2745 break;
2746 case ETHTOOL_GRXCLSRLALL:
2747 /* Buffer size is variable */
2748 if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
2749 return -EFAULT;
2750 if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
2751 return -ENOMEM;
2752 buf_size += rule_cnt * sizeof(u32);
2753 /* fall through */
2754 case ETHTOOL_GRXRINGS:
2755 case ETHTOOL_GRXCLSRLCNT:
2756 case ETHTOOL_GRXCLSRULE:
2757 case ETHTOOL_SRXCLSRLINS:
2758 convert_out = true;
2759 /* fall through */
2760 case ETHTOOL_SRXCLSRLDEL:
2761 buf_size += sizeof(struct ethtool_rxnfc);
2762 convert_in = true;
2763 break;
2764 }
2765
2766 ifr = compat_alloc_user_space(buf_size);
2767 rxnfc = (void __user *)ifr + ALIGN(sizeof(struct ifreq), 8);
2768
2769 if (copy_in_user(&ifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2770 return -EFAULT;
2771
2772 if (put_user(convert_in ? rxnfc : compat_ptr(data),
2773 &ifr->ifr_ifru.ifru_data))
2774 return -EFAULT;
2775
2776 if (convert_in) {
2777 /* We expect there to be holes between fs.m_ext and
2778 * fs.ring_cookie and at the end of fs, but nowhere else.
2779 */
2780 BUILD_BUG_ON(offsetof(struct compat_ethtool_rxnfc, fs.m_ext) +
2781 sizeof(compat_rxnfc->fs.m_ext) !=
2782 offsetof(struct ethtool_rxnfc, fs.m_ext) +
2783 sizeof(rxnfc->fs.m_ext));
2784 BUILD_BUG_ON(
2785 offsetof(struct compat_ethtool_rxnfc, fs.location) -
2786 offsetof(struct compat_ethtool_rxnfc, fs.ring_cookie) !=
2787 offsetof(struct ethtool_rxnfc, fs.location) -
2788 offsetof(struct ethtool_rxnfc, fs.ring_cookie));
2789
2790 if (copy_in_user(rxnfc, compat_rxnfc,
2791 (void __user *)(&rxnfc->fs.m_ext + 1) -
2792 (void __user *)rxnfc) ||
2793 copy_in_user(&rxnfc->fs.ring_cookie,
2794 &compat_rxnfc->fs.ring_cookie,
2795 (void __user *)(&rxnfc->fs.location + 1) -
2796 (void __user *)&rxnfc->fs.ring_cookie) ||
2797 copy_in_user(&rxnfc->rule_cnt, &compat_rxnfc->rule_cnt,
2798 sizeof(rxnfc->rule_cnt)))
2799 return -EFAULT;
2800 }
2801
2802 ret = dev_ioctl(net, SIOCETHTOOL, ifr);
2803 if (ret)
2804 return ret;
2805
2806 if (convert_out) {
2807 if (copy_in_user(compat_rxnfc, rxnfc,
2808 (const void __user *)(&rxnfc->fs.m_ext + 1) -
2809 (const void __user *)rxnfc) ||
2810 copy_in_user(&compat_rxnfc->fs.ring_cookie,
2811 &rxnfc->fs.ring_cookie,
2812 (const void __user *)(&rxnfc->fs.location + 1) -
2813 (const void __user *)&rxnfc->fs.ring_cookie) ||
2814 copy_in_user(&compat_rxnfc->rule_cnt, &rxnfc->rule_cnt,
2815 sizeof(rxnfc->rule_cnt)))
2816 return -EFAULT;
2817
2818 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
2819 /* As an optimisation, we only copy the actual
2820 * number of rules that the underlying
2821 * function returned. Since Mallory might
2822 * change the rule count in user memory, we
2823 * check that it is less than the rule count
2824 * originally given (as the user buffer size),
2825 * which has been range-checked.
2826 */
2827 if (get_user(actual_rule_cnt, &rxnfc->rule_cnt))
2828 return -EFAULT;
2829 if (actual_rule_cnt < rule_cnt)
2830 rule_cnt = actual_rule_cnt;
2831 if (copy_in_user(&compat_rxnfc->rule_locs[0],
2832 &rxnfc->rule_locs[0],
2833 rule_cnt * sizeof(u32)))
2834 return -EFAULT;
2835 }
2836 }
2837
2838 return 0;
2839}
2840
2841static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
2842{
2843 void __user *uptr;
2844 compat_uptr_t uptr32;
2845 struct ifreq __user *uifr;
2846
2847 uifr = compat_alloc_user_space(sizeof(*uifr));
2848 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2849 return -EFAULT;
2850
2851 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
2852 return -EFAULT;
2853
2854 uptr = compat_ptr(uptr32);
2855
2856 if (put_user(uptr, &uifr->ifr_settings.ifs_ifsu.raw_hdlc))
2857 return -EFAULT;
2858
2859 return dev_ioctl(net, SIOCWANDEV, uifr);
2860}
2861
2862static int bond_ioctl(struct net *net, unsigned int cmd,
2863 struct compat_ifreq __user *ifr32)
2864{
2865 struct ifreq kifr;
2866 mm_segment_t old_fs;
2867 int err;
2868
2869 switch (cmd) {
2870 case SIOCBONDENSLAVE:
2871 case SIOCBONDRELEASE:
2872 case SIOCBONDSETHWADDR:
2873 case SIOCBONDCHANGEACTIVE:
2874 if (copy_from_user(&kifr, ifr32, sizeof(struct compat_ifreq)))
2875 return -EFAULT;
2876
2877 old_fs = get_fs();
2878 set_fs(KERNEL_DS);
2879 err = dev_ioctl(net, cmd,
2880 (struct ifreq __user __force *) &kifr);
2881 set_fs(old_fs);
2882
2883 return err;
2884 default:
2885 return -ENOIOCTLCMD;
2886 }
2887}
2888
2889/* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
2890static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
2891 struct compat_ifreq __user *u_ifreq32)
2892{
2893 struct ifreq __user *u_ifreq64;
2894 char tmp_buf[IFNAMSIZ];
2895 void __user *data64;
2896 u32 data32;
2897
2898 if (copy_from_user(&tmp_buf[0], &(u_ifreq32->ifr_ifrn.ifrn_name[0]),
2899 IFNAMSIZ))
2900 return -EFAULT;
2901 if (get_user(data32, &u_ifreq32->ifr_ifru.ifru_data))
2902 return -EFAULT;
2903 data64 = compat_ptr(data32);
2904
2905 u_ifreq64 = compat_alloc_user_space(sizeof(*u_ifreq64));
2906
2907 if (copy_to_user(&u_ifreq64->ifr_ifrn.ifrn_name[0], &tmp_buf[0],
2908 IFNAMSIZ))
2909 return -EFAULT;
2910 if (put_user(data64, &u_ifreq64->ifr_ifru.ifru_data))
2911 return -EFAULT;
2912
2913 return dev_ioctl(net, cmd, u_ifreq64);
2914}
2915
2916static int dev_ifsioc(struct net *net, struct socket *sock,
2917 unsigned int cmd, struct compat_ifreq __user *uifr32)
2918{
2919 struct ifreq __user *uifr;
2920 int err;
2921
2922 uifr = compat_alloc_user_space(sizeof(*uifr));
2923 if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
2924 return -EFAULT;
2925
2926 err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
2927
2928 if (!err) {
2929 switch (cmd) {
2930 case SIOCGIFFLAGS:
2931 case SIOCGIFMETRIC:
2932 case SIOCGIFMTU:
2933 case SIOCGIFMEM:
2934 case SIOCGIFHWADDR:
2935 case SIOCGIFINDEX:
2936 case SIOCGIFADDR:
2937 case SIOCGIFBRDADDR:
2938 case SIOCGIFDSTADDR:
2939 case SIOCGIFNETMASK:
2940 case SIOCGIFPFLAGS:
2941 case SIOCGIFTXQLEN:
2942 case SIOCGMIIPHY:
2943 case SIOCGMIIREG:
2944 if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
2945 err = -EFAULT;
2946 break;
2947 }
2948 }
2949 return err;
2950}
2951
2952static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
2953 struct compat_ifreq __user *uifr32)
2954{
2955 struct ifreq ifr;
2956 struct compat_ifmap __user *uifmap32;
2957 mm_segment_t old_fs;
2958 int err;
2959
2960 uifmap32 = &uifr32->ifr_ifru.ifru_map;
2961 err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
2962 err |= get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2963 err |= get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2964 err |= get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2965 err |= get_user(ifr.ifr_map.irq, &uifmap32->irq);
2966 err |= get_user(ifr.ifr_map.dma, &uifmap32->dma);
2967 err |= get_user(ifr.ifr_map.port, &uifmap32->port);
2968 if (err)
2969 return -EFAULT;
2970
2971 old_fs = get_fs();
2972 set_fs(KERNEL_DS);
2973 err = dev_ioctl(net, cmd, (void __user __force *)&ifr);
2974 set_fs(old_fs);
2975
2976 if (cmd == SIOCGIFMAP && !err) {
2977 err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
2978 err |= put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2979 err |= put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2980 err |= put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2981 err |= put_user(ifr.ifr_map.irq, &uifmap32->irq);
2982 err |= put_user(ifr.ifr_map.dma, &uifmap32->dma);
2983 err |= put_user(ifr.ifr_map.port, &uifmap32->port);
2984 if (err)
2985 err = -EFAULT;
2986 }
2987 return err;
2988}
2989
2990struct rtentry32 {
2991 u32 rt_pad1;
2992 struct sockaddr rt_dst; /* target address */
2993 struct sockaddr rt_gateway; /* gateway addr (RTF_GATEWAY) */
2994 struct sockaddr rt_genmask; /* target network mask (IP) */
2995 unsigned short rt_flags;
2996 short rt_pad2;
2997 u32 rt_pad3;
2998 unsigned char rt_tos;
2999 unsigned char rt_class;
3000 short rt_pad4;
3001 short rt_metric; /* +1 for binary compatibility! */
3002 /* char * */ u32 rt_dev; /* forcing the device at add */
3003 u32 rt_mtu; /* per route MTU/Window */
3004 u32 rt_window; /* Window clamping */
3005 unsigned short rt_irtt; /* Initial RTT */
3006};
3007
3008struct in6_rtmsg32 {
3009 struct in6_addr rtmsg_dst;
3010 struct in6_addr rtmsg_src;
3011 struct in6_addr rtmsg_gateway;
3012 u32 rtmsg_type;
3013 u16 rtmsg_dst_len;
3014 u16 rtmsg_src_len;
3015 u32 rtmsg_metric;
3016 u32 rtmsg_info;
3017 u32 rtmsg_flags;
3018 s32 rtmsg_ifindex;
3019};
3020
3021static int routing_ioctl(struct net *net, struct socket *sock,
3022 unsigned int cmd, void __user *argp)
3023{
3024 int ret;
3025 void *r = NULL;
3026 struct in6_rtmsg r6;
3027 struct rtentry r4;
3028 char devname[16];
3029 u32 rtdev;
3030 mm_segment_t old_fs = get_fs();
3031
3032 if (sock && sock->sk && sock->sk->sk_family == AF_INET6) { /* ipv6 */
3033 struct in6_rtmsg32 __user *ur6 = argp;
3034 ret = copy_from_user(&r6.rtmsg_dst, &(ur6->rtmsg_dst),
3035 3 * sizeof(struct in6_addr));
3036 ret |= get_user(r6.rtmsg_type, &(ur6->rtmsg_type));
3037 ret |= get_user(r6.rtmsg_dst_len, &(ur6->rtmsg_dst_len));
3038 ret |= get_user(r6.rtmsg_src_len, &(ur6->rtmsg_src_len));
3039 ret |= get_user(r6.rtmsg_metric, &(ur6->rtmsg_metric));
3040 ret |= get_user(r6.rtmsg_info, &(ur6->rtmsg_info));
3041 ret |= get_user(r6.rtmsg_flags, &(ur6->rtmsg_flags));
3042 ret |= get_user(r6.rtmsg_ifindex, &(ur6->rtmsg_ifindex));
3043
3044 r = (void *) &r6;
3045 } else { /* ipv4 */
3046 struct rtentry32 __user *ur4 = argp;
3047 ret = copy_from_user(&r4.rt_dst, &(ur4->rt_dst),
3048 3 * sizeof(struct sockaddr));
3049 ret |= get_user(r4.rt_flags, &(ur4->rt_flags));
3050 ret |= get_user(r4.rt_metric, &(ur4->rt_metric));
3051 ret |= get_user(r4.rt_mtu, &(ur4->rt_mtu));
3052 ret |= get_user(r4.rt_window, &(ur4->rt_window));
3053 ret |= get_user(r4.rt_irtt, &(ur4->rt_irtt));
3054 ret |= get_user(rtdev, &(ur4->rt_dev));
3055 if (rtdev) {
3056 ret |= copy_from_user(devname, compat_ptr(rtdev), 15);
3057 r4.rt_dev = (char __user __force *)devname;
3058 devname[15] = 0;
3059 } else
3060 r4.rt_dev = NULL;
3061
3062 r = (void *) &r4;
3063 }
3064
3065 if (ret) {
3066 ret = -EFAULT;
3067 goto out;
3068 }
3069
3070 set_fs(KERNEL_DS);
3071 ret = sock_do_ioctl(net, sock, cmd, (unsigned long) r);
3072 set_fs(old_fs);
3073
3074out:
3075 return ret;
3076}
3077
3078/* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3079 * for some operations; this forces use of the newer bridge-utils that
3080 * use compatible ioctls
3081 */
3082static int old_bridge_ioctl(compat_ulong_t __user *argp)
3083{
3084 compat_ulong_t tmp;
3085
3086 if (get_user(tmp, argp))
3087 return -EFAULT;
3088 if (tmp == BRCTL_GET_VERSION)
3089 return BRCTL_VERSION + 1;
3090 return -EINVAL;
3091}
3092
3093static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3094 unsigned int cmd, unsigned long arg)
3095{
3096 void __user *argp = compat_ptr(arg);
3097 struct sock *sk = sock->sk;
3098 struct net *net = sock_net(sk);
3099
3100 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3101 return compat_ifr_data_ioctl(net, cmd, argp);
3102
3103 switch (cmd) {
3104 case SIOCSIFBR:
3105 case SIOCGIFBR:
3106 return old_bridge_ioctl(argp);
3107 case SIOCGIFNAME:
3108 return dev_ifname32(net, argp);
3109 case SIOCGIFCONF:
3110 return dev_ifconf(net, argp);
3111 case SIOCETHTOOL:
3112 return ethtool_ioctl(net, argp);
3113 case SIOCWANDEV:
3114 return compat_siocwandev(net, argp);
3115 case SIOCGIFMAP:
3116 case SIOCSIFMAP:
3117 return compat_sioc_ifmap(net, cmd, argp);
3118 case SIOCBONDENSLAVE:
3119 case SIOCBONDRELEASE:
3120 case SIOCBONDSETHWADDR:
3121 case SIOCBONDCHANGEACTIVE:
3122 return bond_ioctl(net, cmd, argp);
3123 case SIOCADDRT:
3124 case SIOCDELRT:
3125 return routing_ioctl(net, sock, cmd, argp);
3126 case SIOCGSTAMP:
3127 return do_siocgstamp(net, sock, cmd, argp);
3128 case SIOCGSTAMPNS:
3129 return do_siocgstampns(net, sock, cmd, argp);
3130 case SIOCBONDSLAVEINFOQUERY:
3131 case SIOCBONDINFOQUERY:
3132 case SIOCSHWTSTAMP:
3133 case SIOCGHWTSTAMP:
3134 return compat_ifr_data_ioctl(net, cmd, argp);
3135
3136 case FIOSETOWN:
3137 case SIOCSPGRP:
3138 case FIOGETOWN:
3139 case SIOCGPGRP:
3140 case SIOCBRADDBR:
3141 case SIOCBRDELBR:
3142 case SIOCGIFVLAN:
3143 case SIOCSIFVLAN:
3144 case SIOCADDDLCI:
3145 case SIOCDELDLCI:
3146 case SIOCGSKNS:
3147 return sock_ioctl(file, cmd, arg);
3148
3149 case SIOCGIFFLAGS:
3150 case SIOCSIFFLAGS:
3151 case SIOCGIFMETRIC:
3152 case SIOCSIFMETRIC:
3153 case SIOCGIFMTU:
3154 case SIOCSIFMTU:
3155 case SIOCGIFMEM:
3156 case SIOCSIFMEM:
3157 case SIOCGIFHWADDR:
3158 case SIOCSIFHWADDR:
3159 case SIOCADDMULTI:
3160 case SIOCDELMULTI:
3161 case SIOCGIFINDEX:
3162 case SIOCGIFADDR:
3163 case SIOCSIFADDR:
3164 case SIOCSIFHWBROADCAST:
3165 case SIOCDIFADDR:
3166 case SIOCGIFBRDADDR:
3167 case SIOCSIFBRDADDR:
3168 case SIOCGIFDSTADDR:
3169 case SIOCSIFDSTADDR:
3170 case SIOCGIFNETMASK:
3171 case SIOCSIFNETMASK:
3172 case SIOCSIFPFLAGS:
3173 case SIOCGIFPFLAGS:
3174 case SIOCGIFTXQLEN:
3175 case SIOCSIFTXQLEN:
3176 case SIOCBRADDIF:
3177 case SIOCBRDELIF:
3178 case SIOCSIFNAME:
3179 case SIOCGMIIPHY:
3180 case SIOCGMIIREG:
3181 case SIOCSMIIREG:
3182 return dev_ifsioc(net, sock, cmd, argp);
3183
3184 case SIOCSARP:
3185 case SIOCGARP:
3186 case SIOCDARP:
3187 case SIOCATMARK:
3188 return sock_do_ioctl(net, sock, cmd, arg);
3189 }
3190
3191 return -ENOIOCTLCMD;
3192}
3193
3194static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3195 unsigned long arg)
3196{
3197 struct socket *sock = file->private_data;
3198 int ret = -ENOIOCTLCMD;
3199 struct sock *sk;
3200 struct net *net;
3201
3202 sk = sock->sk;
3203 net = sock_net(sk);
3204
3205 if (sock->ops->compat_ioctl)
3206 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3207
3208 if (ret == -ENOIOCTLCMD &&
3209 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3210 ret = compat_wext_handle_ioctl(net, cmd, arg);
3211
3212 if (ret == -ENOIOCTLCMD)
3213 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3214
3215 return ret;
3216}
3217#endif
3218
3219int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3220{
3221 return sock->ops->bind(sock, addr, addrlen);
3222}
3223EXPORT_SYMBOL(kernel_bind);
3224
3225int kernel_listen(struct socket *sock, int backlog)
3226{
3227 return sock->ops->listen(sock, backlog);
3228}
3229EXPORT_SYMBOL(kernel_listen);
3230
3231int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3232{
3233 struct sock *sk = sock->sk;
3234 int err;
3235
3236 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3237 newsock);
3238 if (err < 0)
3239 goto done;
3240
3241 err = sock->ops->accept(sock, *newsock, flags);
3242 if (err < 0) {
3243 sock_release(*newsock);
3244 *newsock = NULL;
3245 goto done;
3246 }
3247
3248 (*newsock)->ops = sock->ops;
3249 __module_get((*newsock)->ops->owner);
3250
3251done:
3252 return err;
3253}
3254EXPORT_SYMBOL(kernel_accept);
3255
3256int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3257 int flags)
3258{
3259 return sock->ops->connect(sock, addr, addrlen, flags);
3260}
3261EXPORT_SYMBOL(kernel_connect);
3262
3263int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
3264 int *addrlen)
3265{
3266 return sock->ops->getname(sock, addr, addrlen, 0);
3267}
3268EXPORT_SYMBOL(kernel_getsockname);
3269
3270int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
3271 int *addrlen)
3272{
3273 return sock->ops->getname(sock, addr, addrlen, 1);
3274}
3275EXPORT_SYMBOL(kernel_getpeername);
3276
3277int kernel_getsockopt(struct socket *sock, int level, int optname,
3278 char *optval, int *optlen)
3279{
3280 mm_segment_t oldfs = get_fs();
3281 char __user *uoptval;
3282 int __user *uoptlen;
3283 int err;
3284
3285 uoptval = (char __user __force *) optval;
3286 uoptlen = (int __user __force *) optlen;
3287
3288 set_fs(KERNEL_DS);
3289 if (level == SOL_SOCKET)
3290 err = sock_getsockopt(sock, level, optname, uoptval, uoptlen);
3291 else
3292 err = sock->ops->getsockopt(sock, level, optname, uoptval,
3293 uoptlen);
3294 set_fs(oldfs);
3295 return err;
3296}
3297EXPORT_SYMBOL(kernel_getsockopt);
3298
3299int kernel_setsockopt(struct socket *sock, int level, int optname,
3300 char *optval, unsigned int optlen)
3301{
3302 mm_segment_t oldfs = get_fs();
3303 char __user *uoptval;
3304 int err;
3305
3306 uoptval = (char __user __force *) optval;
3307
3308 set_fs(KERNEL_DS);
3309 if (level == SOL_SOCKET)
3310 err = sock_setsockopt(sock, level, optname, uoptval, optlen);
3311 else
3312 err = sock->ops->setsockopt(sock, level, optname, uoptval,
3313 optlen);
3314 set_fs(oldfs);
3315 return err;
3316}
3317EXPORT_SYMBOL(kernel_setsockopt);
3318
3319int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3320 size_t size, int flags)
3321{
3322 if (sock->ops->sendpage)
3323 return sock->ops->sendpage(sock, page, offset, size, flags);
3324
3325 return sock_no_sendpage(sock, page, offset, size, flags);
3326}
3327EXPORT_SYMBOL(kernel_sendpage);
3328
3329int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
3330{
3331 mm_segment_t oldfs = get_fs();
3332 int err;
3333
3334 set_fs(KERNEL_DS);
3335 err = sock->ops->ioctl(sock, cmd, arg);
3336 set_fs(oldfs);
3337
3338 return err;
3339}
3340EXPORT_SYMBOL(kernel_sock_ioctl);
3341
3342int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3343{
3344 return sock->ops->shutdown(sock, how);
3345}
3346EXPORT_SYMBOL(kernel_sock_shutdown);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET An implementation of the SOCKET network access protocol.
4 *
5 * Version: @(#)socket.c 1.1.93 18/02/95
6 *
7 * Authors: Orest Zborowski, <obz@Kodak.COM>
8 * Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 *
11 * Fixes:
12 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
13 * shutdown()
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
18 * top level.
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
23 * tty drivers).
24 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
25 * Jeff Uphoff : Made max number of sockets command-line
26 * configurable.
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
35 * stuff.
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
41 * moment.
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
48 *
49 * This module is effectively the top level interface to the BSD socket
50 * paradigm.
51 *
52 * Based upon Swansea University Computer Society NET3.039
53 */
54
55#include <linux/mm.h>
56#include <linux/socket.h>
57#include <linux/file.h>
58#include <linux/net.h>
59#include <linux/interrupt.h>
60#include <linux/thread_info.h>
61#include <linux/rcupdate.h>
62#include <linux/netdevice.h>
63#include <linux/proc_fs.h>
64#include <linux/seq_file.h>
65#include <linux/mutex.h>
66#include <linux/if_bridge.h>
67#include <linux/if_frad.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>
89
90#include <linux/uaccess.h>
91#include <asm/unistd.h>
92
93#include <net/compat.h>
94#include <net/wext.h>
95#include <net/cls_cgroup.h>
96
97#include <net/sock.h>
98#include <linux/netfilter.h>
99
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
108#ifdef CONFIG_NET_RX_BUSY_POLL
109unsigned int sysctl_net_busy_read __read_mostly;
110unsigned int sysctl_net_busy_poll __read_mostly;
111#endif
112
113static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
114static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
115static int sock_mmap(struct file *file, struct vm_area_struct *vma);
116
117static int sock_close(struct inode *inode, struct file *file);
118static __poll_t sock_poll(struct file *file,
119 struct poll_table_struct *wait);
120static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
121#ifdef CONFIG_COMPAT
122static long compat_sock_ioctl(struct file *file,
123 unsigned int cmd, unsigned long arg);
124#endif
125static int sock_fasync(int fd, struct file *filp, int on);
126static ssize_t sock_sendpage(struct file *file, struct page *page,
127 int offset, size_t size, loff_t *ppos, int more);
128static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
129 struct pipe_inode_info *pipe, size_t len,
130 unsigned int flags);
131
132#ifdef CONFIG_PROC_FS
133static void sock_show_fdinfo(struct seq_file *m, struct file *f)
134{
135 struct socket *sock = f->private_data;
136
137 if (sock->ops->show_fdinfo)
138 sock->ops->show_fdinfo(m, sock);
139}
140#else
141#define sock_show_fdinfo NULL
142#endif
143
144/*
145 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
146 * in the operation structures but are done directly via the socketcall() multiplexor.
147 */
148
149static const struct file_operations socket_file_ops = {
150 .owner = THIS_MODULE,
151 .llseek = no_llseek,
152 .read_iter = sock_read_iter,
153 .write_iter = sock_write_iter,
154 .poll = sock_poll,
155 .unlocked_ioctl = sock_ioctl,
156#ifdef CONFIG_COMPAT
157 .compat_ioctl = compat_sock_ioctl,
158#endif
159 .mmap = sock_mmap,
160 .release = sock_close,
161 .fasync = sock_fasync,
162 .sendpage = sock_sendpage,
163 .splice_write = generic_splice_sendpage,
164 .splice_read = sock_splice_read,
165 .show_fdinfo = sock_show_fdinfo,
166};
167
168/*
169 * The protocol list. Each protocol is registered in here.
170 */
171
172static DEFINE_SPINLOCK(net_family_lock);
173static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
174
175/*
176 * Support routines.
177 * Move socket addresses back and forth across the kernel/user
178 * divide and look after the messy bits.
179 */
180
181/**
182 * move_addr_to_kernel - copy a socket address into kernel space
183 * @uaddr: Address in user space
184 * @kaddr: Address in kernel space
185 * @ulen: Length in user space
186 *
187 * The address is copied into kernel space. If the provided address is
188 * too long an error code of -EINVAL is returned. If the copy gives
189 * invalid addresses -EFAULT is returned. On a success 0 is returned.
190 */
191
192int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
193{
194 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
195 return -EINVAL;
196 if (ulen == 0)
197 return 0;
198 if (copy_from_user(kaddr, uaddr, ulen))
199 return -EFAULT;
200 return audit_sockaddr(ulen, kaddr);
201}
202
203/**
204 * move_addr_to_user - copy an address to user space
205 * @kaddr: kernel space address
206 * @klen: length of address in kernel
207 * @uaddr: user space address
208 * @ulen: pointer to user length field
209 *
210 * The value pointed to by ulen on entry is the buffer length available.
211 * This is overwritten with the buffer space used. -EINVAL is returned
212 * if an overlong buffer is specified or a negative buffer size. -EFAULT
213 * is returned if either the buffer or the length field are not
214 * accessible.
215 * After copying the data up to the limit the user specifies, the true
216 * length of the data is written over the length limit the user
217 * specified. Zero is returned for a success.
218 */
219
220static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
221 void __user *uaddr, int __user *ulen)
222{
223 int err;
224 int len;
225
226 BUG_ON(klen > sizeof(struct sockaddr_storage));
227 err = get_user(len, ulen);
228 if (err)
229 return err;
230 if (len > klen)
231 len = klen;
232 if (len < 0)
233 return -EINVAL;
234 if (len) {
235 if (audit_sockaddr(klen, kaddr))
236 return -ENOMEM;
237 if (copy_to_user(uaddr, kaddr, len))
238 return -EFAULT;
239 }
240 /*
241 * "fromlen shall refer to the value before truncation.."
242 * 1003.1g
243 */
244 return __put_user(klen, ulen);
245}
246
247static struct kmem_cache *sock_inode_cachep __ro_after_init;
248
249static struct inode *sock_alloc_inode(struct super_block *sb)
250{
251 struct socket_alloc *ei;
252
253 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
254 if (!ei)
255 return NULL;
256 init_waitqueue_head(&ei->socket.wq.wait);
257 ei->socket.wq.fasync_list = NULL;
258 ei->socket.wq.flags = 0;
259
260 ei->socket.state = SS_UNCONNECTED;
261 ei->socket.flags = 0;
262 ei->socket.ops = NULL;
263 ei->socket.sk = NULL;
264 ei->socket.file = NULL;
265
266 return &ei->vfs_inode;
267}
268
269static void sock_free_inode(struct inode *inode)
270{
271 struct socket_alloc *ei;
272
273 ei = container_of(inode, struct socket_alloc, vfs_inode);
274 kmem_cache_free(sock_inode_cachep, ei);
275}
276
277static void init_once(void *foo)
278{
279 struct socket_alloc *ei = (struct socket_alloc *)foo;
280
281 inode_init_once(&ei->vfs_inode);
282}
283
284static void init_inodecache(void)
285{
286 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
287 sizeof(struct socket_alloc),
288 0,
289 (SLAB_HWCACHE_ALIGN |
290 SLAB_RECLAIM_ACCOUNT |
291 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
292 init_once);
293 BUG_ON(sock_inode_cachep == NULL);
294}
295
296static const struct super_operations sockfs_ops = {
297 .alloc_inode = sock_alloc_inode,
298 .free_inode = sock_free_inode,
299 .statfs = simple_statfs,
300};
301
302/*
303 * sockfs_dname() is called from d_path().
304 */
305static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
306{
307 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
308 d_inode(dentry)->i_ino);
309}
310
311static const struct dentry_operations sockfs_dentry_operations = {
312 .d_dname = sockfs_dname,
313};
314
315static int sockfs_xattr_get(const struct xattr_handler *handler,
316 struct dentry *dentry, struct inode *inode,
317 const char *suffix, void *value, size_t size)
318{
319 if (value) {
320 if (dentry->d_name.len + 1 > size)
321 return -ERANGE;
322 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
323 }
324 return dentry->d_name.len + 1;
325}
326
327#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
328#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
329#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
330
331static const struct xattr_handler sockfs_xattr_handler = {
332 .name = XATTR_NAME_SOCKPROTONAME,
333 .get = sockfs_xattr_get,
334};
335
336static int sockfs_security_xattr_set(const struct xattr_handler *handler,
337 struct dentry *dentry, struct inode *inode,
338 const char *suffix, const void *value,
339 size_t size, int flags)
340{
341 /* Handled by LSM. */
342 return -EAGAIN;
343}
344
345static const struct xattr_handler sockfs_security_xattr_handler = {
346 .prefix = XATTR_SECURITY_PREFIX,
347 .set = sockfs_security_xattr_set,
348};
349
350static const struct xattr_handler *sockfs_xattr_handlers[] = {
351 &sockfs_xattr_handler,
352 &sockfs_security_xattr_handler,
353 NULL
354};
355
356static int sockfs_init_fs_context(struct fs_context *fc)
357{
358 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
359 if (!ctx)
360 return -ENOMEM;
361 ctx->ops = &sockfs_ops;
362 ctx->dops = &sockfs_dentry_operations;
363 ctx->xattr = sockfs_xattr_handlers;
364 return 0;
365}
366
367static struct vfsmount *sock_mnt __read_mostly;
368
369static struct file_system_type sock_fs_type = {
370 .name = "sockfs",
371 .init_fs_context = sockfs_init_fs_context,
372 .kill_sb = kill_anon_super,
373};
374
375/*
376 * Obtains the first available file descriptor and sets it up for use.
377 *
378 * These functions create file structures and maps them to fd space
379 * of the current process. On success it returns file descriptor
380 * and file struct implicitly stored in sock->file.
381 * Note that another thread may close file descriptor before we return
382 * from this function. We use the fact that now we do not refer
383 * to socket after mapping. If one day we will need it, this
384 * function will increment ref. count on file by 1.
385 *
386 * In any case returned fd MAY BE not valid!
387 * This race condition is unavoidable
388 * with shared fd spaces, we cannot solve it inside kernel,
389 * but we take care of internal coherence yet.
390 */
391
392/**
393 * sock_alloc_file - Bind a &socket to a &file
394 * @sock: socket
395 * @flags: file status flags
396 * @dname: protocol name
397 *
398 * Returns the &file bound with @sock, implicitly storing it
399 * in sock->file. If dname is %NULL, sets to "".
400 * On failure the return is a ERR pointer (see linux/err.h).
401 * This function uses GFP_KERNEL internally.
402 */
403
404struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
405{
406 struct file *file;
407
408 if (!dname)
409 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
410
411 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
412 O_RDWR | (flags & O_NONBLOCK),
413 &socket_file_ops);
414 if (IS_ERR(file)) {
415 sock_release(sock);
416 return file;
417 }
418
419 sock->file = file;
420 file->private_data = sock;
421 stream_open(SOCK_INODE(sock), file);
422 return file;
423}
424EXPORT_SYMBOL(sock_alloc_file);
425
426static int sock_map_fd(struct socket *sock, int flags)
427{
428 struct file *newfile;
429 int fd = get_unused_fd_flags(flags);
430 if (unlikely(fd < 0)) {
431 sock_release(sock);
432 return fd;
433 }
434
435 newfile = sock_alloc_file(sock, flags, NULL);
436 if (!IS_ERR(newfile)) {
437 fd_install(fd, newfile);
438 return fd;
439 }
440
441 put_unused_fd(fd);
442 return PTR_ERR(newfile);
443}
444
445/**
446 * sock_from_file - Return the &socket bounded to @file.
447 * @file: file
448 * @err: pointer to an error code return
449 *
450 * On failure returns %NULL and assigns -ENOTSOCK to @err.
451 */
452
453struct socket *sock_from_file(struct file *file, int *err)
454{
455 if (file->f_op == &socket_file_ops)
456 return file->private_data; /* set in sock_map_fd */
457
458 *err = -ENOTSOCK;
459 return NULL;
460}
461EXPORT_SYMBOL(sock_from_file);
462
463/**
464 * sockfd_lookup - Go from a file number to its socket slot
465 * @fd: file handle
466 * @err: pointer to an error code return
467 *
468 * The file handle passed in is locked and the socket it is bound
469 * to is returned. If an error occurs the err pointer is overwritten
470 * with a negative errno code and NULL is returned. The function checks
471 * for both invalid handles and passing a handle which is not a socket.
472 *
473 * On a success the socket object pointer is returned.
474 */
475
476struct socket *sockfd_lookup(int fd, int *err)
477{
478 struct file *file;
479 struct socket *sock;
480
481 file = fget(fd);
482 if (!file) {
483 *err = -EBADF;
484 return NULL;
485 }
486
487 sock = sock_from_file(file, err);
488 if (!sock)
489 fput(file);
490 return sock;
491}
492EXPORT_SYMBOL(sockfd_lookup);
493
494static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
495{
496 struct fd f = fdget(fd);
497 struct socket *sock;
498
499 *err = -EBADF;
500 if (f.file) {
501 sock = sock_from_file(f.file, err);
502 if (likely(sock)) {
503 *fput_needed = f.flags & FDPUT_FPUT;
504 return sock;
505 }
506 fdput(f);
507 }
508 return NULL;
509}
510
511static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
512 size_t size)
513{
514 ssize_t len;
515 ssize_t used = 0;
516
517 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
518 if (len < 0)
519 return len;
520 used += len;
521 if (buffer) {
522 if (size < used)
523 return -ERANGE;
524 buffer += len;
525 }
526
527 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
528 used += len;
529 if (buffer) {
530 if (size < used)
531 return -ERANGE;
532 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
533 buffer += len;
534 }
535
536 return used;
537}
538
539static int sockfs_setattr(struct dentry *dentry, struct iattr *iattr)
540{
541 int err = simple_setattr(dentry, iattr);
542
543 if (!err && (iattr->ia_valid & ATTR_UID)) {
544 struct socket *sock = SOCKET_I(d_inode(dentry));
545
546 if (sock->sk)
547 sock->sk->sk_uid = iattr->ia_uid;
548 else
549 err = -ENOENT;
550 }
551
552 return err;
553}
554
555static const struct inode_operations sockfs_inode_ops = {
556 .listxattr = sockfs_listxattr,
557 .setattr = sockfs_setattr,
558};
559
560/**
561 * sock_alloc - allocate a socket
562 *
563 * Allocate a new inode and socket object. The two are bound together
564 * and initialised. The socket is then returned. If we are out of inodes
565 * NULL is returned. This functions uses GFP_KERNEL internally.
566 */
567
568struct socket *sock_alloc(void)
569{
570 struct inode *inode;
571 struct socket *sock;
572
573 inode = new_inode_pseudo(sock_mnt->mnt_sb);
574 if (!inode)
575 return NULL;
576
577 sock = SOCKET_I(inode);
578
579 inode->i_ino = get_next_ino();
580 inode->i_mode = S_IFSOCK | S_IRWXUGO;
581 inode->i_uid = current_fsuid();
582 inode->i_gid = current_fsgid();
583 inode->i_op = &sockfs_inode_ops;
584
585 return sock;
586}
587EXPORT_SYMBOL(sock_alloc);
588
589static void __sock_release(struct socket *sock, struct inode *inode)
590{
591 if (sock->ops) {
592 struct module *owner = sock->ops->owner;
593
594 if (inode)
595 inode_lock(inode);
596 sock->ops->release(sock);
597 sock->sk = NULL;
598 if (inode)
599 inode_unlock(inode);
600 sock->ops = NULL;
601 module_put(owner);
602 }
603
604 if (sock->wq.fasync_list)
605 pr_err("%s: fasync list not empty!\n", __func__);
606
607 if (!sock->file) {
608 iput(SOCK_INODE(sock));
609 return;
610 }
611 sock->file = NULL;
612}
613
614/**
615 * sock_release - close a socket
616 * @sock: socket to close
617 *
618 * The socket is released from the protocol stack if it has a release
619 * callback, and the inode is then released if the socket is bound to
620 * an inode not a file.
621 */
622void sock_release(struct socket *sock)
623{
624 __sock_release(sock, NULL);
625}
626EXPORT_SYMBOL(sock_release);
627
628void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
629{
630 u8 flags = *tx_flags;
631
632 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
633 flags |= SKBTX_HW_TSTAMP;
634
635 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
636 flags |= SKBTX_SW_TSTAMP;
637
638 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
639 flags |= SKBTX_SCHED_TSTAMP;
640
641 *tx_flags = flags;
642}
643EXPORT_SYMBOL(__sock_tx_timestamp);
644
645INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
646 size_t));
647INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
648 size_t));
649static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
650{
651 int ret = INDIRECT_CALL_INET(sock->ops->sendmsg, inet6_sendmsg,
652 inet_sendmsg, sock, msg,
653 msg_data_left(msg));
654 BUG_ON(ret == -EIOCBQUEUED);
655 return ret;
656}
657
658/**
659 * sock_sendmsg - send a message through @sock
660 * @sock: socket
661 * @msg: message to send
662 *
663 * Sends @msg through @sock, passing through LSM.
664 * Returns the number of bytes sent, or an error code.
665 */
666int sock_sendmsg(struct socket *sock, struct msghdr *msg)
667{
668 int err = security_socket_sendmsg(sock, msg,
669 msg_data_left(msg));
670
671 return err ?: sock_sendmsg_nosec(sock, msg);
672}
673EXPORT_SYMBOL(sock_sendmsg);
674
675/**
676 * kernel_sendmsg - send a message through @sock (kernel-space)
677 * @sock: socket
678 * @msg: message header
679 * @vec: kernel vec
680 * @num: vec array length
681 * @size: total message data size
682 *
683 * Builds the message data with @vec and sends it through @sock.
684 * Returns the number of bytes sent, or an error code.
685 */
686
687int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
688 struct kvec *vec, size_t num, size_t size)
689{
690 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
691 return sock_sendmsg(sock, msg);
692}
693EXPORT_SYMBOL(kernel_sendmsg);
694
695/**
696 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
697 * @sk: sock
698 * @msg: message header
699 * @vec: output s/g array
700 * @num: output s/g array length
701 * @size: total message data size
702 *
703 * Builds the message data with @vec and sends it through @sock.
704 * Returns the number of bytes sent, or an error code.
705 * Caller must hold @sk.
706 */
707
708int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
709 struct kvec *vec, size_t num, size_t size)
710{
711 struct socket *sock = sk->sk_socket;
712
713 if (!sock->ops->sendmsg_locked)
714 return sock_no_sendmsg_locked(sk, msg, size);
715
716 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
717
718 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
719}
720EXPORT_SYMBOL(kernel_sendmsg_locked);
721
722static bool skb_is_err_queue(const struct sk_buff *skb)
723{
724 /* pkt_type of skbs enqueued on the error queue are set to
725 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
726 * in recvmsg, since skbs received on a local socket will never
727 * have a pkt_type of PACKET_OUTGOING.
728 */
729 return skb->pkt_type == PACKET_OUTGOING;
730}
731
732/* On transmit, software and hardware timestamps are returned independently.
733 * As the two skb clones share the hardware timestamp, which may be updated
734 * before the software timestamp is received, a hardware TX timestamp may be
735 * returned only if there is no software TX timestamp. Ignore false software
736 * timestamps, which may be made in the __sock_recv_timestamp() call when the
737 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
738 * hardware timestamp.
739 */
740static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
741{
742 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
743}
744
745static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
746{
747 struct scm_ts_pktinfo ts_pktinfo;
748 struct net_device *orig_dev;
749
750 if (!skb_mac_header_was_set(skb))
751 return;
752
753 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
754
755 rcu_read_lock();
756 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
757 if (orig_dev)
758 ts_pktinfo.if_index = orig_dev->ifindex;
759 rcu_read_unlock();
760
761 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
762 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
763 sizeof(ts_pktinfo), &ts_pktinfo);
764}
765
766/*
767 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
768 */
769void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
770 struct sk_buff *skb)
771{
772 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
773 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
774 struct scm_timestamping_internal tss;
775
776 int empty = 1, false_tstamp = 0;
777 struct skb_shared_hwtstamps *shhwtstamps =
778 skb_hwtstamps(skb);
779
780 /* Race occurred between timestamp enabling and packet
781 receiving. Fill in the current time for now. */
782 if (need_software_tstamp && skb->tstamp == 0) {
783 __net_timestamp(skb);
784 false_tstamp = 1;
785 }
786
787 if (need_software_tstamp) {
788 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
789 if (new_tstamp) {
790 struct __kernel_sock_timeval tv;
791
792 skb_get_new_timestamp(skb, &tv);
793 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
794 sizeof(tv), &tv);
795 } else {
796 struct __kernel_old_timeval tv;
797
798 skb_get_timestamp(skb, &tv);
799 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
800 sizeof(tv), &tv);
801 }
802 } else {
803 if (new_tstamp) {
804 struct __kernel_timespec ts;
805
806 skb_get_new_timestampns(skb, &ts);
807 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
808 sizeof(ts), &ts);
809 } else {
810 struct __kernel_old_timespec ts;
811
812 skb_get_timestampns(skb, &ts);
813 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
814 sizeof(ts), &ts);
815 }
816 }
817 }
818
819 memset(&tss, 0, sizeof(tss));
820 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
821 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
822 empty = 0;
823 if (shhwtstamps &&
824 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
825 !skb_is_swtx_tstamp(skb, false_tstamp) &&
826 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, tss.ts + 2)) {
827 empty = 0;
828 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
829 !skb_is_err_queue(skb))
830 put_ts_pktinfo(msg, skb);
831 }
832 if (!empty) {
833 if (sock_flag(sk, SOCK_TSTAMP_NEW))
834 put_cmsg_scm_timestamping64(msg, &tss);
835 else
836 put_cmsg_scm_timestamping(msg, &tss);
837
838 if (skb_is_err_queue(skb) && skb->len &&
839 SKB_EXT_ERR(skb)->opt_stats)
840 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
841 skb->len, skb->data);
842 }
843}
844EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
845
846void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
847 struct sk_buff *skb)
848{
849 int ack;
850
851 if (!sock_flag(sk, SOCK_WIFI_STATUS))
852 return;
853 if (!skb->wifi_acked_valid)
854 return;
855
856 ack = skb->wifi_acked;
857
858 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
859}
860EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
861
862static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
863 struct sk_buff *skb)
864{
865 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
866 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
867 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
868}
869
870void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
871 struct sk_buff *skb)
872{
873 sock_recv_timestamp(msg, sk, skb);
874 sock_recv_drops(msg, sk, skb);
875}
876EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
877
878INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
879 size_t, int));
880INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
881 size_t, int));
882static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
883 int flags)
884{
885 return INDIRECT_CALL_INET(sock->ops->recvmsg, inet6_recvmsg,
886 inet_recvmsg, sock, msg, msg_data_left(msg),
887 flags);
888}
889
890/**
891 * sock_recvmsg - receive a message from @sock
892 * @sock: socket
893 * @msg: message to receive
894 * @flags: message flags
895 *
896 * Receives @msg from @sock, passing through LSM. Returns the total number
897 * of bytes received, or an error.
898 */
899int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
900{
901 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
902
903 return err ?: sock_recvmsg_nosec(sock, msg, flags);
904}
905EXPORT_SYMBOL(sock_recvmsg);
906
907/**
908 * kernel_recvmsg - Receive a message from a socket (kernel space)
909 * @sock: The socket to receive the message from
910 * @msg: Received message
911 * @vec: Input s/g array for message data
912 * @num: Size of input s/g array
913 * @size: Number of bytes to read
914 * @flags: Message flags (MSG_DONTWAIT, etc...)
915 *
916 * On return the msg structure contains the scatter/gather array passed in the
917 * vec argument. The array is modified so that it consists of the unfilled
918 * portion of the original array.
919 *
920 * The returned value is the total number of bytes received, or an error.
921 */
922
923int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
924 struct kvec *vec, size_t num, size_t size, int flags)
925{
926 msg->msg_control_is_user = false;
927 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
928 return sock_recvmsg(sock, msg, flags);
929}
930EXPORT_SYMBOL(kernel_recvmsg);
931
932static ssize_t sock_sendpage(struct file *file, struct page *page,
933 int offset, size_t size, loff_t *ppos, int more)
934{
935 struct socket *sock;
936 int flags;
937
938 sock = file->private_data;
939
940 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
941 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
942 flags |= more;
943
944 return kernel_sendpage(sock, page, offset, size, flags);
945}
946
947static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
948 struct pipe_inode_info *pipe, size_t len,
949 unsigned int flags)
950{
951 struct socket *sock = file->private_data;
952
953 if (unlikely(!sock->ops->splice_read))
954 return generic_file_splice_read(file, ppos, pipe, len, flags);
955
956 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
957}
958
959static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
960{
961 struct file *file = iocb->ki_filp;
962 struct socket *sock = file->private_data;
963 struct msghdr msg = {.msg_iter = *to,
964 .msg_iocb = iocb};
965 ssize_t res;
966
967 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
968 msg.msg_flags = MSG_DONTWAIT;
969
970 if (iocb->ki_pos != 0)
971 return -ESPIPE;
972
973 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
974 return 0;
975
976 res = sock_recvmsg(sock, &msg, msg.msg_flags);
977 *to = msg.msg_iter;
978 return res;
979}
980
981static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
982{
983 struct file *file = iocb->ki_filp;
984 struct socket *sock = file->private_data;
985 struct msghdr msg = {.msg_iter = *from,
986 .msg_iocb = iocb};
987 ssize_t res;
988
989 if (iocb->ki_pos != 0)
990 return -ESPIPE;
991
992 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
993 msg.msg_flags = MSG_DONTWAIT;
994
995 if (sock->type == SOCK_SEQPACKET)
996 msg.msg_flags |= MSG_EOR;
997
998 res = sock_sendmsg(sock, &msg);
999 *from = msg.msg_iter;
1000 return res;
1001}
1002
1003/*
1004 * Atomic setting of ioctl hooks to avoid race
1005 * with module unload.
1006 */
1007
1008static DEFINE_MUTEX(br_ioctl_mutex);
1009static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
1010
1011void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
1012{
1013 mutex_lock(&br_ioctl_mutex);
1014 br_ioctl_hook = hook;
1015 mutex_unlock(&br_ioctl_mutex);
1016}
1017EXPORT_SYMBOL(brioctl_set);
1018
1019static DEFINE_MUTEX(vlan_ioctl_mutex);
1020static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1021
1022void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1023{
1024 mutex_lock(&vlan_ioctl_mutex);
1025 vlan_ioctl_hook = hook;
1026 mutex_unlock(&vlan_ioctl_mutex);
1027}
1028EXPORT_SYMBOL(vlan_ioctl_set);
1029
1030static DEFINE_MUTEX(dlci_ioctl_mutex);
1031static int (*dlci_ioctl_hook) (unsigned int, void __user *);
1032
1033void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
1034{
1035 mutex_lock(&dlci_ioctl_mutex);
1036 dlci_ioctl_hook = hook;
1037 mutex_unlock(&dlci_ioctl_mutex);
1038}
1039EXPORT_SYMBOL(dlci_ioctl_set);
1040
1041static long sock_do_ioctl(struct net *net, struct socket *sock,
1042 unsigned int cmd, unsigned long arg)
1043{
1044 int err;
1045 void __user *argp = (void __user *)arg;
1046
1047 err = sock->ops->ioctl(sock, cmd, arg);
1048
1049 /*
1050 * If this ioctl is unknown try to hand it down
1051 * to the NIC driver.
1052 */
1053 if (err != -ENOIOCTLCMD)
1054 return err;
1055
1056 if (cmd == SIOCGIFCONF) {
1057 struct ifconf ifc;
1058 if (copy_from_user(&ifc, argp, sizeof(struct ifconf)))
1059 return -EFAULT;
1060 rtnl_lock();
1061 err = dev_ifconf(net, &ifc, sizeof(struct ifreq));
1062 rtnl_unlock();
1063 if (!err && copy_to_user(argp, &ifc, sizeof(struct ifconf)))
1064 err = -EFAULT;
1065 } else {
1066 struct ifreq ifr;
1067 bool need_copyout;
1068 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1069 return -EFAULT;
1070 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1071 if (!err && need_copyout)
1072 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1073 return -EFAULT;
1074 }
1075 return err;
1076}
1077
1078/*
1079 * With an ioctl, arg may well be a user mode pointer, but we don't know
1080 * what to do with it - that's up to the protocol still.
1081 */
1082
1083/**
1084 * get_net_ns - increment the refcount of the network namespace
1085 * @ns: common namespace (net)
1086 *
1087 * Returns the net's common namespace.
1088 */
1089
1090struct ns_common *get_net_ns(struct ns_common *ns)
1091{
1092 return &get_net(container_of(ns, struct net, ns))->ns;
1093}
1094EXPORT_SYMBOL_GPL(get_net_ns);
1095
1096static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1097{
1098 struct socket *sock;
1099 struct sock *sk;
1100 void __user *argp = (void __user *)arg;
1101 int pid, err;
1102 struct net *net;
1103
1104 sock = file->private_data;
1105 sk = sock->sk;
1106 net = sock_net(sk);
1107 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1108 struct ifreq ifr;
1109 bool need_copyout;
1110 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1111 return -EFAULT;
1112 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1113 if (!err && need_copyout)
1114 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1115 return -EFAULT;
1116 } else
1117#ifdef CONFIG_WEXT_CORE
1118 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1119 err = wext_handle_ioctl(net, cmd, argp);
1120 } else
1121#endif
1122 switch (cmd) {
1123 case FIOSETOWN:
1124 case SIOCSPGRP:
1125 err = -EFAULT;
1126 if (get_user(pid, (int __user *)argp))
1127 break;
1128 err = f_setown(sock->file, pid, 1);
1129 break;
1130 case FIOGETOWN:
1131 case SIOCGPGRP:
1132 err = put_user(f_getown(sock->file),
1133 (int __user *)argp);
1134 break;
1135 case SIOCGIFBR:
1136 case SIOCSIFBR:
1137 case SIOCBRADDBR:
1138 case SIOCBRDELBR:
1139 err = -ENOPKG;
1140 if (!br_ioctl_hook)
1141 request_module("bridge");
1142
1143 mutex_lock(&br_ioctl_mutex);
1144 if (br_ioctl_hook)
1145 err = br_ioctl_hook(net, cmd, argp);
1146 mutex_unlock(&br_ioctl_mutex);
1147 break;
1148 case SIOCGIFVLAN:
1149 case SIOCSIFVLAN:
1150 err = -ENOPKG;
1151 if (!vlan_ioctl_hook)
1152 request_module("8021q");
1153
1154 mutex_lock(&vlan_ioctl_mutex);
1155 if (vlan_ioctl_hook)
1156 err = vlan_ioctl_hook(net, argp);
1157 mutex_unlock(&vlan_ioctl_mutex);
1158 break;
1159 case SIOCADDDLCI:
1160 case SIOCDELDLCI:
1161 err = -ENOPKG;
1162 if (!dlci_ioctl_hook)
1163 request_module("dlci");
1164
1165 mutex_lock(&dlci_ioctl_mutex);
1166 if (dlci_ioctl_hook)
1167 err = dlci_ioctl_hook(cmd, argp);
1168 mutex_unlock(&dlci_ioctl_mutex);
1169 break;
1170 case SIOCGSKNS:
1171 err = -EPERM;
1172 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1173 break;
1174
1175 err = open_related_ns(&net->ns, get_net_ns);
1176 break;
1177 case SIOCGSTAMP_OLD:
1178 case SIOCGSTAMPNS_OLD:
1179 if (!sock->ops->gettstamp) {
1180 err = -ENOIOCTLCMD;
1181 break;
1182 }
1183 err = sock->ops->gettstamp(sock, argp,
1184 cmd == SIOCGSTAMP_OLD,
1185 !IS_ENABLED(CONFIG_64BIT));
1186 break;
1187 case SIOCGSTAMP_NEW:
1188 case SIOCGSTAMPNS_NEW:
1189 if (!sock->ops->gettstamp) {
1190 err = -ENOIOCTLCMD;
1191 break;
1192 }
1193 err = sock->ops->gettstamp(sock, argp,
1194 cmd == SIOCGSTAMP_NEW,
1195 false);
1196 break;
1197 default:
1198 err = sock_do_ioctl(net, sock, cmd, arg);
1199 break;
1200 }
1201 return err;
1202}
1203
1204/**
1205 * sock_create_lite - creates a socket
1206 * @family: protocol family (AF_INET, ...)
1207 * @type: communication type (SOCK_STREAM, ...)
1208 * @protocol: protocol (0, ...)
1209 * @res: new socket
1210 *
1211 * Creates a new socket and assigns it to @res, passing through LSM.
1212 * The new socket initialization is not complete, see kernel_accept().
1213 * Returns 0 or an error. On failure @res is set to %NULL.
1214 * This function internally uses GFP_KERNEL.
1215 */
1216
1217int sock_create_lite(int family, int type, int protocol, struct socket **res)
1218{
1219 int err;
1220 struct socket *sock = NULL;
1221
1222 err = security_socket_create(family, type, protocol, 1);
1223 if (err)
1224 goto out;
1225
1226 sock = sock_alloc();
1227 if (!sock) {
1228 err = -ENOMEM;
1229 goto out;
1230 }
1231
1232 sock->type = type;
1233 err = security_socket_post_create(sock, family, type, protocol, 1);
1234 if (err)
1235 goto out_release;
1236
1237out:
1238 *res = sock;
1239 return err;
1240out_release:
1241 sock_release(sock);
1242 sock = NULL;
1243 goto out;
1244}
1245EXPORT_SYMBOL(sock_create_lite);
1246
1247/* No kernel lock held - perfect */
1248static __poll_t sock_poll(struct file *file, poll_table *wait)
1249{
1250 struct socket *sock = file->private_data;
1251 __poll_t events = poll_requested_events(wait), flag = 0;
1252
1253 if (!sock->ops->poll)
1254 return 0;
1255
1256 if (sk_can_busy_loop(sock->sk)) {
1257 /* poll once if requested by the syscall */
1258 if (events & POLL_BUSY_LOOP)
1259 sk_busy_loop(sock->sk, 1);
1260
1261 /* if this socket can poll_ll, tell the system call */
1262 flag = POLL_BUSY_LOOP;
1263 }
1264
1265 return sock->ops->poll(file, sock, wait) | flag;
1266}
1267
1268static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1269{
1270 struct socket *sock = file->private_data;
1271
1272 return sock->ops->mmap(file, sock, vma);
1273}
1274
1275static int sock_close(struct inode *inode, struct file *filp)
1276{
1277 __sock_release(SOCKET_I(inode), inode);
1278 return 0;
1279}
1280
1281/*
1282 * Update the socket async list
1283 *
1284 * Fasync_list locking strategy.
1285 *
1286 * 1. fasync_list is modified only under process context socket lock
1287 * i.e. under semaphore.
1288 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1289 * or under socket lock
1290 */
1291
1292static int sock_fasync(int fd, struct file *filp, int on)
1293{
1294 struct socket *sock = filp->private_data;
1295 struct sock *sk = sock->sk;
1296 struct socket_wq *wq = &sock->wq;
1297
1298 if (sk == NULL)
1299 return -EINVAL;
1300
1301 lock_sock(sk);
1302 fasync_helper(fd, filp, on, &wq->fasync_list);
1303
1304 if (!wq->fasync_list)
1305 sock_reset_flag(sk, SOCK_FASYNC);
1306 else
1307 sock_set_flag(sk, SOCK_FASYNC);
1308
1309 release_sock(sk);
1310 return 0;
1311}
1312
1313/* This function may be called only under rcu_lock */
1314
1315int sock_wake_async(struct socket_wq *wq, int how, int band)
1316{
1317 if (!wq || !wq->fasync_list)
1318 return -1;
1319
1320 switch (how) {
1321 case SOCK_WAKE_WAITD:
1322 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1323 break;
1324 goto call_kill;
1325 case SOCK_WAKE_SPACE:
1326 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1327 break;
1328 fallthrough;
1329 case SOCK_WAKE_IO:
1330call_kill:
1331 kill_fasync(&wq->fasync_list, SIGIO, band);
1332 break;
1333 case SOCK_WAKE_URG:
1334 kill_fasync(&wq->fasync_list, SIGURG, band);
1335 }
1336
1337 return 0;
1338}
1339EXPORT_SYMBOL(sock_wake_async);
1340
1341/**
1342 * __sock_create - creates a socket
1343 * @net: net namespace
1344 * @family: protocol family (AF_INET, ...)
1345 * @type: communication type (SOCK_STREAM, ...)
1346 * @protocol: protocol (0, ...)
1347 * @res: new socket
1348 * @kern: boolean for kernel space sockets
1349 *
1350 * Creates a new socket and assigns it to @res, passing through LSM.
1351 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1352 * be set to true if the socket resides in kernel space.
1353 * This function internally uses GFP_KERNEL.
1354 */
1355
1356int __sock_create(struct net *net, int family, int type, int protocol,
1357 struct socket **res, int kern)
1358{
1359 int err;
1360 struct socket *sock;
1361 const struct net_proto_family *pf;
1362
1363 /*
1364 * Check protocol is in range
1365 */
1366 if (family < 0 || family >= NPROTO)
1367 return -EAFNOSUPPORT;
1368 if (type < 0 || type >= SOCK_MAX)
1369 return -EINVAL;
1370
1371 /* Compatibility.
1372
1373 This uglymoron is moved from INET layer to here to avoid
1374 deadlock in module load.
1375 */
1376 if (family == PF_INET && type == SOCK_PACKET) {
1377 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1378 current->comm);
1379 family = PF_PACKET;
1380 }
1381
1382 err = security_socket_create(family, type, protocol, kern);
1383 if (err)
1384 return err;
1385
1386 /*
1387 * Allocate the socket and allow the family to set things up. if
1388 * the protocol is 0, the family is instructed to select an appropriate
1389 * default.
1390 */
1391 sock = sock_alloc();
1392 if (!sock) {
1393 net_warn_ratelimited("socket: no more sockets\n");
1394 return -ENFILE; /* Not exactly a match, but its the
1395 closest posix thing */
1396 }
1397
1398 sock->type = type;
1399
1400#ifdef CONFIG_MODULES
1401 /* Attempt to load a protocol module if the find failed.
1402 *
1403 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1404 * requested real, full-featured networking support upon configuration.
1405 * Otherwise module support will break!
1406 */
1407 if (rcu_access_pointer(net_families[family]) == NULL)
1408 request_module("net-pf-%d", family);
1409#endif
1410
1411 rcu_read_lock();
1412 pf = rcu_dereference(net_families[family]);
1413 err = -EAFNOSUPPORT;
1414 if (!pf)
1415 goto out_release;
1416
1417 /*
1418 * We will call the ->create function, that possibly is in a loadable
1419 * module, so we have to bump that loadable module refcnt first.
1420 */
1421 if (!try_module_get(pf->owner))
1422 goto out_release;
1423
1424 /* Now protected by module ref count */
1425 rcu_read_unlock();
1426
1427 err = pf->create(net, sock, protocol, kern);
1428 if (err < 0)
1429 goto out_module_put;
1430
1431 /*
1432 * Now to bump the refcnt of the [loadable] module that owns this
1433 * socket at sock_release time we decrement its refcnt.
1434 */
1435 if (!try_module_get(sock->ops->owner))
1436 goto out_module_busy;
1437
1438 /*
1439 * Now that we're done with the ->create function, the [loadable]
1440 * module can have its refcnt decremented
1441 */
1442 module_put(pf->owner);
1443 err = security_socket_post_create(sock, family, type, protocol, kern);
1444 if (err)
1445 goto out_sock_release;
1446 *res = sock;
1447
1448 return 0;
1449
1450out_module_busy:
1451 err = -EAFNOSUPPORT;
1452out_module_put:
1453 sock->ops = NULL;
1454 module_put(pf->owner);
1455out_sock_release:
1456 sock_release(sock);
1457 return err;
1458
1459out_release:
1460 rcu_read_unlock();
1461 goto out_sock_release;
1462}
1463EXPORT_SYMBOL(__sock_create);
1464
1465/**
1466 * sock_create - creates a socket
1467 * @family: protocol family (AF_INET, ...)
1468 * @type: communication type (SOCK_STREAM, ...)
1469 * @protocol: protocol (0, ...)
1470 * @res: new socket
1471 *
1472 * A wrapper around __sock_create().
1473 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1474 */
1475
1476int sock_create(int family, int type, int protocol, struct socket **res)
1477{
1478 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1479}
1480EXPORT_SYMBOL(sock_create);
1481
1482/**
1483 * sock_create_kern - creates a socket (kernel space)
1484 * @net: net namespace
1485 * @family: protocol family (AF_INET, ...)
1486 * @type: communication type (SOCK_STREAM, ...)
1487 * @protocol: protocol (0, ...)
1488 * @res: new socket
1489 *
1490 * A wrapper around __sock_create().
1491 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1492 */
1493
1494int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1495{
1496 return __sock_create(net, family, type, protocol, res, 1);
1497}
1498EXPORT_SYMBOL(sock_create_kern);
1499
1500int __sys_socket(int family, int type, int protocol)
1501{
1502 int retval;
1503 struct socket *sock;
1504 int flags;
1505
1506 /* Check the SOCK_* constants for consistency. */
1507 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1508 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1509 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1510 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1511
1512 flags = type & ~SOCK_TYPE_MASK;
1513 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1514 return -EINVAL;
1515 type &= SOCK_TYPE_MASK;
1516
1517 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1518 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1519
1520 retval = sock_create(family, type, protocol, &sock);
1521 if (retval < 0)
1522 return retval;
1523
1524 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1525}
1526
1527SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1528{
1529 return __sys_socket(family, type, protocol);
1530}
1531
1532/*
1533 * Create a pair of connected sockets.
1534 */
1535
1536int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1537{
1538 struct socket *sock1, *sock2;
1539 int fd1, fd2, err;
1540 struct file *newfile1, *newfile2;
1541 int flags;
1542
1543 flags = type & ~SOCK_TYPE_MASK;
1544 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1545 return -EINVAL;
1546 type &= SOCK_TYPE_MASK;
1547
1548 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1549 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1550
1551 /*
1552 * reserve descriptors and make sure we won't fail
1553 * to return them to userland.
1554 */
1555 fd1 = get_unused_fd_flags(flags);
1556 if (unlikely(fd1 < 0))
1557 return fd1;
1558
1559 fd2 = get_unused_fd_flags(flags);
1560 if (unlikely(fd2 < 0)) {
1561 put_unused_fd(fd1);
1562 return fd2;
1563 }
1564
1565 err = put_user(fd1, &usockvec[0]);
1566 if (err)
1567 goto out;
1568
1569 err = put_user(fd2, &usockvec[1]);
1570 if (err)
1571 goto out;
1572
1573 /*
1574 * Obtain the first socket and check if the underlying protocol
1575 * supports the socketpair call.
1576 */
1577
1578 err = sock_create(family, type, protocol, &sock1);
1579 if (unlikely(err < 0))
1580 goto out;
1581
1582 err = sock_create(family, type, protocol, &sock2);
1583 if (unlikely(err < 0)) {
1584 sock_release(sock1);
1585 goto out;
1586 }
1587
1588 err = security_socket_socketpair(sock1, sock2);
1589 if (unlikely(err)) {
1590 sock_release(sock2);
1591 sock_release(sock1);
1592 goto out;
1593 }
1594
1595 err = sock1->ops->socketpair(sock1, sock2);
1596 if (unlikely(err < 0)) {
1597 sock_release(sock2);
1598 sock_release(sock1);
1599 goto out;
1600 }
1601
1602 newfile1 = sock_alloc_file(sock1, flags, NULL);
1603 if (IS_ERR(newfile1)) {
1604 err = PTR_ERR(newfile1);
1605 sock_release(sock2);
1606 goto out;
1607 }
1608
1609 newfile2 = sock_alloc_file(sock2, flags, NULL);
1610 if (IS_ERR(newfile2)) {
1611 err = PTR_ERR(newfile2);
1612 fput(newfile1);
1613 goto out;
1614 }
1615
1616 audit_fd_pair(fd1, fd2);
1617
1618 fd_install(fd1, newfile1);
1619 fd_install(fd2, newfile2);
1620 return 0;
1621
1622out:
1623 put_unused_fd(fd2);
1624 put_unused_fd(fd1);
1625 return err;
1626}
1627
1628SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1629 int __user *, usockvec)
1630{
1631 return __sys_socketpair(family, type, protocol, usockvec);
1632}
1633
1634/*
1635 * Bind a name to a socket. Nothing much to do here since it's
1636 * the protocol's responsibility to handle the local address.
1637 *
1638 * We move the socket address to kernel space before we call
1639 * the protocol layer (having also checked the address is ok).
1640 */
1641
1642int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1643{
1644 struct socket *sock;
1645 struct sockaddr_storage address;
1646 int err, fput_needed;
1647
1648 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1649 if (sock) {
1650 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1651 if (!err) {
1652 err = security_socket_bind(sock,
1653 (struct sockaddr *)&address,
1654 addrlen);
1655 if (!err)
1656 err = sock->ops->bind(sock,
1657 (struct sockaddr *)
1658 &address, addrlen);
1659 }
1660 fput_light(sock->file, fput_needed);
1661 }
1662 return err;
1663}
1664
1665SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1666{
1667 return __sys_bind(fd, umyaddr, addrlen);
1668}
1669
1670/*
1671 * Perform a listen. Basically, we allow the protocol to do anything
1672 * necessary for a listen, and if that works, we mark the socket as
1673 * ready for listening.
1674 */
1675
1676int __sys_listen(int fd, int backlog)
1677{
1678 struct socket *sock;
1679 int err, fput_needed;
1680 int somaxconn;
1681
1682 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1683 if (sock) {
1684 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1685 if ((unsigned int)backlog > somaxconn)
1686 backlog = somaxconn;
1687
1688 err = security_socket_listen(sock, backlog);
1689 if (!err)
1690 err = sock->ops->listen(sock, backlog);
1691
1692 fput_light(sock->file, fput_needed);
1693 }
1694 return err;
1695}
1696
1697SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1698{
1699 return __sys_listen(fd, backlog);
1700}
1701
1702int __sys_accept4_file(struct file *file, unsigned file_flags,
1703 struct sockaddr __user *upeer_sockaddr,
1704 int __user *upeer_addrlen, int flags,
1705 unsigned long nofile)
1706{
1707 struct socket *sock, *newsock;
1708 struct file *newfile;
1709 int err, len, newfd;
1710 struct sockaddr_storage address;
1711
1712 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1713 return -EINVAL;
1714
1715 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1716 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1717
1718 sock = sock_from_file(file, &err);
1719 if (!sock)
1720 goto out;
1721
1722 err = -ENFILE;
1723 newsock = sock_alloc();
1724 if (!newsock)
1725 goto out;
1726
1727 newsock->type = sock->type;
1728 newsock->ops = sock->ops;
1729
1730 /*
1731 * We don't need try_module_get here, as the listening socket (sock)
1732 * has the protocol module (sock->ops->owner) held.
1733 */
1734 __module_get(newsock->ops->owner);
1735
1736 newfd = __get_unused_fd_flags(flags, nofile);
1737 if (unlikely(newfd < 0)) {
1738 err = newfd;
1739 sock_release(newsock);
1740 goto out;
1741 }
1742 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1743 if (IS_ERR(newfile)) {
1744 err = PTR_ERR(newfile);
1745 put_unused_fd(newfd);
1746 goto out;
1747 }
1748
1749 err = security_socket_accept(sock, newsock);
1750 if (err)
1751 goto out_fd;
1752
1753 err = sock->ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1754 false);
1755 if (err < 0)
1756 goto out_fd;
1757
1758 if (upeer_sockaddr) {
1759 len = newsock->ops->getname(newsock,
1760 (struct sockaddr *)&address, 2);
1761 if (len < 0) {
1762 err = -ECONNABORTED;
1763 goto out_fd;
1764 }
1765 err = move_addr_to_user(&address,
1766 len, upeer_sockaddr, upeer_addrlen);
1767 if (err < 0)
1768 goto out_fd;
1769 }
1770
1771 /* File flags are not inherited via accept() unlike another OSes. */
1772
1773 fd_install(newfd, newfile);
1774 err = newfd;
1775out:
1776 return err;
1777out_fd:
1778 fput(newfile);
1779 put_unused_fd(newfd);
1780 goto out;
1781
1782}
1783
1784/*
1785 * For accept, we attempt to create a new socket, set up the link
1786 * with the client, wake up the client, then return the new
1787 * connected fd. We collect the address of the connector in kernel
1788 * space and move it to user at the very end. This is unclean because
1789 * we open the socket then return an error.
1790 *
1791 * 1003.1g adds the ability to recvmsg() to query connection pending
1792 * status to recvmsg. We need to add that support in a way thats
1793 * clean when we restructure accept also.
1794 */
1795
1796int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1797 int __user *upeer_addrlen, int flags)
1798{
1799 int ret = -EBADF;
1800 struct fd f;
1801
1802 f = fdget(fd);
1803 if (f.file) {
1804 ret = __sys_accept4_file(f.file, 0, upeer_sockaddr,
1805 upeer_addrlen, flags,
1806 rlimit(RLIMIT_NOFILE));
1807 fdput(f);
1808 }
1809
1810 return ret;
1811}
1812
1813SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1814 int __user *, upeer_addrlen, int, flags)
1815{
1816 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1817}
1818
1819SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1820 int __user *, upeer_addrlen)
1821{
1822 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1823}
1824
1825/*
1826 * Attempt to connect to a socket with the server address. The address
1827 * is in user space so we verify it is OK and move it to kernel space.
1828 *
1829 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1830 * break bindings
1831 *
1832 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1833 * other SEQPACKET protocols that take time to connect() as it doesn't
1834 * include the -EINPROGRESS status for such sockets.
1835 */
1836
1837int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
1838 int addrlen, int file_flags)
1839{
1840 struct socket *sock;
1841 int err;
1842
1843 sock = sock_from_file(file, &err);
1844 if (!sock)
1845 goto out;
1846
1847 err =
1848 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1849 if (err)
1850 goto out;
1851
1852 err = sock->ops->connect(sock, (struct sockaddr *)address, addrlen,
1853 sock->file->f_flags | file_flags);
1854out:
1855 return err;
1856}
1857
1858int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1859{
1860 int ret = -EBADF;
1861 struct fd f;
1862
1863 f = fdget(fd);
1864 if (f.file) {
1865 struct sockaddr_storage address;
1866
1867 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
1868 if (!ret)
1869 ret = __sys_connect_file(f.file, &address, addrlen, 0);
1870 fdput(f);
1871 }
1872
1873 return ret;
1874}
1875
1876SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1877 int, addrlen)
1878{
1879 return __sys_connect(fd, uservaddr, addrlen);
1880}
1881
1882/*
1883 * Get the local address ('name') of a socket object. Move the obtained
1884 * name to user space.
1885 */
1886
1887int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1888 int __user *usockaddr_len)
1889{
1890 struct socket *sock;
1891 struct sockaddr_storage address;
1892 int err, fput_needed;
1893
1894 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1895 if (!sock)
1896 goto out;
1897
1898 err = security_socket_getsockname(sock);
1899 if (err)
1900 goto out_put;
1901
1902 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1903 if (err < 0)
1904 goto out_put;
1905 /* "err" is actually length in this case */
1906 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1907
1908out_put:
1909 fput_light(sock->file, fput_needed);
1910out:
1911 return err;
1912}
1913
1914SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1915 int __user *, usockaddr_len)
1916{
1917 return __sys_getsockname(fd, usockaddr, usockaddr_len);
1918}
1919
1920/*
1921 * Get the remote address ('name') of a socket object. Move the obtained
1922 * name to user space.
1923 */
1924
1925int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1926 int __user *usockaddr_len)
1927{
1928 struct socket *sock;
1929 struct sockaddr_storage address;
1930 int err, fput_needed;
1931
1932 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1933 if (sock != NULL) {
1934 err = security_socket_getpeername(sock);
1935 if (err) {
1936 fput_light(sock->file, fput_needed);
1937 return err;
1938 }
1939
1940 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
1941 if (err >= 0)
1942 /* "err" is actually length in this case */
1943 err = move_addr_to_user(&address, err, usockaddr,
1944 usockaddr_len);
1945 fput_light(sock->file, fput_needed);
1946 }
1947 return err;
1948}
1949
1950SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1951 int __user *, usockaddr_len)
1952{
1953 return __sys_getpeername(fd, usockaddr, usockaddr_len);
1954}
1955
1956/*
1957 * Send a datagram to a given address. We move the address into kernel
1958 * space and check the user space data area is readable before invoking
1959 * the protocol.
1960 */
1961int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
1962 struct sockaddr __user *addr, int addr_len)
1963{
1964 struct socket *sock;
1965 struct sockaddr_storage address;
1966 int err;
1967 struct msghdr msg;
1968 struct iovec iov;
1969 int fput_needed;
1970
1971 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
1972 if (unlikely(err))
1973 return err;
1974 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1975 if (!sock)
1976 goto out;
1977
1978 msg.msg_name = NULL;
1979 msg.msg_control = NULL;
1980 msg.msg_controllen = 0;
1981 msg.msg_namelen = 0;
1982 if (addr) {
1983 err = move_addr_to_kernel(addr, addr_len, &address);
1984 if (err < 0)
1985 goto out_put;
1986 msg.msg_name = (struct sockaddr *)&address;
1987 msg.msg_namelen = addr_len;
1988 }
1989 if (sock->file->f_flags & O_NONBLOCK)
1990 flags |= MSG_DONTWAIT;
1991 msg.msg_flags = flags;
1992 err = sock_sendmsg(sock, &msg);
1993
1994out_put:
1995 fput_light(sock->file, fput_needed);
1996out:
1997 return err;
1998}
1999
2000SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2001 unsigned int, flags, struct sockaddr __user *, addr,
2002 int, addr_len)
2003{
2004 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2005}
2006
2007/*
2008 * Send a datagram down a socket.
2009 */
2010
2011SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2012 unsigned int, flags)
2013{
2014 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2015}
2016
2017/*
2018 * Receive a frame from the socket and optionally record the address of the
2019 * sender. We verify the buffers are writable and if needed move the
2020 * sender address from kernel to user space.
2021 */
2022int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2023 struct sockaddr __user *addr, int __user *addr_len)
2024{
2025 struct socket *sock;
2026 struct iovec iov;
2027 struct msghdr msg;
2028 struct sockaddr_storage address;
2029 int err, err2;
2030 int fput_needed;
2031
2032 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
2033 if (unlikely(err))
2034 return err;
2035 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2036 if (!sock)
2037 goto out;
2038
2039 msg.msg_control = NULL;
2040 msg.msg_controllen = 0;
2041 /* Save some cycles and don't copy the address if not needed */
2042 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
2043 /* We assume all kernel code knows the size of sockaddr_storage */
2044 msg.msg_namelen = 0;
2045 msg.msg_iocb = NULL;
2046 msg.msg_flags = 0;
2047 if (sock->file->f_flags & O_NONBLOCK)
2048 flags |= MSG_DONTWAIT;
2049 err = sock_recvmsg(sock, &msg, flags);
2050
2051 if (err >= 0 && addr != NULL) {
2052 err2 = move_addr_to_user(&address,
2053 msg.msg_namelen, addr, addr_len);
2054 if (err2 < 0)
2055 err = err2;
2056 }
2057
2058 fput_light(sock->file, fput_needed);
2059out:
2060 return err;
2061}
2062
2063SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2064 unsigned int, flags, struct sockaddr __user *, addr,
2065 int __user *, addr_len)
2066{
2067 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2068}
2069
2070/*
2071 * Receive a datagram from a socket.
2072 */
2073
2074SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2075 unsigned int, flags)
2076{
2077 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2078}
2079
2080static bool sock_use_custom_sol_socket(const struct socket *sock)
2081{
2082 const struct sock *sk = sock->sk;
2083
2084 /* Use sock->ops->setsockopt() for MPTCP */
2085 return IS_ENABLED(CONFIG_MPTCP) &&
2086 sk->sk_protocol == IPPROTO_MPTCP &&
2087 sk->sk_type == SOCK_STREAM &&
2088 (sk->sk_family == AF_INET || sk->sk_family == AF_INET6);
2089}
2090
2091/*
2092 * Set a socket option. Because we don't know the option lengths we have
2093 * to pass the user mode parameter for the protocols to sort out.
2094 */
2095int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2096 int optlen)
2097{
2098 sockptr_t optval = USER_SOCKPTR(user_optval);
2099 char *kernel_optval = NULL;
2100 int err, fput_needed;
2101 struct socket *sock;
2102
2103 if (optlen < 0)
2104 return -EINVAL;
2105
2106 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2107 if (!sock)
2108 return err;
2109
2110 err = security_socket_setsockopt(sock, level, optname);
2111 if (err)
2112 goto out_put;
2113
2114 if (!in_compat_syscall())
2115 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2116 user_optval, &optlen,
2117 &kernel_optval);
2118 if (err < 0)
2119 goto out_put;
2120 if (err > 0) {
2121 err = 0;
2122 goto out_put;
2123 }
2124
2125 if (kernel_optval)
2126 optval = KERNEL_SOCKPTR(kernel_optval);
2127 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2128 err = sock_setsockopt(sock, level, optname, optval, optlen);
2129 else if (unlikely(!sock->ops->setsockopt))
2130 err = -EOPNOTSUPP;
2131 else
2132 err = sock->ops->setsockopt(sock, level, optname, optval,
2133 optlen);
2134 kfree(kernel_optval);
2135out_put:
2136 fput_light(sock->file, fput_needed);
2137 return err;
2138}
2139
2140SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2141 char __user *, optval, int, optlen)
2142{
2143 return __sys_setsockopt(fd, level, optname, optval, optlen);
2144}
2145
2146/*
2147 * Get a socket option. Because we don't know the option lengths we have
2148 * to pass a user mode parameter for the protocols to sort out.
2149 */
2150int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2151 int __user *optlen)
2152{
2153 int err, fput_needed;
2154 struct socket *sock;
2155 int max_optlen;
2156
2157 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2158 if (!sock)
2159 return err;
2160
2161 err = security_socket_getsockopt(sock, level, optname);
2162 if (err)
2163 goto out_put;
2164
2165 if (!in_compat_syscall())
2166 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2167
2168 if (level == SOL_SOCKET)
2169 err = sock_getsockopt(sock, level, optname, optval, optlen);
2170 else if (unlikely(!sock->ops->getsockopt))
2171 err = -EOPNOTSUPP;
2172 else
2173 err = sock->ops->getsockopt(sock, level, optname, optval,
2174 optlen);
2175
2176 if (!in_compat_syscall())
2177 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2178 optval, optlen, max_optlen,
2179 err);
2180out_put:
2181 fput_light(sock->file, fput_needed);
2182 return err;
2183}
2184
2185SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2186 char __user *, optval, int __user *, optlen)
2187{
2188 return __sys_getsockopt(fd, level, optname, optval, optlen);
2189}
2190
2191/*
2192 * Shutdown a socket.
2193 */
2194
2195int __sys_shutdown(int fd, int how)
2196{
2197 int err, fput_needed;
2198 struct socket *sock;
2199
2200 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2201 if (sock != NULL) {
2202 err = security_socket_shutdown(sock, how);
2203 if (!err)
2204 err = sock->ops->shutdown(sock, how);
2205 fput_light(sock->file, fput_needed);
2206 }
2207 return err;
2208}
2209
2210SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2211{
2212 return __sys_shutdown(fd, how);
2213}
2214
2215/* A couple of helpful macros for getting the address of the 32/64 bit
2216 * fields which are the same type (int / unsigned) on our platforms.
2217 */
2218#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2219#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2220#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2221
2222struct used_address {
2223 struct sockaddr_storage name;
2224 unsigned int name_len;
2225};
2226
2227int __copy_msghdr_from_user(struct msghdr *kmsg,
2228 struct user_msghdr __user *umsg,
2229 struct sockaddr __user **save_addr,
2230 struct iovec __user **uiov, size_t *nsegs)
2231{
2232 struct user_msghdr msg;
2233 ssize_t err;
2234
2235 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2236 return -EFAULT;
2237
2238 kmsg->msg_control_is_user = true;
2239 kmsg->msg_control_user = msg.msg_control;
2240 kmsg->msg_controllen = msg.msg_controllen;
2241 kmsg->msg_flags = msg.msg_flags;
2242
2243 kmsg->msg_namelen = msg.msg_namelen;
2244 if (!msg.msg_name)
2245 kmsg->msg_namelen = 0;
2246
2247 if (kmsg->msg_namelen < 0)
2248 return -EINVAL;
2249
2250 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2251 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2252
2253 if (save_addr)
2254 *save_addr = msg.msg_name;
2255
2256 if (msg.msg_name && kmsg->msg_namelen) {
2257 if (!save_addr) {
2258 err = move_addr_to_kernel(msg.msg_name,
2259 kmsg->msg_namelen,
2260 kmsg->msg_name);
2261 if (err < 0)
2262 return err;
2263 }
2264 } else {
2265 kmsg->msg_name = NULL;
2266 kmsg->msg_namelen = 0;
2267 }
2268
2269 if (msg.msg_iovlen > UIO_MAXIOV)
2270 return -EMSGSIZE;
2271
2272 kmsg->msg_iocb = NULL;
2273 *uiov = msg.msg_iov;
2274 *nsegs = msg.msg_iovlen;
2275 return 0;
2276}
2277
2278static int copy_msghdr_from_user(struct msghdr *kmsg,
2279 struct user_msghdr __user *umsg,
2280 struct sockaddr __user **save_addr,
2281 struct iovec **iov)
2282{
2283 struct user_msghdr msg;
2284 ssize_t err;
2285
2286 err = __copy_msghdr_from_user(kmsg, umsg, save_addr, &msg.msg_iov,
2287 &msg.msg_iovlen);
2288 if (err)
2289 return err;
2290
2291 err = import_iovec(save_addr ? READ : WRITE,
2292 msg.msg_iov, msg.msg_iovlen,
2293 UIO_FASTIOV, iov, &kmsg->msg_iter);
2294 return err < 0 ? err : 0;
2295}
2296
2297static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2298 unsigned int flags, struct used_address *used_address,
2299 unsigned int allowed_msghdr_flags)
2300{
2301 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2302 __aligned(sizeof(__kernel_size_t));
2303 /* 20 is size of ipv6_pktinfo */
2304 unsigned char *ctl_buf = ctl;
2305 int ctl_len;
2306 ssize_t err;
2307
2308 err = -ENOBUFS;
2309
2310 if (msg_sys->msg_controllen > INT_MAX)
2311 goto out;
2312 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2313 ctl_len = msg_sys->msg_controllen;
2314 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2315 err =
2316 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2317 sizeof(ctl));
2318 if (err)
2319 goto out;
2320 ctl_buf = msg_sys->msg_control;
2321 ctl_len = msg_sys->msg_controllen;
2322 } else if (ctl_len) {
2323 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2324 CMSG_ALIGN(sizeof(struct cmsghdr)));
2325 if (ctl_len > sizeof(ctl)) {
2326 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2327 if (ctl_buf == NULL)
2328 goto out;
2329 }
2330 err = -EFAULT;
2331 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2332 goto out_freectl;
2333 msg_sys->msg_control = ctl_buf;
2334 msg_sys->msg_control_is_user = false;
2335 }
2336 msg_sys->msg_flags = flags;
2337
2338 if (sock->file->f_flags & O_NONBLOCK)
2339 msg_sys->msg_flags |= MSG_DONTWAIT;
2340 /*
2341 * If this is sendmmsg() and current destination address is same as
2342 * previously succeeded address, omit asking LSM's decision.
2343 * used_address->name_len is initialized to UINT_MAX so that the first
2344 * destination address never matches.
2345 */
2346 if (used_address && msg_sys->msg_name &&
2347 used_address->name_len == msg_sys->msg_namelen &&
2348 !memcmp(&used_address->name, msg_sys->msg_name,
2349 used_address->name_len)) {
2350 err = sock_sendmsg_nosec(sock, msg_sys);
2351 goto out_freectl;
2352 }
2353 err = sock_sendmsg(sock, msg_sys);
2354 /*
2355 * If this is sendmmsg() and sending to current destination address was
2356 * successful, remember it.
2357 */
2358 if (used_address && err >= 0) {
2359 used_address->name_len = msg_sys->msg_namelen;
2360 if (msg_sys->msg_name)
2361 memcpy(&used_address->name, msg_sys->msg_name,
2362 used_address->name_len);
2363 }
2364
2365out_freectl:
2366 if (ctl_buf != ctl)
2367 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2368out:
2369 return err;
2370}
2371
2372int sendmsg_copy_msghdr(struct msghdr *msg,
2373 struct user_msghdr __user *umsg, unsigned flags,
2374 struct iovec **iov)
2375{
2376 int err;
2377
2378 if (flags & MSG_CMSG_COMPAT) {
2379 struct compat_msghdr __user *msg_compat;
2380
2381 msg_compat = (struct compat_msghdr __user *) umsg;
2382 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2383 } else {
2384 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2385 }
2386 if (err < 0)
2387 return err;
2388
2389 return 0;
2390}
2391
2392static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2393 struct msghdr *msg_sys, unsigned int flags,
2394 struct used_address *used_address,
2395 unsigned int allowed_msghdr_flags)
2396{
2397 struct sockaddr_storage address;
2398 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2399 ssize_t err;
2400
2401 msg_sys->msg_name = &address;
2402
2403 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2404 if (err < 0)
2405 return err;
2406
2407 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2408 allowed_msghdr_flags);
2409 kfree(iov);
2410 return err;
2411}
2412
2413/*
2414 * BSD sendmsg interface
2415 */
2416long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2417 unsigned int flags)
2418{
2419 /* disallow ancillary data requests from this path */
2420 if (msg->msg_control || msg->msg_controllen)
2421 return -EINVAL;
2422
2423 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2424}
2425
2426long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2427 bool forbid_cmsg_compat)
2428{
2429 int fput_needed, err;
2430 struct msghdr msg_sys;
2431 struct socket *sock;
2432
2433 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2434 return -EINVAL;
2435
2436 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2437 if (!sock)
2438 goto out;
2439
2440 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2441
2442 fput_light(sock->file, fput_needed);
2443out:
2444 return err;
2445}
2446
2447SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2448{
2449 return __sys_sendmsg(fd, msg, flags, true);
2450}
2451
2452/*
2453 * Linux sendmmsg interface
2454 */
2455
2456int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2457 unsigned int flags, bool forbid_cmsg_compat)
2458{
2459 int fput_needed, err, datagrams;
2460 struct socket *sock;
2461 struct mmsghdr __user *entry;
2462 struct compat_mmsghdr __user *compat_entry;
2463 struct msghdr msg_sys;
2464 struct used_address used_address;
2465 unsigned int oflags = flags;
2466
2467 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2468 return -EINVAL;
2469
2470 if (vlen > UIO_MAXIOV)
2471 vlen = UIO_MAXIOV;
2472
2473 datagrams = 0;
2474
2475 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2476 if (!sock)
2477 return err;
2478
2479 used_address.name_len = UINT_MAX;
2480 entry = mmsg;
2481 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2482 err = 0;
2483 flags |= MSG_BATCH;
2484
2485 while (datagrams < vlen) {
2486 if (datagrams == vlen - 1)
2487 flags = oflags;
2488
2489 if (MSG_CMSG_COMPAT & flags) {
2490 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2491 &msg_sys, flags, &used_address, MSG_EOR);
2492 if (err < 0)
2493 break;
2494 err = __put_user(err, &compat_entry->msg_len);
2495 ++compat_entry;
2496 } else {
2497 err = ___sys_sendmsg(sock,
2498 (struct user_msghdr __user *)entry,
2499 &msg_sys, flags, &used_address, MSG_EOR);
2500 if (err < 0)
2501 break;
2502 err = put_user(err, &entry->msg_len);
2503 ++entry;
2504 }
2505
2506 if (err)
2507 break;
2508 ++datagrams;
2509 if (msg_data_left(&msg_sys))
2510 break;
2511 cond_resched();
2512 }
2513
2514 fput_light(sock->file, fput_needed);
2515
2516 /* We only return an error if no datagrams were able to be sent */
2517 if (datagrams != 0)
2518 return datagrams;
2519
2520 return err;
2521}
2522
2523SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2524 unsigned int, vlen, unsigned int, flags)
2525{
2526 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2527}
2528
2529int recvmsg_copy_msghdr(struct msghdr *msg,
2530 struct user_msghdr __user *umsg, unsigned flags,
2531 struct sockaddr __user **uaddr,
2532 struct iovec **iov)
2533{
2534 ssize_t err;
2535
2536 if (MSG_CMSG_COMPAT & flags) {
2537 struct compat_msghdr __user *msg_compat;
2538
2539 msg_compat = (struct compat_msghdr __user *) umsg;
2540 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2541 } else {
2542 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2543 }
2544 if (err < 0)
2545 return err;
2546
2547 return 0;
2548}
2549
2550static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2551 struct user_msghdr __user *msg,
2552 struct sockaddr __user *uaddr,
2553 unsigned int flags, int nosec)
2554{
2555 struct compat_msghdr __user *msg_compat =
2556 (struct compat_msghdr __user *) msg;
2557 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2558 struct sockaddr_storage addr;
2559 unsigned long cmsg_ptr;
2560 int len;
2561 ssize_t err;
2562
2563 msg_sys->msg_name = &addr;
2564 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2565 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2566
2567 /* We assume all kernel code knows the size of sockaddr_storage */
2568 msg_sys->msg_namelen = 0;
2569
2570 if (sock->file->f_flags & O_NONBLOCK)
2571 flags |= MSG_DONTWAIT;
2572
2573 if (unlikely(nosec))
2574 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2575 else
2576 err = sock_recvmsg(sock, msg_sys, flags);
2577
2578 if (err < 0)
2579 goto out;
2580 len = err;
2581
2582 if (uaddr != NULL) {
2583 err = move_addr_to_user(&addr,
2584 msg_sys->msg_namelen, uaddr,
2585 uaddr_len);
2586 if (err < 0)
2587 goto out;
2588 }
2589 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2590 COMPAT_FLAGS(msg));
2591 if (err)
2592 goto out;
2593 if (MSG_CMSG_COMPAT & flags)
2594 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2595 &msg_compat->msg_controllen);
2596 else
2597 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2598 &msg->msg_controllen);
2599 if (err)
2600 goto out;
2601 err = len;
2602out:
2603 return err;
2604}
2605
2606static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2607 struct msghdr *msg_sys, unsigned int flags, int nosec)
2608{
2609 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2610 /* user mode address pointers */
2611 struct sockaddr __user *uaddr;
2612 ssize_t err;
2613
2614 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2615 if (err < 0)
2616 return err;
2617
2618 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2619 kfree(iov);
2620 return err;
2621}
2622
2623/*
2624 * BSD recvmsg interface
2625 */
2626
2627long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2628 struct user_msghdr __user *umsg,
2629 struct sockaddr __user *uaddr, unsigned int flags)
2630{
2631 /* disallow ancillary data requests from this path */
2632 if (msg->msg_control || msg->msg_controllen)
2633 return -EINVAL;
2634
2635 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2636}
2637
2638long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2639 bool forbid_cmsg_compat)
2640{
2641 int fput_needed, err;
2642 struct msghdr msg_sys;
2643 struct socket *sock;
2644
2645 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2646 return -EINVAL;
2647
2648 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2649 if (!sock)
2650 goto out;
2651
2652 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2653
2654 fput_light(sock->file, fput_needed);
2655out:
2656 return err;
2657}
2658
2659SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2660 unsigned int, flags)
2661{
2662 return __sys_recvmsg(fd, msg, flags, true);
2663}
2664
2665/*
2666 * Linux recvmmsg interface
2667 */
2668
2669static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2670 unsigned int vlen, unsigned int flags,
2671 struct timespec64 *timeout)
2672{
2673 int fput_needed, err, datagrams;
2674 struct socket *sock;
2675 struct mmsghdr __user *entry;
2676 struct compat_mmsghdr __user *compat_entry;
2677 struct msghdr msg_sys;
2678 struct timespec64 end_time;
2679 struct timespec64 timeout64;
2680
2681 if (timeout &&
2682 poll_select_set_timeout(&end_time, timeout->tv_sec,
2683 timeout->tv_nsec))
2684 return -EINVAL;
2685
2686 datagrams = 0;
2687
2688 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2689 if (!sock)
2690 return err;
2691
2692 if (likely(!(flags & MSG_ERRQUEUE))) {
2693 err = sock_error(sock->sk);
2694 if (err) {
2695 datagrams = err;
2696 goto out_put;
2697 }
2698 }
2699
2700 entry = mmsg;
2701 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2702
2703 while (datagrams < vlen) {
2704 /*
2705 * No need to ask LSM for more than the first datagram.
2706 */
2707 if (MSG_CMSG_COMPAT & flags) {
2708 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2709 &msg_sys, flags & ~MSG_WAITFORONE,
2710 datagrams);
2711 if (err < 0)
2712 break;
2713 err = __put_user(err, &compat_entry->msg_len);
2714 ++compat_entry;
2715 } else {
2716 err = ___sys_recvmsg(sock,
2717 (struct user_msghdr __user *)entry,
2718 &msg_sys, flags & ~MSG_WAITFORONE,
2719 datagrams);
2720 if (err < 0)
2721 break;
2722 err = put_user(err, &entry->msg_len);
2723 ++entry;
2724 }
2725
2726 if (err)
2727 break;
2728 ++datagrams;
2729
2730 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2731 if (flags & MSG_WAITFORONE)
2732 flags |= MSG_DONTWAIT;
2733
2734 if (timeout) {
2735 ktime_get_ts64(&timeout64);
2736 *timeout = timespec64_sub(end_time, timeout64);
2737 if (timeout->tv_sec < 0) {
2738 timeout->tv_sec = timeout->tv_nsec = 0;
2739 break;
2740 }
2741
2742 /* Timeout, return less than vlen datagrams */
2743 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2744 break;
2745 }
2746
2747 /* Out of band data, return right away */
2748 if (msg_sys.msg_flags & MSG_OOB)
2749 break;
2750 cond_resched();
2751 }
2752
2753 if (err == 0)
2754 goto out_put;
2755
2756 if (datagrams == 0) {
2757 datagrams = err;
2758 goto out_put;
2759 }
2760
2761 /*
2762 * We may return less entries than requested (vlen) if the
2763 * sock is non block and there aren't enough datagrams...
2764 */
2765 if (err != -EAGAIN) {
2766 /*
2767 * ... or if recvmsg returns an error after we
2768 * received some datagrams, where we record the
2769 * error to return on the next call or if the
2770 * app asks about it using getsockopt(SO_ERROR).
2771 */
2772 sock->sk->sk_err = -err;
2773 }
2774out_put:
2775 fput_light(sock->file, fput_needed);
2776
2777 return datagrams;
2778}
2779
2780int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2781 unsigned int vlen, unsigned int flags,
2782 struct __kernel_timespec __user *timeout,
2783 struct old_timespec32 __user *timeout32)
2784{
2785 int datagrams;
2786 struct timespec64 timeout_sys;
2787
2788 if (timeout && get_timespec64(&timeout_sys, timeout))
2789 return -EFAULT;
2790
2791 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2792 return -EFAULT;
2793
2794 if (!timeout && !timeout32)
2795 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2796
2797 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2798
2799 if (datagrams <= 0)
2800 return datagrams;
2801
2802 if (timeout && put_timespec64(&timeout_sys, timeout))
2803 datagrams = -EFAULT;
2804
2805 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2806 datagrams = -EFAULT;
2807
2808 return datagrams;
2809}
2810
2811SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2812 unsigned int, vlen, unsigned int, flags,
2813 struct __kernel_timespec __user *, timeout)
2814{
2815 if (flags & MSG_CMSG_COMPAT)
2816 return -EINVAL;
2817
2818 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2819}
2820
2821#ifdef CONFIG_COMPAT_32BIT_TIME
2822SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2823 unsigned int, vlen, unsigned int, flags,
2824 struct old_timespec32 __user *, timeout)
2825{
2826 if (flags & MSG_CMSG_COMPAT)
2827 return -EINVAL;
2828
2829 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2830}
2831#endif
2832
2833#ifdef __ARCH_WANT_SYS_SOCKETCALL
2834/* Argument list sizes for sys_socketcall */
2835#define AL(x) ((x) * sizeof(unsigned long))
2836static const unsigned char nargs[21] = {
2837 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2838 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2839 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2840 AL(4), AL(5), AL(4)
2841};
2842
2843#undef AL
2844
2845/*
2846 * System call vectors.
2847 *
2848 * Argument checking cleaned up. Saved 20% in size.
2849 * This function doesn't need to set the kernel lock because
2850 * it is set by the callees.
2851 */
2852
2853SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2854{
2855 unsigned long a[AUDITSC_ARGS];
2856 unsigned long a0, a1;
2857 int err;
2858 unsigned int len;
2859
2860 if (call < 1 || call > SYS_SENDMMSG)
2861 return -EINVAL;
2862 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2863
2864 len = nargs[call];
2865 if (len > sizeof(a))
2866 return -EINVAL;
2867
2868 /* copy_from_user should be SMP safe. */
2869 if (copy_from_user(a, args, len))
2870 return -EFAULT;
2871
2872 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2873 if (err)
2874 return err;
2875
2876 a0 = a[0];
2877 a1 = a[1];
2878
2879 switch (call) {
2880 case SYS_SOCKET:
2881 err = __sys_socket(a0, a1, a[2]);
2882 break;
2883 case SYS_BIND:
2884 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2885 break;
2886 case SYS_CONNECT:
2887 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2888 break;
2889 case SYS_LISTEN:
2890 err = __sys_listen(a0, a1);
2891 break;
2892 case SYS_ACCEPT:
2893 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2894 (int __user *)a[2], 0);
2895 break;
2896 case SYS_GETSOCKNAME:
2897 err =
2898 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2899 (int __user *)a[2]);
2900 break;
2901 case SYS_GETPEERNAME:
2902 err =
2903 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2904 (int __user *)a[2]);
2905 break;
2906 case SYS_SOCKETPAIR:
2907 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2908 break;
2909 case SYS_SEND:
2910 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2911 NULL, 0);
2912 break;
2913 case SYS_SENDTO:
2914 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2915 (struct sockaddr __user *)a[4], a[5]);
2916 break;
2917 case SYS_RECV:
2918 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2919 NULL, NULL);
2920 break;
2921 case SYS_RECVFROM:
2922 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2923 (struct sockaddr __user *)a[4],
2924 (int __user *)a[5]);
2925 break;
2926 case SYS_SHUTDOWN:
2927 err = __sys_shutdown(a0, a1);
2928 break;
2929 case SYS_SETSOCKOPT:
2930 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
2931 a[4]);
2932 break;
2933 case SYS_GETSOCKOPT:
2934 err =
2935 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2936 (int __user *)a[4]);
2937 break;
2938 case SYS_SENDMSG:
2939 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
2940 a[2], true);
2941 break;
2942 case SYS_SENDMMSG:
2943 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
2944 a[3], true);
2945 break;
2946 case SYS_RECVMSG:
2947 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
2948 a[2], true);
2949 break;
2950 case SYS_RECVMMSG:
2951 if (IS_ENABLED(CONFIG_64BIT))
2952 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2953 a[2], a[3],
2954 (struct __kernel_timespec __user *)a[4],
2955 NULL);
2956 else
2957 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2958 a[2], a[3], NULL,
2959 (struct old_timespec32 __user *)a[4]);
2960 break;
2961 case SYS_ACCEPT4:
2962 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2963 (int __user *)a[2], a[3]);
2964 break;
2965 default:
2966 err = -EINVAL;
2967 break;
2968 }
2969 return err;
2970}
2971
2972#endif /* __ARCH_WANT_SYS_SOCKETCALL */
2973
2974/**
2975 * sock_register - add a socket protocol handler
2976 * @ops: description of protocol
2977 *
2978 * This function is called by a protocol handler that wants to
2979 * advertise its address family, and have it linked into the
2980 * socket interface. The value ops->family corresponds to the
2981 * socket system call protocol family.
2982 */
2983int sock_register(const struct net_proto_family *ops)
2984{
2985 int err;
2986
2987 if (ops->family >= NPROTO) {
2988 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
2989 return -ENOBUFS;
2990 }
2991
2992 spin_lock(&net_family_lock);
2993 if (rcu_dereference_protected(net_families[ops->family],
2994 lockdep_is_held(&net_family_lock)))
2995 err = -EEXIST;
2996 else {
2997 rcu_assign_pointer(net_families[ops->family], ops);
2998 err = 0;
2999 }
3000 spin_unlock(&net_family_lock);
3001
3002 pr_info("NET: Registered protocol family %d\n", ops->family);
3003 return err;
3004}
3005EXPORT_SYMBOL(sock_register);
3006
3007/**
3008 * sock_unregister - remove a protocol handler
3009 * @family: protocol family to remove
3010 *
3011 * This function is called by a protocol handler that wants to
3012 * remove its address family, and have it unlinked from the
3013 * new socket creation.
3014 *
3015 * If protocol handler is a module, then it can use module reference
3016 * counts to protect against new references. If protocol handler is not
3017 * a module then it needs to provide its own protection in
3018 * the ops->create routine.
3019 */
3020void sock_unregister(int family)
3021{
3022 BUG_ON(family < 0 || family >= NPROTO);
3023
3024 spin_lock(&net_family_lock);
3025 RCU_INIT_POINTER(net_families[family], NULL);
3026 spin_unlock(&net_family_lock);
3027
3028 synchronize_rcu();
3029
3030 pr_info("NET: Unregistered protocol family %d\n", family);
3031}
3032EXPORT_SYMBOL(sock_unregister);
3033
3034bool sock_is_registered(int family)
3035{
3036 return family < NPROTO && rcu_access_pointer(net_families[family]);
3037}
3038
3039static int __init sock_init(void)
3040{
3041 int err;
3042 /*
3043 * Initialize the network sysctl infrastructure.
3044 */
3045 err = net_sysctl_init();
3046 if (err)
3047 goto out;
3048
3049 /*
3050 * Initialize skbuff SLAB cache
3051 */
3052 skb_init();
3053
3054 /*
3055 * Initialize the protocols module.
3056 */
3057
3058 init_inodecache();
3059
3060 err = register_filesystem(&sock_fs_type);
3061 if (err)
3062 goto out;
3063 sock_mnt = kern_mount(&sock_fs_type);
3064 if (IS_ERR(sock_mnt)) {
3065 err = PTR_ERR(sock_mnt);
3066 goto out_mount;
3067 }
3068
3069 /* The real protocol initialization is performed in later initcalls.
3070 */
3071
3072#ifdef CONFIG_NETFILTER
3073 err = netfilter_init();
3074 if (err)
3075 goto out;
3076#endif
3077
3078 ptp_classifier_init();
3079
3080out:
3081 return err;
3082
3083out_mount:
3084 unregister_filesystem(&sock_fs_type);
3085 goto out;
3086}
3087
3088core_initcall(sock_init); /* early initcall */
3089
3090#ifdef CONFIG_PROC_FS
3091void socket_seq_show(struct seq_file *seq)
3092{
3093 seq_printf(seq, "sockets: used %d\n",
3094 sock_inuse_get(seq->private));
3095}
3096#endif /* CONFIG_PROC_FS */
3097
3098#ifdef CONFIG_COMPAT
3099static int compat_dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
3100{
3101 struct compat_ifconf ifc32;
3102 struct ifconf ifc;
3103 int err;
3104
3105 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
3106 return -EFAULT;
3107
3108 ifc.ifc_len = ifc32.ifc_len;
3109 ifc.ifc_req = compat_ptr(ifc32.ifcbuf);
3110
3111 rtnl_lock();
3112 err = dev_ifconf(net, &ifc, sizeof(struct compat_ifreq));
3113 rtnl_unlock();
3114 if (err)
3115 return err;
3116
3117 ifc32.ifc_len = ifc.ifc_len;
3118 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
3119 return -EFAULT;
3120
3121 return 0;
3122}
3123
3124static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
3125{
3126 struct compat_ethtool_rxnfc __user *compat_rxnfc;
3127 bool convert_in = false, convert_out = false;
3128 size_t buf_size = 0;
3129 struct ethtool_rxnfc __user *rxnfc = NULL;
3130 struct ifreq ifr;
3131 u32 rule_cnt = 0, actual_rule_cnt;
3132 u32 ethcmd;
3133 u32 data;
3134 int ret;
3135
3136 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
3137 return -EFAULT;
3138
3139 compat_rxnfc = compat_ptr(data);
3140
3141 if (get_user(ethcmd, &compat_rxnfc->cmd))
3142 return -EFAULT;
3143
3144 /* Most ethtool structures are defined without padding.
3145 * Unfortunately struct ethtool_rxnfc is an exception.
3146 */
3147 switch (ethcmd) {
3148 default:
3149 break;
3150 case ETHTOOL_GRXCLSRLALL:
3151 /* Buffer size is variable */
3152 if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
3153 return -EFAULT;
3154 if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
3155 return -ENOMEM;
3156 buf_size += rule_cnt * sizeof(u32);
3157 fallthrough;
3158 case ETHTOOL_GRXRINGS:
3159 case ETHTOOL_GRXCLSRLCNT:
3160 case ETHTOOL_GRXCLSRULE:
3161 case ETHTOOL_SRXCLSRLINS:
3162 convert_out = true;
3163 fallthrough;
3164 case ETHTOOL_SRXCLSRLDEL:
3165 buf_size += sizeof(struct ethtool_rxnfc);
3166 convert_in = true;
3167 rxnfc = compat_alloc_user_space(buf_size);
3168 break;
3169 }
3170
3171 if (copy_from_user(&ifr.ifr_name, &ifr32->ifr_name, IFNAMSIZ))
3172 return -EFAULT;
3173
3174 ifr.ifr_data = convert_in ? rxnfc : (void __user *)compat_rxnfc;
3175
3176 if (convert_in) {
3177 /* We expect there to be holes between fs.m_ext and
3178 * fs.ring_cookie and at the end of fs, but nowhere else.
3179 */
3180 BUILD_BUG_ON(offsetof(struct compat_ethtool_rxnfc, fs.m_ext) +
3181 sizeof(compat_rxnfc->fs.m_ext) !=
3182 offsetof(struct ethtool_rxnfc, fs.m_ext) +
3183 sizeof(rxnfc->fs.m_ext));
3184 BUILD_BUG_ON(
3185 offsetof(struct compat_ethtool_rxnfc, fs.location) -
3186 offsetof(struct compat_ethtool_rxnfc, fs.ring_cookie) !=
3187 offsetof(struct ethtool_rxnfc, fs.location) -
3188 offsetof(struct ethtool_rxnfc, fs.ring_cookie));
3189
3190 if (copy_in_user(rxnfc, compat_rxnfc,
3191 (void __user *)(&rxnfc->fs.m_ext + 1) -
3192 (void __user *)rxnfc) ||
3193 copy_in_user(&rxnfc->fs.ring_cookie,
3194 &compat_rxnfc->fs.ring_cookie,
3195 (void __user *)(&rxnfc->fs.location + 1) -
3196 (void __user *)&rxnfc->fs.ring_cookie))
3197 return -EFAULT;
3198 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
3199 if (put_user(rule_cnt, &rxnfc->rule_cnt))
3200 return -EFAULT;
3201 } else if (copy_in_user(&rxnfc->rule_cnt,
3202 &compat_rxnfc->rule_cnt,
3203 sizeof(rxnfc->rule_cnt)))
3204 return -EFAULT;
3205 }
3206
3207 ret = dev_ioctl(net, SIOCETHTOOL, &ifr, NULL);
3208 if (ret)
3209 return ret;
3210
3211 if (convert_out) {
3212 if (copy_in_user(compat_rxnfc, rxnfc,
3213 (const void __user *)(&rxnfc->fs.m_ext + 1) -
3214 (const void __user *)rxnfc) ||
3215 copy_in_user(&compat_rxnfc->fs.ring_cookie,
3216 &rxnfc->fs.ring_cookie,
3217 (const void __user *)(&rxnfc->fs.location + 1) -
3218 (const void __user *)&rxnfc->fs.ring_cookie) ||
3219 copy_in_user(&compat_rxnfc->rule_cnt, &rxnfc->rule_cnt,
3220 sizeof(rxnfc->rule_cnt)))
3221 return -EFAULT;
3222
3223 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
3224 /* As an optimisation, we only copy the actual
3225 * number of rules that the underlying
3226 * function returned. Since Mallory might
3227 * change the rule count in user memory, we
3228 * check that it is less than the rule count
3229 * originally given (as the user buffer size),
3230 * which has been range-checked.
3231 */
3232 if (get_user(actual_rule_cnt, &rxnfc->rule_cnt))
3233 return -EFAULT;
3234 if (actual_rule_cnt < rule_cnt)
3235 rule_cnt = actual_rule_cnt;
3236 if (copy_in_user(&compat_rxnfc->rule_locs[0],
3237 &rxnfc->rule_locs[0],
3238 rule_cnt * sizeof(u32)))
3239 return -EFAULT;
3240 }
3241 }
3242
3243 return 0;
3244}
3245
3246static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3247{
3248 compat_uptr_t uptr32;
3249 struct ifreq ifr;
3250 void __user *saved;
3251 int err;
3252
3253 if (copy_from_user(&ifr, uifr32, sizeof(struct compat_ifreq)))
3254 return -EFAULT;
3255
3256 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3257 return -EFAULT;
3258
3259 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3260 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3261
3262 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL);
3263 if (!err) {
3264 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3265 if (copy_to_user(uifr32, &ifr, sizeof(struct compat_ifreq)))
3266 err = -EFAULT;
3267 }
3268 return err;
3269}
3270
3271/* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3272static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3273 struct compat_ifreq __user *u_ifreq32)
3274{
3275 struct ifreq ifreq;
3276 u32 data32;
3277
3278 if (copy_from_user(ifreq.ifr_name, u_ifreq32->ifr_name, IFNAMSIZ))
3279 return -EFAULT;
3280 if (get_user(data32, &u_ifreq32->ifr_data))
3281 return -EFAULT;
3282 ifreq.ifr_data = compat_ptr(data32);
3283
3284 return dev_ioctl(net, cmd, &ifreq, NULL);
3285}
3286
3287static int compat_ifreq_ioctl(struct net *net, struct socket *sock,
3288 unsigned int cmd,
3289 struct compat_ifreq __user *uifr32)
3290{
3291 struct ifreq __user *uifr;
3292 int err;
3293
3294 /* Handle the fact that while struct ifreq has the same *layout* on
3295 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3296 * which are handled elsewhere, it still has different *size* due to
3297 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3298 * resulting in struct ifreq being 32 and 40 bytes respectively).
3299 * As a result, if the struct happens to be at the end of a page and
3300 * the next page isn't readable/writable, we get a fault. To prevent
3301 * that, copy back and forth to the full size.
3302 */
3303
3304 uifr = compat_alloc_user_space(sizeof(*uifr));
3305 if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
3306 return -EFAULT;
3307
3308 err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
3309
3310 if (!err) {
3311 switch (cmd) {
3312 case SIOCGIFFLAGS:
3313 case SIOCGIFMETRIC:
3314 case SIOCGIFMTU:
3315 case SIOCGIFMEM:
3316 case SIOCGIFHWADDR:
3317 case SIOCGIFINDEX:
3318 case SIOCGIFADDR:
3319 case SIOCGIFBRDADDR:
3320 case SIOCGIFDSTADDR:
3321 case SIOCGIFNETMASK:
3322 case SIOCGIFPFLAGS:
3323 case SIOCGIFTXQLEN:
3324 case SIOCGMIIPHY:
3325 case SIOCGMIIREG:
3326 case SIOCGIFNAME:
3327 if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
3328 err = -EFAULT;
3329 break;
3330 }
3331 }
3332 return err;
3333}
3334
3335static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
3336 struct compat_ifreq __user *uifr32)
3337{
3338 struct ifreq ifr;
3339 struct compat_ifmap __user *uifmap32;
3340 int err;
3341
3342 uifmap32 = &uifr32->ifr_ifru.ifru_map;
3343 err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
3344 err |= get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3345 err |= get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3346 err |= get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3347 err |= get_user(ifr.ifr_map.irq, &uifmap32->irq);
3348 err |= get_user(ifr.ifr_map.dma, &uifmap32->dma);
3349 err |= get_user(ifr.ifr_map.port, &uifmap32->port);
3350 if (err)
3351 return -EFAULT;
3352
3353 err = dev_ioctl(net, cmd, &ifr, NULL);
3354
3355 if (cmd == SIOCGIFMAP && !err) {
3356 err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
3357 err |= put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3358 err |= put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3359 err |= put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3360 err |= put_user(ifr.ifr_map.irq, &uifmap32->irq);
3361 err |= put_user(ifr.ifr_map.dma, &uifmap32->dma);
3362 err |= put_user(ifr.ifr_map.port, &uifmap32->port);
3363 if (err)
3364 err = -EFAULT;
3365 }
3366 return err;
3367}
3368
3369/* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3370 * for some operations; this forces use of the newer bridge-utils that
3371 * use compatible ioctls
3372 */
3373static int old_bridge_ioctl(compat_ulong_t __user *argp)
3374{
3375 compat_ulong_t tmp;
3376
3377 if (get_user(tmp, argp))
3378 return -EFAULT;
3379 if (tmp == BRCTL_GET_VERSION)
3380 return BRCTL_VERSION + 1;
3381 return -EINVAL;
3382}
3383
3384static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3385 unsigned int cmd, unsigned long arg)
3386{
3387 void __user *argp = compat_ptr(arg);
3388 struct sock *sk = sock->sk;
3389 struct net *net = sock_net(sk);
3390
3391 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3392 return compat_ifr_data_ioctl(net, cmd, argp);
3393
3394 switch (cmd) {
3395 case SIOCSIFBR:
3396 case SIOCGIFBR:
3397 return old_bridge_ioctl(argp);
3398 case SIOCGIFCONF:
3399 return compat_dev_ifconf(net, argp);
3400 case SIOCETHTOOL:
3401 return ethtool_ioctl(net, argp);
3402 case SIOCWANDEV:
3403 return compat_siocwandev(net, argp);
3404 case SIOCGIFMAP:
3405 case SIOCSIFMAP:
3406 return compat_sioc_ifmap(net, cmd, argp);
3407 case SIOCGSTAMP_OLD:
3408 case SIOCGSTAMPNS_OLD:
3409 if (!sock->ops->gettstamp)
3410 return -ENOIOCTLCMD;
3411 return sock->ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3412 !COMPAT_USE_64BIT_TIME);
3413
3414 case SIOCBONDSLAVEINFOQUERY:
3415 case SIOCBONDINFOQUERY:
3416 case SIOCSHWTSTAMP:
3417 case SIOCGHWTSTAMP:
3418 return compat_ifr_data_ioctl(net, cmd, argp);
3419
3420 case FIOSETOWN:
3421 case SIOCSPGRP:
3422 case FIOGETOWN:
3423 case SIOCGPGRP:
3424 case SIOCBRADDBR:
3425 case SIOCBRDELBR:
3426 case SIOCGIFVLAN:
3427 case SIOCSIFVLAN:
3428 case SIOCADDDLCI:
3429 case SIOCDELDLCI:
3430 case SIOCGSKNS:
3431 case SIOCGSTAMP_NEW:
3432 case SIOCGSTAMPNS_NEW:
3433 return sock_ioctl(file, cmd, arg);
3434
3435 case SIOCGIFFLAGS:
3436 case SIOCSIFFLAGS:
3437 case SIOCGIFMETRIC:
3438 case SIOCSIFMETRIC:
3439 case SIOCGIFMTU:
3440 case SIOCSIFMTU:
3441 case SIOCGIFMEM:
3442 case SIOCSIFMEM:
3443 case SIOCGIFHWADDR:
3444 case SIOCSIFHWADDR:
3445 case SIOCADDMULTI:
3446 case SIOCDELMULTI:
3447 case SIOCGIFINDEX:
3448 case SIOCGIFADDR:
3449 case SIOCSIFADDR:
3450 case SIOCSIFHWBROADCAST:
3451 case SIOCDIFADDR:
3452 case SIOCGIFBRDADDR:
3453 case SIOCSIFBRDADDR:
3454 case SIOCGIFDSTADDR:
3455 case SIOCSIFDSTADDR:
3456 case SIOCGIFNETMASK:
3457 case SIOCSIFNETMASK:
3458 case SIOCSIFPFLAGS:
3459 case SIOCGIFPFLAGS:
3460 case SIOCGIFTXQLEN:
3461 case SIOCSIFTXQLEN:
3462 case SIOCBRADDIF:
3463 case SIOCBRDELIF:
3464 case SIOCGIFNAME:
3465 case SIOCSIFNAME:
3466 case SIOCGMIIPHY:
3467 case SIOCGMIIREG:
3468 case SIOCSMIIREG:
3469 case SIOCBONDENSLAVE:
3470 case SIOCBONDRELEASE:
3471 case SIOCBONDSETHWADDR:
3472 case SIOCBONDCHANGEACTIVE:
3473 return compat_ifreq_ioctl(net, sock, cmd, argp);
3474
3475 case SIOCSARP:
3476 case SIOCGARP:
3477 case SIOCDARP:
3478 case SIOCOUTQ:
3479 case SIOCOUTQNSD:
3480 case SIOCATMARK:
3481 return sock_do_ioctl(net, sock, cmd, arg);
3482 }
3483
3484 return -ENOIOCTLCMD;
3485}
3486
3487static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3488 unsigned long arg)
3489{
3490 struct socket *sock = file->private_data;
3491 int ret = -ENOIOCTLCMD;
3492 struct sock *sk;
3493 struct net *net;
3494
3495 sk = sock->sk;
3496 net = sock_net(sk);
3497
3498 if (sock->ops->compat_ioctl)
3499 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3500
3501 if (ret == -ENOIOCTLCMD &&
3502 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3503 ret = compat_wext_handle_ioctl(net, cmd, arg);
3504
3505 if (ret == -ENOIOCTLCMD)
3506 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3507
3508 return ret;
3509}
3510#endif
3511
3512/**
3513 * kernel_bind - bind an address to a socket (kernel space)
3514 * @sock: socket
3515 * @addr: address
3516 * @addrlen: length of address
3517 *
3518 * Returns 0 or an error.
3519 */
3520
3521int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3522{
3523 return sock->ops->bind(sock, addr, addrlen);
3524}
3525EXPORT_SYMBOL(kernel_bind);
3526
3527/**
3528 * kernel_listen - move socket to listening state (kernel space)
3529 * @sock: socket
3530 * @backlog: pending connections queue size
3531 *
3532 * Returns 0 or an error.
3533 */
3534
3535int kernel_listen(struct socket *sock, int backlog)
3536{
3537 return sock->ops->listen(sock, backlog);
3538}
3539EXPORT_SYMBOL(kernel_listen);
3540
3541/**
3542 * kernel_accept - accept a connection (kernel space)
3543 * @sock: listening socket
3544 * @newsock: new connected socket
3545 * @flags: flags
3546 *
3547 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3548 * If it fails, @newsock is guaranteed to be %NULL.
3549 * Returns 0 or an error.
3550 */
3551
3552int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3553{
3554 struct sock *sk = sock->sk;
3555 int err;
3556
3557 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3558 newsock);
3559 if (err < 0)
3560 goto done;
3561
3562 err = sock->ops->accept(sock, *newsock, flags, true);
3563 if (err < 0) {
3564 sock_release(*newsock);
3565 *newsock = NULL;
3566 goto done;
3567 }
3568
3569 (*newsock)->ops = sock->ops;
3570 __module_get((*newsock)->ops->owner);
3571
3572done:
3573 return err;
3574}
3575EXPORT_SYMBOL(kernel_accept);
3576
3577/**
3578 * kernel_connect - connect a socket (kernel space)
3579 * @sock: socket
3580 * @addr: address
3581 * @addrlen: address length
3582 * @flags: flags (O_NONBLOCK, ...)
3583 *
3584 * For datagram sockets, @addr is the addres to which datagrams are sent
3585 * by default, and the only address from which datagrams are received.
3586 * For stream sockets, attempts to connect to @addr.
3587 * Returns 0 or an error code.
3588 */
3589
3590int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3591 int flags)
3592{
3593 return sock->ops->connect(sock, addr, addrlen, flags);
3594}
3595EXPORT_SYMBOL(kernel_connect);
3596
3597/**
3598 * kernel_getsockname - get the address which the socket is bound (kernel space)
3599 * @sock: socket
3600 * @addr: address holder
3601 *
3602 * Fills the @addr pointer with the address which the socket is bound.
3603 * Returns 0 or an error code.
3604 */
3605
3606int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3607{
3608 return sock->ops->getname(sock, addr, 0);
3609}
3610EXPORT_SYMBOL(kernel_getsockname);
3611
3612/**
3613 * kernel_getpeername - get the address which the socket is connected (kernel space)
3614 * @sock: socket
3615 * @addr: address holder
3616 *
3617 * Fills the @addr pointer with the address which the socket is connected.
3618 * Returns 0 or an error code.
3619 */
3620
3621int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3622{
3623 return sock->ops->getname(sock, addr, 1);
3624}
3625EXPORT_SYMBOL(kernel_getpeername);
3626
3627/**
3628 * kernel_sendpage - send a &page through a socket (kernel space)
3629 * @sock: socket
3630 * @page: page
3631 * @offset: page offset
3632 * @size: total size in bytes
3633 * @flags: flags (MSG_DONTWAIT, ...)
3634 *
3635 * Returns the total amount sent in bytes or an error.
3636 */
3637
3638int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3639 size_t size, int flags)
3640{
3641 if (sock->ops->sendpage) {
3642 /* Warn in case the improper page to zero-copy send */
3643 WARN_ONCE(!sendpage_ok(page), "improper page for zero-copy send");
3644 return sock->ops->sendpage(sock, page, offset, size, flags);
3645 }
3646 return sock_no_sendpage(sock, page, offset, size, flags);
3647}
3648EXPORT_SYMBOL(kernel_sendpage);
3649
3650/**
3651 * kernel_sendpage_locked - send a &page through the locked sock (kernel space)
3652 * @sk: sock
3653 * @page: page
3654 * @offset: page offset
3655 * @size: total size in bytes
3656 * @flags: flags (MSG_DONTWAIT, ...)
3657 *
3658 * Returns the total amount sent in bytes or an error.
3659 * Caller must hold @sk.
3660 */
3661
3662int kernel_sendpage_locked(struct sock *sk, struct page *page, int offset,
3663 size_t size, int flags)
3664{
3665 struct socket *sock = sk->sk_socket;
3666
3667 if (sock->ops->sendpage_locked)
3668 return sock->ops->sendpage_locked(sk, page, offset, size,
3669 flags);
3670
3671 return sock_no_sendpage_locked(sk, page, offset, size, flags);
3672}
3673EXPORT_SYMBOL(kernel_sendpage_locked);
3674
3675/**
3676 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3677 * @sock: socket
3678 * @how: connection part
3679 *
3680 * Returns 0 or an error.
3681 */
3682
3683int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3684{
3685 return sock->ops->shutdown(sock, how);
3686}
3687EXPORT_SYMBOL(kernel_sock_shutdown);
3688
3689/**
3690 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3691 * @sk: socket
3692 *
3693 * This routine returns the IP overhead imposed by a socket i.e.
3694 * the length of the underlying IP header, depending on whether
3695 * this is an IPv4 or IPv6 socket and the length from IP options turned
3696 * on at the socket. Assumes that the caller has a lock on the socket.
3697 */
3698
3699u32 kernel_sock_ip_overhead(struct sock *sk)
3700{
3701 struct inet_sock *inet;
3702 struct ip_options_rcu *opt;
3703 u32 overhead = 0;
3704#if IS_ENABLED(CONFIG_IPV6)
3705 struct ipv6_pinfo *np;
3706 struct ipv6_txoptions *optv6 = NULL;
3707#endif /* IS_ENABLED(CONFIG_IPV6) */
3708
3709 if (!sk)
3710 return overhead;
3711
3712 switch (sk->sk_family) {
3713 case AF_INET:
3714 inet = inet_sk(sk);
3715 overhead += sizeof(struct iphdr);
3716 opt = rcu_dereference_protected(inet->inet_opt,
3717 sock_owned_by_user(sk));
3718 if (opt)
3719 overhead += opt->opt.optlen;
3720 return overhead;
3721#if IS_ENABLED(CONFIG_IPV6)
3722 case AF_INET6:
3723 np = inet6_sk(sk);
3724 overhead += sizeof(struct ipv6hdr);
3725 if (np)
3726 optv6 = rcu_dereference_protected(np->opt,
3727 sock_owned_by_user(sk));
3728 if (optv6)
3729 overhead += (optv6->opt_flen + optv6->opt_nflen);
3730 return overhead;
3731#endif /* IS_ENABLED(CONFIG_IPV6) */
3732 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3733 return overhead;
3734 }
3735}
3736EXPORT_SYMBOL(kernel_sock_ip_overhead);