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/bpf-cgroup.h>
56#include <linux/ethtool.h>
57#include <linux/mm.h>
58#include <linux/socket.h>
59#include <linux/file.h>
60#include <linux/splice.h>
61#include <linux/net.h>
62#include <linux/interrupt.h>
63#include <linux/thread_info.h>
64#include <linux/rcupdate.h>
65#include <linux/netdevice.h>
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68#include <linux/mutex.h>
69#include <linux/if_bridge.h>
70#include <linux/if_vlan.h>
71#include <linux/ptp_classify.h>
72#include <linux/init.h>
73#include <linux/poll.h>
74#include <linux/cache.h>
75#include <linux/module.h>
76#include <linux/highmem.h>
77#include <linux/mount.h>
78#include <linux/pseudo_fs.h>
79#include <linux/security.h>
80#include <linux/syscalls.h>
81#include <linux/compat.h>
82#include <linux/kmod.h>
83#include <linux/audit.h>
84#include <linux/wireless.h>
85#include <linux/nsproxy.h>
86#include <linux/magic.h>
87#include <linux/slab.h>
88#include <linux/xattr.h>
89#include <linux/nospec.h>
90#include <linux/indirect_call_wrapper.h>
91#include <linux/io_uring.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/termios.h>
107#include <linux/sockios.h>
108#include <net/busy_poll.h>
109#include <linux/errqueue.h>
110#include <linux/ptp_clock_kernel.h>
111#include <trace/events/sock.h>
112
113#ifdef CONFIG_NET_RX_BUSY_POLL
114unsigned int sysctl_net_busy_read __read_mostly;
115unsigned int sysctl_net_busy_poll __read_mostly;
116#endif
117
118static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
119static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
120static int sock_mmap(struct file *file, struct vm_area_struct *vma);
121
122static int sock_close(struct inode *inode, struct file *file);
123static __poll_t sock_poll(struct file *file,
124 struct poll_table_struct *wait);
125static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
126#ifdef CONFIG_COMPAT
127static long compat_sock_ioctl(struct file *file,
128 unsigned int cmd, unsigned long arg);
129#endif
130static int sock_fasync(int fd, struct file *filp, int on);
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);
134static void sock_splice_eof(struct file *file);
135
136#ifdef CONFIG_PROC_FS
137static void sock_show_fdinfo(struct seq_file *m, struct file *f)
138{
139 struct socket *sock = f->private_data;
140 const struct proto_ops *ops = READ_ONCE(sock->ops);
141
142 if (ops->show_fdinfo)
143 ops->show_fdinfo(m, sock);
144}
145#else
146#define sock_show_fdinfo NULL
147#endif
148
149/*
150 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
151 * in the operation structures but are done directly via the socketcall() multiplexor.
152 */
153
154static const struct file_operations socket_file_ops = {
155 .owner = THIS_MODULE,
156 .llseek = no_llseek,
157 .read_iter = sock_read_iter,
158 .write_iter = sock_write_iter,
159 .poll = sock_poll,
160 .unlocked_ioctl = sock_ioctl,
161#ifdef CONFIG_COMPAT
162 .compat_ioctl = compat_sock_ioctl,
163#endif
164 .uring_cmd = io_uring_cmd_sock,
165 .mmap = sock_mmap,
166 .release = sock_close,
167 .fasync = sock_fasync,
168 .splice_write = splice_to_socket,
169 .splice_read = sock_splice_read,
170 .splice_eof = sock_splice_eof,
171 .show_fdinfo = sock_show_fdinfo,
172};
173
174static const char * const pf_family_names[] = {
175 [PF_UNSPEC] = "PF_UNSPEC",
176 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
177 [PF_INET] = "PF_INET",
178 [PF_AX25] = "PF_AX25",
179 [PF_IPX] = "PF_IPX",
180 [PF_APPLETALK] = "PF_APPLETALK",
181 [PF_NETROM] = "PF_NETROM",
182 [PF_BRIDGE] = "PF_BRIDGE",
183 [PF_ATMPVC] = "PF_ATMPVC",
184 [PF_X25] = "PF_X25",
185 [PF_INET6] = "PF_INET6",
186 [PF_ROSE] = "PF_ROSE",
187 [PF_DECnet] = "PF_DECnet",
188 [PF_NETBEUI] = "PF_NETBEUI",
189 [PF_SECURITY] = "PF_SECURITY",
190 [PF_KEY] = "PF_KEY",
191 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
192 [PF_PACKET] = "PF_PACKET",
193 [PF_ASH] = "PF_ASH",
194 [PF_ECONET] = "PF_ECONET",
195 [PF_ATMSVC] = "PF_ATMSVC",
196 [PF_RDS] = "PF_RDS",
197 [PF_SNA] = "PF_SNA",
198 [PF_IRDA] = "PF_IRDA",
199 [PF_PPPOX] = "PF_PPPOX",
200 [PF_WANPIPE] = "PF_WANPIPE",
201 [PF_LLC] = "PF_LLC",
202 [PF_IB] = "PF_IB",
203 [PF_MPLS] = "PF_MPLS",
204 [PF_CAN] = "PF_CAN",
205 [PF_TIPC] = "PF_TIPC",
206 [PF_BLUETOOTH] = "PF_BLUETOOTH",
207 [PF_IUCV] = "PF_IUCV",
208 [PF_RXRPC] = "PF_RXRPC",
209 [PF_ISDN] = "PF_ISDN",
210 [PF_PHONET] = "PF_PHONET",
211 [PF_IEEE802154] = "PF_IEEE802154",
212 [PF_CAIF] = "PF_CAIF",
213 [PF_ALG] = "PF_ALG",
214 [PF_NFC] = "PF_NFC",
215 [PF_VSOCK] = "PF_VSOCK",
216 [PF_KCM] = "PF_KCM",
217 [PF_QIPCRTR] = "PF_QIPCRTR",
218 [PF_SMC] = "PF_SMC",
219 [PF_XDP] = "PF_XDP",
220 [PF_MCTP] = "PF_MCTP",
221};
222
223/*
224 * The protocol list. Each protocol is registered in here.
225 */
226
227static DEFINE_SPINLOCK(net_family_lock);
228static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
229
230/*
231 * Support routines.
232 * Move socket addresses back and forth across the kernel/user
233 * divide and look after the messy bits.
234 */
235
236/**
237 * move_addr_to_kernel - copy a socket address into kernel space
238 * @uaddr: Address in user space
239 * @kaddr: Address in kernel space
240 * @ulen: Length in user space
241 *
242 * The address is copied into kernel space. If the provided address is
243 * too long an error code of -EINVAL is returned. If the copy gives
244 * invalid addresses -EFAULT is returned. On a success 0 is returned.
245 */
246
247int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
248{
249 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
250 return -EINVAL;
251 if (ulen == 0)
252 return 0;
253 if (copy_from_user(kaddr, uaddr, ulen))
254 return -EFAULT;
255 return audit_sockaddr(ulen, kaddr);
256}
257
258/**
259 * move_addr_to_user - copy an address to user space
260 * @kaddr: kernel space address
261 * @klen: length of address in kernel
262 * @uaddr: user space address
263 * @ulen: pointer to user length field
264 *
265 * The value pointed to by ulen on entry is the buffer length available.
266 * This is overwritten with the buffer space used. -EINVAL is returned
267 * if an overlong buffer is specified or a negative buffer size. -EFAULT
268 * is returned if either the buffer or the length field are not
269 * accessible.
270 * After copying the data up to the limit the user specifies, the true
271 * length of the data is written over the length limit the user
272 * specified. Zero is returned for a success.
273 */
274
275static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
276 void __user *uaddr, int __user *ulen)
277{
278 int err;
279 int len;
280
281 BUG_ON(klen > sizeof(struct sockaddr_storage));
282 err = get_user(len, ulen);
283 if (err)
284 return err;
285 if (len > klen)
286 len = klen;
287 if (len < 0)
288 return -EINVAL;
289 if (len) {
290 if (audit_sockaddr(klen, kaddr))
291 return -ENOMEM;
292 if (copy_to_user(uaddr, kaddr, len))
293 return -EFAULT;
294 }
295 /*
296 * "fromlen shall refer to the value before truncation.."
297 * 1003.1g
298 */
299 return __put_user(klen, ulen);
300}
301
302static struct kmem_cache *sock_inode_cachep __ro_after_init;
303
304static struct inode *sock_alloc_inode(struct super_block *sb)
305{
306 struct socket_alloc *ei;
307
308 ei = alloc_inode_sb(sb, sock_inode_cachep, GFP_KERNEL);
309 if (!ei)
310 return NULL;
311 init_waitqueue_head(&ei->socket.wq.wait);
312 ei->socket.wq.fasync_list = NULL;
313 ei->socket.wq.flags = 0;
314
315 ei->socket.state = SS_UNCONNECTED;
316 ei->socket.flags = 0;
317 ei->socket.ops = NULL;
318 ei->socket.sk = NULL;
319 ei->socket.file = NULL;
320
321 return &ei->vfs_inode;
322}
323
324static void sock_free_inode(struct inode *inode)
325{
326 struct socket_alloc *ei;
327
328 ei = container_of(inode, struct socket_alloc, vfs_inode);
329 kmem_cache_free(sock_inode_cachep, ei);
330}
331
332static void init_once(void *foo)
333{
334 struct socket_alloc *ei = (struct socket_alloc *)foo;
335
336 inode_init_once(&ei->vfs_inode);
337}
338
339static void init_inodecache(void)
340{
341 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
342 sizeof(struct socket_alloc),
343 0,
344 (SLAB_HWCACHE_ALIGN |
345 SLAB_RECLAIM_ACCOUNT |
346 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
347 init_once);
348 BUG_ON(sock_inode_cachep == NULL);
349}
350
351static const struct super_operations sockfs_ops = {
352 .alloc_inode = sock_alloc_inode,
353 .free_inode = sock_free_inode,
354 .statfs = simple_statfs,
355};
356
357/*
358 * sockfs_dname() is called from d_path().
359 */
360static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
361{
362 return dynamic_dname(buffer, buflen, "socket:[%lu]",
363 d_inode(dentry)->i_ino);
364}
365
366static const struct dentry_operations sockfs_dentry_operations = {
367 .d_dname = sockfs_dname,
368};
369
370static int sockfs_xattr_get(const struct xattr_handler *handler,
371 struct dentry *dentry, struct inode *inode,
372 const char *suffix, void *value, size_t size)
373{
374 if (value) {
375 if (dentry->d_name.len + 1 > size)
376 return -ERANGE;
377 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
378 }
379 return dentry->d_name.len + 1;
380}
381
382#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
383#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
384#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
385
386static const struct xattr_handler sockfs_xattr_handler = {
387 .name = XATTR_NAME_SOCKPROTONAME,
388 .get = sockfs_xattr_get,
389};
390
391static int sockfs_security_xattr_set(const struct xattr_handler *handler,
392 struct mnt_idmap *idmap,
393 struct dentry *dentry, struct inode *inode,
394 const char *suffix, const void *value,
395 size_t size, int flags)
396{
397 /* Handled by LSM. */
398 return -EAGAIN;
399}
400
401static const struct xattr_handler sockfs_security_xattr_handler = {
402 .prefix = XATTR_SECURITY_PREFIX,
403 .set = sockfs_security_xattr_set,
404};
405
406static const struct xattr_handler * const sockfs_xattr_handlers[] = {
407 &sockfs_xattr_handler,
408 &sockfs_security_xattr_handler,
409 NULL
410};
411
412static int sockfs_init_fs_context(struct fs_context *fc)
413{
414 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
415 if (!ctx)
416 return -ENOMEM;
417 ctx->ops = &sockfs_ops;
418 ctx->dops = &sockfs_dentry_operations;
419 ctx->xattr = sockfs_xattr_handlers;
420 return 0;
421}
422
423static struct vfsmount *sock_mnt __read_mostly;
424
425static struct file_system_type sock_fs_type = {
426 .name = "sockfs",
427 .init_fs_context = sockfs_init_fs_context,
428 .kill_sb = kill_anon_super,
429};
430
431/*
432 * Obtains the first available file descriptor and sets it up for use.
433 *
434 * These functions create file structures and maps them to fd space
435 * of the current process. On success it returns file descriptor
436 * and file struct implicitly stored in sock->file.
437 * Note that another thread may close file descriptor before we return
438 * from this function. We use the fact that now we do not refer
439 * to socket after mapping. If one day we will need it, this
440 * function will increment ref. count on file by 1.
441 *
442 * In any case returned fd MAY BE not valid!
443 * This race condition is unavoidable
444 * with shared fd spaces, we cannot solve it inside kernel,
445 * but we take care of internal coherence yet.
446 */
447
448/**
449 * sock_alloc_file - Bind a &socket to a &file
450 * @sock: socket
451 * @flags: file status flags
452 * @dname: protocol name
453 *
454 * Returns the &file bound with @sock, implicitly storing it
455 * in sock->file. If dname is %NULL, sets to "".
456 *
457 * On failure @sock is released, and an ERR pointer is returned.
458 *
459 * This function uses GFP_KERNEL internally.
460 */
461
462struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
463{
464 struct file *file;
465
466 if (!dname)
467 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
468
469 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
470 O_RDWR | (flags & O_NONBLOCK),
471 &socket_file_ops);
472 if (IS_ERR(file)) {
473 sock_release(sock);
474 return file;
475 }
476
477 file->f_mode |= FMODE_NOWAIT;
478 sock->file = file;
479 file->private_data = sock;
480 stream_open(SOCK_INODE(sock), file);
481 return file;
482}
483EXPORT_SYMBOL(sock_alloc_file);
484
485static int sock_map_fd(struct socket *sock, int flags)
486{
487 struct file *newfile;
488 int fd = get_unused_fd_flags(flags);
489 if (unlikely(fd < 0)) {
490 sock_release(sock);
491 return fd;
492 }
493
494 newfile = sock_alloc_file(sock, flags, NULL);
495 if (!IS_ERR(newfile)) {
496 fd_install(fd, newfile);
497 return fd;
498 }
499
500 put_unused_fd(fd);
501 return PTR_ERR(newfile);
502}
503
504/**
505 * sock_from_file - Return the &socket bounded to @file.
506 * @file: file
507 *
508 * On failure returns %NULL.
509 */
510
511struct socket *sock_from_file(struct file *file)
512{
513 if (file->f_op == &socket_file_ops)
514 return file->private_data; /* set in sock_alloc_file */
515
516 return NULL;
517}
518EXPORT_SYMBOL(sock_from_file);
519
520/**
521 * sockfd_lookup - Go from a file number to its socket slot
522 * @fd: file handle
523 * @err: pointer to an error code return
524 *
525 * The file handle passed in is locked and the socket it is bound
526 * to is returned. If an error occurs the err pointer is overwritten
527 * with a negative errno code and NULL is returned. The function checks
528 * for both invalid handles and passing a handle which is not a socket.
529 *
530 * On a success the socket object pointer is returned.
531 */
532
533struct socket *sockfd_lookup(int fd, int *err)
534{
535 struct file *file;
536 struct socket *sock;
537
538 file = fget(fd);
539 if (!file) {
540 *err = -EBADF;
541 return NULL;
542 }
543
544 sock = sock_from_file(file);
545 if (!sock) {
546 *err = -ENOTSOCK;
547 fput(file);
548 }
549 return sock;
550}
551EXPORT_SYMBOL(sockfd_lookup);
552
553static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
554{
555 struct fd f = fdget(fd);
556 struct socket *sock;
557
558 *err = -EBADF;
559 if (f.file) {
560 sock = sock_from_file(f.file);
561 if (likely(sock)) {
562 *fput_needed = f.flags & FDPUT_FPUT;
563 return sock;
564 }
565 *err = -ENOTSOCK;
566 fdput(f);
567 }
568 return NULL;
569}
570
571static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
572 size_t size)
573{
574 ssize_t len;
575 ssize_t used = 0;
576
577 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
578 if (len < 0)
579 return len;
580 used += len;
581 if (buffer) {
582 if (size < used)
583 return -ERANGE;
584 buffer += len;
585 }
586
587 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
588 used += len;
589 if (buffer) {
590 if (size < used)
591 return -ERANGE;
592 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
593 buffer += len;
594 }
595
596 return used;
597}
598
599static int sockfs_setattr(struct mnt_idmap *idmap,
600 struct dentry *dentry, struct iattr *iattr)
601{
602 int err = simple_setattr(&nop_mnt_idmap, dentry, iattr);
603
604 if (!err && (iattr->ia_valid & ATTR_UID)) {
605 struct socket *sock = SOCKET_I(d_inode(dentry));
606
607 if (sock->sk)
608 sock->sk->sk_uid = iattr->ia_uid;
609 else
610 err = -ENOENT;
611 }
612
613 return err;
614}
615
616static const struct inode_operations sockfs_inode_ops = {
617 .listxattr = sockfs_listxattr,
618 .setattr = sockfs_setattr,
619};
620
621/**
622 * sock_alloc - allocate a socket
623 *
624 * Allocate a new inode and socket object. The two are bound together
625 * and initialised. The socket is then returned. If we are out of inodes
626 * NULL is returned. This functions uses GFP_KERNEL internally.
627 */
628
629struct socket *sock_alloc(void)
630{
631 struct inode *inode;
632 struct socket *sock;
633
634 inode = new_inode_pseudo(sock_mnt->mnt_sb);
635 if (!inode)
636 return NULL;
637
638 sock = SOCKET_I(inode);
639
640 inode->i_ino = get_next_ino();
641 inode->i_mode = S_IFSOCK | S_IRWXUGO;
642 inode->i_uid = current_fsuid();
643 inode->i_gid = current_fsgid();
644 inode->i_op = &sockfs_inode_ops;
645
646 return sock;
647}
648EXPORT_SYMBOL(sock_alloc);
649
650static void __sock_release(struct socket *sock, struct inode *inode)
651{
652 const struct proto_ops *ops = READ_ONCE(sock->ops);
653
654 if (ops) {
655 struct module *owner = ops->owner;
656
657 if (inode)
658 inode_lock(inode);
659 ops->release(sock);
660 sock->sk = NULL;
661 if (inode)
662 inode_unlock(inode);
663 sock->ops = NULL;
664 module_put(owner);
665 }
666
667 if (sock->wq.fasync_list)
668 pr_err("%s: fasync list not empty!\n", __func__);
669
670 if (!sock->file) {
671 iput(SOCK_INODE(sock));
672 return;
673 }
674 sock->file = NULL;
675}
676
677/**
678 * sock_release - close a socket
679 * @sock: socket to close
680 *
681 * The socket is released from the protocol stack if it has a release
682 * callback, and the inode is then released if the socket is bound to
683 * an inode not a file.
684 */
685void sock_release(struct socket *sock)
686{
687 __sock_release(sock, NULL);
688}
689EXPORT_SYMBOL(sock_release);
690
691void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
692{
693 u8 flags = *tx_flags;
694
695 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE) {
696 flags |= SKBTX_HW_TSTAMP;
697
698 /* PTP hardware clocks can provide a free running cycle counter
699 * as a time base for virtual clocks. Tell driver to use the
700 * free running cycle counter for timestamp if socket is bound
701 * to virtual clock.
702 */
703 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
704 flags |= SKBTX_HW_TSTAMP_USE_CYCLES;
705 }
706
707 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
708 flags |= SKBTX_SW_TSTAMP;
709
710 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
711 flags |= SKBTX_SCHED_TSTAMP;
712
713 *tx_flags = flags;
714}
715EXPORT_SYMBOL(__sock_tx_timestamp);
716
717INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
718 size_t));
719INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
720 size_t));
721
722static noinline void call_trace_sock_send_length(struct sock *sk, int ret,
723 int flags)
724{
725 trace_sock_send_length(sk, ret, 0);
726}
727
728static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
729{
730 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->sendmsg, inet6_sendmsg,
731 inet_sendmsg, sock, msg,
732 msg_data_left(msg));
733 BUG_ON(ret == -EIOCBQUEUED);
734
735 if (trace_sock_send_length_enabled())
736 call_trace_sock_send_length(sock->sk, ret, 0);
737 return ret;
738}
739
740static int __sock_sendmsg(struct socket *sock, struct msghdr *msg)
741{
742 int err = security_socket_sendmsg(sock, msg,
743 msg_data_left(msg));
744
745 return err ?: sock_sendmsg_nosec(sock, msg);
746}
747
748/**
749 * sock_sendmsg - send a message through @sock
750 * @sock: socket
751 * @msg: message to send
752 *
753 * Sends @msg through @sock, passing through LSM.
754 * Returns the number of bytes sent, or an error code.
755 */
756int sock_sendmsg(struct socket *sock, struct msghdr *msg)
757{
758 struct sockaddr_storage *save_addr = (struct sockaddr_storage *)msg->msg_name;
759 struct sockaddr_storage address;
760 int save_len = msg->msg_namelen;
761 int ret;
762
763 if (msg->msg_name) {
764 memcpy(&address, msg->msg_name, msg->msg_namelen);
765 msg->msg_name = &address;
766 }
767
768 ret = __sock_sendmsg(sock, msg);
769 msg->msg_name = save_addr;
770 msg->msg_namelen = save_len;
771
772 return ret;
773}
774EXPORT_SYMBOL(sock_sendmsg);
775
776/**
777 * kernel_sendmsg - send a message through @sock (kernel-space)
778 * @sock: socket
779 * @msg: message header
780 * @vec: kernel vec
781 * @num: vec array length
782 * @size: total message data size
783 *
784 * Builds the message data with @vec and sends it through @sock.
785 * Returns the number of bytes sent, or an error code.
786 */
787
788int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
789 struct kvec *vec, size_t num, size_t size)
790{
791 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
792 return sock_sendmsg(sock, msg);
793}
794EXPORT_SYMBOL(kernel_sendmsg);
795
796/**
797 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
798 * @sk: sock
799 * @msg: message header
800 * @vec: output s/g array
801 * @num: output s/g array length
802 * @size: total message data size
803 *
804 * Builds the message data with @vec and sends it through @sock.
805 * Returns the number of bytes sent, or an error code.
806 * Caller must hold @sk.
807 */
808
809int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
810 struct kvec *vec, size_t num, size_t size)
811{
812 struct socket *sock = sk->sk_socket;
813 const struct proto_ops *ops = READ_ONCE(sock->ops);
814
815 if (!ops->sendmsg_locked)
816 return sock_no_sendmsg_locked(sk, msg, size);
817
818 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
819
820 return ops->sendmsg_locked(sk, msg, msg_data_left(msg));
821}
822EXPORT_SYMBOL(kernel_sendmsg_locked);
823
824static bool skb_is_err_queue(const struct sk_buff *skb)
825{
826 /* pkt_type of skbs enqueued on the error queue are set to
827 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
828 * in recvmsg, since skbs received on a local socket will never
829 * have a pkt_type of PACKET_OUTGOING.
830 */
831 return skb->pkt_type == PACKET_OUTGOING;
832}
833
834/* On transmit, software and hardware timestamps are returned independently.
835 * As the two skb clones share the hardware timestamp, which may be updated
836 * before the software timestamp is received, a hardware TX timestamp may be
837 * returned only if there is no software TX timestamp. Ignore false software
838 * timestamps, which may be made in the __sock_recv_timestamp() call when the
839 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
840 * hardware timestamp.
841 */
842static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
843{
844 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
845}
846
847static ktime_t get_timestamp(struct sock *sk, struct sk_buff *skb, int *if_index)
848{
849 bool cycles = READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_BIND_PHC;
850 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
851 struct net_device *orig_dev;
852 ktime_t hwtstamp;
853
854 rcu_read_lock();
855 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
856 if (orig_dev) {
857 *if_index = orig_dev->ifindex;
858 hwtstamp = netdev_get_tstamp(orig_dev, shhwtstamps, cycles);
859 } else {
860 hwtstamp = shhwtstamps->hwtstamp;
861 }
862 rcu_read_unlock();
863
864 return hwtstamp;
865}
866
867static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb,
868 int if_index)
869{
870 struct scm_ts_pktinfo ts_pktinfo;
871 struct net_device *orig_dev;
872
873 if (!skb_mac_header_was_set(skb))
874 return;
875
876 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
877
878 if (!if_index) {
879 rcu_read_lock();
880 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
881 if (orig_dev)
882 if_index = orig_dev->ifindex;
883 rcu_read_unlock();
884 }
885 ts_pktinfo.if_index = if_index;
886
887 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
888 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
889 sizeof(ts_pktinfo), &ts_pktinfo);
890}
891
892/*
893 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
894 */
895void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
896 struct sk_buff *skb)
897{
898 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
899 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
900 struct scm_timestamping_internal tss;
901 int empty = 1, false_tstamp = 0;
902 struct skb_shared_hwtstamps *shhwtstamps =
903 skb_hwtstamps(skb);
904 int if_index;
905 ktime_t hwtstamp;
906 u32 tsflags;
907
908 /* Race occurred between timestamp enabling and packet
909 receiving. Fill in the current time for now. */
910 if (need_software_tstamp && skb->tstamp == 0) {
911 __net_timestamp(skb);
912 false_tstamp = 1;
913 }
914
915 if (need_software_tstamp) {
916 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
917 if (new_tstamp) {
918 struct __kernel_sock_timeval tv;
919
920 skb_get_new_timestamp(skb, &tv);
921 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
922 sizeof(tv), &tv);
923 } else {
924 struct __kernel_old_timeval tv;
925
926 skb_get_timestamp(skb, &tv);
927 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
928 sizeof(tv), &tv);
929 }
930 } else {
931 if (new_tstamp) {
932 struct __kernel_timespec ts;
933
934 skb_get_new_timestampns(skb, &ts);
935 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
936 sizeof(ts), &ts);
937 } else {
938 struct __kernel_old_timespec ts;
939
940 skb_get_timestampns(skb, &ts);
941 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
942 sizeof(ts), &ts);
943 }
944 }
945 }
946
947 memset(&tss, 0, sizeof(tss));
948 tsflags = READ_ONCE(sk->sk_tsflags);
949 if ((tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
950 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
951 empty = 0;
952 if (shhwtstamps &&
953 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
954 !skb_is_swtx_tstamp(skb, false_tstamp)) {
955 if_index = 0;
956 if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP_NETDEV)
957 hwtstamp = get_timestamp(sk, skb, &if_index);
958 else
959 hwtstamp = shhwtstamps->hwtstamp;
960
961 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
962 hwtstamp = ptp_convert_timestamp(&hwtstamp,
963 READ_ONCE(sk->sk_bind_phc));
964
965 if (ktime_to_timespec64_cond(hwtstamp, tss.ts + 2)) {
966 empty = 0;
967
968 if ((tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
969 !skb_is_err_queue(skb))
970 put_ts_pktinfo(msg, skb, if_index);
971 }
972 }
973 if (!empty) {
974 if (sock_flag(sk, SOCK_TSTAMP_NEW))
975 put_cmsg_scm_timestamping64(msg, &tss);
976 else
977 put_cmsg_scm_timestamping(msg, &tss);
978
979 if (skb_is_err_queue(skb) && skb->len &&
980 SKB_EXT_ERR(skb)->opt_stats)
981 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
982 skb->len, skb->data);
983 }
984}
985EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
986
987#ifdef CONFIG_WIRELESS
988void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
989 struct sk_buff *skb)
990{
991 int ack;
992
993 if (!sock_flag(sk, SOCK_WIFI_STATUS))
994 return;
995 if (!skb->wifi_acked_valid)
996 return;
997
998 ack = skb->wifi_acked;
999
1000 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
1001}
1002EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
1003#endif
1004
1005static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
1006 struct sk_buff *skb)
1007{
1008 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
1009 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
1010 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
1011}
1012
1013static void sock_recv_mark(struct msghdr *msg, struct sock *sk,
1014 struct sk_buff *skb)
1015{
1016 if (sock_flag(sk, SOCK_RCVMARK) && skb) {
1017 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */
1018 __u32 mark = skb->mark;
1019
1020 put_cmsg(msg, SOL_SOCKET, SO_MARK, sizeof(__u32), &mark);
1021 }
1022}
1023
1024void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
1025 struct sk_buff *skb)
1026{
1027 sock_recv_timestamp(msg, sk, skb);
1028 sock_recv_drops(msg, sk, skb);
1029 sock_recv_mark(msg, sk, skb);
1030}
1031EXPORT_SYMBOL_GPL(__sock_recv_cmsgs);
1032
1033INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
1034 size_t, int));
1035INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
1036 size_t, int));
1037
1038static noinline void call_trace_sock_recv_length(struct sock *sk, int ret, int flags)
1039{
1040 trace_sock_recv_length(sk, ret, flags);
1041}
1042
1043static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
1044 int flags)
1045{
1046 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->recvmsg,
1047 inet6_recvmsg,
1048 inet_recvmsg, sock, msg,
1049 msg_data_left(msg), flags);
1050 if (trace_sock_recv_length_enabled())
1051 call_trace_sock_recv_length(sock->sk, ret, flags);
1052 return ret;
1053}
1054
1055/**
1056 * sock_recvmsg - receive a message from @sock
1057 * @sock: socket
1058 * @msg: message to receive
1059 * @flags: message flags
1060 *
1061 * Receives @msg from @sock, passing through LSM. Returns the total number
1062 * of bytes received, or an error.
1063 */
1064int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
1065{
1066 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
1067
1068 return err ?: sock_recvmsg_nosec(sock, msg, flags);
1069}
1070EXPORT_SYMBOL(sock_recvmsg);
1071
1072/**
1073 * kernel_recvmsg - Receive a message from a socket (kernel space)
1074 * @sock: The socket to receive the message from
1075 * @msg: Received message
1076 * @vec: Input s/g array for message data
1077 * @num: Size of input s/g array
1078 * @size: Number of bytes to read
1079 * @flags: Message flags (MSG_DONTWAIT, etc...)
1080 *
1081 * On return the msg structure contains the scatter/gather array passed in the
1082 * vec argument. The array is modified so that it consists of the unfilled
1083 * portion of the original array.
1084 *
1085 * The returned value is the total number of bytes received, or an error.
1086 */
1087
1088int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
1089 struct kvec *vec, size_t num, size_t size, int flags)
1090{
1091 msg->msg_control_is_user = false;
1092 iov_iter_kvec(&msg->msg_iter, ITER_DEST, vec, num, size);
1093 return sock_recvmsg(sock, msg, flags);
1094}
1095EXPORT_SYMBOL(kernel_recvmsg);
1096
1097static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1098 struct pipe_inode_info *pipe, size_t len,
1099 unsigned int flags)
1100{
1101 struct socket *sock = file->private_data;
1102 const struct proto_ops *ops;
1103
1104 ops = READ_ONCE(sock->ops);
1105 if (unlikely(!ops->splice_read))
1106 return copy_splice_read(file, ppos, pipe, len, flags);
1107
1108 return ops->splice_read(sock, ppos, pipe, len, flags);
1109}
1110
1111static void sock_splice_eof(struct file *file)
1112{
1113 struct socket *sock = file->private_data;
1114 const struct proto_ops *ops;
1115
1116 ops = READ_ONCE(sock->ops);
1117 if (ops->splice_eof)
1118 ops->splice_eof(sock);
1119}
1120
1121static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1122{
1123 struct file *file = iocb->ki_filp;
1124 struct socket *sock = file->private_data;
1125 struct msghdr msg = {.msg_iter = *to,
1126 .msg_iocb = iocb};
1127 ssize_t res;
1128
1129 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1130 msg.msg_flags = MSG_DONTWAIT;
1131
1132 if (iocb->ki_pos != 0)
1133 return -ESPIPE;
1134
1135 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1136 return 0;
1137
1138 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1139 *to = msg.msg_iter;
1140 return res;
1141}
1142
1143static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1144{
1145 struct file *file = iocb->ki_filp;
1146 struct socket *sock = file->private_data;
1147 struct msghdr msg = {.msg_iter = *from,
1148 .msg_iocb = iocb};
1149 ssize_t res;
1150
1151 if (iocb->ki_pos != 0)
1152 return -ESPIPE;
1153
1154 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1155 msg.msg_flags = MSG_DONTWAIT;
1156
1157 if (sock->type == SOCK_SEQPACKET)
1158 msg.msg_flags |= MSG_EOR;
1159
1160 res = __sock_sendmsg(sock, &msg);
1161 *from = msg.msg_iter;
1162 return res;
1163}
1164
1165/*
1166 * Atomic setting of ioctl hooks to avoid race
1167 * with module unload.
1168 */
1169
1170static DEFINE_MUTEX(br_ioctl_mutex);
1171static int (*br_ioctl_hook)(struct net *net, struct net_bridge *br,
1172 unsigned int cmd, struct ifreq *ifr,
1173 void __user *uarg);
1174
1175void brioctl_set(int (*hook)(struct net *net, struct net_bridge *br,
1176 unsigned int cmd, struct ifreq *ifr,
1177 void __user *uarg))
1178{
1179 mutex_lock(&br_ioctl_mutex);
1180 br_ioctl_hook = hook;
1181 mutex_unlock(&br_ioctl_mutex);
1182}
1183EXPORT_SYMBOL(brioctl_set);
1184
1185int br_ioctl_call(struct net *net, struct net_bridge *br, unsigned int cmd,
1186 struct ifreq *ifr, void __user *uarg)
1187{
1188 int err = -ENOPKG;
1189
1190 if (!br_ioctl_hook)
1191 request_module("bridge");
1192
1193 mutex_lock(&br_ioctl_mutex);
1194 if (br_ioctl_hook)
1195 err = br_ioctl_hook(net, br, cmd, ifr, uarg);
1196 mutex_unlock(&br_ioctl_mutex);
1197
1198 return err;
1199}
1200
1201static DEFINE_MUTEX(vlan_ioctl_mutex);
1202static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1203
1204void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1205{
1206 mutex_lock(&vlan_ioctl_mutex);
1207 vlan_ioctl_hook = hook;
1208 mutex_unlock(&vlan_ioctl_mutex);
1209}
1210EXPORT_SYMBOL(vlan_ioctl_set);
1211
1212static long sock_do_ioctl(struct net *net, struct socket *sock,
1213 unsigned int cmd, unsigned long arg)
1214{
1215 const struct proto_ops *ops = READ_ONCE(sock->ops);
1216 struct ifreq ifr;
1217 bool need_copyout;
1218 int err;
1219 void __user *argp = (void __user *)arg;
1220 void __user *data;
1221
1222 err = ops->ioctl(sock, cmd, arg);
1223
1224 /*
1225 * If this ioctl is unknown try to hand it down
1226 * to the NIC driver.
1227 */
1228 if (err != -ENOIOCTLCMD)
1229 return err;
1230
1231 if (!is_socket_ioctl_cmd(cmd))
1232 return -ENOTTY;
1233
1234 if (get_user_ifreq(&ifr, &data, argp))
1235 return -EFAULT;
1236 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1237 if (!err && need_copyout)
1238 if (put_user_ifreq(&ifr, argp))
1239 return -EFAULT;
1240
1241 return err;
1242}
1243
1244/*
1245 * With an ioctl, arg may well be a user mode pointer, but we don't know
1246 * what to do with it - that's up to the protocol still.
1247 */
1248
1249static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1250{
1251 const struct proto_ops *ops;
1252 struct socket *sock;
1253 struct sock *sk;
1254 void __user *argp = (void __user *)arg;
1255 int pid, err;
1256 struct net *net;
1257
1258 sock = file->private_data;
1259 ops = READ_ONCE(sock->ops);
1260 sk = sock->sk;
1261 net = sock_net(sk);
1262 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1263 struct ifreq ifr;
1264 void __user *data;
1265 bool need_copyout;
1266 if (get_user_ifreq(&ifr, &data, argp))
1267 return -EFAULT;
1268 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1269 if (!err && need_copyout)
1270 if (put_user_ifreq(&ifr, argp))
1271 return -EFAULT;
1272 } else
1273#ifdef CONFIG_WEXT_CORE
1274 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1275 err = wext_handle_ioctl(net, cmd, argp);
1276 } else
1277#endif
1278 switch (cmd) {
1279 case FIOSETOWN:
1280 case SIOCSPGRP:
1281 err = -EFAULT;
1282 if (get_user(pid, (int __user *)argp))
1283 break;
1284 err = f_setown(sock->file, pid, 1);
1285 break;
1286 case FIOGETOWN:
1287 case SIOCGPGRP:
1288 err = put_user(f_getown(sock->file),
1289 (int __user *)argp);
1290 break;
1291 case SIOCGIFBR:
1292 case SIOCSIFBR:
1293 case SIOCBRADDBR:
1294 case SIOCBRDELBR:
1295 err = br_ioctl_call(net, NULL, cmd, NULL, argp);
1296 break;
1297 case SIOCGIFVLAN:
1298 case SIOCSIFVLAN:
1299 err = -ENOPKG;
1300 if (!vlan_ioctl_hook)
1301 request_module("8021q");
1302
1303 mutex_lock(&vlan_ioctl_mutex);
1304 if (vlan_ioctl_hook)
1305 err = vlan_ioctl_hook(net, argp);
1306 mutex_unlock(&vlan_ioctl_mutex);
1307 break;
1308 case SIOCGSKNS:
1309 err = -EPERM;
1310 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1311 break;
1312
1313 err = open_related_ns(&net->ns, get_net_ns);
1314 break;
1315 case SIOCGSTAMP_OLD:
1316 case SIOCGSTAMPNS_OLD:
1317 if (!ops->gettstamp) {
1318 err = -ENOIOCTLCMD;
1319 break;
1320 }
1321 err = ops->gettstamp(sock, argp,
1322 cmd == SIOCGSTAMP_OLD,
1323 !IS_ENABLED(CONFIG_64BIT));
1324 break;
1325 case SIOCGSTAMP_NEW:
1326 case SIOCGSTAMPNS_NEW:
1327 if (!ops->gettstamp) {
1328 err = -ENOIOCTLCMD;
1329 break;
1330 }
1331 err = ops->gettstamp(sock, argp,
1332 cmd == SIOCGSTAMP_NEW,
1333 false);
1334 break;
1335
1336 case SIOCGIFCONF:
1337 err = dev_ifconf(net, argp);
1338 break;
1339
1340 default:
1341 err = sock_do_ioctl(net, sock, cmd, arg);
1342 break;
1343 }
1344 return err;
1345}
1346
1347/**
1348 * sock_create_lite - creates a socket
1349 * @family: protocol family (AF_INET, ...)
1350 * @type: communication type (SOCK_STREAM, ...)
1351 * @protocol: protocol (0, ...)
1352 * @res: new socket
1353 *
1354 * Creates a new socket and assigns it to @res, passing through LSM.
1355 * The new socket initialization is not complete, see kernel_accept().
1356 * Returns 0 or an error. On failure @res is set to %NULL.
1357 * This function internally uses GFP_KERNEL.
1358 */
1359
1360int sock_create_lite(int family, int type, int protocol, struct socket **res)
1361{
1362 int err;
1363 struct socket *sock = NULL;
1364
1365 err = security_socket_create(family, type, protocol, 1);
1366 if (err)
1367 goto out;
1368
1369 sock = sock_alloc();
1370 if (!sock) {
1371 err = -ENOMEM;
1372 goto out;
1373 }
1374
1375 sock->type = type;
1376 err = security_socket_post_create(sock, family, type, protocol, 1);
1377 if (err)
1378 goto out_release;
1379
1380out:
1381 *res = sock;
1382 return err;
1383out_release:
1384 sock_release(sock);
1385 sock = NULL;
1386 goto out;
1387}
1388EXPORT_SYMBOL(sock_create_lite);
1389
1390/* No kernel lock held - perfect */
1391static __poll_t sock_poll(struct file *file, poll_table *wait)
1392{
1393 struct socket *sock = file->private_data;
1394 const struct proto_ops *ops = READ_ONCE(sock->ops);
1395 __poll_t events = poll_requested_events(wait), flag = 0;
1396
1397 if (!ops->poll)
1398 return 0;
1399
1400 if (sk_can_busy_loop(sock->sk)) {
1401 /* poll once if requested by the syscall */
1402 if (events & POLL_BUSY_LOOP)
1403 sk_busy_loop(sock->sk, 1);
1404
1405 /* if this socket can poll_ll, tell the system call */
1406 flag = POLL_BUSY_LOOP;
1407 }
1408
1409 return ops->poll(file, sock, wait) | flag;
1410}
1411
1412static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1413{
1414 struct socket *sock = file->private_data;
1415
1416 return READ_ONCE(sock->ops)->mmap(file, sock, vma);
1417}
1418
1419static int sock_close(struct inode *inode, struct file *filp)
1420{
1421 __sock_release(SOCKET_I(inode), inode);
1422 return 0;
1423}
1424
1425/*
1426 * Update the socket async list
1427 *
1428 * Fasync_list locking strategy.
1429 *
1430 * 1. fasync_list is modified only under process context socket lock
1431 * i.e. under semaphore.
1432 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1433 * or under socket lock
1434 */
1435
1436static int sock_fasync(int fd, struct file *filp, int on)
1437{
1438 struct socket *sock = filp->private_data;
1439 struct sock *sk = sock->sk;
1440 struct socket_wq *wq = &sock->wq;
1441
1442 if (sk == NULL)
1443 return -EINVAL;
1444
1445 lock_sock(sk);
1446 fasync_helper(fd, filp, on, &wq->fasync_list);
1447
1448 if (!wq->fasync_list)
1449 sock_reset_flag(sk, SOCK_FASYNC);
1450 else
1451 sock_set_flag(sk, SOCK_FASYNC);
1452
1453 release_sock(sk);
1454 return 0;
1455}
1456
1457/* This function may be called only under rcu_lock */
1458
1459int sock_wake_async(struct socket_wq *wq, int how, int band)
1460{
1461 if (!wq || !wq->fasync_list)
1462 return -1;
1463
1464 switch (how) {
1465 case SOCK_WAKE_WAITD:
1466 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1467 break;
1468 goto call_kill;
1469 case SOCK_WAKE_SPACE:
1470 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1471 break;
1472 fallthrough;
1473 case SOCK_WAKE_IO:
1474call_kill:
1475 kill_fasync(&wq->fasync_list, SIGIO, band);
1476 break;
1477 case SOCK_WAKE_URG:
1478 kill_fasync(&wq->fasync_list, SIGURG, band);
1479 }
1480
1481 return 0;
1482}
1483EXPORT_SYMBOL(sock_wake_async);
1484
1485/**
1486 * __sock_create - creates a socket
1487 * @net: net namespace
1488 * @family: protocol family (AF_INET, ...)
1489 * @type: communication type (SOCK_STREAM, ...)
1490 * @protocol: protocol (0, ...)
1491 * @res: new socket
1492 * @kern: boolean for kernel space sockets
1493 *
1494 * Creates a new socket and assigns it to @res, passing through LSM.
1495 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1496 * be set to true if the socket resides in kernel space.
1497 * This function internally uses GFP_KERNEL.
1498 */
1499
1500int __sock_create(struct net *net, int family, int type, int protocol,
1501 struct socket **res, int kern)
1502{
1503 int err;
1504 struct socket *sock;
1505 const struct net_proto_family *pf;
1506
1507 /*
1508 * Check protocol is in range
1509 */
1510 if (family < 0 || family >= NPROTO)
1511 return -EAFNOSUPPORT;
1512 if (type < 0 || type >= SOCK_MAX)
1513 return -EINVAL;
1514
1515 /* Compatibility.
1516
1517 This uglymoron is moved from INET layer to here to avoid
1518 deadlock in module load.
1519 */
1520 if (family == PF_INET && type == SOCK_PACKET) {
1521 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1522 current->comm);
1523 family = PF_PACKET;
1524 }
1525
1526 err = security_socket_create(family, type, protocol, kern);
1527 if (err)
1528 return err;
1529
1530 /*
1531 * Allocate the socket and allow the family to set things up. if
1532 * the protocol is 0, the family is instructed to select an appropriate
1533 * default.
1534 */
1535 sock = sock_alloc();
1536 if (!sock) {
1537 net_warn_ratelimited("socket: no more sockets\n");
1538 return -ENFILE; /* Not exactly a match, but its the
1539 closest posix thing */
1540 }
1541
1542 sock->type = type;
1543
1544#ifdef CONFIG_MODULES
1545 /* Attempt to load a protocol module if the find failed.
1546 *
1547 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1548 * requested real, full-featured networking support upon configuration.
1549 * Otherwise module support will break!
1550 */
1551 if (rcu_access_pointer(net_families[family]) == NULL)
1552 request_module("net-pf-%d", family);
1553#endif
1554
1555 rcu_read_lock();
1556 pf = rcu_dereference(net_families[family]);
1557 err = -EAFNOSUPPORT;
1558 if (!pf)
1559 goto out_release;
1560
1561 /*
1562 * We will call the ->create function, that possibly is in a loadable
1563 * module, so we have to bump that loadable module refcnt first.
1564 */
1565 if (!try_module_get(pf->owner))
1566 goto out_release;
1567
1568 /* Now protected by module ref count */
1569 rcu_read_unlock();
1570
1571 err = pf->create(net, sock, protocol, kern);
1572 if (err < 0)
1573 goto out_module_put;
1574
1575 /*
1576 * Now to bump the refcnt of the [loadable] module that owns this
1577 * socket at sock_release time we decrement its refcnt.
1578 */
1579 if (!try_module_get(sock->ops->owner))
1580 goto out_module_busy;
1581
1582 /*
1583 * Now that we're done with the ->create function, the [loadable]
1584 * module can have its refcnt decremented
1585 */
1586 module_put(pf->owner);
1587 err = security_socket_post_create(sock, family, type, protocol, kern);
1588 if (err)
1589 goto out_sock_release;
1590 *res = sock;
1591
1592 return 0;
1593
1594out_module_busy:
1595 err = -EAFNOSUPPORT;
1596out_module_put:
1597 sock->ops = NULL;
1598 module_put(pf->owner);
1599out_sock_release:
1600 sock_release(sock);
1601 return err;
1602
1603out_release:
1604 rcu_read_unlock();
1605 goto out_sock_release;
1606}
1607EXPORT_SYMBOL(__sock_create);
1608
1609/**
1610 * sock_create - creates a socket
1611 * @family: protocol family (AF_INET, ...)
1612 * @type: communication type (SOCK_STREAM, ...)
1613 * @protocol: protocol (0, ...)
1614 * @res: new socket
1615 *
1616 * A wrapper around __sock_create().
1617 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1618 */
1619
1620int sock_create(int family, int type, int protocol, struct socket **res)
1621{
1622 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1623}
1624EXPORT_SYMBOL(sock_create);
1625
1626/**
1627 * sock_create_kern - creates a socket (kernel space)
1628 * @net: net namespace
1629 * @family: protocol family (AF_INET, ...)
1630 * @type: communication type (SOCK_STREAM, ...)
1631 * @protocol: protocol (0, ...)
1632 * @res: new socket
1633 *
1634 * A wrapper around __sock_create().
1635 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1636 */
1637
1638int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1639{
1640 return __sock_create(net, family, type, protocol, res, 1);
1641}
1642EXPORT_SYMBOL(sock_create_kern);
1643
1644static struct socket *__sys_socket_create(int family, int type, int protocol)
1645{
1646 struct socket *sock;
1647 int retval;
1648
1649 /* Check the SOCK_* constants for consistency. */
1650 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1651 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1652 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1653 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1654
1655 if ((type & ~SOCK_TYPE_MASK) & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1656 return ERR_PTR(-EINVAL);
1657 type &= SOCK_TYPE_MASK;
1658
1659 retval = sock_create(family, type, protocol, &sock);
1660 if (retval < 0)
1661 return ERR_PTR(retval);
1662
1663 return sock;
1664}
1665
1666struct file *__sys_socket_file(int family, int type, int protocol)
1667{
1668 struct socket *sock;
1669 int flags;
1670
1671 sock = __sys_socket_create(family, type, protocol);
1672 if (IS_ERR(sock))
1673 return ERR_CAST(sock);
1674
1675 flags = type & ~SOCK_TYPE_MASK;
1676 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1677 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1678
1679 return sock_alloc_file(sock, flags, NULL);
1680}
1681
1682/* A hook for bpf progs to attach to and update socket protocol.
1683 *
1684 * A static noinline declaration here could cause the compiler to
1685 * optimize away the function. A global noinline declaration will
1686 * keep the definition, but may optimize away the callsite.
1687 * Therefore, __weak is needed to ensure that the call is still
1688 * emitted, by telling the compiler that we don't know what the
1689 * function might eventually be.
1690 */
1691
1692__bpf_hook_start();
1693
1694__weak noinline int update_socket_protocol(int family, int type, int protocol)
1695{
1696 return protocol;
1697}
1698
1699__bpf_hook_end();
1700
1701int __sys_socket(int family, int type, int protocol)
1702{
1703 struct socket *sock;
1704 int flags;
1705
1706 sock = __sys_socket_create(family, type,
1707 update_socket_protocol(family, type, protocol));
1708 if (IS_ERR(sock))
1709 return PTR_ERR(sock);
1710
1711 flags = type & ~SOCK_TYPE_MASK;
1712 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1713 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1714
1715 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1716}
1717
1718SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1719{
1720 return __sys_socket(family, type, protocol);
1721}
1722
1723/*
1724 * Create a pair of connected sockets.
1725 */
1726
1727int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1728{
1729 struct socket *sock1, *sock2;
1730 int fd1, fd2, err;
1731 struct file *newfile1, *newfile2;
1732 int flags;
1733
1734 flags = type & ~SOCK_TYPE_MASK;
1735 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1736 return -EINVAL;
1737 type &= SOCK_TYPE_MASK;
1738
1739 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1740 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1741
1742 /*
1743 * reserve descriptors and make sure we won't fail
1744 * to return them to userland.
1745 */
1746 fd1 = get_unused_fd_flags(flags);
1747 if (unlikely(fd1 < 0))
1748 return fd1;
1749
1750 fd2 = get_unused_fd_flags(flags);
1751 if (unlikely(fd2 < 0)) {
1752 put_unused_fd(fd1);
1753 return fd2;
1754 }
1755
1756 err = put_user(fd1, &usockvec[0]);
1757 if (err)
1758 goto out;
1759
1760 err = put_user(fd2, &usockvec[1]);
1761 if (err)
1762 goto out;
1763
1764 /*
1765 * Obtain the first socket and check if the underlying protocol
1766 * supports the socketpair call.
1767 */
1768
1769 err = sock_create(family, type, protocol, &sock1);
1770 if (unlikely(err < 0))
1771 goto out;
1772
1773 err = sock_create(family, type, protocol, &sock2);
1774 if (unlikely(err < 0)) {
1775 sock_release(sock1);
1776 goto out;
1777 }
1778
1779 err = security_socket_socketpair(sock1, sock2);
1780 if (unlikely(err)) {
1781 sock_release(sock2);
1782 sock_release(sock1);
1783 goto out;
1784 }
1785
1786 err = READ_ONCE(sock1->ops)->socketpair(sock1, sock2);
1787 if (unlikely(err < 0)) {
1788 sock_release(sock2);
1789 sock_release(sock1);
1790 goto out;
1791 }
1792
1793 newfile1 = sock_alloc_file(sock1, flags, NULL);
1794 if (IS_ERR(newfile1)) {
1795 err = PTR_ERR(newfile1);
1796 sock_release(sock2);
1797 goto out;
1798 }
1799
1800 newfile2 = sock_alloc_file(sock2, flags, NULL);
1801 if (IS_ERR(newfile2)) {
1802 err = PTR_ERR(newfile2);
1803 fput(newfile1);
1804 goto out;
1805 }
1806
1807 audit_fd_pair(fd1, fd2);
1808
1809 fd_install(fd1, newfile1);
1810 fd_install(fd2, newfile2);
1811 return 0;
1812
1813out:
1814 put_unused_fd(fd2);
1815 put_unused_fd(fd1);
1816 return err;
1817}
1818
1819SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1820 int __user *, usockvec)
1821{
1822 return __sys_socketpair(family, type, protocol, usockvec);
1823}
1824
1825/*
1826 * Bind a name to a socket. Nothing much to do here since it's
1827 * the protocol's responsibility to handle the local address.
1828 *
1829 * We move the socket address to kernel space before we call
1830 * the protocol layer (having also checked the address is ok).
1831 */
1832
1833int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1834{
1835 struct socket *sock;
1836 struct sockaddr_storage address;
1837 int err, fput_needed;
1838
1839 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1840 if (sock) {
1841 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1842 if (!err) {
1843 err = security_socket_bind(sock,
1844 (struct sockaddr *)&address,
1845 addrlen);
1846 if (!err)
1847 err = READ_ONCE(sock->ops)->bind(sock,
1848 (struct sockaddr *)
1849 &address, addrlen);
1850 }
1851 fput_light(sock->file, fput_needed);
1852 }
1853 return err;
1854}
1855
1856SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1857{
1858 return __sys_bind(fd, umyaddr, addrlen);
1859}
1860
1861/*
1862 * Perform a listen. Basically, we allow the protocol to do anything
1863 * necessary for a listen, and if that works, we mark the socket as
1864 * ready for listening.
1865 */
1866
1867int __sys_listen(int fd, int backlog)
1868{
1869 struct socket *sock;
1870 int err, fput_needed;
1871 int somaxconn;
1872
1873 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1874 if (sock) {
1875 somaxconn = READ_ONCE(sock_net(sock->sk)->core.sysctl_somaxconn);
1876 if ((unsigned int)backlog > somaxconn)
1877 backlog = somaxconn;
1878
1879 err = security_socket_listen(sock, backlog);
1880 if (!err)
1881 err = READ_ONCE(sock->ops)->listen(sock, backlog);
1882
1883 fput_light(sock->file, fput_needed);
1884 }
1885 return err;
1886}
1887
1888SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1889{
1890 return __sys_listen(fd, backlog);
1891}
1892
1893struct file *do_accept(struct file *file, unsigned file_flags,
1894 struct sockaddr __user *upeer_sockaddr,
1895 int __user *upeer_addrlen, int flags)
1896{
1897 struct socket *sock, *newsock;
1898 struct file *newfile;
1899 int err, len;
1900 struct sockaddr_storage address;
1901 const struct proto_ops *ops;
1902
1903 sock = sock_from_file(file);
1904 if (!sock)
1905 return ERR_PTR(-ENOTSOCK);
1906
1907 newsock = sock_alloc();
1908 if (!newsock)
1909 return ERR_PTR(-ENFILE);
1910 ops = READ_ONCE(sock->ops);
1911
1912 newsock->type = sock->type;
1913 newsock->ops = ops;
1914
1915 /*
1916 * We don't need try_module_get here, as the listening socket (sock)
1917 * has the protocol module (sock->ops->owner) held.
1918 */
1919 __module_get(ops->owner);
1920
1921 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1922 if (IS_ERR(newfile))
1923 return newfile;
1924
1925 err = security_socket_accept(sock, newsock);
1926 if (err)
1927 goto out_fd;
1928
1929 err = ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1930 false);
1931 if (err < 0)
1932 goto out_fd;
1933
1934 if (upeer_sockaddr) {
1935 len = ops->getname(newsock, (struct sockaddr *)&address, 2);
1936 if (len < 0) {
1937 err = -ECONNABORTED;
1938 goto out_fd;
1939 }
1940 err = move_addr_to_user(&address,
1941 len, upeer_sockaddr, upeer_addrlen);
1942 if (err < 0)
1943 goto out_fd;
1944 }
1945
1946 /* File flags are not inherited via accept() unlike another OSes. */
1947 return newfile;
1948out_fd:
1949 fput(newfile);
1950 return ERR_PTR(err);
1951}
1952
1953static int __sys_accept4_file(struct file *file, struct sockaddr __user *upeer_sockaddr,
1954 int __user *upeer_addrlen, int flags)
1955{
1956 struct file *newfile;
1957 int newfd;
1958
1959 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1960 return -EINVAL;
1961
1962 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1963 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1964
1965 newfd = get_unused_fd_flags(flags);
1966 if (unlikely(newfd < 0))
1967 return newfd;
1968
1969 newfile = do_accept(file, 0, upeer_sockaddr, upeer_addrlen,
1970 flags);
1971 if (IS_ERR(newfile)) {
1972 put_unused_fd(newfd);
1973 return PTR_ERR(newfile);
1974 }
1975 fd_install(newfd, newfile);
1976 return newfd;
1977}
1978
1979/*
1980 * For accept, we attempt to create a new socket, set up the link
1981 * with the client, wake up the client, then return the new
1982 * connected fd. We collect the address of the connector in kernel
1983 * space and move it to user at the very end. This is unclean because
1984 * we open the socket then return an error.
1985 *
1986 * 1003.1g adds the ability to recvmsg() to query connection pending
1987 * status to recvmsg. We need to add that support in a way thats
1988 * clean when we restructure accept also.
1989 */
1990
1991int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1992 int __user *upeer_addrlen, int flags)
1993{
1994 int ret = -EBADF;
1995 struct fd f;
1996
1997 f = fdget(fd);
1998 if (f.file) {
1999 ret = __sys_accept4_file(f.file, upeer_sockaddr,
2000 upeer_addrlen, flags);
2001 fdput(f);
2002 }
2003
2004 return ret;
2005}
2006
2007SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
2008 int __user *, upeer_addrlen, int, flags)
2009{
2010 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
2011}
2012
2013SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
2014 int __user *, upeer_addrlen)
2015{
2016 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
2017}
2018
2019/*
2020 * Attempt to connect to a socket with the server address. The address
2021 * is in user space so we verify it is OK and move it to kernel space.
2022 *
2023 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
2024 * break bindings
2025 *
2026 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
2027 * other SEQPACKET protocols that take time to connect() as it doesn't
2028 * include the -EINPROGRESS status for such sockets.
2029 */
2030
2031int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
2032 int addrlen, int file_flags)
2033{
2034 struct socket *sock;
2035 int err;
2036
2037 sock = sock_from_file(file);
2038 if (!sock) {
2039 err = -ENOTSOCK;
2040 goto out;
2041 }
2042
2043 err =
2044 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
2045 if (err)
2046 goto out;
2047
2048 err = READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)address,
2049 addrlen, sock->file->f_flags | file_flags);
2050out:
2051 return err;
2052}
2053
2054int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
2055{
2056 int ret = -EBADF;
2057 struct fd f;
2058
2059 f = fdget(fd);
2060 if (f.file) {
2061 struct sockaddr_storage address;
2062
2063 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
2064 if (!ret)
2065 ret = __sys_connect_file(f.file, &address, addrlen, 0);
2066 fdput(f);
2067 }
2068
2069 return ret;
2070}
2071
2072SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
2073 int, addrlen)
2074{
2075 return __sys_connect(fd, uservaddr, addrlen);
2076}
2077
2078/*
2079 * Get the local address ('name') of a socket object. Move the obtained
2080 * name to user space.
2081 */
2082
2083int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
2084 int __user *usockaddr_len)
2085{
2086 struct socket *sock;
2087 struct sockaddr_storage address;
2088 int err, fput_needed;
2089
2090 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2091 if (!sock)
2092 goto out;
2093
2094 err = security_socket_getsockname(sock);
2095 if (err)
2096 goto out_put;
2097
2098 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 0);
2099 if (err < 0)
2100 goto out_put;
2101 /* "err" is actually length in this case */
2102 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
2103
2104out_put:
2105 fput_light(sock->file, fput_needed);
2106out:
2107 return err;
2108}
2109
2110SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
2111 int __user *, usockaddr_len)
2112{
2113 return __sys_getsockname(fd, usockaddr, usockaddr_len);
2114}
2115
2116/*
2117 * Get the remote address ('name') of a socket object. Move the obtained
2118 * name to user space.
2119 */
2120
2121int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
2122 int __user *usockaddr_len)
2123{
2124 struct socket *sock;
2125 struct sockaddr_storage address;
2126 int err, fput_needed;
2127
2128 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2129 if (sock != NULL) {
2130 const struct proto_ops *ops = READ_ONCE(sock->ops);
2131
2132 err = security_socket_getpeername(sock);
2133 if (err) {
2134 fput_light(sock->file, fput_needed);
2135 return err;
2136 }
2137
2138 err = ops->getname(sock, (struct sockaddr *)&address, 1);
2139 if (err >= 0)
2140 /* "err" is actually length in this case */
2141 err = move_addr_to_user(&address, err, usockaddr,
2142 usockaddr_len);
2143 fput_light(sock->file, fput_needed);
2144 }
2145 return err;
2146}
2147
2148SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
2149 int __user *, usockaddr_len)
2150{
2151 return __sys_getpeername(fd, usockaddr, usockaddr_len);
2152}
2153
2154/*
2155 * Send a datagram to a given address. We move the address into kernel
2156 * space and check the user space data area is readable before invoking
2157 * the protocol.
2158 */
2159int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
2160 struct sockaddr __user *addr, int addr_len)
2161{
2162 struct socket *sock;
2163 struct sockaddr_storage address;
2164 int err;
2165 struct msghdr msg;
2166 int fput_needed;
2167
2168 err = import_ubuf(ITER_SOURCE, buff, len, &msg.msg_iter);
2169 if (unlikely(err))
2170 return err;
2171 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2172 if (!sock)
2173 goto out;
2174
2175 msg.msg_name = NULL;
2176 msg.msg_control = NULL;
2177 msg.msg_controllen = 0;
2178 msg.msg_namelen = 0;
2179 msg.msg_ubuf = NULL;
2180 if (addr) {
2181 err = move_addr_to_kernel(addr, addr_len, &address);
2182 if (err < 0)
2183 goto out_put;
2184 msg.msg_name = (struct sockaddr *)&address;
2185 msg.msg_namelen = addr_len;
2186 }
2187 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2188 if (sock->file->f_flags & O_NONBLOCK)
2189 flags |= MSG_DONTWAIT;
2190 msg.msg_flags = flags;
2191 err = __sock_sendmsg(sock, &msg);
2192
2193out_put:
2194 fput_light(sock->file, fput_needed);
2195out:
2196 return err;
2197}
2198
2199SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2200 unsigned int, flags, struct sockaddr __user *, addr,
2201 int, addr_len)
2202{
2203 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2204}
2205
2206/*
2207 * Send a datagram down a socket.
2208 */
2209
2210SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2211 unsigned int, flags)
2212{
2213 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2214}
2215
2216/*
2217 * Receive a frame from the socket and optionally record the address of the
2218 * sender. We verify the buffers are writable and if needed move the
2219 * sender address from kernel to user space.
2220 */
2221int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2222 struct sockaddr __user *addr, int __user *addr_len)
2223{
2224 struct sockaddr_storage address;
2225 struct msghdr msg = {
2226 /* Save some cycles and don't copy the address if not needed */
2227 .msg_name = addr ? (struct sockaddr *)&address : NULL,
2228 };
2229 struct socket *sock;
2230 int err, err2;
2231 int fput_needed;
2232
2233 err = import_ubuf(ITER_DEST, ubuf, size, &msg.msg_iter);
2234 if (unlikely(err))
2235 return err;
2236 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2237 if (!sock)
2238 goto out;
2239
2240 if (sock->file->f_flags & O_NONBLOCK)
2241 flags |= MSG_DONTWAIT;
2242 err = sock_recvmsg(sock, &msg, flags);
2243
2244 if (err >= 0 && addr != NULL) {
2245 err2 = move_addr_to_user(&address,
2246 msg.msg_namelen, addr, addr_len);
2247 if (err2 < 0)
2248 err = err2;
2249 }
2250
2251 fput_light(sock->file, fput_needed);
2252out:
2253 return err;
2254}
2255
2256SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2257 unsigned int, flags, struct sockaddr __user *, addr,
2258 int __user *, addr_len)
2259{
2260 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2261}
2262
2263/*
2264 * Receive a datagram from a socket.
2265 */
2266
2267SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2268 unsigned int, flags)
2269{
2270 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2271}
2272
2273static bool sock_use_custom_sol_socket(const struct socket *sock)
2274{
2275 return test_bit(SOCK_CUSTOM_SOCKOPT, &sock->flags);
2276}
2277
2278int do_sock_setsockopt(struct socket *sock, bool compat, int level,
2279 int optname, sockptr_t optval, int optlen)
2280{
2281 const struct proto_ops *ops;
2282 char *kernel_optval = NULL;
2283 int err;
2284
2285 if (optlen < 0)
2286 return -EINVAL;
2287
2288 err = security_socket_setsockopt(sock, level, optname);
2289 if (err)
2290 goto out_put;
2291
2292 if (!compat)
2293 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2294 optval, &optlen,
2295 &kernel_optval);
2296 if (err < 0)
2297 goto out_put;
2298 if (err > 0) {
2299 err = 0;
2300 goto out_put;
2301 }
2302
2303 if (kernel_optval)
2304 optval = KERNEL_SOCKPTR(kernel_optval);
2305 ops = READ_ONCE(sock->ops);
2306 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2307 err = sock_setsockopt(sock, level, optname, optval, optlen);
2308 else if (unlikely(!ops->setsockopt))
2309 err = -EOPNOTSUPP;
2310 else
2311 err = ops->setsockopt(sock, level, optname, optval,
2312 optlen);
2313 kfree(kernel_optval);
2314out_put:
2315 return err;
2316}
2317EXPORT_SYMBOL(do_sock_setsockopt);
2318
2319/* Set a socket option. Because we don't know the option lengths we have
2320 * to pass the user mode parameter for the protocols to sort out.
2321 */
2322int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2323 int optlen)
2324{
2325 sockptr_t optval = USER_SOCKPTR(user_optval);
2326 bool compat = in_compat_syscall();
2327 int err, fput_needed;
2328 struct socket *sock;
2329
2330 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2331 if (!sock)
2332 return err;
2333
2334 err = do_sock_setsockopt(sock, compat, level, optname, optval, optlen);
2335
2336 fput_light(sock->file, fput_needed);
2337 return err;
2338}
2339
2340SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2341 char __user *, optval, int, optlen)
2342{
2343 return __sys_setsockopt(fd, level, optname, optval, optlen);
2344}
2345
2346INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2347 int optname));
2348
2349int do_sock_getsockopt(struct socket *sock, bool compat, int level,
2350 int optname, sockptr_t optval, sockptr_t optlen)
2351{
2352 int max_optlen __maybe_unused;
2353 const struct proto_ops *ops;
2354 int err;
2355
2356 err = security_socket_getsockopt(sock, level, optname);
2357 if (err)
2358 return err;
2359
2360 if (!compat)
2361 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2362
2363 ops = READ_ONCE(sock->ops);
2364 if (level == SOL_SOCKET) {
2365 err = sk_getsockopt(sock->sk, level, optname, optval, optlen);
2366 } else if (unlikely(!ops->getsockopt)) {
2367 err = -EOPNOTSUPP;
2368 } else {
2369 if (WARN_ONCE(optval.is_kernel || optlen.is_kernel,
2370 "Invalid argument type"))
2371 return -EOPNOTSUPP;
2372
2373 err = ops->getsockopt(sock, level, optname, optval.user,
2374 optlen.user);
2375 }
2376
2377 if (!compat)
2378 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2379 optval, optlen, max_optlen,
2380 err);
2381
2382 return err;
2383}
2384EXPORT_SYMBOL(do_sock_getsockopt);
2385
2386/*
2387 * Get a socket option. Because we don't know the option lengths we have
2388 * to pass a user mode parameter for the protocols to sort out.
2389 */
2390int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2391 int __user *optlen)
2392{
2393 int err, fput_needed;
2394 struct socket *sock;
2395 bool compat;
2396
2397 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2398 if (!sock)
2399 return err;
2400
2401 compat = in_compat_syscall();
2402 err = do_sock_getsockopt(sock, compat, level, optname,
2403 USER_SOCKPTR(optval), USER_SOCKPTR(optlen));
2404
2405 fput_light(sock->file, fput_needed);
2406 return err;
2407}
2408
2409SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2410 char __user *, optval, int __user *, optlen)
2411{
2412 return __sys_getsockopt(fd, level, optname, optval, optlen);
2413}
2414
2415/*
2416 * Shutdown a socket.
2417 */
2418
2419int __sys_shutdown_sock(struct socket *sock, int how)
2420{
2421 int err;
2422
2423 err = security_socket_shutdown(sock, how);
2424 if (!err)
2425 err = READ_ONCE(sock->ops)->shutdown(sock, how);
2426
2427 return err;
2428}
2429
2430int __sys_shutdown(int fd, int how)
2431{
2432 int err, fput_needed;
2433 struct socket *sock;
2434
2435 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2436 if (sock != NULL) {
2437 err = __sys_shutdown_sock(sock, how);
2438 fput_light(sock->file, fput_needed);
2439 }
2440 return err;
2441}
2442
2443SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2444{
2445 return __sys_shutdown(fd, how);
2446}
2447
2448/* A couple of helpful macros for getting the address of the 32/64 bit
2449 * fields which are the same type (int / unsigned) on our platforms.
2450 */
2451#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2452#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2453#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2454
2455struct used_address {
2456 struct sockaddr_storage name;
2457 unsigned int name_len;
2458};
2459
2460int __copy_msghdr(struct msghdr *kmsg,
2461 struct user_msghdr *msg,
2462 struct sockaddr __user **save_addr)
2463{
2464 ssize_t err;
2465
2466 kmsg->msg_control_is_user = true;
2467 kmsg->msg_get_inq = 0;
2468 kmsg->msg_control_user = msg->msg_control;
2469 kmsg->msg_controllen = msg->msg_controllen;
2470 kmsg->msg_flags = msg->msg_flags;
2471
2472 kmsg->msg_namelen = msg->msg_namelen;
2473 if (!msg->msg_name)
2474 kmsg->msg_namelen = 0;
2475
2476 if (kmsg->msg_namelen < 0)
2477 return -EINVAL;
2478
2479 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2480 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2481
2482 if (save_addr)
2483 *save_addr = msg->msg_name;
2484
2485 if (msg->msg_name && kmsg->msg_namelen) {
2486 if (!save_addr) {
2487 err = move_addr_to_kernel(msg->msg_name,
2488 kmsg->msg_namelen,
2489 kmsg->msg_name);
2490 if (err < 0)
2491 return err;
2492 }
2493 } else {
2494 kmsg->msg_name = NULL;
2495 kmsg->msg_namelen = 0;
2496 }
2497
2498 if (msg->msg_iovlen > UIO_MAXIOV)
2499 return -EMSGSIZE;
2500
2501 kmsg->msg_iocb = NULL;
2502 kmsg->msg_ubuf = NULL;
2503 return 0;
2504}
2505
2506static int copy_msghdr_from_user(struct msghdr *kmsg,
2507 struct user_msghdr __user *umsg,
2508 struct sockaddr __user **save_addr,
2509 struct iovec **iov)
2510{
2511 struct user_msghdr msg;
2512 ssize_t err;
2513
2514 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2515 return -EFAULT;
2516
2517 err = __copy_msghdr(kmsg, &msg, save_addr);
2518 if (err)
2519 return err;
2520
2521 err = import_iovec(save_addr ? ITER_DEST : ITER_SOURCE,
2522 msg.msg_iov, msg.msg_iovlen,
2523 UIO_FASTIOV, iov, &kmsg->msg_iter);
2524 return err < 0 ? err : 0;
2525}
2526
2527static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2528 unsigned int flags, struct used_address *used_address,
2529 unsigned int allowed_msghdr_flags)
2530{
2531 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2532 __aligned(sizeof(__kernel_size_t));
2533 /* 20 is size of ipv6_pktinfo */
2534 unsigned char *ctl_buf = ctl;
2535 int ctl_len;
2536 ssize_t err;
2537
2538 err = -ENOBUFS;
2539
2540 if (msg_sys->msg_controllen > INT_MAX)
2541 goto out;
2542 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2543 ctl_len = msg_sys->msg_controllen;
2544 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2545 err =
2546 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2547 sizeof(ctl));
2548 if (err)
2549 goto out;
2550 ctl_buf = msg_sys->msg_control;
2551 ctl_len = msg_sys->msg_controllen;
2552 } else if (ctl_len) {
2553 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2554 CMSG_ALIGN(sizeof(struct cmsghdr)));
2555 if (ctl_len > sizeof(ctl)) {
2556 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2557 if (ctl_buf == NULL)
2558 goto out;
2559 }
2560 err = -EFAULT;
2561 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2562 goto out_freectl;
2563 msg_sys->msg_control = ctl_buf;
2564 msg_sys->msg_control_is_user = false;
2565 }
2566 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2567 msg_sys->msg_flags = flags;
2568
2569 if (sock->file->f_flags & O_NONBLOCK)
2570 msg_sys->msg_flags |= MSG_DONTWAIT;
2571 /*
2572 * If this is sendmmsg() and current destination address is same as
2573 * previously succeeded address, omit asking LSM's decision.
2574 * used_address->name_len is initialized to UINT_MAX so that the first
2575 * destination address never matches.
2576 */
2577 if (used_address && msg_sys->msg_name &&
2578 used_address->name_len == msg_sys->msg_namelen &&
2579 !memcmp(&used_address->name, msg_sys->msg_name,
2580 used_address->name_len)) {
2581 err = sock_sendmsg_nosec(sock, msg_sys);
2582 goto out_freectl;
2583 }
2584 err = __sock_sendmsg(sock, msg_sys);
2585 /*
2586 * If this is sendmmsg() and sending to current destination address was
2587 * successful, remember it.
2588 */
2589 if (used_address && err >= 0) {
2590 used_address->name_len = msg_sys->msg_namelen;
2591 if (msg_sys->msg_name)
2592 memcpy(&used_address->name, msg_sys->msg_name,
2593 used_address->name_len);
2594 }
2595
2596out_freectl:
2597 if (ctl_buf != ctl)
2598 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2599out:
2600 return err;
2601}
2602
2603int sendmsg_copy_msghdr(struct msghdr *msg,
2604 struct user_msghdr __user *umsg, unsigned flags,
2605 struct iovec **iov)
2606{
2607 int err;
2608
2609 if (flags & MSG_CMSG_COMPAT) {
2610 struct compat_msghdr __user *msg_compat;
2611
2612 msg_compat = (struct compat_msghdr __user *) umsg;
2613 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2614 } else {
2615 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2616 }
2617 if (err < 0)
2618 return err;
2619
2620 return 0;
2621}
2622
2623static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2624 struct msghdr *msg_sys, unsigned int flags,
2625 struct used_address *used_address,
2626 unsigned int allowed_msghdr_flags)
2627{
2628 struct sockaddr_storage address;
2629 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2630 ssize_t err;
2631
2632 msg_sys->msg_name = &address;
2633
2634 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2635 if (err < 0)
2636 return err;
2637
2638 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2639 allowed_msghdr_flags);
2640 kfree(iov);
2641 return err;
2642}
2643
2644/*
2645 * BSD sendmsg interface
2646 */
2647long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2648 unsigned int flags)
2649{
2650 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2651}
2652
2653long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2654 bool forbid_cmsg_compat)
2655{
2656 int fput_needed, err;
2657 struct msghdr msg_sys;
2658 struct socket *sock;
2659
2660 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2661 return -EINVAL;
2662
2663 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2664 if (!sock)
2665 goto out;
2666
2667 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2668
2669 fput_light(sock->file, fput_needed);
2670out:
2671 return err;
2672}
2673
2674SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2675{
2676 return __sys_sendmsg(fd, msg, flags, true);
2677}
2678
2679/*
2680 * Linux sendmmsg interface
2681 */
2682
2683int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2684 unsigned int flags, bool forbid_cmsg_compat)
2685{
2686 int fput_needed, err, datagrams;
2687 struct socket *sock;
2688 struct mmsghdr __user *entry;
2689 struct compat_mmsghdr __user *compat_entry;
2690 struct msghdr msg_sys;
2691 struct used_address used_address;
2692 unsigned int oflags = flags;
2693
2694 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2695 return -EINVAL;
2696
2697 if (vlen > UIO_MAXIOV)
2698 vlen = UIO_MAXIOV;
2699
2700 datagrams = 0;
2701
2702 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2703 if (!sock)
2704 return err;
2705
2706 used_address.name_len = UINT_MAX;
2707 entry = mmsg;
2708 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2709 err = 0;
2710 flags |= MSG_BATCH;
2711
2712 while (datagrams < vlen) {
2713 if (datagrams == vlen - 1)
2714 flags = oflags;
2715
2716 if (MSG_CMSG_COMPAT & flags) {
2717 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2718 &msg_sys, flags, &used_address, MSG_EOR);
2719 if (err < 0)
2720 break;
2721 err = __put_user(err, &compat_entry->msg_len);
2722 ++compat_entry;
2723 } else {
2724 err = ___sys_sendmsg(sock,
2725 (struct user_msghdr __user *)entry,
2726 &msg_sys, flags, &used_address, MSG_EOR);
2727 if (err < 0)
2728 break;
2729 err = put_user(err, &entry->msg_len);
2730 ++entry;
2731 }
2732
2733 if (err)
2734 break;
2735 ++datagrams;
2736 if (msg_data_left(&msg_sys))
2737 break;
2738 cond_resched();
2739 }
2740
2741 fput_light(sock->file, fput_needed);
2742
2743 /* We only return an error if no datagrams were able to be sent */
2744 if (datagrams != 0)
2745 return datagrams;
2746
2747 return err;
2748}
2749
2750SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2751 unsigned int, vlen, unsigned int, flags)
2752{
2753 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2754}
2755
2756int recvmsg_copy_msghdr(struct msghdr *msg,
2757 struct user_msghdr __user *umsg, unsigned flags,
2758 struct sockaddr __user **uaddr,
2759 struct iovec **iov)
2760{
2761 ssize_t err;
2762
2763 if (MSG_CMSG_COMPAT & flags) {
2764 struct compat_msghdr __user *msg_compat;
2765
2766 msg_compat = (struct compat_msghdr __user *) umsg;
2767 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2768 } else {
2769 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2770 }
2771 if (err < 0)
2772 return err;
2773
2774 return 0;
2775}
2776
2777static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2778 struct user_msghdr __user *msg,
2779 struct sockaddr __user *uaddr,
2780 unsigned int flags, int nosec)
2781{
2782 struct compat_msghdr __user *msg_compat =
2783 (struct compat_msghdr __user *) msg;
2784 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2785 struct sockaddr_storage addr;
2786 unsigned long cmsg_ptr;
2787 int len;
2788 ssize_t err;
2789
2790 msg_sys->msg_name = &addr;
2791 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2792 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2793
2794 /* We assume all kernel code knows the size of sockaddr_storage */
2795 msg_sys->msg_namelen = 0;
2796
2797 if (sock->file->f_flags & O_NONBLOCK)
2798 flags |= MSG_DONTWAIT;
2799
2800 if (unlikely(nosec))
2801 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2802 else
2803 err = sock_recvmsg(sock, msg_sys, flags);
2804
2805 if (err < 0)
2806 goto out;
2807 len = err;
2808
2809 if (uaddr != NULL) {
2810 err = move_addr_to_user(&addr,
2811 msg_sys->msg_namelen, uaddr,
2812 uaddr_len);
2813 if (err < 0)
2814 goto out;
2815 }
2816 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2817 COMPAT_FLAGS(msg));
2818 if (err)
2819 goto out;
2820 if (MSG_CMSG_COMPAT & flags)
2821 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2822 &msg_compat->msg_controllen);
2823 else
2824 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2825 &msg->msg_controllen);
2826 if (err)
2827 goto out;
2828 err = len;
2829out:
2830 return err;
2831}
2832
2833static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2834 struct msghdr *msg_sys, unsigned int flags, int nosec)
2835{
2836 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2837 /* user mode address pointers */
2838 struct sockaddr __user *uaddr;
2839 ssize_t err;
2840
2841 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2842 if (err < 0)
2843 return err;
2844
2845 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2846 kfree(iov);
2847 return err;
2848}
2849
2850/*
2851 * BSD recvmsg interface
2852 */
2853
2854long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2855 struct user_msghdr __user *umsg,
2856 struct sockaddr __user *uaddr, unsigned int flags)
2857{
2858 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2859}
2860
2861long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2862 bool forbid_cmsg_compat)
2863{
2864 int fput_needed, err;
2865 struct msghdr msg_sys;
2866 struct socket *sock;
2867
2868 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2869 return -EINVAL;
2870
2871 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2872 if (!sock)
2873 goto out;
2874
2875 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2876
2877 fput_light(sock->file, fput_needed);
2878out:
2879 return err;
2880}
2881
2882SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2883 unsigned int, flags)
2884{
2885 return __sys_recvmsg(fd, msg, flags, true);
2886}
2887
2888/*
2889 * Linux recvmmsg interface
2890 */
2891
2892static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2893 unsigned int vlen, unsigned int flags,
2894 struct timespec64 *timeout)
2895{
2896 int fput_needed, err, datagrams;
2897 struct socket *sock;
2898 struct mmsghdr __user *entry;
2899 struct compat_mmsghdr __user *compat_entry;
2900 struct msghdr msg_sys;
2901 struct timespec64 end_time;
2902 struct timespec64 timeout64;
2903
2904 if (timeout &&
2905 poll_select_set_timeout(&end_time, timeout->tv_sec,
2906 timeout->tv_nsec))
2907 return -EINVAL;
2908
2909 datagrams = 0;
2910
2911 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2912 if (!sock)
2913 return err;
2914
2915 if (likely(!(flags & MSG_ERRQUEUE))) {
2916 err = sock_error(sock->sk);
2917 if (err) {
2918 datagrams = err;
2919 goto out_put;
2920 }
2921 }
2922
2923 entry = mmsg;
2924 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2925
2926 while (datagrams < vlen) {
2927 /*
2928 * No need to ask LSM for more than the first datagram.
2929 */
2930 if (MSG_CMSG_COMPAT & flags) {
2931 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2932 &msg_sys, flags & ~MSG_WAITFORONE,
2933 datagrams);
2934 if (err < 0)
2935 break;
2936 err = __put_user(err, &compat_entry->msg_len);
2937 ++compat_entry;
2938 } else {
2939 err = ___sys_recvmsg(sock,
2940 (struct user_msghdr __user *)entry,
2941 &msg_sys, flags & ~MSG_WAITFORONE,
2942 datagrams);
2943 if (err < 0)
2944 break;
2945 err = put_user(err, &entry->msg_len);
2946 ++entry;
2947 }
2948
2949 if (err)
2950 break;
2951 ++datagrams;
2952
2953 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2954 if (flags & MSG_WAITFORONE)
2955 flags |= MSG_DONTWAIT;
2956
2957 if (timeout) {
2958 ktime_get_ts64(&timeout64);
2959 *timeout = timespec64_sub(end_time, timeout64);
2960 if (timeout->tv_sec < 0) {
2961 timeout->tv_sec = timeout->tv_nsec = 0;
2962 break;
2963 }
2964
2965 /* Timeout, return less than vlen datagrams */
2966 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2967 break;
2968 }
2969
2970 /* Out of band data, return right away */
2971 if (msg_sys.msg_flags & MSG_OOB)
2972 break;
2973 cond_resched();
2974 }
2975
2976 if (err == 0)
2977 goto out_put;
2978
2979 if (datagrams == 0) {
2980 datagrams = err;
2981 goto out_put;
2982 }
2983
2984 /*
2985 * We may return less entries than requested (vlen) if the
2986 * sock is non block and there aren't enough datagrams...
2987 */
2988 if (err != -EAGAIN) {
2989 /*
2990 * ... or if recvmsg returns an error after we
2991 * received some datagrams, where we record the
2992 * error to return on the next call or if the
2993 * app asks about it using getsockopt(SO_ERROR).
2994 */
2995 WRITE_ONCE(sock->sk->sk_err, -err);
2996 }
2997out_put:
2998 fput_light(sock->file, fput_needed);
2999
3000 return datagrams;
3001}
3002
3003int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
3004 unsigned int vlen, unsigned int flags,
3005 struct __kernel_timespec __user *timeout,
3006 struct old_timespec32 __user *timeout32)
3007{
3008 int datagrams;
3009 struct timespec64 timeout_sys;
3010
3011 if (timeout && get_timespec64(&timeout_sys, timeout))
3012 return -EFAULT;
3013
3014 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
3015 return -EFAULT;
3016
3017 if (!timeout && !timeout32)
3018 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
3019
3020 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
3021
3022 if (datagrams <= 0)
3023 return datagrams;
3024
3025 if (timeout && put_timespec64(&timeout_sys, timeout))
3026 datagrams = -EFAULT;
3027
3028 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
3029 datagrams = -EFAULT;
3030
3031 return datagrams;
3032}
3033
3034SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
3035 unsigned int, vlen, unsigned int, flags,
3036 struct __kernel_timespec __user *, timeout)
3037{
3038 if (flags & MSG_CMSG_COMPAT)
3039 return -EINVAL;
3040
3041 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
3042}
3043
3044#ifdef CONFIG_COMPAT_32BIT_TIME
3045SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
3046 unsigned int, vlen, unsigned int, flags,
3047 struct old_timespec32 __user *, timeout)
3048{
3049 if (flags & MSG_CMSG_COMPAT)
3050 return -EINVAL;
3051
3052 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
3053}
3054#endif
3055
3056#ifdef __ARCH_WANT_SYS_SOCKETCALL
3057/* Argument list sizes for sys_socketcall */
3058#define AL(x) ((x) * sizeof(unsigned long))
3059static const unsigned char nargs[21] = {
3060 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
3061 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
3062 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
3063 AL(4), AL(5), AL(4)
3064};
3065
3066#undef AL
3067
3068/*
3069 * System call vectors.
3070 *
3071 * Argument checking cleaned up. Saved 20% in size.
3072 * This function doesn't need to set the kernel lock because
3073 * it is set by the callees.
3074 */
3075
3076SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
3077{
3078 unsigned long a[AUDITSC_ARGS];
3079 unsigned long a0, a1;
3080 int err;
3081 unsigned int len;
3082
3083 if (call < 1 || call > SYS_SENDMMSG)
3084 return -EINVAL;
3085 call = array_index_nospec(call, SYS_SENDMMSG + 1);
3086
3087 len = nargs[call];
3088 if (len > sizeof(a))
3089 return -EINVAL;
3090
3091 /* copy_from_user should be SMP safe. */
3092 if (copy_from_user(a, args, len))
3093 return -EFAULT;
3094
3095 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
3096 if (err)
3097 return err;
3098
3099 a0 = a[0];
3100 a1 = a[1];
3101
3102 switch (call) {
3103 case SYS_SOCKET:
3104 err = __sys_socket(a0, a1, a[2]);
3105 break;
3106 case SYS_BIND:
3107 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
3108 break;
3109 case SYS_CONNECT:
3110 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
3111 break;
3112 case SYS_LISTEN:
3113 err = __sys_listen(a0, a1);
3114 break;
3115 case SYS_ACCEPT:
3116 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3117 (int __user *)a[2], 0);
3118 break;
3119 case SYS_GETSOCKNAME:
3120 err =
3121 __sys_getsockname(a0, (struct sockaddr __user *)a1,
3122 (int __user *)a[2]);
3123 break;
3124 case SYS_GETPEERNAME:
3125 err =
3126 __sys_getpeername(a0, (struct sockaddr __user *)a1,
3127 (int __user *)a[2]);
3128 break;
3129 case SYS_SOCKETPAIR:
3130 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
3131 break;
3132 case SYS_SEND:
3133 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3134 NULL, 0);
3135 break;
3136 case SYS_SENDTO:
3137 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3138 (struct sockaddr __user *)a[4], a[5]);
3139 break;
3140 case SYS_RECV:
3141 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3142 NULL, NULL);
3143 break;
3144 case SYS_RECVFROM:
3145 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3146 (struct sockaddr __user *)a[4],
3147 (int __user *)a[5]);
3148 break;
3149 case SYS_SHUTDOWN:
3150 err = __sys_shutdown(a0, a1);
3151 break;
3152 case SYS_SETSOCKOPT:
3153 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
3154 a[4]);
3155 break;
3156 case SYS_GETSOCKOPT:
3157 err =
3158 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
3159 (int __user *)a[4]);
3160 break;
3161 case SYS_SENDMSG:
3162 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
3163 a[2], true);
3164 break;
3165 case SYS_SENDMMSG:
3166 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
3167 a[3], true);
3168 break;
3169 case SYS_RECVMSG:
3170 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
3171 a[2], true);
3172 break;
3173 case SYS_RECVMMSG:
3174 if (IS_ENABLED(CONFIG_64BIT))
3175 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3176 a[2], a[3],
3177 (struct __kernel_timespec __user *)a[4],
3178 NULL);
3179 else
3180 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3181 a[2], a[3], NULL,
3182 (struct old_timespec32 __user *)a[4]);
3183 break;
3184 case SYS_ACCEPT4:
3185 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3186 (int __user *)a[2], a[3]);
3187 break;
3188 default:
3189 err = -EINVAL;
3190 break;
3191 }
3192 return err;
3193}
3194
3195#endif /* __ARCH_WANT_SYS_SOCKETCALL */
3196
3197/**
3198 * sock_register - add a socket protocol handler
3199 * @ops: description of protocol
3200 *
3201 * This function is called by a protocol handler that wants to
3202 * advertise its address family, and have it linked into the
3203 * socket interface. The value ops->family corresponds to the
3204 * socket system call protocol family.
3205 */
3206int sock_register(const struct net_proto_family *ops)
3207{
3208 int err;
3209
3210 if (ops->family >= NPROTO) {
3211 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3212 return -ENOBUFS;
3213 }
3214
3215 spin_lock(&net_family_lock);
3216 if (rcu_dereference_protected(net_families[ops->family],
3217 lockdep_is_held(&net_family_lock)))
3218 err = -EEXIST;
3219 else {
3220 rcu_assign_pointer(net_families[ops->family], ops);
3221 err = 0;
3222 }
3223 spin_unlock(&net_family_lock);
3224
3225 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3226 return err;
3227}
3228EXPORT_SYMBOL(sock_register);
3229
3230/**
3231 * sock_unregister - remove a protocol handler
3232 * @family: protocol family to remove
3233 *
3234 * This function is called by a protocol handler that wants to
3235 * remove its address family, and have it unlinked from the
3236 * new socket creation.
3237 *
3238 * If protocol handler is a module, then it can use module reference
3239 * counts to protect against new references. If protocol handler is not
3240 * a module then it needs to provide its own protection in
3241 * the ops->create routine.
3242 */
3243void sock_unregister(int family)
3244{
3245 BUG_ON(family < 0 || family >= NPROTO);
3246
3247 spin_lock(&net_family_lock);
3248 RCU_INIT_POINTER(net_families[family], NULL);
3249 spin_unlock(&net_family_lock);
3250
3251 synchronize_rcu();
3252
3253 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3254}
3255EXPORT_SYMBOL(sock_unregister);
3256
3257bool sock_is_registered(int family)
3258{
3259 return family < NPROTO && rcu_access_pointer(net_families[family]);
3260}
3261
3262static int __init sock_init(void)
3263{
3264 int err;
3265 /*
3266 * Initialize the network sysctl infrastructure.
3267 */
3268 err = net_sysctl_init();
3269 if (err)
3270 goto out;
3271
3272 /*
3273 * Initialize skbuff SLAB cache
3274 */
3275 skb_init();
3276
3277 /*
3278 * Initialize the protocols module.
3279 */
3280
3281 init_inodecache();
3282
3283 err = register_filesystem(&sock_fs_type);
3284 if (err)
3285 goto out;
3286 sock_mnt = kern_mount(&sock_fs_type);
3287 if (IS_ERR(sock_mnt)) {
3288 err = PTR_ERR(sock_mnt);
3289 goto out_mount;
3290 }
3291
3292 /* The real protocol initialization is performed in later initcalls.
3293 */
3294
3295#ifdef CONFIG_NETFILTER
3296 err = netfilter_init();
3297 if (err)
3298 goto out;
3299#endif
3300
3301 ptp_classifier_init();
3302
3303out:
3304 return err;
3305
3306out_mount:
3307 unregister_filesystem(&sock_fs_type);
3308 goto out;
3309}
3310
3311core_initcall(sock_init); /* early initcall */
3312
3313#ifdef CONFIG_PROC_FS
3314void socket_seq_show(struct seq_file *seq)
3315{
3316 seq_printf(seq, "sockets: used %d\n",
3317 sock_inuse_get(seq->private));
3318}
3319#endif /* CONFIG_PROC_FS */
3320
3321/* Handle the fact that while struct ifreq has the same *layout* on
3322 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3323 * which are handled elsewhere, it still has different *size* due to
3324 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3325 * resulting in struct ifreq being 32 and 40 bytes respectively).
3326 * As a result, if the struct happens to be at the end of a page and
3327 * the next page isn't readable/writable, we get a fault. To prevent
3328 * that, copy back and forth to the full size.
3329 */
3330int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3331{
3332 if (in_compat_syscall()) {
3333 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3334
3335 memset(ifr, 0, sizeof(*ifr));
3336 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3337 return -EFAULT;
3338
3339 if (ifrdata)
3340 *ifrdata = compat_ptr(ifr32->ifr_data);
3341
3342 return 0;
3343 }
3344
3345 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3346 return -EFAULT;
3347
3348 if (ifrdata)
3349 *ifrdata = ifr->ifr_data;
3350
3351 return 0;
3352}
3353EXPORT_SYMBOL(get_user_ifreq);
3354
3355int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3356{
3357 size_t size = sizeof(*ifr);
3358
3359 if (in_compat_syscall())
3360 size = sizeof(struct compat_ifreq);
3361
3362 if (copy_to_user(arg, ifr, size))
3363 return -EFAULT;
3364
3365 return 0;
3366}
3367EXPORT_SYMBOL(put_user_ifreq);
3368
3369#ifdef CONFIG_COMPAT
3370static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3371{
3372 compat_uptr_t uptr32;
3373 struct ifreq ifr;
3374 void __user *saved;
3375 int err;
3376
3377 if (get_user_ifreq(&ifr, NULL, uifr32))
3378 return -EFAULT;
3379
3380 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3381 return -EFAULT;
3382
3383 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3384 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3385
3386 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3387 if (!err) {
3388 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3389 if (put_user_ifreq(&ifr, uifr32))
3390 err = -EFAULT;
3391 }
3392 return err;
3393}
3394
3395/* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3396static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3397 struct compat_ifreq __user *u_ifreq32)
3398{
3399 struct ifreq ifreq;
3400 void __user *data;
3401
3402 if (!is_socket_ioctl_cmd(cmd))
3403 return -ENOTTY;
3404 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3405 return -EFAULT;
3406 ifreq.ifr_data = data;
3407
3408 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3409}
3410
3411static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3412 unsigned int cmd, unsigned long arg)
3413{
3414 void __user *argp = compat_ptr(arg);
3415 struct sock *sk = sock->sk;
3416 struct net *net = sock_net(sk);
3417 const struct proto_ops *ops;
3418
3419 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3420 return sock_ioctl(file, cmd, (unsigned long)argp);
3421
3422 switch (cmd) {
3423 case SIOCWANDEV:
3424 return compat_siocwandev(net, argp);
3425 case SIOCGSTAMP_OLD:
3426 case SIOCGSTAMPNS_OLD:
3427 ops = READ_ONCE(sock->ops);
3428 if (!ops->gettstamp)
3429 return -ENOIOCTLCMD;
3430 return ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3431 !COMPAT_USE_64BIT_TIME);
3432
3433 case SIOCETHTOOL:
3434 case SIOCBONDSLAVEINFOQUERY:
3435 case SIOCBONDINFOQUERY:
3436 case SIOCSHWTSTAMP:
3437 case SIOCGHWTSTAMP:
3438 return compat_ifr_data_ioctl(net, cmd, argp);
3439
3440 case FIOSETOWN:
3441 case SIOCSPGRP:
3442 case FIOGETOWN:
3443 case SIOCGPGRP:
3444 case SIOCBRADDBR:
3445 case SIOCBRDELBR:
3446 case SIOCGIFVLAN:
3447 case SIOCSIFVLAN:
3448 case SIOCGSKNS:
3449 case SIOCGSTAMP_NEW:
3450 case SIOCGSTAMPNS_NEW:
3451 case SIOCGIFCONF:
3452 case SIOCSIFBR:
3453 case SIOCGIFBR:
3454 return sock_ioctl(file, cmd, arg);
3455
3456 case SIOCGIFFLAGS:
3457 case SIOCSIFFLAGS:
3458 case SIOCGIFMAP:
3459 case SIOCSIFMAP:
3460 case SIOCGIFMETRIC:
3461 case SIOCSIFMETRIC:
3462 case SIOCGIFMTU:
3463 case SIOCSIFMTU:
3464 case SIOCGIFMEM:
3465 case SIOCSIFMEM:
3466 case SIOCGIFHWADDR:
3467 case SIOCSIFHWADDR:
3468 case SIOCADDMULTI:
3469 case SIOCDELMULTI:
3470 case SIOCGIFINDEX:
3471 case SIOCGIFADDR:
3472 case SIOCSIFADDR:
3473 case SIOCSIFHWBROADCAST:
3474 case SIOCDIFADDR:
3475 case SIOCGIFBRDADDR:
3476 case SIOCSIFBRDADDR:
3477 case SIOCGIFDSTADDR:
3478 case SIOCSIFDSTADDR:
3479 case SIOCGIFNETMASK:
3480 case SIOCSIFNETMASK:
3481 case SIOCSIFPFLAGS:
3482 case SIOCGIFPFLAGS:
3483 case SIOCGIFTXQLEN:
3484 case SIOCSIFTXQLEN:
3485 case SIOCBRADDIF:
3486 case SIOCBRDELIF:
3487 case SIOCGIFNAME:
3488 case SIOCSIFNAME:
3489 case SIOCGMIIPHY:
3490 case SIOCGMIIREG:
3491 case SIOCSMIIREG:
3492 case SIOCBONDENSLAVE:
3493 case SIOCBONDRELEASE:
3494 case SIOCBONDSETHWADDR:
3495 case SIOCBONDCHANGEACTIVE:
3496 case SIOCSARP:
3497 case SIOCGARP:
3498 case SIOCDARP:
3499 case SIOCOUTQ:
3500 case SIOCOUTQNSD:
3501 case SIOCATMARK:
3502 return sock_do_ioctl(net, sock, cmd, arg);
3503 }
3504
3505 return -ENOIOCTLCMD;
3506}
3507
3508static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3509 unsigned long arg)
3510{
3511 struct socket *sock = file->private_data;
3512 const struct proto_ops *ops = READ_ONCE(sock->ops);
3513 int ret = -ENOIOCTLCMD;
3514 struct sock *sk;
3515 struct net *net;
3516
3517 sk = sock->sk;
3518 net = sock_net(sk);
3519
3520 if (ops->compat_ioctl)
3521 ret = ops->compat_ioctl(sock, cmd, arg);
3522
3523 if (ret == -ENOIOCTLCMD &&
3524 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3525 ret = compat_wext_handle_ioctl(net, cmd, arg);
3526
3527 if (ret == -ENOIOCTLCMD)
3528 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3529
3530 return ret;
3531}
3532#endif
3533
3534/**
3535 * kernel_bind - bind an address to a socket (kernel space)
3536 * @sock: socket
3537 * @addr: address
3538 * @addrlen: length of address
3539 *
3540 * Returns 0 or an error.
3541 */
3542
3543int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3544{
3545 struct sockaddr_storage address;
3546
3547 memcpy(&address, addr, addrlen);
3548
3549 return READ_ONCE(sock->ops)->bind(sock, (struct sockaddr *)&address,
3550 addrlen);
3551}
3552EXPORT_SYMBOL(kernel_bind);
3553
3554/**
3555 * kernel_listen - move socket to listening state (kernel space)
3556 * @sock: socket
3557 * @backlog: pending connections queue size
3558 *
3559 * Returns 0 or an error.
3560 */
3561
3562int kernel_listen(struct socket *sock, int backlog)
3563{
3564 return READ_ONCE(sock->ops)->listen(sock, backlog);
3565}
3566EXPORT_SYMBOL(kernel_listen);
3567
3568/**
3569 * kernel_accept - accept a connection (kernel space)
3570 * @sock: listening socket
3571 * @newsock: new connected socket
3572 * @flags: flags
3573 *
3574 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3575 * If it fails, @newsock is guaranteed to be %NULL.
3576 * Returns 0 or an error.
3577 */
3578
3579int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3580{
3581 struct sock *sk = sock->sk;
3582 const struct proto_ops *ops = READ_ONCE(sock->ops);
3583 int err;
3584
3585 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3586 newsock);
3587 if (err < 0)
3588 goto done;
3589
3590 err = ops->accept(sock, *newsock, flags, true);
3591 if (err < 0) {
3592 sock_release(*newsock);
3593 *newsock = NULL;
3594 goto done;
3595 }
3596
3597 (*newsock)->ops = ops;
3598 __module_get(ops->owner);
3599
3600done:
3601 return err;
3602}
3603EXPORT_SYMBOL(kernel_accept);
3604
3605/**
3606 * kernel_connect - connect a socket (kernel space)
3607 * @sock: socket
3608 * @addr: address
3609 * @addrlen: address length
3610 * @flags: flags (O_NONBLOCK, ...)
3611 *
3612 * For datagram sockets, @addr is the address to which datagrams are sent
3613 * by default, and the only address from which datagrams are received.
3614 * For stream sockets, attempts to connect to @addr.
3615 * Returns 0 or an error code.
3616 */
3617
3618int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3619 int flags)
3620{
3621 struct sockaddr_storage address;
3622
3623 memcpy(&address, addr, addrlen);
3624
3625 return READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)&address,
3626 addrlen, flags);
3627}
3628EXPORT_SYMBOL(kernel_connect);
3629
3630/**
3631 * kernel_getsockname - get the address which the socket is bound (kernel space)
3632 * @sock: socket
3633 * @addr: address holder
3634 *
3635 * Fills the @addr pointer with the address which the socket is bound.
3636 * Returns the length of the address in bytes or an error code.
3637 */
3638
3639int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3640{
3641 return READ_ONCE(sock->ops)->getname(sock, addr, 0);
3642}
3643EXPORT_SYMBOL(kernel_getsockname);
3644
3645/**
3646 * kernel_getpeername - get the address which the socket is connected (kernel space)
3647 * @sock: socket
3648 * @addr: address holder
3649 *
3650 * Fills the @addr pointer with the address which the socket is connected.
3651 * Returns the length of the address in bytes or an error code.
3652 */
3653
3654int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3655{
3656 return READ_ONCE(sock->ops)->getname(sock, addr, 1);
3657}
3658EXPORT_SYMBOL(kernel_getpeername);
3659
3660/**
3661 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3662 * @sock: socket
3663 * @how: connection part
3664 *
3665 * Returns 0 or an error.
3666 */
3667
3668int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3669{
3670 return READ_ONCE(sock->ops)->shutdown(sock, how);
3671}
3672EXPORT_SYMBOL(kernel_sock_shutdown);
3673
3674/**
3675 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3676 * @sk: socket
3677 *
3678 * This routine returns the IP overhead imposed by a socket i.e.
3679 * the length of the underlying IP header, depending on whether
3680 * this is an IPv4 or IPv6 socket and the length from IP options turned
3681 * on at the socket. Assumes that the caller has a lock on the socket.
3682 */
3683
3684u32 kernel_sock_ip_overhead(struct sock *sk)
3685{
3686 struct inet_sock *inet;
3687 struct ip_options_rcu *opt;
3688 u32 overhead = 0;
3689#if IS_ENABLED(CONFIG_IPV6)
3690 struct ipv6_pinfo *np;
3691 struct ipv6_txoptions *optv6 = NULL;
3692#endif /* IS_ENABLED(CONFIG_IPV6) */
3693
3694 if (!sk)
3695 return overhead;
3696
3697 switch (sk->sk_family) {
3698 case AF_INET:
3699 inet = inet_sk(sk);
3700 overhead += sizeof(struct iphdr);
3701 opt = rcu_dereference_protected(inet->inet_opt,
3702 sock_owned_by_user(sk));
3703 if (opt)
3704 overhead += opt->opt.optlen;
3705 return overhead;
3706#if IS_ENABLED(CONFIG_IPV6)
3707 case AF_INET6:
3708 np = inet6_sk(sk);
3709 overhead += sizeof(struct ipv6hdr);
3710 if (np)
3711 optv6 = rcu_dereference_protected(np->opt,
3712 sock_owned_by_user(sk));
3713 if (optv6)
3714 overhead += (optv6->opt_flen + optv6->opt_nflen);
3715 return overhead;
3716#endif /* IS_ENABLED(CONFIG_IPV6) */
3717 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3718 return overhead;
3719 }
3720}
3721EXPORT_SYMBOL(kernel_sock_ip_overhead);