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