1/*
   2 *  linux/fs/compat.c
   3 *
   4 *  Kernel compatibililty routines for e.g. 32 bit syscall support
   5 *  on 64 bit kernels.
   6 *
   7 *  Copyright (C) 2002       Stephen Rothwell, IBM Corporation
   8 *  Copyright (C) 1997-2000  Jakub Jelinek  (jakub@redhat.com)
   9 *  Copyright (C) 1998       Eddie C. Dost  (ecd@skynet.be)
  10 *  Copyright (C) 2001,2002  Andi Kleen, SuSE Labs 
  11 *  Copyright (C) 2003       Pavel Machek (pavel@ucw.cz)
  12 *
  13 *  This program is free software; you can redistribute it and/or modify
  14 *  it under the terms of the GNU General Public License version 2 as
  15 *  published by the Free Software Foundation.
  16 */
  17
  18#include <linux/stddef.h>
  19#include <linux/kernel.h>
  20#include <linux/linkage.h>
  21#include <linux/compat.h>
  22#include <linux/errno.h>
  23#include <linux/time.h>
  24#include <linux/fs.h>
  25#include <linux/fcntl.h>
  26#include <linux/namei.h>
  27#include <linux/file.h>
  28#include <linux/fdtable.h>
  29#include <linux/vfs.h>
  30#include <linux/ioctl.h>
  31#include <linux/init.h>
  32#include <linux/ncp_mount.h>
  33#include <linux/nfs4_mount.h>
  34#include <linux/syscalls.h>
  35#include <linux/ctype.h>
  36#include <linux/module.h>
  37#include <linux/dirent.h>
  38#include <linux/fsnotify.h>
  39#include <linux/highuid.h>
  40#include <linux/nfsd/syscall.h>
  41#include <linux/personality.h>
  42#include <linux/rwsem.h>
  43#include <linux/tsacct_kern.h>
  44#include <linux/security.h>
  45#include <linux/highmem.h>
  46#include <linux/signal.h>
  47#include <linux/poll.h>
  48#include <linux/mm.h>
  49#include <linux/eventpoll.h>
  50#include <linux/fs_struct.h>
  51#include <linux/slab.h>
  52#include <linux/pagemap.h>
  53
  54#include <asm/uaccess.h>
  55#include <asm/mmu_context.h>
  56#include <asm/ioctls.h>
  57#include "internal.h"
  58
  59int compat_log = 1;
  60
  61int compat_printk(const char *fmt, ...)
  62{
  63	va_list ap;
  64	int ret;
  65	if (!compat_log)
  66		return 0;
  67	va_start(ap, fmt);
  68	ret = vprintk(fmt, ap);
  69	va_end(ap);
  70	return ret;
  71}
  72
  73#include "read_write.h"
  74
  75/*
  76 * Not all architectures have sys_utime, so implement this in terms
  77 * of sys_utimes.
  78 */
  79asmlinkage long compat_sys_utime(const char __user *filename,
  80				 struct compat_utimbuf __user *t)
  81{
  82	struct timespec tv[2];
  83
  84	if (t) {
  85		if (get_user(tv[0].tv_sec, &t->actime) ||
  86		    get_user(tv[1].tv_sec, &t->modtime))
  87			return -EFAULT;
  88		tv[0].tv_nsec = 0;
  89		tv[1].tv_nsec = 0;
  90	}
  91	return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
  92}
  93
  94asmlinkage long compat_sys_utimensat(unsigned int dfd, const char __user *filename, struct compat_timespec __user *t, int flags)
  95{
  96	struct timespec tv[2];
  97
  98	if  (t) {
  99		if (get_compat_timespec(&tv[0], &t[0]) ||
 100		    get_compat_timespec(&tv[1], &t[1]))
 101			return -EFAULT;
 102
 103		if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
 104			return 0;
 105	}
 106	return do_utimes(dfd, filename, t ? tv : NULL, flags);
 107}
 108
 109asmlinkage long compat_sys_futimesat(unsigned int dfd, const char __user *filename, struct compat_timeval __user *t)
 110{
 111	struct timespec tv[2];
 112
 113	if (t) {
 114		if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
 115		    get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
 116		    get_user(tv[1].tv_sec, &t[1].tv_sec) ||
 117		    get_user(tv[1].tv_nsec, &t[1].tv_usec))
 118			return -EFAULT;
 119		if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
 120		    tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
 121			return -EINVAL;
 122		tv[0].tv_nsec *= 1000;
 123		tv[1].tv_nsec *= 1000;
 124	}
 125	return do_utimes(dfd, filename, t ? tv : NULL, 0);
 126}
 127
 128asmlinkage long compat_sys_utimes(const char __user *filename, struct compat_timeval __user *t)
 129{
 130	return compat_sys_futimesat(AT_FDCWD, filename, t);
 131}
 132
 133static int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
 134{
 135	compat_ino_t ino = stat->ino;
 136	typeof(ubuf->st_uid) uid = 0;
 137	typeof(ubuf->st_gid) gid = 0;
 138	int err;
 139
 140	SET_UID(uid, stat->uid);
 141	SET_GID(gid, stat->gid);
 142
 143	if ((u64) stat->size > MAX_NON_LFS ||
 144	    !old_valid_dev(stat->dev) ||
 145	    !old_valid_dev(stat->rdev))
 146		return -EOVERFLOW;
 147	if (sizeof(ino) < sizeof(stat->ino) && ino != stat->ino)
 148		return -EOVERFLOW;
 149
 150	if (clear_user(ubuf, sizeof(*ubuf)))
 151		return -EFAULT;
 152
 153	err  = __put_user(old_encode_dev(stat->dev), &ubuf->st_dev);
 154	err |= __put_user(ino, &ubuf->st_ino);
 155	err |= __put_user(stat->mode, &ubuf->st_mode);
 156	err |= __put_user(stat->nlink, &ubuf->st_nlink);
 157	err |= __put_user(uid, &ubuf->st_uid);
 158	err |= __put_user(gid, &ubuf->st_gid);
 159	err |= __put_user(old_encode_dev(stat->rdev), &ubuf->st_rdev);
 160	err |= __put_user(stat->size, &ubuf->st_size);
 161	err |= __put_user(stat->atime.tv_sec, &ubuf->st_atime);
 162	err |= __put_user(stat->atime.tv_nsec, &ubuf->st_atime_nsec);
 163	err |= __put_user(stat->mtime.tv_sec, &ubuf->st_mtime);
 164	err |= __put_user(stat->mtime.tv_nsec, &ubuf->st_mtime_nsec);
 165	err |= __put_user(stat->ctime.tv_sec, &ubuf->st_ctime);
 166	err |= __put_user(stat->ctime.tv_nsec, &ubuf->st_ctime_nsec);
 167	err |= __put_user(stat->blksize, &ubuf->st_blksize);
 168	err |= __put_user(stat->blocks, &ubuf->st_blocks);
 169	return err;
 170}
 171
 172asmlinkage long compat_sys_newstat(const char __user * filename,
 173		struct compat_stat __user *statbuf)
 174{
 175	struct kstat stat;
 176	int error;
 177
 178	error = vfs_stat(filename, &stat);
 179	if (error)
 180		return error;
 181	return cp_compat_stat(&stat, statbuf);
 182}
 183
 184asmlinkage long compat_sys_newlstat(const char __user * filename,
 185		struct compat_stat __user *statbuf)
 186{
 187	struct kstat stat;
 188	int error;
 189
 190	error = vfs_lstat(filename, &stat);
 191	if (error)
 192		return error;
 193	return cp_compat_stat(&stat, statbuf);
 194}
 195
 196#ifndef __ARCH_WANT_STAT64
 197asmlinkage long compat_sys_newfstatat(unsigned int dfd,
 198		const char __user *filename,
 199		struct compat_stat __user *statbuf, int flag)
 200{
 201	struct kstat stat;
 202	int error;
 203
 204	error = vfs_fstatat(dfd, filename, &stat, flag);
 205	if (error)
 206		return error;
 207	return cp_compat_stat(&stat, statbuf);
 208}
 209#endif
 210
 211asmlinkage long compat_sys_newfstat(unsigned int fd,
 212		struct compat_stat __user * statbuf)
 213{
 214	struct kstat stat;
 215	int error = vfs_fstat(fd, &stat);
 216
 217	if (!error)
 218		error = cp_compat_stat(&stat, statbuf);
 219	return error;
 220}
 221
 222static int put_compat_statfs(struct compat_statfs __user *ubuf, struct kstatfs *kbuf)
 223{
 224	
 225	if (sizeof ubuf->f_blocks == 4) {
 226		if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
 227		     kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
 228			return -EOVERFLOW;
 229		/* f_files and f_ffree may be -1; it's okay
 230		 * to stuff that into 32 bits */
 231		if (kbuf->f_files != 0xffffffffffffffffULL
 232		 && (kbuf->f_files & 0xffffffff00000000ULL))
 233			return -EOVERFLOW;
 234		if (kbuf->f_ffree != 0xffffffffffffffffULL
 235		 && (kbuf->f_ffree & 0xffffffff00000000ULL))
 236			return -EOVERFLOW;
 237	}
 238	if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
 239	    __put_user(kbuf->f_type, &ubuf->f_type) ||
 240	    __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
 241	    __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
 242	    __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
 243	    __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
 244	    __put_user(kbuf->f_files, &ubuf->f_files) ||
 245	    __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
 246	    __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
 247	    __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
 248	    __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
 249	    __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
 250	    __put_user(0, &ubuf->f_spare[0]) || 
 251	    __put_user(0, &ubuf->f_spare[1]) || 
 252	    __put_user(0, &ubuf->f_spare[2]) || 
 253	    __put_user(0, &ubuf->f_spare[3]) || 
 254	    __put_user(0, &ubuf->f_spare[4]))
 255		return -EFAULT;
 256	return 0;
 257}
 258
 259/*
 260 * The following statfs calls are copies of code from fs/statfs.c and
 261 * should be checked against those from time to time
 262 */
 263asmlinkage long compat_sys_statfs(const char __user *pathname, struct compat_statfs __user *buf)
 264{
 265	struct kstatfs tmp;
 266	int error = user_statfs(pathname, &tmp);
 267	if (!error)
 268		error = put_compat_statfs(buf, &tmp);
 269	return error;
 270}
 271
 272asmlinkage long compat_sys_fstatfs(unsigned int fd, struct compat_statfs __user *buf)
 273{
 274	struct kstatfs tmp;
 275	int error = fd_statfs(fd, &tmp);
 276	if (!error)
 277		error = put_compat_statfs(buf, &tmp);
 278	return error;
 279}
 280
 281static int put_compat_statfs64(struct compat_statfs64 __user *ubuf, struct kstatfs *kbuf)
 282{
 283	if (sizeof ubuf->f_blocks == 4) {
 284		if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
 285		     kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
 286			return -EOVERFLOW;
 287		/* f_files and f_ffree may be -1; it's okay
 288		 * to stuff that into 32 bits */
 289		if (kbuf->f_files != 0xffffffffffffffffULL
 290		 && (kbuf->f_files & 0xffffffff00000000ULL))
 291			return -EOVERFLOW;
 292		if (kbuf->f_ffree != 0xffffffffffffffffULL
 293		 && (kbuf->f_ffree & 0xffffffff00000000ULL))
 294			return -EOVERFLOW;
 295	}
 296	if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
 297	    __put_user(kbuf->f_type, &ubuf->f_type) ||
 298	    __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
 299	    __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
 300	    __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
 301	    __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
 302	    __put_user(kbuf->f_files, &ubuf->f_files) ||
 303	    __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
 304	    __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
 305	    __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
 306	    __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
 307	    __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
 308	    __put_user(kbuf->f_flags, &ubuf->f_flags) ||
 309	    __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
 310		return -EFAULT;
 311	return 0;
 312}
 313
 314asmlinkage long compat_sys_statfs64(const char __user *pathname, compat_size_t sz, struct compat_statfs64 __user *buf)
 315{
 316	struct kstatfs tmp;
 317	int error;
 318
 319	if (sz != sizeof(*buf))
 320		return -EINVAL;
 321
 322	error = user_statfs(pathname, &tmp);
 323	if (!error)
 324		error = put_compat_statfs64(buf, &tmp);
 325	return error;
 326}
 327
 328asmlinkage long compat_sys_fstatfs64(unsigned int fd, compat_size_t sz, struct compat_statfs64 __user *buf)
 329{
 330	struct kstatfs tmp;
 331	int error;
 332
 333	if (sz != sizeof(*buf))
 334		return -EINVAL;
 335
 336	error = fd_statfs(fd, &tmp);
 337	if (!error)
 338		error = put_compat_statfs64(buf, &tmp);
 339	return error;
 340}
 341
 342/*
 343 * This is a copy of sys_ustat, just dealing with a structure layout.
 344 * Given how simple this syscall is that apporach is more maintainable
 345 * than the various conversion hacks.
 346 */
 347asmlinkage long compat_sys_ustat(unsigned dev, struct compat_ustat __user *u)
 348{
 349	struct super_block *sb;
 350	struct compat_ustat tmp;
 351	struct kstatfs sbuf;
 352	int err;
 353
 354	sb = user_get_super(new_decode_dev(dev));
 355	if (!sb)
 356		return -EINVAL;
 357	err = statfs_by_dentry(sb->s_root, &sbuf);
 358	drop_super(sb);
 359	if (err)
 360		return err;
 361
 362	memset(&tmp, 0, sizeof(struct compat_ustat));
 363	tmp.f_tfree = sbuf.f_bfree;
 364	tmp.f_tinode = sbuf.f_ffree;
 365	if (copy_to_user(u, &tmp, sizeof(struct compat_ustat)))
 366		return -EFAULT;
 367	return 0;
 368}
 369
 370static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 371{
 372	if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 373	    __get_user(kfl->l_type, &ufl->l_type) ||
 374	    __get_user(kfl->l_whence, &ufl->l_whence) ||
 375	    __get_user(kfl->l_start, &ufl->l_start) ||
 376	    __get_user(kfl->l_len, &ufl->l_len) ||
 377	    __get_user(kfl->l_pid, &ufl->l_pid))
 378		return -EFAULT;
 379	return 0;
 380}
 381
 382static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 383{
 384	if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 385	    __put_user(kfl->l_type, &ufl->l_type) ||
 386	    __put_user(kfl->l_whence, &ufl->l_whence) ||
 387	    __put_user(kfl->l_start, &ufl->l_start) ||
 388	    __put_user(kfl->l_len, &ufl->l_len) ||
 389	    __put_user(kfl->l_pid, &ufl->l_pid))
 390		return -EFAULT;
 391	return 0;
 392}
 393
 394#ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
 395static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 396{
 397	if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 398	    __get_user(kfl->l_type, &ufl->l_type) ||
 399	    __get_user(kfl->l_whence, &ufl->l_whence) ||
 400	    __get_user(kfl->l_start, &ufl->l_start) ||
 401	    __get_user(kfl->l_len, &ufl->l_len) ||
 402	    __get_user(kfl->l_pid, &ufl->l_pid))
 403		return -EFAULT;
 404	return 0;
 405}
 406#endif
 407
 408#ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
 409static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 410{
 411	if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 412	    __put_user(kfl->l_type, &ufl->l_type) ||
 413	    __put_user(kfl->l_whence, &ufl->l_whence) ||
 414	    __put_user(kfl->l_start, &ufl->l_start) ||
 415	    __put_user(kfl->l_len, &ufl->l_len) ||
 416	    __put_user(kfl->l_pid, &ufl->l_pid))
 417		return -EFAULT;
 418	return 0;
 419}
 420#endif
 421
 422asmlinkage long compat_sys_fcntl64(unsigned int fd, unsigned int cmd,
 423		unsigned long arg)
 424{
 425	mm_segment_t old_fs;
 426	struct flock f;
 427	long ret;
 428
 429	switch (cmd) {
 430	case F_GETLK:
 431	case F_SETLK:
 432	case F_SETLKW:
 433		ret = get_compat_flock(&f, compat_ptr(arg));
 434		if (ret != 0)
 435			break;
 436		old_fs = get_fs();
 437		set_fs(KERNEL_DS);
 438		ret = sys_fcntl(fd, cmd, (unsigned long)&f);
 439		set_fs(old_fs);
 440		if (cmd == F_GETLK && ret == 0) {
 441			/* GETLK was successful and we need to return the data...
 442			 * but it needs to fit in the compat structure.
 443			 * l_start shouldn't be too big, unless the original
 444			 * start + end is greater than COMPAT_OFF_T_MAX, in which
 445			 * case the app was asking for trouble, so we return
 446			 * -EOVERFLOW in that case.
 447			 * l_len could be too big, in which case we just truncate it,
 448			 * and only allow the app to see that part of the conflicting
 449			 * lock that might make sense to it anyway
 450			 */
 451
 452			if (f.l_start > COMPAT_OFF_T_MAX)
 453				ret = -EOVERFLOW;
 454			if (f.l_len > COMPAT_OFF_T_MAX)
 455				f.l_len = COMPAT_OFF_T_MAX;
 456			if (ret == 0)
 457				ret = put_compat_flock(&f, compat_ptr(arg));
 458		}
 459		break;
 460
 461	case F_GETLK64:
 462	case F_SETLK64:
 463	case F_SETLKW64:
 464		ret = get_compat_flock64(&f, compat_ptr(arg));
 465		if (ret != 0)
 466			break;
 467		old_fs = get_fs();
 468		set_fs(KERNEL_DS);
 469		ret = sys_fcntl(fd, (cmd == F_GETLK64) ? F_GETLK :
 470				((cmd == F_SETLK64) ? F_SETLK : F_SETLKW),
 471				(unsigned long)&f);
 472		set_fs(old_fs);
 473		if (cmd == F_GETLK64 && ret == 0) {
 474			/* need to return lock information - see above for commentary */
 475			if (f.l_start > COMPAT_LOFF_T_MAX)
 476				ret = -EOVERFLOW;
 477			if (f.l_len > COMPAT_LOFF_T_MAX)
 478				f.l_len = COMPAT_LOFF_T_MAX;
 479			if (ret == 0)
 480				ret = put_compat_flock64(&f, compat_ptr(arg));
 481		}
 482		break;
 483
 484	default:
 485		ret = sys_fcntl(fd, cmd, arg);
 486		break;
 487	}
 488	return ret;
 489}
 490
 491asmlinkage long compat_sys_fcntl(unsigned int fd, unsigned int cmd,
 492		unsigned long arg)
 493{
 494	if ((cmd == F_GETLK64) || (cmd == F_SETLK64) || (cmd == F_SETLKW64))
 495		return -EINVAL;
 496	return compat_sys_fcntl64(fd, cmd, arg);
 497}
 498
 499asmlinkage long
 500compat_sys_io_setup(unsigned nr_reqs, u32 __user *ctx32p)
 501{
 502	long ret;
 503	aio_context_t ctx64;
 504
 505	mm_segment_t oldfs = get_fs();
 506	if (unlikely(get_user(ctx64, ctx32p)))
 507		return -EFAULT;
 508
 509	set_fs(KERNEL_DS);
 510	/* The __user pointer cast is valid because of the set_fs() */
 511	ret = sys_io_setup(nr_reqs, (aio_context_t __user *) &ctx64);
 512	set_fs(oldfs);
 513	/* truncating is ok because it's a user address */
 514	if (!ret)
 515		ret = put_user((u32) ctx64, ctx32p);
 516	return ret;
 517}
 518
 519asmlinkage long
 520compat_sys_io_getevents(aio_context_t ctx_id,
 521				 unsigned long min_nr,
 522				 unsigned long nr,
 523				 struct io_event __user *events,
 524				 struct compat_timespec __user *timeout)
 525{
 526	long ret;
 527	struct timespec t;
 528	struct timespec __user *ut = NULL;
 529
 530	ret = -EFAULT;
 531	if (unlikely(!access_ok(VERIFY_WRITE, events, 
 532				nr * sizeof(struct io_event))))
 533		goto out;
 534	if (timeout) {
 535		if (get_compat_timespec(&t, timeout))
 536			goto out;
 537
 538		ut = compat_alloc_user_space(sizeof(*ut));
 539		if (copy_to_user(ut, &t, sizeof(t)) )
 540			goto out;
 541	} 
 542	ret = sys_io_getevents(ctx_id, min_nr, nr, events, ut);
 543out:
 544	return ret;
 545}
 546
 547/* A write operation does a read from user space and vice versa */
 548#define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 549
 550ssize_t compat_rw_copy_check_uvector(int type,
 551		const struct compat_iovec __user *uvector, unsigned long nr_segs,
 552		unsigned long fast_segs, struct iovec *fast_pointer,
 553		struct iovec **ret_pointer)
 554{
 555	compat_ssize_t tot_len;
 556	struct iovec *iov = *ret_pointer = fast_pointer;
 557	ssize_t ret = 0;
 558	int seg;
 559
 560	/*
 561	 * SuS says "The readv() function *may* fail if the iovcnt argument
 562	 * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 563	 * traditionally returned zero for zero segments, so...
 564	 */
 565	if (nr_segs == 0)
 566		goto out;
 567
 568	ret = -EINVAL;
 569	if (nr_segs > UIO_MAXIOV || nr_segs < 0)
 570		goto out;
 571	if (nr_segs > fast_segs) {
 572		ret = -ENOMEM;
 573		iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 574		if (iov == NULL)
 575			goto out;
 576	}
 577	*ret_pointer = iov;
 578
 579	/*
 580	 * Single unix specification:
 581	 * We should -EINVAL if an element length is not >= 0 and fitting an
 582	 * ssize_t.
 583	 *
 584	 * In Linux, the total length is limited to MAX_RW_COUNT, there is
 585	 * no overflow possibility.
 586	 */
 587	tot_len = 0;
 588	ret = -EINVAL;
 589	for (seg = 0; seg < nr_segs; seg++) {
 590		compat_uptr_t buf;
 591		compat_ssize_t len;
 592
 593		if (__get_user(len, &uvector->iov_len) ||
 594		   __get_user(buf, &uvector->iov_base)) {
 595			ret = -EFAULT;
 596			goto out;
 597		}
 598		if (len < 0)	/* size_t not fitting in compat_ssize_t .. */
 599			goto out;
 600		if (!access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
 601			ret = -EFAULT;
 602			goto out;
 603		}
 604		if (len > MAX_RW_COUNT - tot_len)
 605			len = MAX_RW_COUNT - tot_len;
 606		tot_len += len;
 607		iov->iov_base = compat_ptr(buf);
 608		iov->iov_len = (compat_size_t) len;
 609		uvector++;
 610		iov++;
 611	}
 612	ret = tot_len;
 613
 614out:
 615	return ret;
 616}
 617
 618static inline long
 619copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
 620{
 621	compat_uptr_t uptr;
 622	int i;
 623
 624	for (i = 0; i < nr; ++i) {
 625		if (get_user(uptr, ptr32 + i))
 626			return -EFAULT;
 627		if (put_user(compat_ptr(uptr), ptr64 + i))
 628			return -EFAULT;
 629	}
 630	return 0;
 631}
 632
 633#define MAX_AIO_SUBMITS 	(PAGE_SIZE/sizeof(struct iocb *))
 634
 635asmlinkage long
 636compat_sys_io_submit(aio_context_t ctx_id, int nr, u32 __user *iocb)
 637{
 638	struct iocb __user * __user *iocb64; 
 639	long ret;
 640
 641	if (unlikely(nr < 0))
 642		return -EINVAL;
 643
 644	if (nr > MAX_AIO_SUBMITS)
 645		nr = MAX_AIO_SUBMITS;
 646	
 647	iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
 648	ret = copy_iocb(nr, iocb, iocb64);
 649	if (!ret)
 650		ret = do_io_submit(ctx_id, nr, iocb64, 1);
 651	return ret;
 652}
 653
 654struct compat_ncp_mount_data {
 655	compat_int_t version;
 656	compat_uint_t ncp_fd;
 657	__compat_uid_t mounted_uid;
 658	compat_pid_t wdog_pid;
 659	unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
 660	compat_uint_t time_out;
 661	compat_uint_t retry_count;
 662	compat_uint_t flags;
 663	__compat_uid_t uid;
 664	__compat_gid_t gid;
 665	compat_mode_t file_mode;
 666	compat_mode_t dir_mode;
 667};
 668
 669struct compat_ncp_mount_data_v4 {
 670	compat_int_t version;
 671	compat_ulong_t flags;
 672	compat_ulong_t mounted_uid;
 673	compat_long_t wdog_pid;
 674	compat_uint_t ncp_fd;
 675	compat_uint_t time_out;
 676	compat_uint_t retry_count;
 677	compat_ulong_t uid;
 678	compat_ulong_t gid;
 679	compat_ulong_t file_mode;
 680	compat_ulong_t dir_mode;
 681};
 682
 683static void *do_ncp_super_data_conv(void *raw_data)
 684{
 685	int version = *(unsigned int *)raw_data;
 686
 687	if (version == 3) {
 688		struct compat_ncp_mount_data *c_n = raw_data;
 689		struct ncp_mount_data *n = raw_data;
 690
 691		n->dir_mode = c_n->dir_mode;
 692		n->file_mode = c_n->file_mode;
 693		n->gid = c_n->gid;
 694		n->uid = c_n->uid;
 695		memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
 696		n->wdog_pid = c_n->wdog_pid;
 697		n->mounted_uid = c_n->mounted_uid;
 698	} else if (version == 4) {
 699		struct compat_ncp_mount_data_v4 *c_n = raw_data;
 700		struct ncp_mount_data_v4 *n = raw_data;
 701
 702		n->dir_mode = c_n->dir_mode;
 703		n->file_mode = c_n->file_mode;
 704		n->gid = c_n->gid;
 705		n->uid = c_n->uid;
 706		n->retry_count = c_n->retry_count;
 707		n->time_out = c_n->time_out;
 708		n->ncp_fd = c_n->ncp_fd;
 709		n->wdog_pid = c_n->wdog_pid;
 710		n->mounted_uid = c_n->mounted_uid;
 711		n->flags = c_n->flags;
 712	} else if (version != 5) {
 713		return NULL;
 714	}
 715
 716	return raw_data;
 717}
 718
 719
 720struct compat_nfs_string {
 721	compat_uint_t len;
 722	compat_uptr_t data;
 723};
 724
 725static inline void compat_nfs_string(struct nfs_string *dst,
 726				     struct compat_nfs_string *src)
 727{
 728	dst->data = compat_ptr(src->data);
 729	dst->len = src->len;
 730}
 731
 732struct compat_nfs4_mount_data_v1 {
 733	compat_int_t version;
 734	compat_int_t flags;
 735	compat_int_t rsize;
 736	compat_int_t wsize;
 737	compat_int_t timeo;
 738	compat_int_t retrans;
 739	compat_int_t acregmin;
 740	compat_int_t acregmax;
 741	compat_int_t acdirmin;
 742	compat_int_t acdirmax;
 743	struct compat_nfs_string client_addr;
 744	struct compat_nfs_string mnt_path;
 745	struct compat_nfs_string hostname;
 746	compat_uint_t host_addrlen;
 747	compat_uptr_t host_addr;
 748	compat_int_t proto;
 749	compat_int_t auth_flavourlen;
 750	compat_uptr_t auth_flavours;
 751};
 752
 753static int do_nfs4_super_data_conv(void *raw_data)
 754{
 755	int version = *(compat_uint_t *) raw_data;
 756
 757	if (version == 1) {
 758		struct compat_nfs4_mount_data_v1 *raw = raw_data;
 759		struct nfs4_mount_data *real = raw_data;
 760
 761		/* copy the fields backwards */
 762		real->auth_flavours = compat_ptr(raw->auth_flavours);
 763		real->auth_flavourlen = raw->auth_flavourlen;
 764		real->proto = raw->proto;
 765		real->host_addr = compat_ptr(raw->host_addr);
 766		real->host_addrlen = raw->host_addrlen;
 767		compat_nfs_string(&real->hostname, &raw->hostname);
 768		compat_nfs_string(&real->mnt_path, &raw->mnt_path);
 769		compat_nfs_string(&real->client_addr, &raw->client_addr);
 770		real->acdirmax = raw->acdirmax;
 771		real->acdirmin = raw->acdirmin;
 772		real->acregmax = raw->acregmax;
 773		real->acregmin = raw->acregmin;
 774		real->retrans = raw->retrans;
 775		real->timeo = raw->timeo;
 776		real->wsize = raw->wsize;
 777		real->rsize = raw->rsize;
 778		real->flags = raw->flags;
 779		real->version = raw->version;
 780	}
 781
 782	return 0;
 783}
 784
 785#define NCPFS_NAME      "ncpfs"
 786#define NFS4_NAME	"nfs4"
 787
 788asmlinkage long compat_sys_mount(const char __user * dev_name,
 789				 const char __user * dir_name,
 790				 const char __user * type, unsigned long flags,
 791				 const void __user * data)
 792{
 793	char *kernel_type;
 794	unsigned long data_page;
 795	char *kernel_dev;
 796	char *dir_page;
 797	int retval;
 798
 799	retval = copy_mount_string(type, &kernel_type);
 800	if (retval < 0)
 
 801		goto out;
 802
 803	dir_page = getname(dir_name);
 804	retval = PTR_ERR(dir_page);
 805	if (IS_ERR(dir_page))
 806		goto out1;
 807
 808	retval = copy_mount_string(dev_name, &kernel_dev);
 809	if (retval < 0)
 
 810		goto out2;
 811
 812	retval = copy_mount_options(data, &data_page);
 813	if (retval < 0)
 814		goto out3;
 815
 816	retval = -EINVAL;
 817
 818	if (kernel_type && data_page) {
 819		if (!strcmp(kernel_type, NCPFS_NAME)) {
 820			do_ncp_super_data_conv((void *)data_page);
 821		} else if (!strcmp(kernel_type, NFS4_NAME)) {
 822			if (do_nfs4_super_data_conv((void *) data_page))
 823				goto out4;
 824		}
 825	}
 826
 827	retval = do_mount(kernel_dev, dir_page, kernel_type,
 828			flags, (void*)data_page);
 829
 830 out4:
 831	free_page(data_page);
 832 out3:
 833	kfree(kernel_dev);
 834 out2:
 835	putname(dir_page);
 836 out1:
 837	kfree(kernel_type);
 838 out:
 839	return retval;
 840}
 841
 842struct compat_old_linux_dirent {
 843	compat_ulong_t	d_ino;
 844	compat_ulong_t	d_offset;
 845	unsigned short	d_namlen;
 846	char		d_name[1];
 847};
 848
 849struct compat_readdir_callback {
 850	struct compat_old_linux_dirent __user *dirent;
 851	int result;
 852};
 853
 854static int compat_fillonedir(void *__buf, const char *name, int namlen,
 855			loff_t offset, u64 ino, unsigned int d_type)
 856{
 857	struct compat_readdir_callback *buf = __buf;
 858	struct compat_old_linux_dirent __user *dirent;
 859	compat_ulong_t d_ino;
 860
 861	if (buf->result)
 862		return -EINVAL;
 863	d_ino = ino;
 864	if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
 865		buf->result = -EOVERFLOW;
 866		return -EOVERFLOW;
 867	}
 868	buf->result++;
 869	dirent = buf->dirent;
 870	if (!access_ok(VERIFY_WRITE, dirent,
 871			(unsigned long)(dirent->d_name + namlen + 1) -
 872				(unsigned long)dirent))
 873		goto efault;
 874	if (	__put_user(d_ino, &dirent->d_ino) ||
 875		__put_user(offset, &dirent->d_offset) ||
 876		__put_user(namlen, &dirent->d_namlen) ||
 877		__copy_to_user(dirent->d_name, name, namlen) ||
 878		__put_user(0, dirent->d_name + namlen))
 879		goto efault;
 880	return 0;
 881efault:
 882	buf->result = -EFAULT;
 883	return -EFAULT;
 884}
 885
 886asmlinkage long compat_sys_old_readdir(unsigned int fd,
 887	struct compat_old_linux_dirent __user *dirent, unsigned int count)
 888{
 889	int error;
 890	struct file *file;
 891	struct compat_readdir_callback buf;
 892
 893	error = -EBADF;
 894	file = fget(fd);
 895	if (!file)
 896		goto out;
 897
 898	buf.result = 0;
 899	buf.dirent = dirent;
 900
 901	error = vfs_readdir(file, compat_fillonedir, &buf);
 902	if (buf.result)
 903		error = buf.result;
 904
 905	fput(file);
 906out:
 907	return error;
 908}
 909
 910struct compat_linux_dirent {
 911	compat_ulong_t	d_ino;
 912	compat_ulong_t	d_off;
 913	unsigned short	d_reclen;
 914	char		d_name[1];
 915};
 916
 917struct compat_getdents_callback {
 918	struct compat_linux_dirent __user *current_dir;
 919	struct compat_linux_dirent __user *previous;
 920	int count;
 921	int error;
 922};
 923
 924static int compat_filldir(void *__buf, const char *name, int namlen,
 925		loff_t offset, u64 ino, unsigned int d_type)
 926{
 927	struct compat_linux_dirent __user * dirent;
 928	struct compat_getdents_callback *buf = __buf;
 929	compat_ulong_t d_ino;
 930	int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
 931		namlen + 2, sizeof(compat_long_t));
 932
 933	buf->error = -EINVAL;	/* only used if we fail.. */
 934	if (reclen > buf->count)
 935		return -EINVAL;
 936	d_ino = ino;
 937	if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
 938		buf->error = -EOVERFLOW;
 939		return -EOVERFLOW;
 940	}
 941	dirent = buf->previous;
 942	if (dirent) {
 943		if (__put_user(offset, &dirent->d_off))
 944			goto efault;
 945	}
 946	dirent = buf->current_dir;
 947	if (__put_user(d_ino, &dirent->d_ino))
 948		goto efault;
 949	if (__put_user(reclen, &dirent->d_reclen))
 950		goto efault;
 951	if (copy_to_user(dirent->d_name, name, namlen))
 952		goto efault;
 953	if (__put_user(0, dirent->d_name + namlen))
 954		goto efault;
 955	if (__put_user(d_type, (char  __user *) dirent + reclen - 1))
 956		goto efault;
 957	buf->previous = dirent;
 958	dirent = (void __user *)dirent + reclen;
 959	buf->current_dir = dirent;
 960	buf->count -= reclen;
 961	return 0;
 962efault:
 963	buf->error = -EFAULT;
 964	return -EFAULT;
 965}
 966
 967asmlinkage long compat_sys_getdents(unsigned int fd,
 968		struct compat_linux_dirent __user *dirent, unsigned int count)
 969{
 970	struct file * file;
 971	struct compat_linux_dirent __user * lastdirent;
 972	struct compat_getdents_callback buf;
 973	int error;
 974
 975	error = -EFAULT;
 976	if (!access_ok(VERIFY_WRITE, dirent, count))
 977		goto out;
 978
 979	error = -EBADF;
 980	file = fget(fd);
 981	if (!file)
 982		goto out;
 983
 984	buf.current_dir = dirent;
 985	buf.previous = NULL;
 986	buf.count = count;
 987	buf.error = 0;
 988
 989	error = vfs_readdir(file, compat_filldir, &buf);
 990	if (error >= 0)
 991		error = buf.error;
 992	lastdirent = buf.previous;
 993	if (lastdirent) {
 994		if (put_user(file->f_pos, &lastdirent->d_off))
 995			error = -EFAULT;
 996		else
 997			error = count - buf.count;
 998	}
 999	fput(file);
1000out:
1001	return error;
1002}
1003
1004#ifndef __ARCH_OMIT_COMPAT_SYS_GETDENTS64
1005
1006struct compat_getdents_callback64 {
1007	struct linux_dirent64 __user *current_dir;
1008	struct linux_dirent64 __user *previous;
1009	int count;
1010	int error;
1011};
1012
1013static int compat_filldir64(void * __buf, const char * name, int namlen, loff_t offset,
1014		     u64 ino, unsigned int d_type)
1015{
1016	struct linux_dirent64 __user *dirent;
1017	struct compat_getdents_callback64 *buf = __buf;
1018	int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
1019		sizeof(u64));
1020	u64 off;
1021
1022	buf->error = -EINVAL;	/* only used if we fail.. */
1023	if (reclen > buf->count)
1024		return -EINVAL;
1025	dirent = buf->previous;
1026
1027	if (dirent) {
1028		if (__put_user_unaligned(offset, &dirent->d_off))
1029			goto efault;
1030	}
1031	dirent = buf->current_dir;
1032	if (__put_user_unaligned(ino, &dirent->d_ino))
1033		goto efault;
1034	off = 0;
1035	if (__put_user_unaligned(off, &dirent->d_off))
1036		goto efault;
1037	if (__put_user(reclen, &dirent->d_reclen))
1038		goto efault;
1039	if (__put_user(d_type, &dirent->d_type))
1040		goto efault;
1041	if (copy_to_user(dirent->d_name, name, namlen))
1042		goto efault;
1043	if (__put_user(0, dirent->d_name + namlen))
1044		goto efault;
1045	buf->previous = dirent;
1046	dirent = (void __user *)dirent + reclen;
1047	buf->current_dir = dirent;
1048	buf->count -= reclen;
1049	return 0;
1050efault:
1051	buf->error = -EFAULT;
1052	return -EFAULT;
1053}
1054
1055asmlinkage long compat_sys_getdents64(unsigned int fd,
1056		struct linux_dirent64 __user * dirent, unsigned int count)
1057{
1058	struct file * file;
1059	struct linux_dirent64 __user * lastdirent;
1060	struct compat_getdents_callback64 buf;
1061	int error;
1062
1063	error = -EFAULT;
1064	if (!access_ok(VERIFY_WRITE, dirent, count))
1065		goto out;
1066
1067	error = -EBADF;
1068	file = fget(fd);
1069	if (!file)
1070		goto out;
1071
1072	buf.current_dir = dirent;
1073	buf.previous = NULL;
1074	buf.count = count;
1075	buf.error = 0;
1076
1077	error = vfs_readdir(file, compat_filldir64, &buf);
1078	if (error >= 0)
1079		error = buf.error;
1080	lastdirent = buf.previous;
1081	if (lastdirent) {
1082		typeof(lastdirent->d_off) d_off = file->f_pos;
1083		if (__put_user_unaligned(d_off, &lastdirent->d_off))
1084			error = -EFAULT;
1085		else
1086			error = count - buf.count;
1087	}
1088	fput(file);
1089out:
1090	return error;
1091}
1092#endif /* ! __ARCH_OMIT_COMPAT_SYS_GETDENTS64 */
1093
1094static ssize_t compat_do_readv_writev(int type, struct file *file,
1095			       const struct compat_iovec __user *uvector,
1096			       unsigned long nr_segs, loff_t *pos)
1097{
1098	compat_ssize_t tot_len;
1099	struct iovec iovstack[UIO_FASTIOV];
1100	struct iovec *iov = iovstack;
1101	ssize_t ret;
1102	io_fn_t fn;
1103	iov_fn_t fnv;
1104
1105	ret = -EINVAL;
1106	if (!file->f_op)
1107		goto out;
1108
1109	ret = -EFAULT;
1110	if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
1111		goto out;
1112
1113	tot_len = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1114					       UIO_FASTIOV, iovstack, &iov);
1115	if (tot_len == 0) {
1116		ret = 0;
1117		goto out;
1118	}
1119
1120	ret = rw_verify_area(type, file, pos, tot_len);
1121	if (ret < 0)
1122		goto out;
1123
1124	fnv = NULL;
1125	if (type == READ) {
1126		fn = file->f_op->read;
1127		fnv = file->f_op->aio_read;
1128	} else {
1129		fn = (io_fn_t)file->f_op->write;
1130		fnv = file->f_op->aio_write;
1131	}
1132
1133	if (fnv)
1134		ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1135						pos, fnv);
1136	else
1137		ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1138
1139out:
1140	if (iov != iovstack)
1141		kfree(iov);
1142	if ((ret + (type == READ)) > 0) {
1143		if (type == READ)
1144			fsnotify_access(file);
1145		else
1146			fsnotify_modify(file);
1147	}
1148	return ret;
1149}
1150
1151static size_t compat_readv(struct file *file,
1152			   const struct compat_iovec __user *vec,
1153			   unsigned long vlen, loff_t *pos)
1154{
1155	ssize_t ret = -EBADF;
1156
1157	if (!(file->f_mode & FMODE_READ))
1158		goto out;
1159
1160	ret = -EINVAL;
1161	if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
1162		goto out;
1163
1164	ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1165
1166out:
1167	if (ret > 0)
1168		add_rchar(current, ret);
1169	inc_syscr(current);
1170	return ret;
1171}
1172
1173asmlinkage ssize_t
1174compat_sys_readv(unsigned long fd, const struct compat_iovec __user *vec,
1175		 unsigned long vlen)
1176{
1177	struct file *file;
1178	int fput_needed;
1179	ssize_t ret;
1180
1181	file = fget_light(fd, &fput_needed);
1182	if (!file)
1183		return -EBADF;
1184	ret = compat_readv(file, vec, vlen, &file->f_pos);
1185	fput_light(file, fput_needed);
1186	return ret;
1187}
1188
1189asmlinkage ssize_t
1190compat_sys_preadv(unsigned long fd, const struct compat_iovec __user *vec,
1191		  unsigned long vlen, u32 pos_low, u32 pos_high)
1192{
1193	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1194	struct file *file;
1195	int fput_needed;
1196	ssize_t ret;
1197
1198	if (pos < 0)
1199		return -EINVAL;
1200	file = fget_light(fd, &fput_needed);
1201	if (!file)
1202		return -EBADF;
1203	ret = -ESPIPE;
1204	if (file->f_mode & FMODE_PREAD)
1205		ret = compat_readv(file, vec, vlen, &pos);
1206	fput_light(file, fput_needed);
1207	return ret;
1208}
1209
1210static size_t compat_writev(struct file *file,
1211			    const struct compat_iovec __user *vec,
1212			    unsigned long vlen, loff_t *pos)
1213{
1214	ssize_t ret = -EBADF;
1215
1216	if (!(file->f_mode & FMODE_WRITE))
1217		goto out;
1218
1219	ret = -EINVAL;
1220	if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
1221		goto out;
1222
1223	ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1224
1225out:
1226	if (ret > 0)
1227		add_wchar(current, ret);
1228	inc_syscw(current);
1229	return ret;
1230}
1231
1232asmlinkage ssize_t
1233compat_sys_writev(unsigned long fd, const struct compat_iovec __user *vec,
1234		  unsigned long vlen)
1235{
1236	struct file *file;
1237	int fput_needed;
1238	ssize_t ret;
1239
1240	file = fget_light(fd, &fput_needed);
1241	if (!file)
1242		return -EBADF;
1243	ret = compat_writev(file, vec, vlen, &file->f_pos);
1244	fput_light(file, fput_needed);
1245	return ret;
1246}
1247
1248asmlinkage ssize_t
1249compat_sys_pwritev(unsigned long fd, const struct compat_iovec __user *vec,
1250		   unsigned long vlen, u32 pos_low, u32 pos_high)
1251{
1252	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1253	struct file *file;
1254	int fput_needed;
1255	ssize_t ret;
1256
1257	if (pos < 0)
1258		return -EINVAL;
1259	file = fget_light(fd, &fput_needed);
1260	if (!file)
1261		return -EBADF;
1262	ret = -ESPIPE;
1263	if (file->f_mode & FMODE_PWRITE)
1264		ret = compat_writev(file, vec, vlen, &pos);
1265	fput_light(file, fput_needed);
1266	return ret;
1267}
1268
1269asmlinkage long
1270compat_sys_vmsplice(int fd, const struct compat_iovec __user *iov32,
1271		    unsigned int nr_segs, unsigned int flags)
1272{
1273	unsigned i;
1274	struct iovec __user *iov;
1275	if (nr_segs > UIO_MAXIOV)
1276		return -EINVAL;
1277	iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1278	for (i = 0; i < nr_segs; i++) {
1279		struct compat_iovec v;
1280		if (get_user(v.iov_base, &iov32[i].iov_base) ||
1281		    get_user(v.iov_len, &iov32[i].iov_len) ||
1282		    put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1283		    put_user(v.iov_len, &iov[i].iov_len))
1284			return -EFAULT;
1285	}
1286	return sys_vmsplice(fd, iov, nr_segs, flags);
1287}
1288
1289/*
1290 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1291 * O_LARGEFILE flag.
1292 */
1293asmlinkage long
1294compat_sys_open(const char __user *filename, int flags, int mode)
1295{
1296	return do_sys_open(AT_FDCWD, filename, flags, mode);
1297}
1298
1299/*
1300 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1301 * O_LARGEFILE flag.
1302 */
1303asmlinkage long
1304compat_sys_openat(unsigned int dfd, const char __user *filename, int flags, int mode)
1305{
1306	return do_sys_open(dfd, filename, flags, mode);
1307}
1308
1309#define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1310
1311static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1312				      int timeval, int ret)
1313{
1314	struct timespec ts;
1315
1316	if (!p)
1317		return ret;
1318
1319	if (current->personality & STICKY_TIMEOUTS)
1320		goto sticky;
1321
1322	/* No update for zero timeout */
1323	if (!end_time->tv_sec && !end_time->tv_nsec)
1324		return ret;
1325
1326	ktime_get_ts(&ts);
1327	ts = timespec_sub(*end_time, ts);
1328	if (ts.tv_sec < 0)
1329		ts.tv_sec = ts.tv_nsec = 0;
1330
1331	if (timeval) {
1332		struct compat_timeval rtv;
1333
1334		rtv.tv_sec = ts.tv_sec;
1335		rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1336
1337		if (!copy_to_user(p, &rtv, sizeof(rtv)))
1338			return ret;
1339	} else {
1340		struct compat_timespec rts;
1341
1342		rts.tv_sec = ts.tv_sec;
1343		rts.tv_nsec = ts.tv_nsec;
1344
1345		if (!copy_to_user(p, &rts, sizeof(rts)))
1346			return ret;
1347	}
1348	/*
1349	 * If an application puts its timeval in read-only memory, we
1350	 * don't want the Linux-specific update to the timeval to
1351	 * cause a fault after the select has completed
1352	 * successfully. However, because we're not updating the
1353	 * timeval, we can't restart the system call.
1354	 */
1355
1356sticky:
1357	if (ret == -ERESTARTNOHAND)
1358		ret = -EINTR;
1359	return ret;
1360}
1361
1362/*
1363 * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1364 * 64-bit unsigned longs.
1365 */
1366static
1367int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1368			unsigned long *fdset)
1369{
1370	nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1371	if (ufdset) {
1372		unsigned long odd;
1373
1374		if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1375			return -EFAULT;
1376
1377		odd = nr & 1UL;
1378		nr &= ~1UL;
1379		while (nr) {
1380			unsigned long h, l;
1381			if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1382				return -EFAULT;
1383			ufdset += 2;
1384			*fdset++ = h << 32 | l;
1385			nr -= 2;
1386		}
1387		if (odd && __get_user(*fdset, ufdset))
1388			return -EFAULT;
1389	} else {
1390		/* Tricky, must clear full unsigned long in the
1391		 * kernel fdset at the end, this makes sure that
1392		 * actually happens.
1393		 */
1394		memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1395	}
1396	return 0;
1397}
1398
1399static
1400int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1401		      unsigned long *fdset)
1402{
1403	unsigned long odd;
1404	nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1405
1406	if (!ufdset)
1407		return 0;
1408
1409	odd = nr & 1UL;
1410	nr &= ~1UL;
1411	while (nr) {
1412		unsigned long h, l;
1413		l = *fdset++;
1414		h = l >> 32;
1415		if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1416			return -EFAULT;
1417		ufdset += 2;
1418		nr -= 2;
1419	}
1420	if (odd && __put_user(*fdset, ufdset))
1421		return -EFAULT;
1422	return 0;
1423}
1424
1425
1426/*
1427 * This is a virtual copy of sys_select from fs/select.c and probably
1428 * should be compared to it from time to time
1429 */
1430
1431/*
1432 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1433 * like to be certain this leads to no problems. So I return
1434 * EINTR just for safety.
1435 *
1436 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1437 * I'm trying ERESTARTNOHAND which restart only when you want to.
1438 */
1439int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1440	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1441	struct timespec *end_time)
1442{
1443	fd_set_bits fds;
1444	void *bits;
1445	int size, max_fds, ret = -EINVAL;
1446	struct fdtable *fdt;
1447	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1448
1449	if (n < 0)
1450		goto out_nofds;
1451
1452	/* max_fds can increase, so grab it once to avoid race */
1453	rcu_read_lock();
1454	fdt = files_fdtable(current->files);
1455	max_fds = fdt->max_fds;
1456	rcu_read_unlock();
1457	if (n > max_fds)
1458		n = max_fds;
1459
1460	/*
1461	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1462	 * since we used fdset we need to allocate memory in units of
1463	 * long-words.
1464	 */
1465	size = FDS_BYTES(n);
1466	bits = stack_fds;
1467	if (size > sizeof(stack_fds) / 6) {
1468		bits = kmalloc(6 * size, GFP_KERNEL);
1469		ret = -ENOMEM;
1470		if (!bits)
1471			goto out_nofds;
1472	}
1473	fds.in      = (unsigned long *)  bits;
1474	fds.out     = (unsigned long *) (bits +   size);
1475	fds.ex      = (unsigned long *) (bits + 2*size);
1476	fds.res_in  = (unsigned long *) (bits + 3*size);
1477	fds.res_out = (unsigned long *) (bits + 4*size);
1478	fds.res_ex  = (unsigned long *) (bits + 5*size);
1479
1480	if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1481	    (ret = compat_get_fd_set(n, outp, fds.out)) ||
1482	    (ret = compat_get_fd_set(n, exp, fds.ex)))
1483		goto out;
1484	zero_fd_set(n, fds.res_in);
1485	zero_fd_set(n, fds.res_out);
1486	zero_fd_set(n, fds.res_ex);
1487
1488	ret = do_select(n, &fds, end_time);
1489
1490	if (ret < 0)
1491		goto out;
1492	if (!ret) {
1493		ret = -ERESTARTNOHAND;
1494		if (signal_pending(current))
1495			goto out;
1496		ret = 0;
1497	}
1498
1499	if (compat_set_fd_set(n, inp, fds.res_in) ||
1500	    compat_set_fd_set(n, outp, fds.res_out) ||
1501	    compat_set_fd_set(n, exp, fds.res_ex))
1502		ret = -EFAULT;
1503out:
1504	if (bits != stack_fds)
1505		kfree(bits);
1506out_nofds:
1507	return ret;
1508}
1509
1510asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1511	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1512	struct compat_timeval __user *tvp)
1513{
1514	struct timespec end_time, *to = NULL;
1515	struct compat_timeval tv;
1516	int ret;
1517
1518	if (tvp) {
1519		if (copy_from_user(&tv, tvp, sizeof(tv)))
1520			return -EFAULT;
1521
1522		to = &end_time;
1523		if (poll_select_set_timeout(to,
1524				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1525				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1526			return -EINVAL;
1527	}
1528
1529	ret = compat_core_sys_select(n, inp, outp, exp, to);
1530	ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1531
1532	return ret;
1533}
1534
1535struct compat_sel_arg_struct {
1536	compat_ulong_t n;
1537	compat_uptr_t inp;
1538	compat_uptr_t outp;
1539	compat_uptr_t exp;
1540	compat_uptr_t tvp;
1541};
1542
1543asmlinkage long compat_sys_old_select(struct compat_sel_arg_struct __user *arg)
1544{
1545	struct compat_sel_arg_struct a;
1546
1547	if (copy_from_user(&a, arg, sizeof(a)))
1548		return -EFAULT;
1549	return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1550				 compat_ptr(a.exp), compat_ptr(a.tvp));
1551}
1552
1553#ifdef HAVE_SET_RESTORE_SIGMASK
1554static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1555	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1556	struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1557	compat_size_t sigsetsize)
1558{
1559	compat_sigset_t ss32;
1560	sigset_t ksigmask, sigsaved;
1561	struct compat_timespec ts;
1562	struct timespec end_time, *to = NULL;
1563	int ret;
1564
1565	if (tsp) {
1566		if (copy_from_user(&ts, tsp, sizeof(ts)))
1567			return -EFAULT;
1568
1569		to = &end_time;
1570		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1571			return -EINVAL;
1572	}
1573
1574	if (sigmask) {
1575		if (sigsetsize != sizeof(compat_sigset_t))
1576			return -EINVAL;
1577		if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1578			return -EFAULT;
1579		sigset_from_compat(&ksigmask, &ss32);
1580
1581		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1582		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1583	}
1584
1585	ret = compat_core_sys_select(n, inp, outp, exp, to);
1586	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1587
1588	if (ret == -ERESTARTNOHAND) {
1589		/*
1590		 * Don't restore the signal mask yet. Let do_signal() deliver
1591		 * the signal on the way back to userspace, before the signal
1592		 * mask is restored.
1593		 */
1594		if (sigmask) {
1595			memcpy(&current->saved_sigmask, &sigsaved,
1596					sizeof(sigsaved));
1597			set_restore_sigmask();
1598		}
1599	} else if (sigmask)
1600		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1601
1602	return ret;
1603}
1604
1605asmlinkage long compat_sys_pselect6(int n, compat_ulong_t __user *inp,
1606	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1607	struct compat_timespec __user *tsp, void __user *sig)
1608{
1609	compat_size_t sigsetsize = 0;
1610	compat_uptr_t up = 0;
1611
1612	if (sig) {
1613		if (!access_ok(VERIFY_READ, sig,
1614				sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1615		    	__get_user(up, (compat_uptr_t __user *)sig) ||
1616		    	__get_user(sigsetsize,
1617				(compat_size_t __user *)(sig+sizeof(up))))
1618			return -EFAULT;
1619	}
1620	return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1621				 sigsetsize);
1622}
1623
1624asmlinkage long compat_sys_ppoll(struct pollfd __user *ufds,
1625	unsigned int nfds, struct compat_timespec __user *tsp,
1626	const compat_sigset_t __user *sigmask, compat_size_t sigsetsize)
1627{
1628	compat_sigset_t ss32;
1629	sigset_t ksigmask, sigsaved;
1630	struct compat_timespec ts;
1631	struct timespec end_time, *to = NULL;
1632	int ret;
1633
1634	if (tsp) {
1635		if (copy_from_user(&ts, tsp, sizeof(ts)))
1636			return -EFAULT;
1637
1638		to = &end_time;
1639		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1640			return -EINVAL;
1641	}
1642
1643	if (sigmask) {
1644		if (sigsetsize != sizeof(compat_sigset_t))
1645			return -EINVAL;
1646		if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1647			return -EFAULT;
1648		sigset_from_compat(&ksigmask, &ss32);
1649
1650		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1651		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1652	}
1653
1654	ret = do_sys_poll(ufds, nfds, to);
1655
1656	/* We can restart this syscall, usually */
1657	if (ret == -EINTR) {
1658		/*
1659		 * Don't restore the signal mask yet. Let do_signal() deliver
1660		 * the signal on the way back to userspace, before the signal
1661		 * mask is restored.
1662		 */
1663		if (sigmask) {
1664			memcpy(&current->saved_sigmask, &sigsaved,
1665				sizeof(sigsaved));
1666			set_restore_sigmask();
1667		}
1668		ret = -ERESTARTNOHAND;
1669	} else if (sigmask)
1670		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1671
1672	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1673
1674	return ret;
1675}
1676#endif /* HAVE_SET_RESTORE_SIGMASK */
1677
1678#ifdef CONFIG_EPOLL
1679
1680#ifdef HAVE_SET_RESTORE_SIGMASK
1681asmlinkage long compat_sys_epoll_pwait(int epfd,
1682			struct compat_epoll_event __user *events,
1683			int maxevents, int timeout,
1684			const compat_sigset_t __user *sigmask,
1685			compat_size_t sigsetsize)
1686{
1687	long err;
1688	compat_sigset_t csigmask;
1689	sigset_t ksigmask, sigsaved;
1690
1691	/*
1692	 * If the caller wants a certain signal mask to be set during the wait,
1693	 * we apply it here.
1694	 */
1695	if (sigmask) {
1696		if (sigsetsize != sizeof(compat_sigset_t))
1697			return -EINVAL;
1698		if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
1699			return -EFAULT;
1700		sigset_from_compat(&ksigmask, &csigmask);
1701		sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1702		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1703	}
1704
1705	err = sys_epoll_wait(epfd, events, maxevents, timeout);
1706
1707	/*
1708	 * If we changed the signal mask, we need to restore the original one.
1709	 * In case we've got a signal while waiting, we do not restore the
1710	 * signal mask yet, and we allow do_signal() to deliver the signal on
1711	 * the way back to userspace, before the signal mask is restored.
1712	 */
1713	if (sigmask) {
1714		if (err == -EINTR) {
1715			memcpy(&current->saved_sigmask, &sigsaved,
1716			       sizeof(sigsaved));
1717			set_restore_sigmask();
1718		} else
1719			sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1720	}
1721
1722	return err;
1723}
1724#endif /* HAVE_SET_RESTORE_SIGMASK */
1725
1726#endif /* CONFIG_EPOLL */
1727
1728#ifdef CONFIG_SIGNALFD
1729
1730asmlinkage long compat_sys_signalfd4(int ufd,
1731				     const compat_sigset_t __user *sigmask,
1732				     compat_size_t sigsetsize, int flags)
1733{
1734	compat_sigset_t ss32;
1735	sigset_t tmp;
1736	sigset_t __user *ksigmask;
1737
1738	if (sigsetsize != sizeof(compat_sigset_t))
1739		return -EINVAL;
1740	if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1741		return -EFAULT;
1742	sigset_from_compat(&tmp, &ss32);
1743	ksigmask = compat_alloc_user_space(sizeof(sigset_t));
1744	if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
1745		return -EFAULT;
1746
1747	return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
1748}
1749
1750asmlinkage long compat_sys_signalfd(int ufd,
1751				    const compat_sigset_t __user *sigmask,
1752				    compat_size_t sigsetsize)
1753{
1754	return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
1755}
1756#endif /* CONFIG_SIGNALFD */
1757
1758#ifdef CONFIG_TIMERFD
1759
1760asmlinkage long compat_sys_timerfd_settime(int ufd, int flags,
1761				   const struct compat_itimerspec __user *utmr,
1762				   struct compat_itimerspec __user *otmr)
1763{
1764	int error;
1765	struct itimerspec t;
1766	struct itimerspec __user *ut;
1767
1768	if (get_compat_itimerspec(&t, utmr))
1769		return -EFAULT;
1770	ut = compat_alloc_user_space(2 * sizeof(struct itimerspec));
1771	if (copy_to_user(&ut[0], &t, sizeof(t)))
1772		return -EFAULT;
1773	error = sys_timerfd_settime(ufd, flags, &ut[0], &ut[1]);
1774	if (!error && otmr)
1775		error = (copy_from_user(&t, &ut[1], sizeof(struct itimerspec)) ||
1776			 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1777
1778	return error;
1779}
1780
1781asmlinkage long compat_sys_timerfd_gettime(int ufd,
1782				   struct compat_itimerspec __user *otmr)
1783{
1784	int error;
1785	struct itimerspec t;
1786	struct itimerspec __user *ut;
1787
1788	ut = compat_alloc_user_space(sizeof(struct itimerspec));
1789	error = sys_timerfd_gettime(ufd, ut);
1790	if (!error)
1791		error = (copy_from_user(&t, ut, sizeof(struct itimerspec)) ||
1792			 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1793
1794	return error;
1795}
1796
1797#endif /* CONFIG_TIMERFD */
1798
1799#ifdef CONFIG_FHANDLE
1800/*
1801 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1802 * doesn't set the O_LARGEFILE flag.
1803 */
1804asmlinkage long
1805compat_sys_open_by_handle_at(int mountdirfd,
1806			     struct file_handle __user *handle, int flags)
1807{
1808	return do_handle_open(mountdirfd, handle, flags);
1809}
1810#endif