linux/fs/compat.c
<<
>>
Prefs
   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/dirent.h>
  37#include <linux/fsnotify.h>
  38#include <linux/highuid.h>
  39#include <linux/personality.h>
  40#include <linux/rwsem.h>
  41#include <linux/tsacct_kern.h>
  42#include <linux/security.h>
  43#include <linux/highmem.h>
  44#include <linux/signal.h>
  45#include <linux/poll.h>
  46#include <linux/mm.h>
  47#include <linux/fs_struct.h>
  48#include <linux/slab.h>
  49#include <linux/pagemap.h>
  50#include <linux/aio.h>
  51
  52#include <asm/uaccess.h>
  53#include <asm/mmu_context.h>
  54#include <asm/ioctls.h>
  55#include "internal.h"
  56
  57int compat_log = 1;
  58
  59int compat_printk(const char *fmt, ...)
  60{
  61        va_list ap;
  62        int ret;
  63        if (!compat_log)
  64                return 0;
  65        va_start(ap, fmt);
  66        ret = vprintk(fmt, ap);
  67        va_end(ap);
  68        return ret;
  69}
  70
  71/*
  72 * Not all architectures have sys_utime, so implement this in terms
  73 * of sys_utimes.
  74 */
  75COMPAT_SYSCALL_DEFINE2(utime, const char __user *, filename,
  76                       struct compat_utimbuf __user *, t)
  77{
  78        struct timespec tv[2];
  79
  80        if (t) {
  81                if (get_user(tv[0].tv_sec, &t->actime) ||
  82                    get_user(tv[1].tv_sec, &t->modtime))
  83                        return -EFAULT;
  84                tv[0].tv_nsec = 0;
  85                tv[1].tv_nsec = 0;
  86        }
  87        return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
  88}
  89
  90COMPAT_SYSCALL_DEFINE4(utimensat, unsigned int, dfd, const char __user *, filename, struct compat_timespec __user *, t, int, flags)
  91{
  92        struct timespec tv[2];
  93
  94        if  (t) {
  95                if (compat_get_timespec(&tv[0], &t[0]) ||
  96                    compat_get_timespec(&tv[1], &t[1]))
  97                        return -EFAULT;
  98
  99                if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
 100                        return 0;
 101        }
 102        return do_utimes(dfd, filename, t ? tv : NULL, flags);
 103}
 104
 105COMPAT_SYSCALL_DEFINE3(futimesat, unsigned int, dfd, const char __user *, filename, struct compat_timeval __user *, t)
 106{
 107        struct timespec tv[2];
 108
 109        if (t) {
 110                if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
 111                    get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
 112                    get_user(tv[1].tv_sec, &t[1].tv_sec) ||
 113                    get_user(tv[1].tv_nsec, &t[1].tv_usec))
 114                        return -EFAULT;
 115                if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
 116                    tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
 117                        return -EINVAL;
 118                tv[0].tv_nsec *= 1000;
 119                tv[1].tv_nsec *= 1000;
 120        }
 121        return do_utimes(dfd, filename, t ? tv : NULL, 0);
 122}
 123
 124COMPAT_SYSCALL_DEFINE2(utimes, const char __user *, filename, struct compat_timeval __user *, t)
 125{
 126        return compat_sys_futimesat(AT_FDCWD, filename, t);
 127}
 128
 129static int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
 130{
 131        struct compat_stat tmp;
 132
 133        if (!old_valid_dev(stat->dev) || !old_valid_dev(stat->rdev))
 134                return -EOVERFLOW;
 135
 136        memset(&tmp, 0, sizeof(tmp));
 137        tmp.st_dev = old_encode_dev(stat->dev);
 138        tmp.st_ino = stat->ino;
 139        if (sizeof(tmp.st_ino) < sizeof(stat->ino) && tmp.st_ino != stat->ino)
 140                return -EOVERFLOW;
 141        tmp.st_mode = stat->mode;
 142        tmp.st_nlink = stat->nlink;
 143        if (tmp.st_nlink != stat->nlink)
 144                return -EOVERFLOW;
 145        SET_UID(tmp.st_uid, from_kuid_munged(current_user_ns(), stat->uid));
 146        SET_GID(tmp.st_gid, from_kgid_munged(current_user_ns(), stat->gid));
 147        tmp.st_rdev = old_encode_dev(stat->rdev);
 148        if ((u64) stat->size > MAX_NON_LFS)
 149                return -EOVERFLOW;
 150        tmp.st_size = stat->size;
 151        tmp.st_atime = stat->atime.tv_sec;
 152        tmp.st_atime_nsec = stat->atime.tv_nsec;
 153        tmp.st_mtime = stat->mtime.tv_sec;
 154        tmp.st_mtime_nsec = stat->mtime.tv_nsec;
 155        tmp.st_ctime = stat->ctime.tv_sec;
 156        tmp.st_ctime_nsec = stat->ctime.tv_nsec;
 157        tmp.st_blocks = stat->blocks;
 158        tmp.st_blksize = stat->blksize;
 159        return copy_to_user(ubuf, &tmp, sizeof(tmp)) ? -EFAULT : 0;
 160}
 161
 162COMPAT_SYSCALL_DEFINE2(newstat, const char __user *, filename,
 163                       struct compat_stat __user *, statbuf)
 164{
 165        struct kstat stat;
 166        int error;
 167
 168        error = vfs_stat(filename, &stat);
 169        if (error)
 170                return error;
 171        return cp_compat_stat(&stat, statbuf);
 172}
 173
 174COMPAT_SYSCALL_DEFINE2(newlstat, const char __user *, filename,
 175                       struct compat_stat __user *, statbuf)
 176{
 177        struct kstat stat;
 178        int error;
 179
 180        error = vfs_lstat(filename, &stat);
 181        if (error)
 182                return error;
 183        return cp_compat_stat(&stat, statbuf);
 184}
 185
 186#ifndef __ARCH_WANT_STAT64
 187COMPAT_SYSCALL_DEFINE4(newfstatat, unsigned int, dfd,
 188                       const char __user *, filename,
 189                       struct compat_stat __user *, statbuf, int, flag)
 190{
 191        struct kstat stat;
 192        int error;
 193
 194        error = vfs_fstatat(dfd, filename, &stat, flag);
 195        if (error)
 196                return error;
 197        return cp_compat_stat(&stat, statbuf);
 198}
 199#endif
 200
 201COMPAT_SYSCALL_DEFINE2(newfstat, unsigned int, fd,
 202                       struct compat_stat __user *, statbuf)
 203{
 204        struct kstat stat;
 205        int error = vfs_fstat(fd, &stat);
 206
 207        if (!error)
 208                error = cp_compat_stat(&stat, statbuf);
 209        return error;
 210}
 211
 212static int put_compat_statfs(struct compat_statfs __user *ubuf, struct kstatfs *kbuf)
 213{
 214        
 215        if (sizeof ubuf->f_blocks == 4) {
 216                if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
 217                     kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
 218                        return -EOVERFLOW;
 219                /* f_files and f_ffree may be -1; it's okay
 220                 * to stuff that into 32 bits */
 221                if (kbuf->f_files != 0xffffffffffffffffULL
 222                 && (kbuf->f_files & 0xffffffff00000000ULL))
 223                        return -EOVERFLOW;
 224                if (kbuf->f_ffree != 0xffffffffffffffffULL
 225                 && (kbuf->f_ffree & 0xffffffff00000000ULL))
 226                        return -EOVERFLOW;
 227        }
 228        if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
 229            __put_user(kbuf->f_type, &ubuf->f_type) ||
 230            __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
 231            __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
 232            __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
 233            __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
 234            __put_user(kbuf->f_files, &ubuf->f_files) ||
 235            __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
 236            __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
 237            __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
 238            __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
 239            __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
 240            __put_user(kbuf->f_flags, &ubuf->f_flags) ||
 241            __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
 242                return -EFAULT;
 243        return 0;
 244}
 245
 246/*
 247 * The following statfs calls are copies of code from fs/statfs.c and
 248 * should be checked against those from time to time
 249 */
 250COMPAT_SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct compat_statfs __user *, buf)
 251{
 252        struct kstatfs tmp;
 253        int error = user_statfs(pathname, &tmp);
 254        if (!error)
 255                error = put_compat_statfs(buf, &tmp);
 256        return error;
 257}
 258
 259COMPAT_SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct compat_statfs __user *, buf)
 260{
 261        struct kstatfs tmp;
 262        int error = fd_statfs(fd, &tmp);
 263        if (!error)
 264                error = put_compat_statfs(buf, &tmp);
 265        return error;
 266}
 267
 268static int put_compat_statfs64(struct compat_statfs64 __user *ubuf, struct kstatfs *kbuf)
 269{
 270        if (sizeof ubuf->f_blocks == 4) {
 271                if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
 272                     kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
 273                        return -EOVERFLOW;
 274                /* f_files and f_ffree may be -1; it's okay
 275                 * to stuff that into 32 bits */
 276                if (kbuf->f_files != 0xffffffffffffffffULL
 277                 && (kbuf->f_files & 0xffffffff00000000ULL))
 278                        return -EOVERFLOW;
 279                if (kbuf->f_ffree != 0xffffffffffffffffULL
 280                 && (kbuf->f_ffree & 0xffffffff00000000ULL))
 281                        return -EOVERFLOW;
 282        }
 283        if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
 284            __put_user(kbuf->f_type, &ubuf->f_type) ||
 285            __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
 286            __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
 287            __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
 288            __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
 289            __put_user(kbuf->f_files, &ubuf->f_files) ||
 290            __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
 291            __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
 292            __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
 293            __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
 294            __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
 295            __put_user(kbuf->f_flags, &ubuf->f_flags) ||
 296            __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
 297                return -EFAULT;
 298        return 0;
 299}
 300
 301COMPAT_SYSCALL_DEFINE3(statfs64, const char __user *, pathname, compat_size_t, sz, struct compat_statfs64 __user *, buf)
 302{
 303        struct kstatfs tmp;
 304        int error;
 305
 306        if (sz != sizeof(*buf))
 307                return -EINVAL;
 308
 309        error = user_statfs(pathname, &tmp);
 310        if (!error)
 311                error = put_compat_statfs64(buf, &tmp);
 312        return error;
 313}
 314
 315COMPAT_SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, compat_size_t, sz, struct compat_statfs64 __user *, buf)
 316{
 317        struct kstatfs tmp;
 318        int error;
 319
 320        if (sz != sizeof(*buf))
 321                return -EINVAL;
 322
 323        error = fd_statfs(fd, &tmp);
 324        if (!error)
 325                error = put_compat_statfs64(buf, &tmp);
 326        return error;
 327}
 328
 329/*
 330 * This is a copy of sys_ustat, just dealing with a structure layout.
 331 * Given how simple this syscall is that apporach is more maintainable
 332 * than the various conversion hacks.
 333 */
 334COMPAT_SYSCALL_DEFINE2(ustat, unsigned, dev, struct compat_ustat __user *, u)
 335{
 336        struct compat_ustat tmp;
 337        struct kstatfs sbuf;
 338        int err = vfs_ustat(new_decode_dev(dev), &sbuf);
 339        if (err)
 340                return err;
 341
 342        memset(&tmp, 0, sizeof(struct compat_ustat));
 343        tmp.f_tfree = sbuf.f_bfree;
 344        tmp.f_tinode = sbuf.f_ffree;
 345        if (copy_to_user(u, &tmp, sizeof(struct compat_ustat)))
 346                return -EFAULT;
 347        return 0;
 348}
 349
 350static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 351{
 352        if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 353            __get_user(kfl->l_type, &ufl->l_type) ||
 354            __get_user(kfl->l_whence, &ufl->l_whence) ||
 355            __get_user(kfl->l_start, &ufl->l_start) ||
 356            __get_user(kfl->l_len, &ufl->l_len) ||
 357            __get_user(kfl->l_pid, &ufl->l_pid))
 358                return -EFAULT;
 359        return 0;
 360}
 361
 362static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 363{
 364        if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 365            __put_user(kfl->l_type, &ufl->l_type) ||
 366            __put_user(kfl->l_whence, &ufl->l_whence) ||
 367            __put_user(kfl->l_start, &ufl->l_start) ||
 368            __put_user(kfl->l_len, &ufl->l_len) ||
 369            __put_user(kfl->l_pid, &ufl->l_pid))
 370                return -EFAULT;
 371        return 0;
 372}
 373
 374#ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
 375static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 376{
 377        if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 378            __get_user(kfl->l_type, &ufl->l_type) ||
 379            __get_user(kfl->l_whence, &ufl->l_whence) ||
 380            __get_user(kfl->l_start, &ufl->l_start) ||
 381            __get_user(kfl->l_len, &ufl->l_len) ||
 382            __get_user(kfl->l_pid, &ufl->l_pid))
 383                return -EFAULT;
 384        return 0;
 385}
 386#endif
 387
 388#ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
 389static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 390{
 391        if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 392            __put_user(kfl->l_type, &ufl->l_type) ||
 393            __put_user(kfl->l_whence, &ufl->l_whence) ||
 394            __put_user(kfl->l_start, &ufl->l_start) ||
 395            __put_user(kfl->l_len, &ufl->l_len) ||
 396            __put_user(kfl->l_pid, &ufl->l_pid))
 397                return -EFAULT;
 398        return 0;
 399}
 400#endif
 401
 402static unsigned int
 403convert_fcntl_cmd(unsigned int cmd)
 404{
 405        switch (cmd) {
 406        case F_GETLK64:
 407                return F_GETLK;
 408        case F_SETLK64:
 409                return F_SETLK;
 410        case F_SETLKW64:
 411                return F_SETLKW;
 412        }
 413
 414        return cmd;
 415}
 416
 417COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 418                       compat_ulong_t, arg)
 419{
 420        mm_segment_t old_fs;
 421        struct flock f;
 422        long ret;
 423        unsigned int conv_cmd;
 424
 425        switch (cmd) {
 426        case F_GETLK:
 427        case F_SETLK:
 428        case F_SETLKW:
 429                ret = get_compat_flock(&f, compat_ptr(arg));
 430                if (ret != 0)
 431                        break;
 432                old_fs = get_fs();
 433                set_fs(KERNEL_DS);
 434                ret = sys_fcntl(fd, cmd, (unsigned long)&f);
 435                set_fs(old_fs);
 436                if (cmd == F_GETLK && ret == 0) {
 437                        /* GETLK was successful and we need to return the data...
 438                         * but it needs to fit in the compat structure.
 439                         * l_start shouldn't be too big, unless the original
 440                         * start + end is greater than COMPAT_OFF_T_MAX, in which
 441                         * case the app was asking for trouble, so we return
 442                         * -EOVERFLOW in that case.
 443                         * l_len could be too big, in which case we just truncate it,
 444                         * and only allow the app to see that part of the conflicting
 445                         * lock that might make sense to it anyway
 446                         */
 447
 448                        if (f.l_start > COMPAT_OFF_T_MAX)
 449                                ret = -EOVERFLOW;
 450                        if (f.l_len > COMPAT_OFF_T_MAX)
 451                                f.l_len = COMPAT_OFF_T_MAX;
 452                        if (ret == 0)
 453                                ret = put_compat_flock(&f, compat_ptr(arg));
 454                }
 455                break;
 456
 457        case F_GETLK64:
 458        case F_SETLK64:
 459        case F_SETLKW64:
 460        case F_OFD_GETLK:
 461        case F_OFD_SETLK:
 462        case F_OFD_SETLKW:
 463                ret = get_compat_flock64(&f, compat_ptr(arg));
 464                if (ret != 0)
 465                        break;
 466                old_fs = get_fs();
 467                set_fs(KERNEL_DS);
 468                conv_cmd = convert_fcntl_cmd(cmd);
 469                ret = sys_fcntl(fd, conv_cmd, (unsigned long)&f);
 470                set_fs(old_fs);
 471                if ((conv_cmd == F_GETLK || conv_cmd == F_OFD_GETLK) && ret == 0) {
 472                        /* need to return lock information - see above for commentary */
 473                        if (f.l_start > COMPAT_LOFF_T_MAX)
 474                                ret = -EOVERFLOW;
 475                        if (f.l_len > COMPAT_LOFF_T_MAX)
 476                                f.l_len = COMPAT_LOFF_T_MAX;
 477                        if (ret == 0)
 478                                ret = put_compat_flock64(&f, compat_ptr(arg));
 479                }
 480                break;
 481
 482        default:
 483                ret = sys_fcntl(fd, cmd, arg);
 484                break;
 485        }
 486        return ret;
 487}
 488
 489COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
 490                       compat_ulong_t, arg)
 491{
 492        switch (cmd) {
 493        case F_GETLK64:
 494        case F_SETLK64:
 495        case F_SETLKW64:
 496        case F_OFD_GETLK:
 497        case F_OFD_SETLK:
 498        case F_OFD_SETLKW:
 499                return -EINVAL;
 500        }
 501        return compat_sys_fcntl64(fd, cmd, arg);
 502}
 503
 504COMPAT_SYSCALL_DEFINE2(io_setup, unsigned, nr_reqs, u32 __user *, ctx32p)
 505{
 506        long ret;
 507        aio_context_t ctx64;
 508
 509        mm_segment_t oldfs = get_fs();
 510        if (unlikely(get_user(ctx64, ctx32p)))
 511                return -EFAULT;
 512
 513        set_fs(KERNEL_DS);
 514        /* The __user pointer cast is valid because of the set_fs() */
 515        ret = sys_io_setup(nr_reqs, (aio_context_t __user *) &ctx64);
 516        set_fs(oldfs);
 517        /* truncating is ok because it's a user address */
 518        if (!ret)
 519                ret = put_user((u32) ctx64, ctx32p);
 520        return ret;
 521}
 522
 523COMPAT_SYSCALL_DEFINE5(io_getevents, compat_aio_context_t, ctx_id,
 524                       compat_long_t, min_nr,
 525                       compat_long_t, nr,
 526                       struct io_event __user *, events,
 527                       struct compat_timespec __user *, timeout)
 528{
 529        struct timespec t;
 530        struct timespec __user *ut = NULL;
 531
 532        if (timeout) {
 533                if (compat_get_timespec(&t, timeout))
 534                        return -EFAULT;
 535
 536                ut = compat_alloc_user_space(sizeof(*ut));
 537                if (copy_to_user(ut, &t, sizeof(t)) )
 538                        return -EFAULT;
 539        } 
 540        return sys_io_getevents(ctx_id, min_nr, nr, events, ut);
 541}
 542
 543/* A write operation does a read from user space and vice versa */
 544#define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 545
 546ssize_t compat_rw_copy_check_uvector(int type,
 547                const struct compat_iovec __user *uvector, unsigned long nr_segs,
 548                unsigned long fast_segs, struct iovec *fast_pointer,
 549                struct iovec **ret_pointer)
 550{
 551        compat_ssize_t tot_len;
 552        struct iovec *iov = *ret_pointer = fast_pointer;
 553        ssize_t ret = 0;
 554        int seg;
 555
 556        /*
 557         * SuS says "The readv() function *may* fail if the iovcnt argument
 558         * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 559         * traditionally returned zero for zero segments, so...
 560         */
 561        if (nr_segs == 0)
 562                goto out;
 563
 564        ret = -EINVAL;
 565        if (nr_segs > UIO_MAXIOV || nr_segs < 0)
 566                goto out;
 567        if (nr_segs > fast_segs) {
 568                ret = -ENOMEM;
 569                iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 570                if (iov == NULL)
 571                        goto out;
 572        }
 573        *ret_pointer = iov;
 574
 575        ret = -EFAULT;
 576        if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
 577                goto out;
 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 (type >= 0 &&
 601                    !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
 602                        ret = -EFAULT;
 603                        goto out;
 604                }
 605                if (len > MAX_RW_COUNT - tot_len)
 606                        len = MAX_RW_COUNT - tot_len;
 607                tot_len += len;
 608                iov->iov_base = compat_ptr(buf);
 609                iov->iov_len = (compat_size_t) len;
 610                uvector++;
 611                iov++;
 612        }
 613        ret = tot_len;
 614
 615out:
 616        return ret;
 617}
 618
 619static inline long
 620copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
 621{
 622        compat_uptr_t uptr;
 623        int i;
 624
 625        for (i = 0; i < nr; ++i) {
 626                if (get_user(uptr, ptr32 + i))
 627                        return -EFAULT;
 628                if (put_user(compat_ptr(uptr), ptr64 + i))
 629                        return -EFAULT;
 630        }
 631        return 0;
 632}
 633
 634#define MAX_AIO_SUBMITS         (PAGE_SIZE/sizeof(struct iocb *))
 635
 636COMPAT_SYSCALL_DEFINE3(io_submit, compat_aio_context_t, ctx_id,
 637                       int, nr, u32 __user *, iocb)
 638{
 639        struct iocb __user * __user *iocb64; 
 640        long ret;
 641
 642        if (unlikely(nr < 0))
 643                return -EINVAL;
 644
 645        if (nr > MAX_AIO_SUBMITS)
 646                nr = MAX_AIO_SUBMITS;
 647        
 648        iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
 649        ret = copy_iocb(nr, iocb, iocb64);
 650        if (!ret)
 651                ret = do_io_submit(ctx_id, nr, iocb64, 1);
 652        return ret;
 653}
 654
 655struct compat_ncp_mount_data {
 656        compat_int_t version;
 657        compat_uint_t ncp_fd;
 658        __compat_uid_t mounted_uid;
 659        compat_pid_t wdog_pid;
 660        unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
 661        compat_uint_t time_out;
 662        compat_uint_t retry_count;
 663        compat_uint_t flags;
 664        __compat_uid_t uid;
 665        __compat_gid_t gid;
 666        compat_mode_t file_mode;
 667        compat_mode_t dir_mode;
 668};
 669
 670struct compat_ncp_mount_data_v4 {
 671        compat_int_t version;
 672        compat_ulong_t flags;
 673        compat_ulong_t mounted_uid;
 674        compat_long_t wdog_pid;
 675        compat_uint_t ncp_fd;
 676        compat_uint_t time_out;
 677        compat_uint_t retry_count;
 678        compat_ulong_t uid;
 679        compat_ulong_t gid;
 680        compat_ulong_t file_mode;
 681        compat_ulong_t dir_mode;
 682};
 683
 684static void *do_ncp_super_data_conv(void *raw_data)
 685{
 686        int version = *(unsigned int *)raw_data;
 687
 688        if (version == 3) {
 689                struct compat_ncp_mount_data *c_n = raw_data;
 690                struct ncp_mount_data *n = raw_data;
 691
 692                n->dir_mode = c_n->dir_mode;
 693                n->file_mode = c_n->file_mode;
 694                n->gid = c_n->gid;
 695                n->uid = c_n->uid;
 696                memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
 697                n->wdog_pid = c_n->wdog_pid;
 698                n->mounted_uid = c_n->mounted_uid;
 699        } else if (version == 4) {
 700                struct compat_ncp_mount_data_v4 *c_n = raw_data;
 701                struct ncp_mount_data_v4 *n = raw_data;
 702
 703                n->dir_mode = c_n->dir_mode;
 704                n->file_mode = c_n->file_mode;
 705                n->gid = c_n->gid;
 706                n->uid = c_n->uid;
 707                n->retry_count = c_n->retry_count;
 708                n->time_out = c_n->time_out;
 709                n->ncp_fd = c_n->ncp_fd;
 710                n->wdog_pid = c_n->wdog_pid;
 711                n->mounted_uid = c_n->mounted_uid;
 712                n->flags = c_n->flags;
 713        } else if (version != 5) {
 714                return NULL;
 715        }
 716
 717        return raw_data;
 718}
 719
 720
 721struct compat_nfs_string {
 722        compat_uint_t len;
 723        compat_uptr_t data;
 724};
 725
 726static inline void compat_nfs_string(struct nfs_string *dst,
 727                                     struct compat_nfs_string *src)
 728{
 729        dst->data = compat_ptr(src->data);
 730        dst->len = src->len;
 731}
 732
 733struct compat_nfs4_mount_data_v1 {
 734        compat_int_t version;
 735        compat_int_t flags;
 736        compat_int_t rsize;
 737        compat_int_t wsize;
 738        compat_int_t timeo;
 739        compat_int_t retrans;
 740        compat_int_t acregmin;
 741        compat_int_t acregmax;
 742        compat_int_t acdirmin;
 743        compat_int_t acdirmax;
 744        struct compat_nfs_string client_addr;
 745        struct compat_nfs_string mnt_path;
 746        struct compat_nfs_string hostname;
 747        compat_uint_t host_addrlen;
 748        compat_uptr_t host_addr;
 749        compat_int_t proto;
 750        compat_int_t auth_flavourlen;
 751        compat_uptr_t auth_flavours;
 752};
 753
 754static int do_nfs4_super_data_conv(void *raw_data)
 755{
 756        int version = *(compat_uint_t *) raw_data;
 757
 758        if (version == 1) {
 759                struct compat_nfs4_mount_data_v1 *raw = raw_data;
 760                struct nfs4_mount_data *real = raw_data;
 761
 762                /* copy the fields backwards */
 763                real->auth_flavours = compat_ptr(raw->auth_flavours);
 764                real->auth_flavourlen = raw->auth_flavourlen;
 765                real->proto = raw->proto;
 766                real->host_addr = compat_ptr(raw->host_addr);
 767                real->host_addrlen = raw->host_addrlen;
 768                compat_nfs_string(&real->hostname, &raw->hostname);
 769                compat_nfs_string(&real->mnt_path, &raw->mnt_path);
 770                compat_nfs_string(&real->client_addr, &raw->client_addr);
 771                real->acdirmax = raw->acdirmax;
 772                real->acdirmin = raw->acdirmin;
 773                real->acregmax = raw->acregmax;
 774                real->acregmin = raw->acregmin;
 775                real->retrans = raw->retrans;
 776                real->timeo = raw->timeo;
 777                real->wsize = raw->wsize;
 778                real->rsize = raw->rsize;
 779                real->flags = raw->flags;
 780                real->version = raw->version;
 781        }
 782
 783        return 0;
 784}
 785
 786#define NCPFS_NAME      "ncpfs"
 787#define NFS4_NAME       "nfs4"
 788
 789COMPAT_SYSCALL_DEFINE5(mount, const char __user *, dev_name,
 790                       const char __user *, dir_name,
 791                       const char __user *, type, compat_ulong_t, flags,
 792                       const void __user *, data)
 793{
 794        char *kernel_type;
 795        unsigned long data_page;
 796        char *kernel_dev;
 797        int retval;
 798
 799        kernel_type = copy_mount_string(type);
 800        retval = PTR_ERR(kernel_type);
 801        if (IS_ERR(kernel_type))
 802                goto out;
 803
 804        kernel_dev = copy_mount_string(dev_name);
 805        retval = PTR_ERR(kernel_dev);
 806        if (IS_ERR(kernel_dev))
 807                goto out1;
 808
 809        retval = copy_mount_options(data, &data_page);
 810        if (retval < 0)
 811                goto out2;
 812
 813        retval = -EINVAL;
 814
 815        if (kernel_type && data_page) {
 816                if (!strcmp(kernel_type, NCPFS_NAME)) {
 817                        do_ncp_super_data_conv((void *)data_page);
 818                } else if (!strcmp(kernel_type, NFS4_NAME)) {
 819                        if (do_nfs4_super_data_conv((void *) data_page))
 820                                goto out3;
 821                }
 822        }
 823
 824        retval = do_mount(kernel_dev, dir_name, kernel_type,
 825                        flags, (void*)data_page);
 826
 827 out3:
 828        free_page(data_page);
 829 out2:
 830        kfree(kernel_dev);
 831 out1:
 832        kfree(kernel_type);
 833 out:
 834        return retval;
 835}
 836
 837struct compat_old_linux_dirent {
 838        compat_ulong_t  d_ino;
 839        compat_ulong_t  d_offset;
 840        unsigned short  d_namlen;
 841        char            d_name[1];
 842};
 843
 844struct compat_readdir_callback {
 845        struct dir_context ctx;
 846        struct compat_old_linux_dirent __user *dirent;
 847        int result;
 848};
 849
 850static int compat_fillonedir(struct dir_context *ctx, const char *name,
 851                             int namlen, loff_t offset, u64 ino,
 852                             unsigned int d_type)
 853{
 854        struct compat_readdir_callback *buf =
 855                container_of(ctx, struct compat_readdir_callback, ctx);
 856        struct compat_old_linux_dirent __user *dirent;
 857        compat_ulong_t d_ino;
 858
 859        if (buf->result)
 860                return -EINVAL;
 861        d_ino = ino;
 862        if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
 863                buf->result = -EOVERFLOW;
 864                return -EOVERFLOW;
 865        }
 866        buf->result++;
 867        dirent = buf->dirent;
 868        if (!access_ok(VERIFY_WRITE, dirent,
 869                        (unsigned long)(dirent->d_name + namlen + 1) -
 870                                (unsigned long)dirent))
 871                goto efault;
 872        if (    __put_user(d_ino, &dirent->d_ino) ||
 873                __put_user(offset, &dirent->d_offset) ||
 874                __put_user(namlen, &dirent->d_namlen) ||
 875                __copy_to_user(dirent->d_name, name, namlen) ||
 876                __put_user(0, dirent->d_name + namlen))
 877                goto efault;
 878        return 0;
 879efault:
 880        buf->result = -EFAULT;
 881        return -EFAULT;
 882}
 883
 884COMPAT_SYSCALL_DEFINE3(old_readdir, unsigned int, fd,
 885                struct compat_old_linux_dirent __user *, dirent, unsigned int, count)
 886{
 887        int error;
 888        struct fd f = fdget(fd);
 889        struct compat_readdir_callback buf = {
 890                .ctx.actor = compat_fillonedir,
 891                .dirent = dirent
 892        };
 893
 894        if (!f.file)
 895                return -EBADF;
 896
 897        error = iterate_dir(f.file, &buf.ctx);
 898        if (buf.result)
 899                error = buf.result;
 900
 901        fdput(f);
 902        return error;
 903}
 904
 905struct compat_linux_dirent {
 906        compat_ulong_t  d_ino;
 907        compat_ulong_t  d_off;
 908        unsigned short  d_reclen;
 909        char            d_name[1];
 910};
 911
 912struct compat_getdents_callback {
 913        struct dir_context ctx;
 914        struct compat_linux_dirent __user *current_dir;
 915        struct compat_linux_dirent __user *previous;
 916        int count;
 917        int error;
 918};
 919
 920static int compat_filldir(struct dir_context *ctx, const char *name, int namlen,
 921                loff_t offset, u64 ino, unsigned int d_type)
 922{
 923        struct compat_linux_dirent __user * dirent;
 924        struct compat_getdents_callback *buf =
 925                container_of(ctx, struct compat_getdents_callback, ctx);
 926        compat_ulong_t d_ino;
 927        int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
 928                namlen + 2, sizeof(compat_long_t));
 929
 930        buf->error = -EINVAL;   /* only used if we fail.. */
 931        if (reclen > buf->count)
 932                return -EINVAL;
 933        d_ino = ino;
 934        if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
 935                buf->error = -EOVERFLOW;
 936                return -EOVERFLOW;
 937        }
 938        dirent = buf->previous;
 939        if (dirent) {
 940                if (__put_user(offset, &dirent->d_off))
 941                        goto efault;
 942        }
 943        dirent = buf->current_dir;
 944        if (__put_user(d_ino, &dirent->d_ino))
 945                goto efault;
 946        if (__put_user(reclen, &dirent->d_reclen))
 947                goto efault;
 948        if (copy_to_user(dirent->d_name, name, namlen))
 949                goto efault;
 950        if (__put_user(0, dirent->d_name + namlen))
 951                goto efault;
 952        if (__put_user(d_type, (char  __user *) dirent + reclen - 1))
 953                goto efault;
 954        buf->previous = dirent;
 955        dirent = (void __user *)dirent + reclen;
 956        buf->current_dir = dirent;
 957        buf->count -= reclen;
 958        return 0;
 959efault:
 960        buf->error = -EFAULT;
 961        return -EFAULT;
 962}
 963
 964COMPAT_SYSCALL_DEFINE3(getdents, unsigned int, fd,
 965                struct compat_linux_dirent __user *, dirent, unsigned int, count)
 966{
 967        struct fd f;
 968        struct compat_linux_dirent __user * lastdirent;
 969        struct compat_getdents_callback buf = {
 970                .ctx.actor = compat_filldir,
 971                .current_dir = dirent,
 972                .count = count
 973        };
 974        int error;
 975
 976        if (!access_ok(VERIFY_WRITE, dirent, count))
 977                return -EFAULT;
 978
 979        f = fdget(fd);
 980        if (!f.file)
 981                return -EBADF;
 982
 983        error = iterate_dir(f.file, &buf.ctx);
 984        if (error >= 0)
 985                error = buf.error;
 986        lastdirent = buf.previous;
 987        if (lastdirent) {
 988                if (put_user(buf.ctx.pos, &lastdirent->d_off))
 989                        error = -EFAULT;
 990                else
 991                        error = count - buf.count;
 992        }
 993        fdput(f);
 994        return error;
 995}
 996
 997#ifdef __ARCH_WANT_COMPAT_SYS_GETDENTS64
 998
 999struct compat_getdents_callback64 {
1000        struct dir_context ctx;
1001        struct linux_dirent64 __user *current_dir;
1002        struct linux_dirent64 __user *previous;
1003        int count;
1004        int error;
1005};
1006
1007static int compat_filldir64(struct dir_context *ctx, const char *name,
1008                            int namlen, loff_t offset, u64 ino,
1009                            unsigned int d_type)
1010{
1011        struct linux_dirent64 __user *dirent;
1012        struct compat_getdents_callback64 *buf =
1013                container_of(ctx, struct compat_getdents_callback64, ctx);
1014        int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
1015                sizeof(u64));
1016        u64 off;
1017
1018        buf->error = -EINVAL;   /* only used if we fail.. */
1019        if (reclen > buf->count)
1020                return -EINVAL;
1021        dirent = buf->previous;
1022
1023        if (dirent) {
1024                if (__put_user_unaligned(offset, &dirent->d_off))
1025                        goto efault;
1026        }
1027        dirent = buf->current_dir;
1028        if (__put_user_unaligned(ino, &dirent->d_ino))
1029                goto efault;
1030        off = 0;
1031        if (__put_user_unaligned(off, &dirent->d_off))
1032                goto efault;
1033        if (__put_user(reclen, &dirent->d_reclen))
1034                goto efault;
1035        if (__put_user(d_type, &dirent->d_type))
1036                goto efault;
1037        if (copy_to_user(dirent->d_name, name, namlen))
1038                goto efault;
1039        if (__put_user(0, dirent->d_name + namlen))
1040                goto efault;
1041        buf->previous = dirent;
1042        dirent = (void __user *)dirent + reclen;
1043        buf->current_dir = dirent;
1044        buf->count -= reclen;
1045        return 0;
1046efault:
1047        buf->error = -EFAULT;
1048        return -EFAULT;
1049}
1050
1051COMPAT_SYSCALL_DEFINE3(getdents64, unsigned int, fd,
1052                struct linux_dirent64 __user *, dirent, unsigned int, count)
1053{
1054        struct fd f;
1055        struct linux_dirent64 __user * lastdirent;
1056        struct compat_getdents_callback64 buf = {
1057                .ctx.actor = compat_filldir64,
1058                .current_dir = dirent,
1059                .count = count
1060        };
1061        int error;
1062
1063        if (!access_ok(VERIFY_WRITE, dirent, count))
1064                return -EFAULT;
1065
1066        f = fdget(fd);
1067        if (!f.file)
1068                return -EBADF;
1069
1070        error = iterate_dir(f.file, &buf.ctx);
1071        if (error >= 0)
1072                error = buf.error;
1073        lastdirent = buf.previous;
1074        if (lastdirent) {
1075                typeof(lastdirent->d_off) d_off = buf.ctx.pos;
1076                if (__put_user_unaligned(d_off, &lastdirent->d_off))
1077                        error = -EFAULT;
1078                else
1079                        error = count - buf.count;
1080        }
1081        fdput(f);
1082        return error;
1083}
1084#endif /* __ARCH_WANT_COMPAT_SYS_GETDENTS64 */
1085
1086/*
1087 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1088 * O_LARGEFILE flag.
1089 */
1090COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1091{
1092        return do_sys_open(AT_FDCWD, filename, flags, mode);
1093}
1094
1095/*
1096 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1097 * O_LARGEFILE flag.
1098 */
1099COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1100{
1101        return do_sys_open(dfd, filename, flags, mode);
1102}
1103
1104#define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1105
1106static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1107                                      int timeval, int ret)
1108{
1109        struct timespec ts;
1110
1111        if (!p)
1112                return ret;
1113
1114        if (current->personality & STICKY_TIMEOUTS)
1115                goto sticky;
1116
1117        /* No update for zero timeout */
1118        if (!end_time->tv_sec && !end_time->tv_nsec)
1119                return ret;
1120
1121        ktime_get_ts(&ts);
1122        ts = timespec_sub(*end_time, ts);
1123        if (ts.tv_sec < 0)
1124                ts.tv_sec = ts.tv_nsec = 0;
1125
1126        if (timeval) {
1127                struct compat_timeval rtv;
1128
1129                rtv.tv_sec = ts.tv_sec;
1130                rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1131
1132                if (!copy_to_user(p, &rtv, sizeof(rtv)))
1133                        return ret;
1134        } else {
1135                struct compat_timespec rts;
1136
1137                rts.tv_sec = ts.tv_sec;
1138                rts.tv_nsec = ts.tv_nsec;
1139
1140                if (!copy_to_user(p, &rts, sizeof(rts)))
1141                        return ret;
1142        }
1143        /*
1144         * If an application puts its timeval in read-only memory, we
1145         * don't want the Linux-specific update to the timeval to
1146         * cause a fault after the select has completed
1147         * successfully. However, because we're not updating the
1148         * timeval, we can't restart the system call.
1149         */
1150
1151sticky:
1152        if (ret == -ERESTARTNOHAND)
1153                ret = -EINTR;
1154        return ret;
1155}
1156
1157/*
1158 * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1159 * 64-bit unsigned longs.
1160 */
1161static
1162int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1163                        unsigned long *fdset)
1164{
1165        nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1166        if (ufdset) {
1167                unsigned long odd;
1168
1169                if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1170                        return -EFAULT;
1171
1172                odd = nr & 1UL;
1173                nr &= ~1UL;
1174                while (nr) {
1175                        unsigned long h, l;
1176                        if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1177                                return -EFAULT;
1178                        ufdset += 2;
1179                        *fdset++ = h << 32 | l;
1180                        nr -= 2;
1181                }
1182                if (odd && __get_user(*fdset, ufdset))
1183                        return -EFAULT;
1184        } else {
1185                /* Tricky, must clear full unsigned long in the
1186                 * kernel fdset at the end, this makes sure that
1187                 * actually happens.
1188                 */
1189                memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1190        }
1191        return 0;
1192}
1193
1194static
1195int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1196                      unsigned long *fdset)
1197{
1198        unsigned long odd;
1199        nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1200
1201        if (!ufdset)
1202                return 0;
1203
1204        odd = nr & 1UL;
1205        nr &= ~1UL;
1206        while (nr) {
1207                unsigned long h, l;
1208                l = *fdset++;
1209                h = l >> 32;
1210                if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1211                        return -EFAULT;
1212                ufdset += 2;
1213                nr -= 2;
1214        }
1215        if (odd && __put_user(*fdset, ufdset))
1216                return -EFAULT;
1217        return 0;
1218}
1219
1220
1221/*
1222 * This is a virtual copy of sys_select from fs/select.c and probably
1223 * should be compared to it from time to time
1224 */
1225
1226/*
1227 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1228 * like to be certain this leads to no problems. So I return
1229 * EINTR just for safety.
1230 *
1231 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1232 * I'm trying ERESTARTNOHAND which restart only when you want to.
1233 */
1234int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1235        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1236        struct timespec *end_time)
1237{
1238        fd_set_bits fds;
1239        void *bits;
1240        int size, max_fds, ret = -EINVAL;
1241        struct fdtable *fdt;
1242        long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1243
1244        if (n < 0)
1245                goto out_nofds;
1246
1247        /* max_fds can increase, so grab it once to avoid race */
1248        rcu_read_lock();
1249        fdt = files_fdtable(current->files);
1250        max_fds = fdt->max_fds;
1251        rcu_read_unlock();
1252        if (n > max_fds)
1253                n = max_fds;
1254
1255        /*
1256         * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1257         * since we used fdset we need to allocate memory in units of
1258         * long-words.
1259         */
1260        size = FDS_BYTES(n);
1261        bits = stack_fds;
1262        if (size > sizeof(stack_fds) / 6) {
1263                bits = kmalloc(6 * size, GFP_KERNEL);
1264                ret = -ENOMEM;
1265                if (!bits)
1266                        goto out_nofds;
1267        }
1268        fds.in      = (unsigned long *)  bits;
1269        fds.out     = (unsigned long *) (bits +   size);
1270        fds.ex      = (unsigned long *) (bits + 2*size);
1271        fds.res_in  = (unsigned long *) (bits + 3*size);
1272        fds.res_out = (unsigned long *) (bits + 4*size);
1273        fds.res_ex  = (unsigned long *) (bits + 5*size);
1274
1275        if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1276            (ret = compat_get_fd_set(n, outp, fds.out)) ||
1277            (ret = compat_get_fd_set(n, exp, fds.ex)))
1278                goto out;
1279        zero_fd_set(n, fds.res_in);
1280        zero_fd_set(n, fds.res_out);
1281        zero_fd_set(n, fds.res_ex);
1282
1283        ret = do_select(n, &fds, end_time);
1284
1285        if (ret < 0)
1286                goto out;
1287        if (!ret) {
1288                ret = -ERESTARTNOHAND;
1289                if (signal_pending(current))
1290                        goto out;
1291                ret = 0;
1292        }
1293
1294        if (compat_set_fd_set(n, inp, fds.res_in) ||
1295            compat_set_fd_set(n, outp, fds.res_out) ||
1296            compat_set_fd_set(n, exp, fds.res_ex))
1297                ret = -EFAULT;
1298out:
1299        if (bits != stack_fds)
1300                kfree(bits);
1301out_nofds:
1302        return ret;
1303}
1304
1305COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1306        compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1307        struct compat_timeval __user *, tvp)
1308{
1309        struct timespec end_time, *to = NULL;
1310        struct compat_timeval tv;
1311        int ret;
1312
1313        if (tvp) {
1314                if (copy_from_user(&tv, tvp, sizeof(tv)))
1315                        return -EFAULT;
1316
1317                to = &end_time;
1318                if (poll_select_set_timeout(to,
1319                                tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1320                                (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1321                        return -EINVAL;
1322        }
1323
1324        ret = compat_core_sys_select(n, inp, outp, exp, to);
1325        ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1326
1327        return ret;
1328}
1329
1330struct compat_sel_arg_struct {
1331        compat_ulong_t n;
1332        compat_uptr_t inp;
1333        compat_uptr_t outp;
1334        compat_uptr_t exp;
1335        compat_uptr_t tvp;
1336};
1337
1338COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1339{
1340        struct compat_sel_arg_struct a;
1341
1342        if (copy_from_user(&a, arg, sizeof(a)))
1343                return -EFAULT;
1344        return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1345                                 compat_ptr(a.exp), compat_ptr(a.tvp));
1346}
1347
1348static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1349        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1350        struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1351        compat_size_t sigsetsize)
1352{
1353        compat_sigset_t ss32;
1354        sigset_t ksigmask, sigsaved;
1355        struct compat_timespec ts;
1356        struct timespec end_time, *to = NULL;
1357        int ret;
1358
1359        if (tsp) {
1360                if (copy_from_user(&ts, tsp, sizeof(ts)))
1361                        return -EFAULT;
1362
1363                to = &end_time;
1364                if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1365                        return -EINVAL;
1366        }
1367
1368        if (sigmask) {
1369                if (sigsetsize != sizeof(compat_sigset_t))
1370                        return -EINVAL;
1371                if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1372                        return -EFAULT;
1373                sigset_from_compat(&ksigmask, &ss32);
1374
1375                sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1376                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1377        }
1378
1379        ret = compat_core_sys_select(n, inp, outp, exp, to);
1380        ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1381
1382        if (ret == -ERESTARTNOHAND) {
1383                /*
1384                 * Don't restore the signal mask yet. Let do_signal() deliver
1385                 * the signal on the way back to userspace, before the signal
1386                 * mask is restored.
1387                 */
1388                if (sigmask) {
1389                        memcpy(&current->saved_sigmask, &sigsaved,
1390                                        sizeof(sigsaved));
1391                        set_restore_sigmask();
1392                }
1393        } else if (sigmask)
1394                sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1395
1396        return ret;
1397}
1398
1399COMPAT_SYSCALL_DEFINE6(pselect6, int, n, compat_ulong_t __user *, inp,
1400        compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1401        struct compat_timespec __user *, tsp, void __user *, sig)
1402{
1403        compat_size_t sigsetsize = 0;
1404        compat_uptr_t up = 0;
1405
1406        if (sig) {
1407                if (!access_ok(VERIFY_READ, sig,
1408                                sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1409                        __get_user(up, (compat_uptr_t __user *)sig) ||
1410                        __get_user(sigsetsize,
1411                                (compat_size_t __user *)(sig+sizeof(up))))
1412                        return -EFAULT;
1413        }
1414        return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1415                                 sigsetsize);
1416}
1417
1418COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds,
1419        unsigned int,  nfds, struct compat_timespec __user *, tsp,
1420        const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1421{
1422        compat_sigset_t ss32;
1423        sigset_t ksigmask, sigsaved;
1424        struct compat_timespec ts;
1425        struct timespec end_time, *to = NULL;
1426        int ret;
1427
1428        if (tsp) {
1429                if (copy_from_user(&ts, tsp, sizeof(ts)))
1430                        return -EFAULT;
1431
1432                to = &end_time;
1433                if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1434                        return -EINVAL;
1435        }
1436
1437        if (sigmask) {
1438                if (sigsetsize != sizeof(compat_sigset_t))
1439                        return -EINVAL;
1440                if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1441                        return -EFAULT;
1442                sigset_from_compat(&ksigmask, &ss32);
1443
1444                sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1445                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1446        }
1447
1448        ret = do_sys_poll(ufds, nfds, to);
1449
1450        /* We can restart this syscall, usually */
1451        if (ret == -EINTR) {
1452                /*
1453                 * Don't restore the signal mask yet. Let do_signal() deliver
1454                 * the signal on the way back to userspace, before the signal
1455                 * mask is restored.
1456                 */
1457                if (sigmask) {
1458                        memcpy(&current->saved_sigmask, &sigsaved,
1459                                sizeof(sigsaved));
1460                        set_restore_sigmask();
1461                }
1462                ret = -ERESTARTNOHAND;
1463        } else if (sigmask)
1464                sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1465
1466        ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1467
1468        return ret;
1469}
1470
1471#ifdef CONFIG_FHANDLE
1472/*
1473 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1474 * doesn't set the O_LARGEFILE flag.
1475 */
1476COMPAT_SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
1477                             struct file_handle __user *, handle, int, flags)
1478{
1479        return do_handle_open(mountdirfd, handle, flags);
1480}
1481#endif
1482