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 */
  75asmlinkage long compat_sys_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
  90asmlinkage long compat_sys_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 (get_compat_timespec(&tv[0], &t[0]) ||
  96                    get_compat_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
 105asmlinkage long compat_sys_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
 124asmlinkage long compat_sys_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
 162asmlinkage long compat_sys_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
 174asmlinkage long compat_sys_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
 187asmlinkage long compat_sys_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
 201asmlinkage long compat_sys_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 */
 250asmlinkage long compat_sys_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
 259asmlinkage long compat_sys_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
 301asmlinkage long compat_sys_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
 315asmlinkage long compat_sys_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 */
 334asmlinkage long compat_sys_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
 402asmlinkage long compat_sys_fcntl64(unsigned int fd, unsigned int cmd,
 403                unsigned long arg)
 404{
 405        mm_segment_t old_fs;
 406        struct flock f;
 407        long ret;
 408
 409        switch (cmd) {
 410        case F_GETLK:
 411        case F_SETLK:
 412        case F_SETLKW:
 413                ret = get_compat_flock(&f, compat_ptr(arg));
 414                if (ret != 0)
 415                        break;
 416                old_fs = get_fs();
 417                set_fs(KERNEL_DS);
 418                ret = sys_fcntl(fd, cmd, (unsigned long)&f);
 419                set_fs(old_fs);
 420                if (cmd == F_GETLK && ret == 0) {
 421                        /* GETLK was successful and we need to return the data...
 422                         * but it needs to fit in the compat structure.
 423                         * l_start shouldn't be too big, unless the original
 424                         * start + end is greater than COMPAT_OFF_T_MAX, in which
 425                         * case the app was asking for trouble, so we return
 426                         * -EOVERFLOW in that case.
 427                         * l_len could be too big, in which case we just truncate it,
 428                         * and only allow the app to see that part of the conflicting
 429                         * lock that might make sense to it anyway
 430                         */
 431
 432                        if (f.l_start > COMPAT_OFF_T_MAX)
 433                                ret = -EOVERFLOW;
 434                        if (f.l_len > COMPAT_OFF_T_MAX)
 435                                f.l_len = COMPAT_OFF_T_MAX;
 436                        if (ret == 0)
 437                                ret = put_compat_flock(&f, compat_ptr(arg));
 438                }
 439                break;
 440
 441        case F_GETLK64:
 442        case F_SETLK64:
 443        case F_SETLKW64:
 444                ret = get_compat_flock64(&f, compat_ptr(arg));
 445                if (ret != 0)
 446                        break;
 447                old_fs = get_fs();
 448                set_fs(KERNEL_DS);
 449                ret = sys_fcntl(fd, (cmd == F_GETLK64) ? F_GETLK :
 450                                ((cmd == F_SETLK64) ? F_SETLK : F_SETLKW),
 451                                (unsigned long)&f);
 452                set_fs(old_fs);
 453                if (cmd == F_GETLK64 && ret == 0) {
 454                        /* need to return lock information - see above for commentary */
 455                        if (f.l_start > COMPAT_LOFF_T_MAX)
 456                                ret = -EOVERFLOW;
 457                        if (f.l_len > COMPAT_LOFF_T_MAX)
 458                                f.l_len = COMPAT_LOFF_T_MAX;
 459                        if (ret == 0)
 460                                ret = put_compat_flock64(&f, compat_ptr(arg));
 461                }
 462                break;
 463
 464        default:
 465                ret = sys_fcntl(fd, cmd, arg);
 466                break;
 467        }
 468        return ret;
 469}
 470
 471asmlinkage long compat_sys_fcntl(unsigned int fd, unsigned int cmd,
 472                unsigned long arg)
 473{
 474        if ((cmd == F_GETLK64) || (cmd == F_SETLK64) || (cmd == F_SETLKW64))
 475                return -EINVAL;
 476        return compat_sys_fcntl64(fd, cmd, arg);
 477}
 478
 479asmlinkage long
 480compat_sys_io_setup(unsigned nr_reqs, u32 __user *ctx32p)
 481{
 482        long ret;
 483        aio_context_t ctx64;
 484
 485        mm_segment_t oldfs = get_fs();
 486        if (unlikely(get_user(ctx64, ctx32p)))
 487                return -EFAULT;
 488
 489        set_fs(KERNEL_DS);
 490        /* The __user pointer cast is valid because of the set_fs() */
 491        ret = sys_io_setup(nr_reqs, (aio_context_t __user *) &ctx64);
 492        set_fs(oldfs);
 493        /* truncating is ok because it's a user address */
 494        if (!ret)
 495                ret = put_user((u32) ctx64, ctx32p);
 496        return ret;
 497}
 498
 499asmlinkage long
 500compat_sys_io_getevents(aio_context_t ctx_id,
 501                                 unsigned long min_nr,
 502                                 unsigned long nr,
 503                                 struct io_event __user *events,
 504                                 struct compat_timespec __user *timeout)
 505{
 506        long ret;
 507        struct timespec t;
 508        struct timespec __user *ut = NULL;
 509
 510        ret = -EFAULT;
 511        if (unlikely(!access_ok(VERIFY_WRITE, events, 
 512                                nr * sizeof(struct io_event))))
 513                goto out;
 514        if (timeout) {
 515                if (get_compat_timespec(&t, timeout))
 516                        goto out;
 517
 518                ut = compat_alloc_user_space(sizeof(*ut));
 519                if (copy_to_user(ut, &t, sizeof(t)) )
 520                        goto out;
 521        } 
 522        ret = sys_io_getevents(ctx_id, min_nr, nr, events, ut);
 523out:
 524        return ret;
 525}
 526
 527/* A write operation does a read from user space and vice versa */
 528#define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 529
 530ssize_t compat_rw_copy_check_uvector(int type,
 531                const struct compat_iovec __user *uvector, unsigned long nr_segs,
 532                unsigned long fast_segs, struct iovec *fast_pointer,
 533                struct iovec **ret_pointer)
 534{
 535        compat_ssize_t tot_len;
 536        struct iovec *iov = *ret_pointer = fast_pointer;
 537        ssize_t ret = 0;
 538        int seg;
 539
 540        /*
 541         * SuS says "The readv() function *may* fail if the iovcnt argument
 542         * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 543         * traditionally returned zero for zero segments, so...
 544         */
 545        if (nr_segs == 0)
 546                goto out;
 547
 548        ret = -EINVAL;
 549        if (nr_segs > UIO_MAXIOV || nr_segs < 0)
 550                goto out;
 551        if (nr_segs > fast_segs) {
 552                ret = -ENOMEM;
 553                iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 554                if (iov == NULL)
 555                        goto out;
 556        }
 557        *ret_pointer = iov;
 558
 559        ret = -EFAULT;
 560        if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
 561                goto out;
 562
 563        /*
 564         * Single unix specification:
 565         * We should -EINVAL if an element length is not >= 0 and fitting an
 566         * ssize_t.
 567         *
 568         * In Linux, the total length is limited to MAX_RW_COUNT, there is
 569         * no overflow possibility.
 570         */
 571        tot_len = 0;
 572        ret = -EINVAL;
 573        for (seg = 0; seg < nr_segs; seg++) {
 574                compat_uptr_t buf;
 575                compat_ssize_t len;
 576
 577                if (__get_user(len, &uvector->iov_len) ||
 578                   __get_user(buf, &uvector->iov_base)) {
 579                        ret = -EFAULT;
 580                        goto out;
 581                }
 582                if (len < 0)    /* size_t not fitting in compat_ssize_t .. */
 583                        goto out;
 584                if (type >= 0 &&
 585                    !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
 586                        ret = -EFAULT;
 587                        goto out;
 588                }
 589                if (len > MAX_RW_COUNT - tot_len)
 590                        len = MAX_RW_COUNT - tot_len;
 591                tot_len += len;
 592                iov->iov_base = compat_ptr(buf);
 593                iov->iov_len = (compat_size_t) len;
 594                uvector++;
 595                iov++;
 596        }
 597        ret = tot_len;
 598
 599out:
 600        return ret;
 601}
 602
 603static inline long
 604copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
 605{
 606        compat_uptr_t uptr;
 607        int i;
 608
 609        for (i = 0; i < nr; ++i) {
 610                if (get_user(uptr, ptr32 + i))
 611                        return -EFAULT;
 612                if (put_user(compat_ptr(uptr), ptr64 + i))
 613                        return -EFAULT;
 614        }
 615        return 0;
 616}
 617
 618#define MAX_AIO_SUBMITS         (PAGE_SIZE/sizeof(struct iocb *))
 619
 620asmlinkage long
 621compat_sys_io_submit(aio_context_t ctx_id, int nr, u32 __user *iocb)
 622{
 623        struct iocb __user * __user *iocb64; 
 624        long ret;
 625
 626        if (unlikely(nr < 0))
 627                return -EINVAL;
 628
 629        if (nr > MAX_AIO_SUBMITS)
 630                nr = MAX_AIO_SUBMITS;
 631        
 632        iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
 633        ret = copy_iocb(nr, iocb, iocb64);
 634        if (!ret)
 635                ret = do_io_submit(ctx_id, nr, iocb64, 1);
 636        return ret;
 637}
 638
 639struct compat_ncp_mount_data {
 640        compat_int_t version;
 641        compat_uint_t ncp_fd;
 642        __compat_uid_t mounted_uid;
 643        compat_pid_t wdog_pid;
 644        unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
 645        compat_uint_t time_out;
 646        compat_uint_t retry_count;
 647        compat_uint_t flags;
 648        __compat_uid_t uid;
 649        __compat_gid_t gid;
 650        compat_mode_t file_mode;
 651        compat_mode_t dir_mode;
 652};
 653
 654struct compat_ncp_mount_data_v4 {
 655        compat_int_t version;
 656        compat_ulong_t flags;
 657        compat_ulong_t mounted_uid;
 658        compat_long_t wdog_pid;
 659        compat_uint_t ncp_fd;
 660        compat_uint_t time_out;
 661        compat_uint_t retry_count;
 662        compat_ulong_t uid;
 663        compat_ulong_t gid;
 664        compat_ulong_t file_mode;
 665        compat_ulong_t dir_mode;
 666};
 667
 668static void *do_ncp_super_data_conv(void *raw_data)
 669{
 670        int version = *(unsigned int *)raw_data;
 671
 672        if (version == 3) {
 673                struct compat_ncp_mount_data *c_n = raw_data;
 674                struct ncp_mount_data *n = raw_data;
 675
 676                n->dir_mode = c_n->dir_mode;
 677                n->file_mode = c_n->file_mode;
 678                n->gid = c_n->gid;
 679                n->uid = c_n->uid;
 680                memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
 681                n->wdog_pid = c_n->wdog_pid;
 682                n->mounted_uid = c_n->mounted_uid;
 683        } else if (version == 4) {
 684                struct compat_ncp_mount_data_v4 *c_n = raw_data;
 685                struct ncp_mount_data_v4 *n = raw_data;
 686
 687                n->dir_mode = c_n->dir_mode;
 688                n->file_mode = c_n->file_mode;
 689                n->gid = c_n->gid;
 690                n->uid = c_n->uid;
 691                n->retry_count = c_n->retry_count;
 692                n->time_out = c_n->time_out;
 693                n->ncp_fd = c_n->ncp_fd;
 694                n->wdog_pid = c_n->wdog_pid;
 695                n->mounted_uid = c_n->mounted_uid;
 696                n->flags = c_n->flags;
 697        } else if (version != 5) {
 698                return NULL;
 699        }
 700
 701        return raw_data;
 702}
 703
 704
 705struct compat_nfs_string {
 706        compat_uint_t len;
 707        compat_uptr_t data;
 708};
 709
 710static inline void compat_nfs_string(struct nfs_string *dst,
 711                                     struct compat_nfs_string *src)
 712{
 713        dst->data = compat_ptr(src->data);
 714        dst->len = src->len;
 715}
 716
 717struct compat_nfs4_mount_data_v1 {
 718        compat_int_t version;
 719        compat_int_t flags;
 720        compat_int_t rsize;
 721        compat_int_t wsize;
 722        compat_int_t timeo;
 723        compat_int_t retrans;
 724        compat_int_t acregmin;
 725        compat_int_t acregmax;
 726        compat_int_t acdirmin;
 727        compat_int_t acdirmax;
 728        struct compat_nfs_string client_addr;
 729        struct compat_nfs_string mnt_path;
 730        struct compat_nfs_string hostname;
 731        compat_uint_t host_addrlen;
 732        compat_uptr_t host_addr;
 733        compat_int_t proto;
 734        compat_int_t auth_flavourlen;
 735        compat_uptr_t auth_flavours;
 736};
 737
 738static int do_nfs4_super_data_conv(void *raw_data)
 739{
 740        int version = *(compat_uint_t *) raw_data;
 741
 742        if (version == 1) {
 743                struct compat_nfs4_mount_data_v1 *raw = raw_data;
 744                struct nfs4_mount_data *real = raw_data;
 745
 746                /* copy the fields backwards */
 747                real->auth_flavours = compat_ptr(raw->auth_flavours);
 748                real->auth_flavourlen = raw->auth_flavourlen;
 749                real->proto = raw->proto;
 750                real->host_addr = compat_ptr(raw->host_addr);
 751                real->host_addrlen = raw->host_addrlen;
 752                compat_nfs_string(&real->hostname, &raw->hostname);
 753                compat_nfs_string(&real->mnt_path, &raw->mnt_path);
 754                compat_nfs_string(&real->client_addr, &raw->client_addr);
 755                real->acdirmax = raw->acdirmax;
 756                real->acdirmin = raw->acdirmin;
 757                real->acregmax = raw->acregmax;
 758                real->acregmin = raw->acregmin;
 759                real->retrans = raw->retrans;
 760                real->timeo = raw->timeo;
 761                real->wsize = raw->wsize;
 762                real->rsize = raw->rsize;
 763                real->flags = raw->flags;
 764                real->version = raw->version;
 765        }
 766
 767        return 0;
 768}
 769
 770#define NCPFS_NAME      "ncpfs"
 771#define NFS4_NAME       "nfs4"
 772
 773asmlinkage long compat_sys_mount(const char __user * dev_name,
 774                                 const char __user * dir_name,
 775                                 const char __user * type, unsigned long flags,
 776                                 const void __user * data)
 777{
 778        char *kernel_type;
 779        unsigned long data_page;
 780        char *kernel_dev;
 781        struct filename *dir;
 782        int retval;
 783
 784        retval = copy_mount_string(type, &kernel_type);
 785        if (retval < 0)
 786                goto out;
 787
 788        dir = getname(dir_name);
 789        retval = PTR_ERR(dir);
 790        if (IS_ERR(dir))
 791                goto out1;
 792
 793        retval = copy_mount_string(dev_name, &kernel_dev);
 794        if (retval < 0)
 795                goto out2;
 796
 797        retval = copy_mount_options(data, &data_page);
 798        if (retval < 0)
 799                goto out3;
 800
 801        retval = -EINVAL;
 802
 803        if (kernel_type && data_page) {
 804                if (!strcmp(kernel_type, NCPFS_NAME)) {
 805                        do_ncp_super_data_conv((void *)data_page);
 806                } else if (!strcmp(kernel_type, NFS4_NAME)) {
 807                        if (do_nfs4_super_data_conv((void *) data_page))
 808                                goto out4;
 809                }
 810        }
 811
 812        retval = do_mount(kernel_dev, dir->name, kernel_type,
 813                        flags, (void*)data_page);
 814
 815 out4:
 816        free_page(data_page);
 817 out3:
 818        kfree(kernel_dev);
 819 out2:
 820        putname(dir);
 821 out1:
 822        kfree(kernel_type);
 823 out:
 824        return retval;
 825}
 826
 827struct compat_old_linux_dirent {
 828        compat_ulong_t  d_ino;
 829        compat_ulong_t  d_offset;
 830        unsigned short  d_namlen;
 831        char            d_name[1];
 832};
 833
 834struct compat_readdir_callback {
 835        struct dir_context ctx;
 836        struct compat_old_linux_dirent __user *dirent;
 837        int result;
 838};
 839
 840static int compat_fillonedir(void *__buf, const char *name, int namlen,
 841                        loff_t offset, u64 ino, unsigned int d_type)
 842{
 843        struct compat_readdir_callback *buf = __buf;
 844        struct compat_old_linux_dirent __user *dirent;
 845        compat_ulong_t d_ino;
 846
 847        if (buf->result)
 848                return -EINVAL;
 849        d_ino = ino;
 850        if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
 851                buf->result = -EOVERFLOW;
 852                return -EOVERFLOW;
 853        }
 854        buf->result++;
 855        dirent = buf->dirent;
 856        if (!access_ok(VERIFY_WRITE, dirent,
 857                        (unsigned long)(dirent->d_name + namlen + 1) -
 858                                (unsigned long)dirent))
 859                goto efault;
 860        if (    __put_user(d_ino, &dirent->d_ino) ||
 861                __put_user(offset, &dirent->d_offset) ||
 862                __put_user(namlen, &dirent->d_namlen) ||
 863                __copy_to_user(dirent->d_name, name, namlen) ||
 864                __put_user(0, dirent->d_name + namlen))
 865                goto efault;
 866        return 0;
 867efault:
 868        buf->result = -EFAULT;
 869        return -EFAULT;
 870}
 871
 872asmlinkage long compat_sys_old_readdir(unsigned int fd,
 873        struct compat_old_linux_dirent __user *dirent, unsigned int count)
 874{
 875        int error;
 876        struct fd f = fdget(fd);
 877        struct compat_readdir_callback buf = {
 878                .ctx.actor = compat_fillonedir,
 879                .dirent = dirent
 880        };
 881
 882        if (!f.file)
 883                return -EBADF;
 884
 885        error = iterate_dir(f.file, &buf.ctx);
 886        if (buf.result)
 887                error = buf.result;
 888
 889        fdput(f);
 890        return error;
 891}
 892
 893struct compat_linux_dirent {
 894        compat_ulong_t  d_ino;
 895        compat_ulong_t  d_off;
 896        unsigned short  d_reclen;
 897        char            d_name[1];
 898};
 899
 900struct compat_getdents_callback {
 901        struct dir_context ctx;
 902        struct compat_linux_dirent __user *current_dir;
 903        struct compat_linux_dirent __user *previous;
 904        int count;
 905        int error;
 906};
 907
 908static int compat_filldir(void *__buf, const char *name, int namlen,
 909                loff_t offset, u64 ino, unsigned int d_type)
 910{
 911        struct compat_linux_dirent __user * dirent;
 912        struct compat_getdents_callback *buf = __buf;
 913        compat_ulong_t d_ino;
 914        int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
 915                namlen + 2, sizeof(compat_long_t));
 916
 917        buf->error = -EINVAL;   /* only used if we fail.. */
 918        if (reclen > buf->count)
 919                return -EINVAL;
 920        d_ino = ino;
 921        if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
 922                buf->error = -EOVERFLOW;
 923                return -EOVERFLOW;
 924        }
 925        dirent = buf->previous;
 926        if (dirent) {
 927                if (__put_user(offset, &dirent->d_off))
 928                        goto efault;
 929        }
 930        dirent = buf->current_dir;
 931        if (__put_user(d_ino, &dirent->d_ino))
 932                goto efault;
 933        if (__put_user(reclen, &dirent->d_reclen))
 934                goto efault;
 935        if (copy_to_user(dirent->d_name, name, namlen))
 936                goto efault;
 937        if (__put_user(0, dirent->d_name + namlen))
 938                goto efault;
 939        if (__put_user(d_type, (char  __user *) dirent + reclen - 1))
 940                goto efault;
 941        buf->previous = dirent;
 942        dirent = (void __user *)dirent + reclen;
 943        buf->current_dir = dirent;
 944        buf->count -= reclen;
 945        return 0;
 946efault:
 947        buf->error = -EFAULT;
 948        return -EFAULT;
 949}
 950
 951asmlinkage long compat_sys_getdents(unsigned int fd,
 952                struct compat_linux_dirent __user *dirent, unsigned int count)
 953{
 954        struct fd f;
 955        struct compat_linux_dirent __user * lastdirent;
 956        struct compat_getdents_callback buf = {
 957                .ctx.actor = compat_filldir,
 958                .current_dir = dirent,
 959                .count = count
 960        };
 961        int error;
 962
 963        if (!access_ok(VERIFY_WRITE, dirent, count))
 964                return -EFAULT;
 965
 966        f = fdget(fd);
 967        if (!f.file)
 968                return -EBADF;
 969
 970        error = iterate_dir(f.file, &buf.ctx);
 971        if (error >= 0)
 972                error = buf.error;
 973        lastdirent = buf.previous;
 974        if (lastdirent) {
 975                if (put_user(buf.ctx.pos, &lastdirent->d_off))
 976                        error = -EFAULT;
 977                else
 978                        error = count - buf.count;
 979        }
 980        fdput(f);
 981        return error;
 982}
 983
 984#ifndef __ARCH_OMIT_COMPAT_SYS_GETDENTS64
 985
 986struct compat_getdents_callback64 {
 987        struct dir_context ctx;
 988        struct linux_dirent64 __user *current_dir;
 989        struct linux_dirent64 __user *previous;
 990        int count;
 991        int error;
 992};
 993
 994static int compat_filldir64(void * __buf, const char * name, int namlen, loff_t offset,
 995                     u64 ino, unsigned int d_type)
 996{
 997        struct linux_dirent64 __user *dirent;
 998        struct compat_getdents_callback64 *buf = __buf;
 999        int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
1000                sizeof(u64));
1001        u64 off;
1002
1003        buf->error = -EINVAL;   /* only used if we fail.. */
1004        if (reclen > buf->count)
1005                return -EINVAL;
1006        dirent = buf->previous;
1007
1008        if (dirent) {
1009                if (__put_user_unaligned(offset, &dirent->d_off))
1010                        goto efault;
1011        }
1012        dirent = buf->current_dir;
1013        if (__put_user_unaligned(ino, &dirent->d_ino))
1014                goto efault;
1015        off = 0;
1016        if (__put_user_unaligned(off, &dirent->d_off))
1017                goto efault;
1018        if (__put_user(reclen, &dirent->d_reclen))
1019                goto efault;
1020        if (__put_user(d_type, &dirent->d_type))
1021                goto efault;
1022        if (copy_to_user(dirent->d_name, name, namlen))
1023                goto efault;
1024        if (__put_user(0, dirent->d_name + namlen))
1025                goto efault;
1026        buf->previous = dirent;
1027        dirent = (void __user *)dirent + reclen;
1028        buf->current_dir = dirent;
1029        buf->count -= reclen;
1030        return 0;
1031efault:
1032        buf->error = -EFAULT;
1033        return -EFAULT;
1034}
1035
1036asmlinkage long compat_sys_getdents64(unsigned int fd,
1037                struct linux_dirent64 __user * dirent, unsigned int count)
1038{
1039        struct fd f;
1040        struct linux_dirent64 __user * lastdirent;
1041        struct compat_getdents_callback64 buf = {
1042                .ctx.actor = compat_filldir64,
1043                .current_dir = dirent,
1044                .count = count
1045        };
1046        int error;
1047
1048        if (!access_ok(VERIFY_WRITE, dirent, count))
1049                return -EFAULT;
1050
1051        f = fdget(fd);
1052        if (!f.file)
1053                return -EBADF;
1054
1055        error = iterate_dir(f.file, &buf.ctx);
1056        if (error >= 0)
1057                error = buf.error;
1058        lastdirent = buf.previous;
1059        if (lastdirent) {
1060                typeof(lastdirent->d_off) d_off = buf.ctx.pos;
1061                if (__put_user_unaligned(d_off, &lastdirent->d_off))
1062                        error = -EFAULT;
1063                else
1064                        error = count - buf.count;
1065        }
1066        fdput(f);
1067        return error;
1068}
1069#endif /* ! __ARCH_OMIT_COMPAT_SYS_GETDENTS64 */
1070
1071/*
1072 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1073 * O_LARGEFILE flag.
1074 */
1075COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1076{
1077        return do_sys_open(AT_FDCWD, filename, flags, mode);
1078}
1079
1080/*
1081 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1082 * O_LARGEFILE flag.
1083 */
1084COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1085{
1086        return do_sys_open(dfd, filename, flags, mode);
1087}
1088
1089#define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1090
1091static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1092                                      int timeval, int ret)
1093{
1094        struct timespec ts;
1095
1096        if (!p)
1097                return ret;
1098
1099        if (current->personality & STICKY_TIMEOUTS)
1100                goto sticky;
1101
1102        /* No update for zero timeout */
1103        if (!end_time->tv_sec && !end_time->tv_nsec)
1104                return ret;
1105
1106        ktime_get_ts(&ts);
1107        ts = timespec_sub(*end_time, ts);
1108        if (ts.tv_sec < 0)
1109                ts.tv_sec = ts.tv_nsec = 0;
1110
1111        if (timeval) {
1112                struct compat_timeval rtv;
1113
1114                rtv.tv_sec = ts.tv_sec;
1115                rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1116
1117                if (!copy_to_user(p, &rtv, sizeof(rtv)))
1118                        return ret;
1119        } else {
1120                struct compat_timespec rts;
1121
1122                rts.tv_sec = ts.tv_sec;
1123                rts.tv_nsec = ts.tv_nsec;
1124
1125                if (!copy_to_user(p, &rts, sizeof(rts)))
1126                        return ret;
1127        }
1128        /*
1129         * If an application puts its timeval in read-only memory, we
1130         * don't want the Linux-specific update to the timeval to
1131         * cause a fault after the select has completed
1132         * successfully. However, because we're not updating the
1133         * timeval, we can't restart the system call.
1134         */
1135
1136sticky:
1137        if (ret == -ERESTARTNOHAND)
1138                ret = -EINTR;
1139        return ret;
1140}
1141
1142/*
1143 * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1144 * 64-bit unsigned longs.
1145 */
1146static
1147int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1148                        unsigned long *fdset)
1149{
1150        nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1151        if (ufdset) {
1152                unsigned long odd;
1153
1154                if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1155                        return -EFAULT;
1156
1157                odd = nr & 1UL;
1158                nr &= ~1UL;
1159                while (nr) {
1160                        unsigned long h, l;
1161                        if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1162                                return -EFAULT;
1163                        ufdset += 2;
1164                        *fdset++ = h << 32 | l;
1165                        nr -= 2;
1166                }
1167                if (odd && __get_user(*fdset, ufdset))
1168                        return -EFAULT;
1169        } else {
1170                /* Tricky, must clear full unsigned long in the
1171                 * kernel fdset at the end, this makes sure that
1172                 * actually happens.
1173                 */
1174                memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1175        }
1176        return 0;
1177}
1178
1179static
1180int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1181                      unsigned long *fdset)
1182{
1183        unsigned long odd;
1184        nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1185
1186        if (!ufdset)
1187                return 0;
1188
1189        odd = nr & 1UL;
1190        nr &= ~1UL;
1191        while (nr) {
1192                unsigned long h, l;
1193                l = *fdset++;
1194                h = l >> 32;
1195                if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1196                        return -EFAULT;
1197                ufdset += 2;
1198                nr -= 2;
1199        }
1200        if (odd && __put_user(*fdset, ufdset))
1201                return -EFAULT;
1202        return 0;
1203}
1204
1205
1206/*
1207 * This is a virtual copy of sys_select from fs/select.c and probably
1208 * should be compared to it from time to time
1209 */
1210
1211/*
1212 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1213 * like to be certain this leads to no problems. So I return
1214 * EINTR just for safety.
1215 *
1216 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1217 * I'm trying ERESTARTNOHAND which restart only when you want to.
1218 */
1219int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1220        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1221        struct timespec *end_time)
1222{
1223        fd_set_bits fds;
1224        void *bits;
1225        int size, max_fds, ret = -EINVAL;
1226        struct fdtable *fdt;
1227        long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1228
1229        if (n < 0)
1230                goto out_nofds;
1231
1232        /* max_fds can increase, so grab it once to avoid race */
1233        rcu_read_lock();
1234        fdt = files_fdtable(current->files);
1235        max_fds = fdt->max_fds;
1236        rcu_read_unlock();
1237        if (n > max_fds)
1238                n = max_fds;
1239
1240        /*
1241         * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1242         * since we used fdset we need to allocate memory in units of
1243         * long-words.
1244         */
1245        size = FDS_BYTES(n);
1246        bits = stack_fds;
1247        if (size > sizeof(stack_fds) / 6) {
1248                bits = kmalloc(6 * size, GFP_KERNEL);
1249                ret = -ENOMEM;
1250                if (!bits)
1251                        goto out_nofds;
1252        }
1253        fds.in      = (unsigned long *)  bits;
1254        fds.out     = (unsigned long *) (bits +   size);
1255        fds.ex      = (unsigned long *) (bits + 2*size);
1256        fds.res_in  = (unsigned long *) (bits + 3*size);
1257        fds.res_out = (unsigned long *) (bits + 4*size);
1258        fds.res_ex  = (unsigned long *) (bits + 5*size);
1259
1260        if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1261            (ret = compat_get_fd_set(n, outp, fds.out)) ||
1262            (ret = compat_get_fd_set(n, exp, fds.ex)))
1263                goto out;
1264        zero_fd_set(n, fds.res_in);
1265        zero_fd_set(n, fds.res_out);
1266        zero_fd_set(n, fds.res_ex);
1267
1268        ret = do_select(n, &fds, end_time);
1269
1270        if (ret < 0)
1271                goto out;
1272        if (!ret) {
1273                ret = -ERESTARTNOHAND;
1274                if (signal_pending(current))
1275                        goto out;
1276                ret = 0;
1277        }
1278
1279        if (compat_set_fd_set(n, inp, fds.res_in) ||
1280            compat_set_fd_set(n, outp, fds.res_out) ||
1281            compat_set_fd_set(n, exp, fds.res_ex))
1282                ret = -EFAULT;
1283out:
1284        if (bits != stack_fds)
1285                kfree(bits);
1286out_nofds:
1287        return ret;
1288}
1289
1290asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1291        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1292        struct compat_timeval __user *tvp)
1293{
1294        struct timespec end_time, *to = NULL;
1295        struct compat_timeval tv;
1296        int ret;
1297
1298        if (tvp) {
1299                if (copy_from_user(&tv, tvp, sizeof(tv)))
1300                        return -EFAULT;
1301
1302                to = &end_time;
1303                if (poll_select_set_timeout(to,
1304                                tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1305                                (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1306                        return -EINVAL;
1307        }
1308
1309        ret = compat_core_sys_select(n, inp, outp, exp, to);
1310        ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1311
1312        return ret;
1313}
1314
1315struct compat_sel_arg_struct {
1316        compat_ulong_t n;
1317        compat_uptr_t inp;
1318        compat_uptr_t outp;
1319        compat_uptr_t exp;
1320        compat_uptr_t tvp;
1321};
1322
1323asmlinkage long compat_sys_old_select(struct compat_sel_arg_struct __user *arg)
1324{
1325        struct compat_sel_arg_struct a;
1326
1327        if (copy_from_user(&a, arg, sizeof(a)))
1328                return -EFAULT;
1329        return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1330                                 compat_ptr(a.exp), compat_ptr(a.tvp));
1331}
1332
1333static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1334        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1335        struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1336        compat_size_t sigsetsize)
1337{
1338        compat_sigset_t ss32;
1339        sigset_t ksigmask, sigsaved;
1340        struct compat_timespec ts;
1341        struct timespec end_time, *to = NULL;
1342        int ret;
1343
1344        if (tsp) {
1345                if (copy_from_user(&ts, tsp, sizeof(ts)))
1346                        return -EFAULT;
1347
1348                to = &end_time;
1349                if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1350                        return -EINVAL;
1351        }
1352
1353        if (sigmask) {
1354                if (sigsetsize != sizeof(compat_sigset_t))
1355                        return -EINVAL;
1356                if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1357                        return -EFAULT;
1358                sigset_from_compat(&ksigmask, &ss32);
1359
1360                sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1361                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1362        }
1363
1364        ret = compat_core_sys_select(n, inp, outp, exp, to);
1365        ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1366
1367        if (ret == -ERESTARTNOHAND) {
1368                /*
1369                 * Don't restore the signal mask yet. Let do_signal() deliver
1370                 * the signal on the way back to userspace, before the signal
1371                 * mask is restored.
1372                 */
1373                if (sigmask) {
1374                        memcpy(&current->saved_sigmask, &sigsaved,
1375                                        sizeof(sigsaved));
1376                        set_restore_sigmask();
1377                }
1378        } else if (sigmask)
1379                sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1380
1381        return ret;
1382}
1383
1384asmlinkage long compat_sys_pselect6(int n, compat_ulong_t __user *inp,
1385        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1386        struct compat_timespec __user *tsp, void __user *sig)
1387{
1388        compat_size_t sigsetsize = 0;
1389        compat_uptr_t up = 0;
1390
1391        if (sig) {
1392                if (!access_ok(VERIFY_READ, sig,
1393                                sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1394                        __get_user(up, (compat_uptr_t __user *)sig) ||
1395                        __get_user(sigsetsize,
1396                                (compat_size_t __user *)(sig+sizeof(up))))
1397                        return -EFAULT;
1398        }
1399        return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1400                                 sigsetsize);
1401}
1402
1403asmlinkage long compat_sys_ppoll(struct pollfd __user *ufds,
1404        unsigned int nfds, struct compat_timespec __user *tsp,
1405        const compat_sigset_t __user *sigmask, compat_size_t sigsetsize)
1406{
1407        compat_sigset_t ss32;
1408        sigset_t ksigmask, sigsaved;
1409        struct compat_timespec ts;
1410        struct timespec end_time, *to = NULL;
1411        int ret;
1412
1413        if (tsp) {
1414                if (copy_from_user(&ts, tsp, sizeof(ts)))
1415                        return -EFAULT;
1416
1417                to = &end_time;
1418                if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1419                        return -EINVAL;
1420        }
1421
1422        if (sigmask) {
1423                if (sigsetsize != sizeof(compat_sigset_t))
1424                        return -EINVAL;
1425                if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1426                        return -EFAULT;
1427                sigset_from_compat(&ksigmask, &ss32);
1428
1429                sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1430                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1431        }
1432
1433        ret = do_sys_poll(ufds, nfds, to);
1434
1435        /* We can restart this syscall, usually */
1436        if (ret == -EINTR) {
1437                /*
1438                 * Don't restore the signal mask yet. Let do_signal() deliver
1439                 * the signal on the way back to userspace, before the signal
1440                 * mask is restored.
1441                 */
1442                if (sigmask) {
1443                        memcpy(&current->saved_sigmask, &sigsaved,
1444                                sizeof(sigsaved));
1445                        set_restore_sigmask();
1446                }
1447                ret = -ERESTARTNOHAND;
1448        } else if (sigmask)
1449                sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1450
1451        ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1452
1453        return ret;
1454}
1455
1456#ifdef CONFIG_FHANDLE
1457/*
1458 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1459 * doesn't set the O_LARGEFILE flag.
1460 */
1461COMPAT_SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
1462                             struct file_handle __user *, handle, int, flags)
1463{
1464        return do_handle_open(mountdirfd, handle, flags);
1465}
1466#endif
1467