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