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/eventpoll.h>
  48#include <linux/fs_struct.h>
  49#include <linux/slab.h>
  50#include <linux/pagemap.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#include "read_write.h"
  72
  73/*
  74 * Not all architectures have sys_utime, so implement this in terms
  75 * of sys_utimes.
  76 */
  77asmlinkage long compat_sys_utime(const char __user *filename,
  78                                 struct compat_utimbuf __user *t)
  79{
  80        struct timespec tv[2];
  81
  82        if (t) {
  83                if (get_user(tv[0].tv_sec, &t->actime) ||
  84                    get_user(tv[1].tv_sec, &t->modtime))
  85                        return -EFAULT;
  86                tv[0].tv_nsec = 0;
  87                tv[1].tv_nsec = 0;
  88        }
  89        return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
  90}
  91
  92asmlinkage long compat_sys_utimensat(unsigned int dfd, const char __user *filename, struct compat_timespec __user *t, int flags)
  93{
  94        struct timespec tv[2];
  95
  96        if  (t) {
  97                if (get_compat_timespec(&tv[0], &t[0]) ||
  98                    get_compat_timespec(&tv[1], &t[1]))
  99                        return -EFAULT;
 100
 101                if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
 102                        return 0;
 103        }
 104        return do_utimes(dfd, filename, t ? tv : NULL, flags);
 105}
 106
 107asmlinkage long compat_sys_futimesat(unsigned int dfd, const char __user *filename, struct compat_timeval __user *t)
 108{
 109        struct timespec tv[2];
 110
 111        if (t) {
 112                if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
 113                    get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
 114                    get_user(tv[1].tv_sec, &t[1].tv_sec) ||
 115                    get_user(tv[1].tv_nsec, &t[1].tv_usec))
 116                        return -EFAULT;
 117                if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
 118                    tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
 119                        return -EINVAL;
 120                tv[0].tv_nsec *= 1000;
 121                tv[1].tv_nsec *= 1000;
 122        }
 123        return do_utimes(dfd, filename, t ? tv : NULL, 0);
 124}
 125
 126asmlinkage long compat_sys_utimes(const char __user *filename, struct compat_timeval __user *t)
 127{
 128        return compat_sys_futimesat(AT_FDCWD, filename, t);
 129}
 130
 131static int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
 132{
 133        struct compat_stat tmp;
 134
 135        if (!old_valid_dev(stat->dev) || !old_valid_dev(stat->rdev))
 136                return -EOVERFLOW;
 137
 138        memset(&tmp, 0, sizeof(tmp));
 139        tmp.st_dev = old_encode_dev(stat->dev);
 140        tmp.st_ino = stat->ino;
 141        if (sizeof(tmp.st_ino) < sizeof(stat->ino) && tmp.st_ino != stat->ino)
 142                return -EOVERFLOW;
 143        tmp.st_mode = stat->mode;
 144        tmp.st_nlink = stat->nlink;
 145        if (tmp.st_nlink != stat->nlink)
 146                return -EOVERFLOW;
 147        SET_UID(tmp.st_uid, from_kuid_munged(current_user_ns(), stat->uid));
 148        SET_GID(tmp.st_gid, from_kgid_munged(current_user_ns(), stat->gid));
 149        tmp.st_rdev = old_encode_dev(stat->rdev);
 150        if ((u64) stat->size > MAX_NON_LFS)
 151                return -EOVERFLOW;
 152        tmp.st_size = stat->size;
 153        tmp.st_atime = stat->atime.tv_sec;
 154        tmp.st_atime_nsec = stat->atime.tv_nsec;
 155        tmp.st_mtime = stat->mtime.tv_sec;
 156        tmp.st_mtime_nsec = stat->mtime.tv_nsec;
 157        tmp.st_ctime = stat->ctime.tv_sec;
 158        tmp.st_ctime_nsec = stat->ctime.tv_nsec;
 159        tmp.st_blocks = stat->blocks;
 160        tmp.st_blksize = stat->blksize;
 161        return copy_to_user(ubuf, &tmp, sizeof(tmp)) ? -EFAULT : 0;
 162}
 163
 164asmlinkage long compat_sys_newstat(const char __user * filename,
 165                struct compat_stat __user *statbuf)
 166{
 167        struct kstat stat;
 168        int error;
 169
 170        error = vfs_stat(filename, &stat);
 171        if (error)
 172                return error;
 173        return cp_compat_stat(&stat, statbuf);
 174}
 175
 176asmlinkage long compat_sys_newlstat(const char __user * filename,
 177                struct compat_stat __user *statbuf)
 178{
 179        struct kstat stat;
 180        int error;
 181
 182        error = vfs_lstat(filename, &stat);
 183        if (error)
 184                return error;
 185        return cp_compat_stat(&stat, statbuf);
 186}
 187
 188#ifndef __ARCH_WANT_STAT64
 189asmlinkage long compat_sys_newfstatat(unsigned int dfd,
 190                const char __user *filename,
 191                struct compat_stat __user *statbuf, int flag)
 192{
 193        struct kstat stat;
 194        int error;
 195
 196        error = vfs_fstatat(dfd, filename, &stat, flag);
 197        if (error)
 198                return error;
 199        return cp_compat_stat(&stat, statbuf);
 200}
 201#endif
 202
 203asmlinkage long compat_sys_newfstat(unsigned int fd,
 204                struct compat_stat __user * statbuf)
 205{
 206        struct kstat stat;
 207        int error = vfs_fstat(fd, &stat);
 208
 209        if (!error)
 210                error = cp_compat_stat(&stat, statbuf);
 211        return error;
 212}
 213
 214static int put_compat_statfs(struct compat_statfs __user *ubuf, struct kstatfs *kbuf)
 215{
 216        
 217        if (sizeof ubuf->f_blocks == 4) {
 218                if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
 219                     kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
 220                        return -EOVERFLOW;
 221                /* f_files and f_ffree may be -1; it's okay
 222                 * to stuff that into 32 bits */
 223                if (kbuf->f_files != 0xffffffffffffffffULL
 224                 && (kbuf->f_files & 0xffffffff00000000ULL))
 225                        return -EOVERFLOW;
 226                if (kbuf->f_ffree != 0xffffffffffffffffULL
 227                 && (kbuf->f_ffree & 0xffffffff00000000ULL))
 228                        return -EOVERFLOW;
 229        }
 230        if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
 231            __put_user(kbuf->f_type, &ubuf->f_type) ||
 232            __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
 233            __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
 234            __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
 235            __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
 236            __put_user(kbuf->f_files, &ubuf->f_files) ||
 237            __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
 238            __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
 239            __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
 240            __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
 241            __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
 242            __put_user(kbuf->f_flags, &ubuf->f_flags) ||
 243            __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
 244                return -EFAULT;
 245        return 0;
 246}
 247
 248/*
 249 * The following statfs calls are copies of code from fs/statfs.c and
 250 * should be checked against those from time to time
 251 */
 252asmlinkage long compat_sys_statfs(const char __user *pathname, struct compat_statfs __user *buf)
 253{
 254        struct kstatfs tmp;
 255        int error = user_statfs(pathname, &tmp);
 256        if (!error)
 257                error = put_compat_statfs(buf, &tmp);
 258        return error;
 259}
 260
 261asmlinkage long compat_sys_fstatfs(unsigned int fd, struct compat_statfs __user *buf)
 262{
 263        struct kstatfs tmp;
 264        int error = fd_statfs(fd, &tmp);
 265        if (!error)
 266                error = put_compat_statfs(buf, &tmp);
 267        return error;
 268}
 269
 270static int put_compat_statfs64(struct compat_statfs64 __user *ubuf, struct kstatfs *kbuf)
 271{
 272        if (sizeof ubuf->f_blocks == 4) {
 273                if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
 274                     kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
 275                        return -EOVERFLOW;
 276                /* f_files and f_ffree may be -1; it's okay
 277                 * to stuff that into 32 bits */
 278                if (kbuf->f_files != 0xffffffffffffffffULL
 279                 && (kbuf->f_files & 0xffffffff00000000ULL))
 280                        return -EOVERFLOW;
 281                if (kbuf->f_ffree != 0xffffffffffffffffULL
 282                 && (kbuf->f_ffree & 0xffffffff00000000ULL))
 283                        return -EOVERFLOW;
 284        }
 285        if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
 286            __put_user(kbuf->f_type, &ubuf->f_type) ||
 287            __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
 288            __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
 289            __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
 290            __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
 291            __put_user(kbuf->f_files, &ubuf->f_files) ||
 292            __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
 293            __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
 294            __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
 295            __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
 296            __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
 297            __put_user(kbuf->f_flags, &ubuf->f_flags) ||
 298            __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
 299                return -EFAULT;
 300        return 0;
 301}
 302
 303asmlinkage long compat_sys_statfs64(const char __user *pathname, compat_size_t sz, struct compat_statfs64 __user *buf)
 304{
 305        struct kstatfs tmp;
 306        int error;
 307
 308        if (sz != sizeof(*buf))
 309                return -EINVAL;
 310
 311        error = user_statfs(pathname, &tmp);
 312        if (!error)
 313                error = put_compat_statfs64(buf, &tmp);
 314        return error;
 315}
 316
 317asmlinkage long compat_sys_fstatfs64(unsigned int fd, compat_size_t sz, struct compat_statfs64 __user *buf)
 318{
 319        struct kstatfs tmp;
 320        int error;
 321
 322        if (sz != sizeof(*buf))
 323                return -EINVAL;
 324
 325        error = fd_statfs(fd, &tmp);
 326        if (!error)
 327                error = put_compat_statfs64(buf, &tmp);
 328        return error;
 329}
 330
 331/*
 332 * This is a copy of sys_ustat, just dealing with a structure layout.
 333 * Given how simple this syscall is that apporach is more maintainable
 334 * than the various conversion hacks.
 335 */
 336asmlinkage long compat_sys_ustat(unsigned dev, struct compat_ustat __user *u)
 337{
 338        struct compat_ustat tmp;
 339        struct kstatfs sbuf;
 340        int err = vfs_ustat(new_decode_dev(dev), &sbuf);
 341        if (err)
 342                return err;
 343
 344        memset(&tmp, 0, sizeof(struct compat_ustat));
 345        tmp.f_tfree = sbuf.f_bfree;
 346        tmp.f_tinode = sbuf.f_ffree;
 347        if (copy_to_user(u, &tmp, sizeof(struct compat_ustat)))
 348                return -EFAULT;
 349        return 0;
 350}
 351
 352static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 353{
 354        if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 355            __get_user(kfl->l_type, &ufl->l_type) ||
 356            __get_user(kfl->l_whence, &ufl->l_whence) ||
 357            __get_user(kfl->l_start, &ufl->l_start) ||
 358            __get_user(kfl->l_len, &ufl->l_len) ||
 359            __get_user(kfl->l_pid, &ufl->l_pid))
 360                return -EFAULT;
 361        return 0;
 362}
 363
 364static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 365{
 366        if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 367            __put_user(kfl->l_type, &ufl->l_type) ||
 368            __put_user(kfl->l_whence, &ufl->l_whence) ||
 369            __put_user(kfl->l_start, &ufl->l_start) ||
 370            __put_user(kfl->l_len, &ufl->l_len) ||
 371            __put_user(kfl->l_pid, &ufl->l_pid))
 372                return -EFAULT;
 373        return 0;
 374}
 375
 376#ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
 377static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 378{
 379        if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
 380            __get_user(kfl->l_type, &ufl->l_type) ||
 381            __get_user(kfl->l_whence, &ufl->l_whence) ||
 382            __get_user(kfl->l_start, &ufl->l_start) ||
 383            __get_user(kfl->l_len, &ufl->l_len) ||
 384            __get_user(kfl->l_pid, &ufl->l_pid))
 385                return -EFAULT;
 386        return 0;
 387}
 388#endif
 389
 390#ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
 391static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
 392{
 393        if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
 394            __put_user(kfl->l_type, &ufl->l_type) ||
 395            __put_user(kfl->l_whence, &ufl->l_whence) ||
 396            __put_user(kfl->l_start, &ufl->l_start) ||
 397            __put_user(kfl->l_len, &ufl->l_len) ||
 398            __put_user(kfl->l_pid, &ufl->l_pid))
 399                return -EFAULT;
 400        return 0;
 401}
 402#endif
 403
 404asmlinkage long compat_sys_fcntl64(unsigned int fd, unsigned int cmd,
 405                unsigned long arg)
 406{
 407        mm_segment_t old_fs;
 408        struct flock f;
 409        long ret;
 410
 411        switch (cmd) {
 412        case F_GETLK:
 413        case F_SETLK:
 414        case F_SETLKW:
 415                ret = get_compat_flock(&f, compat_ptr(arg));
 416                if (ret != 0)
 417                        break;
 418                old_fs = get_fs();
 419                set_fs(KERNEL_DS);
 420                ret = sys_fcntl(fd, cmd, (unsigned long)&f);
 421                set_fs(old_fs);
 422                if (cmd == F_GETLK && ret == 0) {
 423                        /* GETLK was successful and we need to return the data...
 424                         * but it needs to fit in the compat structure.
 425                         * l_start shouldn't be too big, unless the original
 426                         * start + end is greater than COMPAT_OFF_T_MAX, in which
 427                         * case the app was asking for trouble, so we return
 428                         * -EOVERFLOW in that case.
 429                         * l_len could be too big, in which case we just truncate it,
 430                         * and only allow the app to see that part of the conflicting
 431                         * lock that might make sense to it anyway
 432                         */
 433
 434                        if (f.l_start > COMPAT_OFF_T_MAX)
 435                                ret = -EOVERFLOW;
 436                        if (f.l_len > COMPAT_OFF_T_MAX)
 437                                f.l_len = COMPAT_OFF_T_MAX;
 438                        if (ret == 0)
 439                                ret = put_compat_flock(&f, compat_ptr(arg));
 440                }
 441                break;
 442
 443        case F_GETLK64:
 444        case F_SETLK64:
 445        case F_SETLKW64:
 446                ret = get_compat_flock64(&f, compat_ptr(arg));
 447                if (ret != 0)
 448                        break;
 449                old_fs = get_fs();
 450                set_fs(KERNEL_DS);
 451                ret = sys_fcntl(fd, (cmd == F_GETLK64) ? F_GETLK :
 452                                ((cmd == F_SETLK64) ? F_SETLK : F_SETLKW),
 453                                (unsigned long)&f);
 454                set_fs(old_fs);
 455                if (cmd == F_GETLK64 && ret == 0) {
 456                        /* need to return lock information - see above for commentary */
 457                        if (f.l_start > COMPAT_LOFF_T_MAX)
 458                                ret = -EOVERFLOW;
 459                        if (f.l_len > COMPAT_LOFF_T_MAX)
 460                                f.l_len = COMPAT_LOFF_T_MAX;
 461                        if (ret == 0)
 462                                ret = put_compat_flock64(&f, compat_ptr(arg));
 463                }
 464                break;
 465
 466        default:
 467                ret = sys_fcntl(fd, cmd, arg);
 468                break;
 469        }
 470        return ret;
 471}
 472
 473asmlinkage long compat_sys_fcntl(unsigned int fd, unsigned int cmd,
 474                unsigned long arg)
 475{
 476        if ((cmd == F_GETLK64) || (cmd == F_SETLK64) || (cmd == F_SETLKW64))
 477                return -EINVAL;
 478        return compat_sys_fcntl64(fd, cmd, arg);
 479}
 480
 481asmlinkage long
 482compat_sys_io_setup(unsigned nr_reqs, u32 __user *ctx32p)
 483{
 484        long ret;
 485        aio_context_t ctx64;
 486
 487        mm_segment_t oldfs = get_fs();
 488        if (unlikely(get_user(ctx64, ctx32p)))
 489                return -EFAULT;
 490
 491        set_fs(KERNEL_DS);
 492        /* The __user pointer cast is valid because of the set_fs() */
 493        ret = sys_io_setup(nr_reqs, (aio_context_t __user *) &ctx64);
 494        set_fs(oldfs);
 495        /* truncating is ok because it's a user address */
 496        if (!ret)
 497                ret = put_user((u32) ctx64, ctx32p);
 498        return ret;
 499}
 500
 501asmlinkage long
 502compat_sys_io_getevents(aio_context_t ctx_id,
 503                                 unsigned long min_nr,
 504                                 unsigned long nr,
 505                                 struct io_event __user *events,
 506                                 struct compat_timespec __user *timeout)
 507{
 508        long ret;
 509        struct timespec t;
 510        struct timespec __user *ut = NULL;
 511
 512        ret = -EFAULT;
 513        if (unlikely(!access_ok(VERIFY_WRITE, events, 
 514                                nr * sizeof(struct io_event))))
 515                goto out;
 516        if (timeout) {
 517                if (get_compat_timespec(&t, timeout))
 518                        goto out;
 519
 520                ut = compat_alloc_user_space(sizeof(*ut));
 521                if (copy_to_user(ut, &t, sizeof(t)) )
 522                        goto out;
 523        } 
 524        ret = sys_io_getevents(ctx_id, min_nr, nr, events, ut);
 525out:
 526        return ret;
 527}
 528
 529/* A write operation does a read from user space and vice versa */
 530#define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 531
 532ssize_t compat_rw_copy_check_uvector(int type,
 533                const struct compat_iovec __user *uvector, unsigned long nr_segs,
 534                unsigned long fast_segs, struct iovec *fast_pointer,
 535                struct iovec **ret_pointer)
 536{
 537        compat_ssize_t tot_len;
 538        struct iovec *iov = *ret_pointer = fast_pointer;
 539        ssize_t ret = 0;
 540        int seg;
 541
 542        /*
 543         * SuS says "The readv() function *may* fail if the iovcnt argument
 544         * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 545         * traditionally returned zero for zero segments, so...
 546         */
 547        if (nr_segs == 0)
 548                goto out;
 549
 550        ret = -EINVAL;
 551        if (nr_segs > UIO_MAXIOV || nr_segs < 0)
 552                goto out;
 553        if (nr_segs > fast_segs) {
 554                ret = -ENOMEM;
 555                iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 556                if (iov == NULL)
 557                        goto out;
 558        }
 559        *ret_pointer = iov;
 560
 561        /*
 562         * Single unix specification:
 563         * We should -EINVAL if an element length is not >= 0 and fitting an
 564         * ssize_t.
 565         *
 566         * In Linux, the total length is limited to MAX_RW_COUNT, there is
 567         * no overflow possibility.
 568         */
 569        tot_len = 0;
 570        ret = -EINVAL;
 571        for (seg = 0; seg < nr_segs; seg++) {
 572                compat_uptr_t buf;
 573                compat_ssize_t len;
 574
 575                if (__get_user(len, &uvector->iov_len) ||
 576                   __get_user(buf, &uvector->iov_base)) {
 577                        ret = -EFAULT;
 578                        goto out;
 579                }
 580                if (len < 0)    /* size_t not fitting in compat_ssize_t .. */
 581                        goto out;
 582                if (type >= 0 &&
 583                    !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
 584                        ret = -EFAULT;
 585                        goto out;
 586                }
 587                if (len > MAX_RW_COUNT - tot_len)
 588                        len = MAX_RW_COUNT - tot_len;
 589                tot_len += len;
 590                iov->iov_base = compat_ptr(buf);
 591                iov->iov_len = (compat_size_t) len;
 592                uvector++;
 593                iov++;
 594        }
 595        ret = tot_len;
 596
 597out:
 598        return ret;
 599}
 600
 601static inline long
 602copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
 603{
 604        compat_uptr_t uptr;
 605        int i;
 606
 607        for (i = 0; i < nr; ++i) {
 608                if (get_user(uptr, ptr32 + i))
 609                        return -EFAULT;
 610                if (put_user(compat_ptr(uptr), ptr64 + i))
 611                        return -EFAULT;
 612        }
 613        return 0;
 614}
 615
 616#define MAX_AIO_SUBMITS         (PAGE_SIZE/sizeof(struct iocb *))
 617
 618asmlinkage long
 619compat_sys_io_submit(aio_context_t ctx_id, int nr, u32 __user *iocb)
 620{
 621        struct iocb __user * __user *iocb64; 
 622        long ret;
 623
 624        if (unlikely(nr < 0))
 625                return -EINVAL;
 626
 627        if (nr > MAX_AIO_SUBMITS)
 628                nr = MAX_AIO_SUBMITS;
 629        
 630        iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
 631        ret = copy_iocb(nr, iocb, iocb64);
 632        if (!ret)
 633                ret = do_io_submit(ctx_id, nr, iocb64, 1);
 634        return ret;
 635}
 636
 637struct compat_ncp_mount_data {
 638        compat_int_t version;
 639        compat_uint_t ncp_fd;
 640        __compat_uid_t mounted_uid;
 641        compat_pid_t wdog_pid;
 642        unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
 643        compat_uint_t time_out;
 644        compat_uint_t retry_count;
 645        compat_uint_t flags;
 646        __compat_uid_t uid;
 647        __compat_gid_t gid;
 648        compat_mode_t file_mode;
 649        compat_mode_t dir_mode;
 650};
 651
 652struct compat_ncp_mount_data_v4 {
 653        compat_int_t version;
 654        compat_ulong_t flags;
 655        compat_ulong_t mounted_uid;
 656        compat_long_t wdog_pid;
 657        compat_uint_t ncp_fd;
 658        compat_uint_t time_out;
 659        compat_uint_t retry_count;
 660        compat_ulong_t uid;
 661        compat_ulong_t gid;
 662        compat_ulong_t file_mode;
 663        compat_ulong_t dir_mode;
 664};
 665
 666static void *do_ncp_super_data_conv(void *raw_data)
 667{
 668        int version = *(unsigned int *)raw_data;
 669
 670        if (version == 3) {
 671                struct compat_ncp_mount_data *c_n = raw_data;
 672                struct ncp_mount_data *n = raw_data;
 673
 674                n->dir_mode = c_n->dir_mode;
 675                n->file_mode = c_n->file_mode;
 676                n->gid = c_n->gid;
 677                n->uid = c_n->uid;
 678                memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
 679                n->wdog_pid = c_n->wdog_pid;
 680                n->mounted_uid = c_n->mounted_uid;
 681        } else if (version == 4) {
 682                struct compat_ncp_mount_data_v4 *c_n = raw_data;
 683                struct ncp_mount_data_v4 *n = raw_data;
 684
 685                n->dir_mode = c_n->dir_mode;
 686                n->file_mode = c_n->file_mode;
 687                n->gid = c_n->gid;
 688                n->uid = c_n->uid;
 689                n->retry_count = c_n->retry_count;
 690                n->time_out = c_n->time_out;
 691                n->ncp_fd = c_n->ncp_fd;
 692                n->wdog_pid = c_n->wdog_pid;
 693                n->mounted_uid = c_n->mounted_uid;
 694                n->flags = c_n->flags;
 695        } else if (version != 5) {
 696                return NULL;
 697        }
 698
 699        return raw_data;
 700}
 701
 702
 703struct compat_nfs_string {
 704        compat_uint_t len;
 705        compat_uptr_t data;
 706};
 707
 708static inline void compat_nfs_string(struct nfs_string *dst,
 709                                     struct compat_nfs_string *src)
 710{
 711        dst->data = compat_ptr(src->data);
 712        dst->len = src->len;
 713}
 714
 715struct compat_nfs4_mount_data_v1 {
 716        compat_int_t version;
 717        compat_int_t flags;
 718        compat_int_t rsize;
 719        compat_int_t wsize;
 720        compat_int_t timeo;
 721        compat_int_t retrans;
 722        compat_int_t acregmin;
 723        compat_int_t acregmax;
 724        compat_int_t acdirmin;
 725        compat_int_t acdirmax;
 726        struct compat_nfs_string client_addr;
 727        struct compat_nfs_string mnt_path;
 728        struct compat_nfs_string hostname;
 729        compat_uint_t host_addrlen;
 730        compat_uptr_t host_addr;
 731        compat_int_t proto;
 732        compat_int_t auth_flavourlen;
 733        compat_uptr_t auth_flavours;
 734};
 735
 736static int do_nfs4_super_data_conv(void *raw_data)
 737{
 738        int version = *(compat_uint_t *) raw_data;
 739
 740        if (version == 1) {
 741                struct compat_nfs4_mount_data_v1 *raw = raw_data;
 742                struct nfs4_mount_data *real = raw_data;
 743
 744                /* copy the fields backwards */
 745                real->auth_flavours = compat_ptr(raw->auth_flavours);
 746                real->auth_flavourlen = raw->auth_flavourlen;
 747                real->proto = raw->proto;
 748                real->host_addr = compat_ptr(raw->host_addr);
 749                real->host_addrlen = raw->host_addrlen;
 750                compat_nfs_string(&real->hostname, &raw->hostname);
 751                compat_nfs_string(&real->mnt_path, &raw->mnt_path);
 752                compat_nfs_string(&real->client_addr, &raw->client_addr);
 753                real->acdirmax = raw->acdirmax;
 754                real->acdirmin = raw->acdirmin;
 755                real->acregmax = raw->acregmax;
 756                real->acregmin = raw->acregmin;
 757                real->retrans = raw->retrans;
 758                real->timeo = raw->timeo;
 759                real->wsize = raw->wsize;
 760                real->rsize = raw->rsize;
 761                real->flags = raw->flags;
 762                real->version = raw->version;
 763        }
 764
 765        return 0;
 766}
 767
 768#define NCPFS_NAME      "ncpfs"
 769#define NFS4_NAME       "nfs4"
 770
 771asmlinkage long compat_sys_mount(const char __user * dev_name,
 772                                 const char __user * dir_name,
 773                                 const char __user * type, unsigned long flags,
 774                                 const void __user * data)
 775{
 776        char *kernel_type;
 777        unsigned long data_page;
 778        char *kernel_dev;
 779        char *dir_page;
 780        int retval;
 781
 782        retval = copy_mount_string(type, &kernel_type);
 783        if (retval < 0)
 784                goto out;
 785
 786        dir_page = getname(dir_name);
 787        retval = PTR_ERR(dir_page);
 788        if (IS_ERR(dir_page))
 789                goto out1;
 790
 791        retval = copy_mount_string(dev_name, &kernel_dev);
 792        if (retval < 0)
 793                goto out2;
 794
 795        retval = copy_mount_options(data, &data_page);
 796        if (retval < 0)
 797                goto out3;
 798
 799        retval = -EINVAL;
 800
 801        if (kernel_type && data_page) {
 802                if (!strcmp(kernel_type, NCPFS_NAME)) {
 803                        do_ncp_super_data_conv((void *)data_page);
 804                } else if (!strcmp(kernel_type, NFS4_NAME)) {
 805                        if (do_nfs4_super_data_conv((void *) data_page))
 806                                goto out4;
 807                }
 808        }
 809
 810        retval = do_mount(kernel_dev, dir_page, kernel_type,
 811                        flags, (void*)data_page);
 812
 813 out4:
 814        free_page(data_page);
 815 out3:
 816        kfree(kernel_dev);
 817 out2:
 818        putname(dir_page);
 819 out1:
 820        kfree(kernel_type);
 821 out:
 822        return retval;
 823}
 824
 825struct compat_old_linux_dirent {
 826        compat_ulong_t  d_ino;
 827        compat_ulong_t  d_offset;
 828        unsigned short  d_namlen;
 829        char            d_name[1];
 830};
 831
 832struct compat_readdir_callback {
 833        struct compat_old_linux_dirent __user *dirent;
 834        int result;
 835};
 836
 837static int compat_fillonedir(void *__buf, const char *name, int namlen,
 838                        loff_t offset, u64 ino, unsigned int d_type)
 839{
 840        struct compat_readdir_callback *buf = __buf;
 841        struct compat_old_linux_dirent __user *dirent;
 842        compat_ulong_t d_ino;
 843
 844        if (buf->result)
 845                return -EINVAL;
 846        d_ino = ino;
 847        if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
 848                buf->result = -EOVERFLOW;
 849                return -EOVERFLOW;
 850        }
 851        buf->result++;
 852        dirent = buf->dirent;
 853        if (!access_ok(VERIFY_WRITE, dirent,
 854                        (unsigned long)(dirent->d_name + namlen + 1) -
 855                                (unsigned long)dirent))
 856                goto efault;
 857        if (    __put_user(d_ino, &dirent->d_ino) ||
 858                __put_user(offset, &dirent->d_offset) ||
 859                __put_user(namlen, &dirent->d_namlen) ||
 860                __copy_to_user(dirent->d_name, name, namlen) ||
 861                __put_user(0, dirent->d_name + namlen))
 862                goto efault;
 863        return 0;
 864efault:
 865        buf->result = -EFAULT;
 866        return -EFAULT;
 867}
 868
 869asmlinkage long compat_sys_old_readdir(unsigned int fd,
 870        struct compat_old_linux_dirent __user *dirent, unsigned int count)
 871{
 872        int error;
 873        struct file *file;
 874        int fput_needed;
 875        struct compat_readdir_callback buf;
 876
 877        file = fget_light(fd, &fput_needed);
 878        if (!file)
 879                return -EBADF;
 880
 881        buf.result = 0;
 882        buf.dirent = dirent;
 883
 884        error = vfs_readdir(file, compat_fillonedir, &buf);
 885        if (buf.result)
 886                error = buf.result;
 887
 888        fput_light(file, fput_needed);
 889        return error;
 890}
 891
 892struct compat_linux_dirent {
 893        compat_ulong_t  d_ino;
 894        compat_ulong_t  d_off;
 895        unsigned short  d_reclen;
 896        char            d_name[1];
 897};
 898
 899struct compat_getdents_callback {
 900        struct compat_linux_dirent __user *current_dir;
 901        struct compat_linux_dirent __user *previous;
 902        int count;
 903        int error;
 904};
 905
 906static int compat_filldir(void *__buf, const char *name, int namlen,
 907                loff_t offset, u64 ino, unsigned int d_type)
 908{
 909        struct compat_linux_dirent __user * dirent;
 910        struct compat_getdents_callback *buf = __buf;
 911        compat_ulong_t d_ino;
 912        int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
 913                namlen + 2, sizeof(compat_long_t));
 914
 915        buf->error = -EINVAL;   /* only used if we fail.. */
 916        if (reclen > buf->count)
 917                return -EINVAL;
 918        d_ino = ino;
 919        if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
 920                buf->error = -EOVERFLOW;
 921                return -EOVERFLOW;
 922        }
 923        dirent = buf->previous;
 924        if (dirent) {
 925                if (__put_user(offset, &dirent->d_off))
 926                        goto efault;
 927        }
 928        dirent = buf->current_dir;
 929        if (__put_user(d_ino, &dirent->d_ino))
 930                goto efault;
 931        if (__put_user(reclen, &dirent->d_reclen))
 932                goto efault;
 933        if (copy_to_user(dirent->d_name, name, namlen))
 934                goto efault;
 935        if (__put_user(0, dirent->d_name + namlen))
 936                goto efault;
 937        if (__put_user(d_type, (char  __user *) dirent + reclen - 1))
 938                goto efault;
 939        buf->previous = dirent;
 940        dirent = (void __user *)dirent + reclen;
 941        buf->current_dir = dirent;
 942        buf->count -= reclen;
 943        return 0;
 944efault:
 945        buf->error = -EFAULT;
 946        return -EFAULT;
 947}
 948
 949asmlinkage long compat_sys_getdents(unsigned int fd,
 950                struct compat_linux_dirent __user *dirent, unsigned int count)
 951{
 952        struct file * file;
 953        struct compat_linux_dirent __user * lastdirent;
 954        struct compat_getdents_callback buf;
 955        int fput_needed;
 956        int error;
 957
 958        if (!access_ok(VERIFY_WRITE, dirent, count))
 959                return -EFAULT;
 960
 961        file = fget_light(fd, &fput_needed);
 962        if (!file)
 963                return -EBADF;
 964
 965        buf.current_dir = dirent;
 966        buf.previous = NULL;
 967        buf.count = count;
 968        buf.error = 0;
 969
 970        error = vfs_readdir(file, compat_filldir, &buf);
 971        if (error >= 0)
 972                error = buf.error;
 973        lastdirent = buf.previous;
 974        if (lastdirent) {
 975                if (put_user(file->f_pos, &lastdirent->d_off))
 976                        error = -EFAULT;
 977                else
 978                        error = count - buf.count;
 979        }
 980        fput_light(file, fput_needed);
 981        return error;
 982}
 983
 984#ifndef __ARCH_OMIT_COMPAT_SYS_GETDENTS64
 985
 986struct compat_getdents_callback64 {
 987        struct linux_dirent64 __user *current_dir;
 988        struct linux_dirent64 __user *previous;
 989        int count;
 990        int error;
 991};
 992
 993static int compat_filldir64(void * __buf, const char * name, int namlen, loff_t offset,
 994                     u64 ino, unsigned int d_type)
 995{
 996        struct linux_dirent64 __user *dirent;
 997        struct compat_getdents_callback64 *buf = __buf;
 998        int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
 999                sizeof(u64));
1000        u64 off;
1001
1002        buf->error = -EINVAL;   /* only used if we fail.. */
1003        if (reclen > buf->count)
1004                return -EINVAL;
1005        dirent = buf->previous;
1006
1007        if (dirent) {
1008                if (__put_user_unaligned(offset, &dirent->d_off))
1009                        goto efault;
1010        }
1011        dirent = buf->current_dir;
1012        if (__put_user_unaligned(ino, &dirent->d_ino))
1013                goto efault;
1014        off = 0;
1015        if (__put_user_unaligned(off, &dirent->d_off))
1016                goto efault;
1017        if (__put_user(reclen, &dirent->d_reclen))
1018                goto efault;
1019        if (__put_user(d_type, &dirent->d_type))
1020                goto efault;
1021        if (copy_to_user(dirent->d_name, name, namlen))
1022                goto efault;
1023        if (__put_user(0, dirent->d_name + namlen))
1024                goto efault;
1025        buf->previous = dirent;
1026        dirent = (void __user *)dirent + reclen;
1027        buf->current_dir = dirent;
1028        buf->count -= reclen;
1029        return 0;
1030efault:
1031        buf->error = -EFAULT;
1032        return -EFAULT;
1033}
1034
1035asmlinkage long compat_sys_getdents64(unsigned int fd,
1036                struct linux_dirent64 __user * dirent, unsigned int count)
1037{
1038        struct file * file;
1039        struct linux_dirent64 __user * lastdirent;
1040        struct compat_getdents_callback64 buf;
1041        int fput_needed;
1042        int error;
1043
1044        if (!access_ok(VERIFY_WRITE, dirent, count))
1045                return -EFAULT;
1046
1047        file = fget_light(fd, &fput_needed);
1048        if (!file)
1049                return -EBADF;
1050
1051        buf.current_dir = dirent;
1052        buf.previous = NULL;
1053        buf.count = count;
1054        buf.error = 0;
1055
1056        error = vfs_readdir(file, compat_filldir64, &buf);
1057        if (error >= 0)
1058                error = buf.error;
1059        lastdirent = buf.previous;
1060        if (lastdirent) {
1061                typeof(lastdirent->d_off) d_off = file->f_pos;
1062                if (__put_user_unaligned(d_off, &lastdirent->d_off))
1063                        error = -EFAULT;
1064                else
1065                        error = count - buf.count;
1066        }
1067        fput_light(file, fput_needed);
1068        return error;
1069}
1070#endif /* ! __ARCH_OMIT_COMPAT_SYS_GETDENTS64 */
1071
1072static ssize_t compat_do_readv_writev(int type, struct file *file,
1073                               const struct compat_iovec __user *uvector,
1074                               unsigned long nr_segs, loff_t *pos)
1075{
1076        compat_ssize_t tot_len;
1077        struct iovec iovstack[UIO_FASTIOV];
1078        struct iovec *iov = iovstack;
1079        ssize_t ret;
1080        io_fn_t fn;
1081        iov_fn_t fnv;
1082
1083        ret = -EINVAL;
1084        if (!file->f_op)
1085                goto out;
1086
1087        ret = -EFAULT;
1088        if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
1089                goto out;
1090
1091        tot_len = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1092                                               UIO_FASTIOV, iovstack, &iov);
1093        if (tot_len == 0) {
1094                ret = 0;
1095                goto out;
1096        }
1097
1098        ret = rw_verify_area(type, file, pos, tot_len);
1099        if (ret < 0)
1100                goto out;
1101
1102        fnv = NULL;
1103        if (type == READ) {
1104                fn = file->f_op->read;
1105                fnv = file->f_op->aio_read;
1106        } else {
1107                fn = (io_fn_t)file->f_op->write;
1108                fnv = file->f_op->aio_write;
1109        }
1110
1111        if (fnv)
1112                ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1113                                                pos, fnv);
1114        else
1115                ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1116
1117out:
1118        if (iov != iovstack)
1119                kfree(iov);
1120        if ((ret + (type == READ)) > 0) {
1121                if (type == READ)
1122                        fsnotify_access(file);
1123                else
1124                        fsnotify_modify(file);
1125        }
1126        return ret;
1127}
1128
1129static size_t compat_readv(struct file *file,
1130                           const struct compat_iovec __user *vec,
1131                           unsigned long vlen, loff_t *pos)
1132{
1133        ssize_t ret = -EBADF;
1134
1135        if (!(file->f_mode & FMODE_READ))
1136                goto out;
1137
1138        ret = -EINVAL;
1139        if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
1140                goto out;
1141
1142        ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1143
1144out:
1145        if (ret > 0)
1146                add_rchar(current, ret);
1147        inc_syscr(current);
1148        return ret;
1149}
1150
1151asmlinkage ssize_t
1152compat_sys_readv(unsigned long fd, const struct compat_iovec __user *vec,
1153                 unsigned long vlen)
1154{
1155        struct file *file;
1156        int fput_needed;
1157        ssize_t ret;
1158        loff_t pos;
1159
1160        file = fget_light(fd, &fput_needed);
1161        if (!file)
1162                return -EBADF;
1163        pos = file->f_pos;
1164        ret = compat_readv(file, vec, vlen, &pos);
1165        file->f_pos = pos;
1166        fput_light(file, fput_needed);
1167        return ret;
1168}
1169
1170asmlinkage ssize_t
1171compat_sys_preadv64(unsigned long fd, const struct compat_iovec __user *vec,
1172                    unsigned long vlen, loff_t pos)
1173{
1174        struct file *file;
1175        int fput_needed;
1176        ssize_t ret;
1177
1178        if (pos < 0)
1179                return -EINVAL;
1180        file = fget_light(fd, &fput_needed);
1181        if (!file)
1182                return -EBADF;
1183        ret = -ESPIPE;
1184        if (file->f_mode & FMODE_PREAD)
1185                ret = compat_readv(file, vec, vlen, &pos);
1186        fput_light(file, fput_needed);
1187        return ret;
1188}
1189
1190asmlinkage ssize_t
1191compat_sys_preadv(unsigned long fd, const struct compat_iovec __user *vec,
1192                  unsigned long vlen, u32 pos_low, u32 pos_high)
1193{
1194        loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1195        return compat_sys_preadv64(fd, vec, vlen, pos);
1196}
1197
1198static size_t compat_writev(struct file *file,
1199                            const struct compat_iovec __user *vec,
1200                            unsigned long vlen, loff_t *pos)
1201{
1202        ssize_t ret = -EBADF;
1203
1204        if (!(file->f_mode & FMODE_WRITE))
1205                goto out;
1206
1207        ret = -EINVAL;
1208        if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
1209                goto out;
1210
1211        ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1212
1213out:
1214        if (ret > 0)
1215                add_wchar(current, ret);
1216        inc_syscw(current);
1217        return ret;
1218}
1219
1220asmlinkage ssize_t
1221compat_sys_writev(unsigned long fd, const struct compat_iovec __user *vec,
1222                  unsigned long vlen)
1223{
1224        struct file *file;
1225        int fput_needed;
1226        ssize_t ret;
1227        loff_t pos;
1228
1229        file = fget_light(fd, &fput_needed);
1230        if (!file)
1231                return -EBADF;
1232        pos = file->f_pos;
1233        ret = compat_writev(file, vec, vlen, &pos);
1234        file->f_pos = pos;
1235        fput_light(file, fput_needed);
1236        return ret;
1237}
1238
1239asmlinkage ssize_t
1240compat_sys_pwritev64(unsigned long fd, const struct compat_iovec __user *vec,
1241                     unsigned long vlen, loff_t pos)
1242{
1243        struct file *file;
1244        int fput_needed;
1245        ssize_t ret;
1246
1247        if (pos < 0)
1248                return -EINVAL;
1249        file = fget_light(fd, &fput_needed);
1250        if (!file)
1251                return -EBADF;
1252        ret = -ESPIPE;
1253        if (file->f_mode & FMODE_PWRITE)
1254                ret = compat_writev(file, vec, vlen, &pos);
1255        fput_light(file, fput_needed);
1256        return ret;
1257}
1258
1259asmlinkage ssize_t
1260compat_sys_pwritev(unsigned long fd, const struct compat_iovec __user *vec,
1261                   unsigned long vlen, u32 pos_low, u32 pos_high)
1262{
1263        loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1264        return compat_sys_pwritev64(fd, vec, vlen, pos);
1265}
1266
1267asmlinkage long
1268compat_sys_vmsplice(int fd, const struct compat_iovec __user *iov32,
1269                    unsigned int nr_segs, unsigned int flags)
1270{
1271        unsigned i;
1272        struct iovec __user *iov;
1273        if (nr_segs > UIO_MAXIOV)
1274                return -EINVAL;
1275        iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1276        for (i = 0; i < nr_segs; i++) {
1277                struct compat_iovec v;
1278                if (get_user(v.iov_base, &iov32[i].iov_base) ||
1279                    get_user(v.iov_len, &iov32[i].iov_len) ||
1280                    put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1281                    put_user(v.iov_len, &iov[i].iov_len))
1282                        return -EFAULT;
1283        }
1284        return sys_vmsplice(fd, iov, nr_segs, flags);
1285}
1286
1287/*
1288 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1289 * O_LARGEFILE flag.
1290 */
1291asmlinkage long
1292compat_sys_open(const char __user *filename, int flags, umode_t mode)
1293{
1294        return do_sys_open(AT_FDCWD, filename, flags, mode);
1295}
1296
1297/*
1298 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1299 * O_LARGEFILE flag.
1300 */
1301asmlinkage long
1302compat_sys_openat(unsigned int dfd, const char __user *filename, int flags, umode_t mode)
1303{
1304        return do_sys_open(dfd, filename, flags, mode);
1305}
1306
1307#define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1308
1309static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1310                                      int timeval, int ret)
1311{
1312        struct timespec ts;
1313
1314        if (!p)
1315                return ret;
1316
1317        if (current->personality & STICKY_TIMEOUTS)
1318                goto sticky;
1319
1320        /* No update for zero timeout */
1321        if (!end_time->tv_sec && !end_time->tv_nsec)
1322                return ret;
1323
1324        ktime_get_ts(&ts);
1325        ts = timespec_sub(*end_time, ts);
1326        if (ts.tv_sec < 0)
1327                ts.tv_sec = ts.tv_nsec = 0;
1328
1329        if (timeval) {
1330                struct compat_timeval rtv;
1331
1332                rtv.tv_sec = ts.tv_sec;
1333                rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1334
1335                if (!copy_to_user(p, &rtv, sizeof(rtv)))
1336                        return ret;
1337        } else {
1338                struct compat_timespec rts;
1339
1340                rts.tv_sec = ts.tv_sec;
1341                rts.tv_nsec = ts.tv_nsec;
1342
1343                if (!copy_to_user(p, &rts, sizeof(rts)))
1344                        return ret;
1345        }
1346        /*
1347         * If an application puts its timeval in read-only memory, we
1348         * don't want the Linux-specific update to the timeval to
1349         * cause a fault after the select has completed
1350         * successfully. However, because we're not updating the
1351         * timeval, we can't restart the system call.
1352         */
1353
1354sticky:
1355        if (ret == -ERESTARTNOHAND)
1356                ret = -EINTR;
1357        return ret;
1358}
1359
1360/*
1361 * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1362 * 64-bit unsigned longs.
1363 */
1364static
1365int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1366                        unsigned long *fdset)
1367{
1368        nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1369        if (ufdset) {
1370                unsigned long odd;
1371
1372                if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1373                        return -EFAULT;
1374
1375                odd = nr & 1UL;
1376                nr &= ~1UL;
1377                while (nr) {
1378                        unsigned long h, l;
1379                        if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1380                                return -EFAULT;
1381                        ufdset += 2;
1382                        *fdset++ = h << 32 | l;
1383                        nr -= 2;
1384                }
1385                if (odd && __get_user(*fdset, ufdset))
1386                        return -EFAULT;
1387        } else {
1388                /* Tricky, must clear full unsigned long in the
1389                 * kernel fdset at the end, this makes sure that
1390                 * actually happens.
1391                 */
1392                memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1393        }
1394        return 0;
1395}
1396
1397static
1398int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1399                      unsigned long *fdset)
1400{
1401        unsigned long odd;
1402        nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1403
1404        if (!ufdset)
1405                return 0;
1406
1407        odd = nr & 1UL;
1408        nr &= ~1UL;
1409        while (nr) {
1410                unsigned long h, l;
1411                l = *fdset++;
1412                h = l >> 32;
1413                if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1414                        return -EFAULT;
1415                ufdset += 2;
1416                nr -= 2;
1417        }
1418        if (odd && __put_user(*fdset, ufdset))
1419                return -EFAULT;
1420        return 0;
1421}
1422
1423
1424/*
1425 * This is a virtual copy of sys_select from fs/select.c and probably
1426 * should be compared to it from time to time
1427 */
1428
1429/*
1430 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1431 * like to be certain this leads to no problems. So I return
1432 * EINTR just for safety.
1433 *
1434 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1435 * I'm trying ERESTARTNOHAND which restart only when you want to.
1436 */
1437int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1438        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1439        struct timespec *end_time)
1440{
1441        fd_set_bits fds;
1442        void *bits;
1443        int size, max_fds, ret = -EINVAL;
1444        struct fdtable *fdt;
1445        long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1446
1447        if (n < 0)
1448                goto out_nofds;
1449
1450        /* max_fds can increase, so grab it once to avoid race */
1451        rcu_read_lock();
1452        fdt = files_fdtable(current->files);
1453        max_fds = fdt->max_fds;
1454        rcu_read_unlock();
1455        if (n > max_fds)
1456                n = max_fds;
1457
1458        /*
1459         * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1460         * since we used fdset we need to allocate memory in units of
1461         * long-words.
1462         */
1463        size = FDS_BYTES(n);
1464        bits = stack_fds;
1465        if (size > sizeof(stack_fds) / 6) {
1466                bits = kmalloc(6 * size, GFP_KERNEL);
1467                ret = -ENOMEM;
1468                if (!bits)
1469                        goto out_nofds;
1470        }
1471        fds.in      = (unsigned long *)  bits;
1472        fds.out     = (unsigned long *) (bits +   size);
1473        fds.ex      = (unsigned long *) (bits + 2*size);
1474        fds.res_in  = (unsigned long *) (bits + 3*size);
1475        fds.res_out = (unsigned long *) (bits + 4*size);
1476        fds.res_ex  = (unsigned long *) (bits + 5*size);
1477
1478        if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1479            (ret = compat_get_fd_set(n, outp, fds.out)) ||
1480            (ret = compat_get_fd_set(n, exp, fds.ex)))
1481                goto out;
1482        zero_fd_set(n, fds.res_in);
1483        zero_fd_set(n, fds.res_out);
1484        zero_fd_set(n, fds.res_ex);
1485
1486        ret = do_select(n, &fds, end_time);
1487
1488        if (ret < 0)
1489                goto out;
1490        if (!ret) {
1491                ret = -ERESTARTNOHAND;
1492                if (signal_pending(current))
1493                        goto out;
1494                ret = 0;
1495        }
1496
1497        if (compat_set_fd_set(n, inp, fds.res_in) ||
1498            compat_set_fd_set(n, outp, fds.res_out) ||
1499            compat_set_fd_set(n, exp, fds.res_ex))
1500                ret = -EFAULT;
1501out:
1502        if (bits != stack_fds)
1503                kfree(bits);
1504out_nofds:
1505        return ret;
1506}
1507
1508asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1509        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1510        struct compat_timeval __user *tvp)
1511{
1512        struct timespec end_time, *to = NULL;
1513        struct compat_timeval tv;
1514        int ret;
1515
1516        if (tvp) {
1517                if (copy_from_user(&tv, tvp, sizeof(tv)))
1518                        return -EFAULT;
1519
1520                to = &end_time;
1521                if (poll_select_set_timeout(to,
1522                                tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1523                                (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1524                        return -EINVAL;
1525        }
1526
1527        ret = compat_core_sys_select(n, inp, outp, exp, to);
1528        ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1529
1530        return ret;
1531}
1532
1533struct compat_sel_arg_struct {
1534        compat_ulong_t n;
1535        compat_uptr_t inp;
1536        compat_uptr_t outp;
1537        compat_uptr_t exp;
1538        compat_uptr_t tvp;
1539};
1540
1541asmlinkage long compat_sys_old_select(struct compat_sel_arg_struct __user *arg)
1542{
1543        struct compat_sel_arg_struct a;
1544
1545        if (copy_from_user(&a, arg, sizeof(a)))
1546                return -EFAULT;
1547        return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1548                                 compat_ptr(a.exp), compat_ptr(a.tvp));
1549}
1550
1551static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1552        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1553        struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1554        compat_size_t sigsetsize)
1555{
1556        compat_sigset_t ss32;
1557        sigset_t ksigmask, sigsaved;
1558        struct compat_timespec ts;
1559        struct timespec end_time, *to = NULL;
1560        int ret;
1561
1562        if (tsp) {
1563                if (copy_from_user(&ts, tsp, sizeof(ts)))
1564                        return -EFAULT;
1565
1566                to = &end_time;
1567                if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1568                        return -EINVAL;
1569        }
1570
1571        if (sigmask) {
1572                if (sigsetsize != sizeof(compat_sigset_t))
1573                        return -EINVAL;
1574                if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1575                        return -EFAULT;
1576                sigset_from_compat(&ksigmask, &ss32);
1577
1578                sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1579                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1580        }
1581
1582        ret = compat_core_sys_select(n, inp, outp, exp, to);
1583        ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1584
1585        if (ret == -ERESTARTNOHAND) {
1586                /*
1587                 * Don't restore the signal mask yet. Let do_signal() deliver
1588                 * the signal on the way back to userspace, before the signal
1589                 * mask is restored.
1590                 */
1591                if (sigmask) {
1592                        memcpy(&current->saved_sigmask, &sigsaved,
1593                                        sizeof(sigsaved));
1594                        set_restore_sigmask();
1595                }
1596        } else if (sigmask)
1597                sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1598
1599        return ret;
1600}
1601
1602asmlinkage long compat_sys_pselect6(int n, compat_ulong_t __user *inp,
1603        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1604        struct compat_timespec __user *tsp, void __user *sig)
1605{
1606        compat_size_t sigsetsize = 0;
1607        compat_uptr_t up = 0;
1608
1609        if (sig) {
1610                if (!access_ok(VERIFY_READ, sig,
1611                                sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1612                        __get_user(up, (compat_uptr_t __user *)sig) ||
1613                        __get_user(sigsetsize,
1614                                (compat_size_t __user *)(sig+sizeof(up))))
1615                        return -EFAULT;
1616        }
1617        return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1618                                 sigsetsize);
1619}
1620
1621asmlinkage long compat_sys_ppoll(struct pollfd __user *ufds,
1622        unsigned int nfds, struct compat_timespec __user *tsp,
1623        const compat_sigset_t __user *sigmask, compat_size_t sigsetsize)
1624{
1625        compat_sigset_t ss32;
1626        sigset_t ksigmask, sigsaved;
1627        struct compat_timespec ts;
1628        struct timespec end_time, *to = NULL;
1629        int ret;
1630
1631        if (tsp) {
1632                if (copy_from_user(&ts, tsp, sizeof(ts)))
1633                        return -EFAULT;
1634
1635                to = &end_time;
1636                if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1637                        return -EINVAL;
1638        }
1639
1640        if (sigmask) {
1641                if (sigsetsize != sizeof(compat_sigset_t))
1642                        return -EINVAL;
1643                if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1644                        return -EFAULT;
1645                sigset_from_compat(&ksigmask, &ss32);
1646
1647                sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1648                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1649        }
1650
1651        ret = do_sys_poll(ufds, nfds, to);
1652
1653        /* We can restart this syscall, usually */
1654        if (ret == -EINTR) {
1655                /*
1656                 * Don't restore the signal mask yet. Let do_signal() deliver
1657                 * the signal on the way back to userspace, before the signal
1658                 * mask is restored.
1659                 */
1660                if (sigmask) {
1661                        memcpy(&current->saved_sigmask, &sigsaved,
1662                                sizeof(sigsaved));
1663                        set_restore_sigmask();
1664                }
1665                ret = -ERESTARTNOHAND;
1666        } else if (sigmask)
1667                sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1668
1669        ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1670
1671        return ret;
1672}
1673
1674#ifdef CONFIG_EPOLL
1675
1676asmlinkage long compat_sys_epoll_pwait(int epfd,
1677                        struct compat_epoll_event __user *events,
1678                        int maxevents, int timeout,
1679                        const compat_sigset_t __user *sigmask,
1680                        compat_size_t sigsetsize)
1681{
1682        long err;
1683        compat_sigset_t csigmask;
1684        sigset_t ksigmask, sigsaved;
1685
1686        /*
1687         * If the caller wants a certain signal mask to be set during the wait,
1688         * we apply it here.
1689         */
1690        if (sigmask) {
1691                if (sigsetsize != sizeof(compat_sigset_t))
1692                        return -EINVAL;
1693                if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
1694                        return -EFAULT;
1695                sigset_from_compat(&ksigmask, &csigmask);
1696                sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1697                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1698        }
1699
1700        err = sys_epoll_wait(epfd, events, maxevents, timeout);
1701
1702        /*
1703         * If we changed the signal mask, we need to restore the original one.
1704         * In case we've got a signal while waiting, we do not restore the
1705         * signal mask yet, and we allow do_signal() to deliver the signal on
1706         * the way back to userspace, before the signal mask is restored.
1707         */
1708        if (sigmask) {
1709                if (err == -EINTR) {
1710                        memcpy(&current->saved_sigmask, &sigsaved,
1711                               sizeof(sigsaved));
1712                        set_restore_sigmask();
1713                } else
1714                        sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1715        }
1716
1717        return err;
1718}
1719
1720#endif /* CONFIG_EPOLL */
1721
1722#ifdef CONFIG_SIGNALFD
1723
1724asmlinkage long compat_sys_signalfd4(int ufd,
1725                                     const compat_sigset_t __user *sigmask,
1726                                     compat_size_t sigsetsize, int flags)
1727{
1728        compat_sigset_t ss32;
1729        sigset_t tmp;
1730        sigset_t __user *ksigmask;
1731
1732        if (sigsetsize != sizeof(compat_sigset_t))
1733                return -EINVAL;
1734        if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1735                return -EFAULT;
1736        sigset_from_compat(&tmp, &ss32);
1737        ksigmask = compat_alloc_user_space(sizeof(sigset_t));
1738        if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
1739                return -EFAULT;
1740
1741        return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
1742}
1743
1744asmlinkage long compat_sys_signalfd(int ufd,
1745                                    const compat_sigset_t __user *sigmask,
1746                                    compat_size_t sigsetsize)
1747{
1748        return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
1749}
1750#endif /* CONFIG_SIGNALFD */
1751
1752#ifdef CONFIG_TIMERFD
1753
1754asmlinkage long compat_sys_timerfd_settime(int ufd, int flags,
1755                                   const struct compat_itimerspec __user *utmr,
1756                                   struct compat_itimerspec __user *otmr)
1757{
1758        int error;
1759        struct itimerspec t;
1760        struct itimerspec __user *ut;
1761
1762        if (get_compat_itimerspec(&t, utmr))
1763                return -EFAULT;
1764        ut = compat_alloc_user_space(2 * sizeof(struct itimerspec));
1765        if (copy_to_user(&ut[0], &t, sizeof(t)))
1766                return -EFAULT;
1767        error = sys_timerfd_settime(ufd, flags, &ut[0], &ut[1]);
1768        if (!error && otmr)
1769                error = (copy_from_user(&t, &ut[1], sizeof(struct itimerspec)) ||
1770                         put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1771
1772        return error;
1773}
1774
1775asmlinkage long compat_sys_timerfd_gettime(int ufd,
1776                                   struct compat_itimerspec __user *otmr)
1777{
1778        int error;
1779        struct itimerspec t;
1780        struct itimerspec __user *ut;
1781
1782        ut = compat_alloc_user_space(sizeof(struct itimerspec));
1783        error = sys_timerfd_gettime(ufd, ut);
1784        if (!error)
1785                error = (copy_from_user(&t, ut, sizeof(struct itimerspec)) ||
1786                         put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1787
1788        return error;
1789}
1790
1791#endif /* CONFIG_TIMERFD */
1792
1793#ifdef CONFIG_FHANDLE
1794/*
1795 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1796 * doesn't set the O_LARGEFILE flag.
1797 */
1798asmlinkage long
1799compat_sys_open_by_handle_at(int mountdirfd,
1800                             struct file_handle __user *handle, int flags)
1801{
1802        return do_handle_open(mountdirfd, handle, flags);
1803}
1804#endif
1805
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.