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        ret = -EFAULT;
 562        if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
 563                goto out;
 564
 565        /*
 566         * Single unix specification:
 567         * We should -EINVAL if an element length is not >= 0 and fitting an
 568         * ssize_t.
 569         *
 570         * In Linux, the total length is limited to MAX_RW_COUNT, there is
 571         * no overflow possibility.
 572         */
 573        tot_len = 0;
 574        ret = -EINVAL;
 575        for (seg = 0; seg < nr_segs; seg++) {
 576                compat_uptr_t buf;
 577                compat_ssize_t len;
 578
 579                if (__get_user(len, &uvector->iov_len) ||
 580                   __get_user(buf, &uvector->iov_base)) {
 581                        ret = -EFAULT;
 582                        goto out;
 583                }
 584                if (len < 0)    /* size_t not fitting in compat_ssize_t .. */
 585                        goto out;
 586                if (type >= 0 &&
 587                    !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
 588                        ret = -EFAULT;
 589                        goto out;
 590                }
 591                if (len > MAX_RW_COUNT - tot_len)
 592                        len = MAX_RW_COUNT - tot_len;
 593                tot_len += len;
 594                iov->iov_base = compat_ptr(buf);
 595                iov->iov_len = (compat_size_t) len;
 596                uvector++;
 597                iov++;
 598        }
 599        ret = tot_len;
 600
 601out:
 602        return ret;
 603}
 604
 605static inline long
 606copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
 607{
 608        compat_uptr_t uptr;
 609        int i;
 610
 611        for (i = 0; i < nr; ++i) {
 612                if (get_user(uptr, ptr32 + i))
 613                        return -EFAULT;
 614                if (put_user(compat_ptr(uptr), ptr64 + i))
 615                        return -EFAULT;
 616        }
 617        return 0;
 618}
 619
 620#define MAX_AIO_SUBMITS         (PAGE_SIZE/sizeof(struct iocb *))
 621
 622asmlinkage long
 623compat_sys_io_submit(aio_context_t ctx_id, int nr, u32 __user *iocb)
 624{
 625        struct iocb __user * __user *iocb64; 
 626        long ret;
 627
 628        if (unlikely(nr < 0))
 629                return -EINVAL;
 630
 631        if (nr > MAX_AIO_SUBMITS)
 632                nr = MAX_AIO_SUBMITS;
 633        
 634        iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
 635        ret = copy_iocb(nr, iocb, iocb64);
 636        if (!ret)
 637                ret = do_io_submit(ctx_id, nr, iocb64, 1);
 638        return ret;
 639}
 640
 641struct compat_ncp_mount_data {
 642        compat_int_t version;
 643        compat_uint_t ncp_fd;
 644        __compat_uid_t mounted_uid;
 645        compat_pid_t wdog_pid;
 646        unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
 647        compat_uint_t time_out;
 648        compat_uint_t retry_count;
 649        compat_uint_t flags;
 650        __compat_uid_t uid;
 651        __compat_gid_t gid;
 652        compat_mode_t file_mode;
 653        compat_mode_t dir_mode;
 654};
 655
 656struct compat_ncp_mount_data_v4 {
 657        compat_int_t version;
 658        compat_ulong_t flags;
 659        compat_ulong_t mounted_uid;
 660        compat_long_t wdog_pid;
 661        compat_uint_t ncp_fd;
 662        compat_uint_t time_out;
 663        compat_uint_t retry_count;
 664        compat_ulong_t uid;
 665        compat_ulong_t gid;
 666        compat_ulong_t file_mode;
 667        compat_ulong_t dir_mode;
 668};
 669
 670static void *do_ncp_super_data_conv(void *raw_data)
 671{
 672        int version = *(unsigned int *)raw_data;
 673
 674        if (version == 3) {
 675                struct compat_ncp_mount_data *c_n = raw_data;
 676                struct ncp_mount_data *n = raw_data;
 677
 678                n->dir_mode = c_n->dir_mode;
 679                n->file_mode = c_n->file_mode;
 680                n->gid = c_n->gid;
 681                n->uid = c_n->uid;
 682                memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
 683                n->wdog_pid = c_n->wdog_pid;
 684                n->mounted_uid = c_n->mounted_uid;
 685        } else if (version == 4) {
 686                struct compat_ncp_mount_data_v4 *c_n = raw_data;
 687                struct ncp_mount_data_v4 *n = raw_data;
 688
 689                n->dir_mode = c_n->dir_mode;
 690                n->file_mode = c_n->file_mode;
 691                n->gid = c_n->gid;
 692                n->uid = c_n->uid;
 693                n->retry_count = c_n->retry_count;
 694                n->time_out = c_n->time_out;
 695                n->ncp_fd = c_n->ncp_fd;
 696                n->wdog_pid = c_n->wdog_pid;
 697                n->mounted_uid = c_n->mounted_uid;
 698                n->flags = c_n->flags;
 699        } else if (version != 5) {
 700                return NULL;
 701        }
 702
 703        return raw_data;
 704}
 705
 706
 707struct compat_nfs_string {
 708        compat_uint_t len;
 709        compat_uptr_t data;
 710};
 711
 712static inline void compat_nfs_string(struct nfs_string *dst,
 713                                     struct compat_nfs_string *src)
 714{
 715        dst->data = compat_ptr(src->data);
 716        dst->len = src->len;
 717}
 718
 719struct compat_nfs4_mount_data_v1 {
 720        compat_int_t version;
 721        compat_int_t flags;
 722        compat_int_t rsize;
 723        compat_int_t wsize;
 724        compat_int_t timeo;
 725        compat_int_t retrans;
 726        compat_int_t acregmin;
 727        compat_int_t acregmax;
 728        compat_int_t acdirmin;
 729        compat_int_t acdirmax;
 730        struct compat_nfs_string client_addr;
 731        struct compat_nfs_string mnt_path;
 732        struct compat_nfs_string hostname;
 733        compat_uint_t host_addrlen;
 734        compat_uptr_t host_addr;
 735        compat_int_t proto;
 736        compat_int_t auth_flavourlen;
 737        compat_uptr_t auth_flavours;
 738};
 739
 740static int do_nfs4_super_data_conv(void *raw_data)
 741{
 742        int version = *(compat_uint_t *) raw_data;
 743
 744        if (version == 1) {
 745                struct compat_nfs4_mount_data_v1 *raw = raw_data;
 746                struct nfs4_mount_data *real = raw_data;
 747
 748                /* copy the fields backwards */
 749                real->auth_flavours = compat_ptr(raw->auth_flavours);
 750                real->auth_flavourlen = raw->auth_flavourlen;
 751                real->proto = raw->proto;
 752                real->host_addr = compat_ptr(raw->host_addr);
 753                real->host_addrlen = raw->host_addrlen;
 754                compat_nfs_string(&real->hostname, &raw->hostname);
 755                compat_nfs_string(&real->mnt_path, &raw->mnt_path);
 756                compat_nfs_string(&real->client_addr, &raw->client_addr);
 757                real->acdirmax = raw->acdirmax;
 758                real->acdirmin = raw->acdirmin;
 759                real->acregmax = raw->acregmax;
 760                real->acregmin = raw->acregmin;
 761                real->retrans = raw->retrans;
 762                real->timeo = raw->timeo;
 763                real->wsize = raw->wsize;
 764                real->rsize = raw->rsize;
 765                real->flags = raw->flags;
 766                real->version = raw->version;
 767        }
 768
 769        return 0;
 770}
 771
 772#define NCPFS_NAME      "ncpfs"
 773#define NFS4_NAME       "nfs4"
 774
 775asmlinkage long compat_sys_mount(const char __user * dev_name,
 776                                 const char __user * dir_name,
 777                                 const char __user * type, unsigned long flags,
 778                                 const void __user * data)
 779{
 780        char *kernel_type;
 781        unsigned long data_page;
 782        char *kernel_dev;
 783        struct filename *dir;
 784        int retval;
 785
 786        retval = copy_mount_string(type, &kernel_type);
 787        if (retval < 0)
 788                goto out;
 789
 790        dir = getname(dir_name);
 791        retval = PTR_ERR(dir);
 792        if (IS_ERR(dir))
 793                goto out1;
 794
 795        retval = copy_mount_string(dev_name, &kernel_dev);
 796        if (retval < 0)
 797                goto out2;
 798
 799        retval = copy_mount_options(data, &data_page);
 800        if (retval < 0)
 801                goto out3;
 802
 803        retval = -EINVAL;
 804
 805        if (kernel_type && data_page) {
 806                if (!strcmp(kernel_type, NCPFS_NAME)) {
 807                        do_ncp_super_data_conv((void *)data_page);
 808                } else if (!strcmp(kernel_type, NFS4_NAME)) {
 809                        if (do_nfs4_super_data_conv((void *) data_page))
 810                                goto out4;
 811                }
 812        }
 813
 814        retval = do_mount(kernel_dev, dir->name, kernel_type,
 815                        flags, (void*)data_page);
 816
 817 out4:
 818        free_page(data_page);
 819 out3:
 820        kfree(kernel_dev);
 821 out2:
 822        putname(dir);
 823 out1:
 824        kfree(kernel_type);
 825 out:
 826        return retval;
 827}
 828
 829struct compat_old_linux_dirent {
 830        compat_ulong_t  d_ino;
 831        compat_ulong_t  d_offset;
 832        unsigned short  d_namlen;
 833        char            d_name[1];
 834};
 835
 836struct compat_readdir_callback {
 837        struct compat_old_linux_dirent __user *dirent;
 838        int result;
 839};
 840
 841static int compat_fillonedir(void *__buf, const char *name, int namlen,
 842                        loff_t offset, u64 ino, unsigned int d_type)
 843{
 844        struct compat_readdir_callback *buf = __buf;
 845        struct compat_old_linux_dirent __user *dirent;
 846        compat_ulong_t d_ino;
 847
 848        if (buf->result)
 849                return -EINVAL;
 850        d_ino = ino;
 851        if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
 852                buf->result = -EOVERFLOW;
 853                return -EOVERFLOW;
 854        }
 855        buf->result++;
 856        dirent = buf->dirent;
 857        if (!access_ok(VERIFY_WRITE, dirent,
 858                        (unsigned long)(dirent->d_name + namlen + 1) -
 859                                (unsigned long)dirent))
 860                goto efault;
 861        if (    __put_user(d_ino, &dirent->d_ino) ||
 862                __put_user(offset, &dirent->d_offset) ||
 863                __put_user(namlen, &dirent->d_namlen) ||
 864                __copy_to_user(dirent->d_name, name, namlen) ||
 865                __put_user(0, dirent->d_name + namlen))
 866                goto efault;
 867        return 0;
 868efault:
 869        buf->result = -EFAULT;
 870        return -EFAULT;
 871}
 872
 873asmlinkage long compat_sys_old_readdir(unsigned int fd,
 874        struct compat_old_linux_dirent __user *dirent, unsigned int count)
 875{
 876        int error;
 877        struct fd f = fdget(fd);
 878        struct compat_readdir_callback buf;
 879
 880        if (!f.file)
 881                return -EBADF;
 882
 883        buf.result = 0;
 884        buf.dirent = dirent;
 885
 886        error = vfs_readdir(f.file, compat_fillonedir, &buf);
 887        if (buf.result)
 888                error = buf.result;
 889
 890        fdput(f);
 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 fd f;
 955        struct compat_linux_dirent __user * lastdirent;
 956        struct compat_getdents_callback buf;
 957        int error;
 958
 959        if (!access_ok(VERIFY_WRITE, dirent, count))
 960                return -EFAULT;
 961
 962        f = fdget(fd);
 963        if (!f.file)
 964                return -EBADF;
 965
 966        buf.current_dir = dirent;
 967        buf.previous = NULL;
 968        buf.count = count;
 969        buf.error = 0;
 970
 971        error = vfs_readdir(f.file, compat_filldir, &buf);
 972        if (error >= 0)
 973                error = buf.error;
 974        lastdirent = buf.previous;
 975        if (lastdirent) {
 976                if (put_user(f.file->f_pos, &lastdirent->d_off))
 977                        error = -EFAULT;
 978                else
 979                        error = count - buf.count;
 980        }
 981        fdput(f);
 982        return error;
 983}
 984
 985#ifndef __ARCH_OMIT_COMPAT_SYS_GETDENTS64
 986
 987struct compat_getdents_callback64 {
 988        struct linux_dirent64 __user *current_dir;
 989        struct linux_dirent64 __user *previous;
 990        int count;
 991        int error;
 992};
 993
 994static int compat_filldir64(void * __buf, const char * name, int namlen, loff_t offset,
 995                     u64 ino, unsigned int d_type)
 996{
 997        struct linux_dirent64 __user *dirent;
 998        struct compat_getdents_callback64 *buf = __buf;
 999        int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
1000                sizeof(u64));
1001        u64 off;
1002
1003        buf->error = -EINVAL;   /* only used if we fail.. */
1004        if (reclen > buf->count)
1005                return -EINVAL;
1006        dirent = buf->previous;
1007
1008        if (dirent) {
1009                if (__put_user_unaligned(offset, &dirent->d_off))
1010                        goto efault;
1011        }
1012        dirent = buf->current_dir;
1013        if (__put_user_unaligned(ino, &dirent->d_ino))
1014                goto efault;
1015        off = 0;
1016        if (__put_user_unaligned(off, &dirent->d_off))
1017                goto efault;
1018        if (__put_user(reclen, &dirent->d_reclen))
1019                goto efault;
1020        if (__put_user(d_type, &dirent->d_type))
1021                goto efault;
1022        if (copy_to_user(dirent->d_name, name, namlen))
1023                goto efault;
1024        if (__put_user(0, dirent->d_name + namlen))
1025                goto efault;
1026        buf->previous = dirent;
1027        dirent = (void __user *)dirent + reclen;
1028        buf->current_dir = dirent;
1029        buf->count -= reclen;
1030        return 0;
1031efault:
1032        buf->error = -EFAULT;
1033        return -EFAULT;
1034}
1035
1036asmlinkage long compat_sys_getdents64(unsigned int fd,
1037                struct linux_dirent64 __user * dirent, unsigned int count)
1038{
1039        struct fd f;
1040        struct linux_dirent64 __user * lastdirent;
1041        struct compat_getdents_callback64 buf;
1042        int error;
1043
1044        if (!access_ok(VERIFY_WRITE, dirent, count))
1045                return -EFAULT;
1046
1047        f = fdget(fd);
1048        if (!f.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(f.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 = f.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        fdput(f);
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 = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1088                                               UIO_FASTIOV, iovstack, &iov);
1089        if (ret <= 0)
1090                goto out;
1091
1092        tot_len = ret;
1093        ret = rw_verify_area(type, file, pos, tot_len);
1094        if (ret < 0)
1095                goto out;
1096
1097        fnv = NULL;
1098        if (type == READ) {
1099                fn = file->f_op->read;
1100                fnv = file->f_op->aio_read;
1101        } else {
1102                fn = (io_fn_t)file->f_op->write;
1103                fnv = file->f_op->aio_write;
1104        }
1105
1106        if (fnv)
1107                ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1108                                                pos, fnv);
1109        else
1110                ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1111
1112out:
1113        if (iov != iovstack)
1114                kfree(iov);
1115        if ((ret + (type == READ)) > 0) {
1116                if (type == READ)
1117                        fsnotify_access(file);
1118                else
1119                        fsnotify_modify(file);
1120        }
1121        return ret;
1122}
1123
1124static size_t compat_readv(struct file *file,
1125                           const struct compat_iovec __user *vec,
1126                           unsigned long vlen, loff_t *pos)
1127{
1128        ssize_t ret = -EBADF;
1129
1130        if (!(file->f_mode & FMODE_READ))
1131                goto out;
1132
1133        ret = -EINVAL;
1134        if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
1135                goto out;
1136
1137        ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1138
1139out:
1140        if (ret > 0)
1141                add_rchar(current, ret);
1142        inc_syscr(current);
1143        return ret;
1144}
1145
1146asmlinkage ssize_t
1147compat_sys_readv(unsigned long fd, const struct compat_iovec __user *vec,
1148                 unsigned long vlen)
1149{
1150        struct fd f = fdget(fd);
1151        ssize_t ret;
1152        loff_t pos;
1153
1154        if (!f.file)
1155                return -EBADF;
1156        pos = f.file->f_pos;
1157        ret = compat_readv(f.file, vec, vlen, &pos);
1158        f.file->f_pos = pos;
1159        fdput(f);
1160        return ret;
1161}
1162
1163asmlinkage ssize_t
1164compat_sys_preadv64(unsigned long fd, const struct compat_iovec __user *vec,
1165                    unsigned long vlen, loff_t pos)
1166{
1167        struct fd f;
1168        ssize_t ret;
1169
1170        if (pos < 0)
1171                return -EINVAL;
1172        f = fdget(fd);
1173        if (!f.file)
1174                return -EBADF;
1175        ret = -ESPIPE;
1176        if (f.file->f_mode & FMODE_PREAD)
1177                ret = compat_readv(f.file, vec, vlen, &pos);
1178        fdput(f);
1179        return ret;
1180}
1181
1182asmlinkage ssize_t
1183compat_sys_preadv(unsigned long fd, const struct compat_iovec __user *vec,
1184                  unsigned long vlen, u32 pos_low, u32 pos_high)
1185{
1186        loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1187        return compat_sys_preadv64(fd, vec, vlen, pos);
1188}
1189
1190static size_t compat_writev(struct file *file,
1191                            const struct compat_iovec __user *vec,
1192                            unsigned long vlen, loff_t *pos)
1193{
1194        ssize_t ret = -EBADF;
1195
1196        if (!(file->f_mode & FMODE_WRITE))
1197                goto out;
1198
1199        ret = -EINVAL;
1200        if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
1201                goto out;
1202
1203        ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1204
1205out:
1206        if (ret > 0)
1207                add_wchar(current, ret);
1208        inc_syscw(current);
1209        return ret;
1210}
1211
1212asmlinkage ssize_t
1213compat_sys_writev(unsigned long fd, const struct compat_iovec __user *vec,
1214                  unsigned long vlen)
1215{
1216        struct fd f = fdget(fd);
1217        ssize_t ret;
1218        loff_t pos;
1219
1220        if (!f.file)
1221                return -EBADF;
1222        pos = f.file->f_pos;
1223        ret = compat_writev(f.file, vec, vlen, &pos);
1224        f.file->f_pos = pos;
1225        fdput(f);
1226        return ret;
1227}
1228
1229asmlinkage ssize_t
1230compat_sys_pwritev64(unsigned long fd, const struct compat_iovec __user *vec,
1231                     unsigned long vlen, loff_t pos)
1232{
1233        struct fd f;
1234        ssize_t ret;
1235
1236        if (pos < 0)
1237                return -EINVAL;
1238        f = fdget(fd);
1239        if (!f.file)
1240                return -EBADF;
1241        ret = -ESPIPE;
1242        if (f.file->f_mode & FMODE_PWRITE)
1243                ret = compat_writev(f.file, vec, vlen, &pos);
1244        fdput(f);
1245        return ret;
1246}
1247
1248asmlinkage ssize_t
1249compat_sys_pwritev(unsigned long fd, const struct compat_iovec __user *vec,
1250                   unsigned long vlen, u32 pos_low, u32 pos_high)
1251{
1252        loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1253        return compat_sys_pwritev64(fd, vec, vlen, pos);
1254}
1255
1256asmlinkage long
1257compat_sys_vmsplice(int fd, const struct compat_iovec __user *iov32,
1258                    unsigned int nr_segs, unsigned int flags)
1259{
1260        unsigned i;
1261        struct iovec __user *iov;
1262        if (nr_segs > UIO_MAXIOV)
1263                return -EINVAL;
1264        iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1265        for (i = 0; i < nr_segs; i++) {
1266                struct compat_iovec v;
1267                if (get_user(v.iov_base, &iov32[i].iov_base) ||
1268                    get_user(v.iov_len, &iov32[i].iov_len) ||
1269                    put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1270                    put_user(v.iov_len, &iov[i].iov_len))
1271                        return -EFAULT;
1272        }
1273        return sys_vmsplice(fd, iov, nr_segs, flags);
1274}
1275
1276/*
1277 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1278 * O_LARGEFILE flag.
1279 */
1280COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1281{
1282        return do_sys_open(AT_FDCWD, filename, flags, mode);
1283}
1284
1285/*
1286 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1287 * O_LARGEFILE flag.
1288 */
1289COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1290{
1291        return do_sys_open(dfd, filename, flags, mode);
1292}
1293
1294#define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1295
1296static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1297                                      int timeval, int ret)
1298{
1299        struct timespec ts;
1300
1301        if (!p)
1302                return ret;
1303
1304        if (current->personality & STICKY_TIMEOUTS)
1305                goto sticky;
1306
1307        /* No update for zero timeout */
1308        if (!end_time->tv_sec && !end_time->tv_nsec)
1309                return ret;
1310
1311        ktime_get_ts(&ts);
1312        ts = timespec_sub(*end_time, ts);
1313        if (ts.tv_sec < 0)
1314                ts.tv_sec = ts.tv_nsec = 0;
1315
1316        if (timeval) {
1317                struct compat_timeval rtv;
1318
1319                rtv.tv_sec = ts.tv_sec;
1320                rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1321
1322                if (!copy_to_user(p, &rtv, sizeof(rtv)))
1323                        return ret;
1324        } else {
1325                struct compat_timespec rts;
1326
1327                rts.tv_sec = ts.tv_sec;
1328                rts.tv_nsec = ts.tv_nsec;
1329
1330                if (!copy_to_user(p, &rts, sizeof(rts)))
1331                        return ret;
1332        }
1333        /*
1334         * If an application puts its timeval in read-only memory, we
1335         * don't want the Linux-specific update to the timeval to
1336         * cause a fault after the select has completed
1337         * successfully. However, because we're not updating the
1338         * timeval, we can't restart the system call.
1339         */
1340
1341sticky:
1342        if (ret == -ERESTARTNOHAND)
1343                ret = -EINTR;
1344        return ret;
1345}
1346
1347/*
1348 * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1349 * 64-bit unsigned longs.
1350 */
1351static
1352int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1353                        unsigned long *fdset)
1354{
1355        nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1356        if (ufdset) {
1357                unsigned long odd;
1358
1359                if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1360                        return -EFAULT;
1361
1362                odd = nr & 1UL;
1363                nr &= ~1UL;
1364                while (nr) {
1365                        unsigned long h, l;
1366                        if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1367                                return -EFAULT;
1368                        ufdset += 2;
1369                        *fdset++ = h << 32 | l;
1370                        nr -= 2;
1371                }
1372                if (odd && __get_user(*fdset, ufdset))
1373                        return -EFAULT;
1374        } else {
1375                /* Tricky, must clear full unsigned long in the
1376                 * kernel fdset at the end, this makes sure that
1377                 * actually happens.
1378                 */
1379                memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1380        }
1381        return 0;
1382}
1383
1384static
1385int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1386                      unsigned long *fdset)
1387{
1388        unsigned long odd;
1389        nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1390
1391        if (!ufdset)
1392                return 0;
1393
1394        odd = nr & 1UL;
1395        nr &= ~1UL;
1396        while (nr) {
1397                unsigned long h, l;
1398                l = *fdset++;
1399                h = l >> 32;
1400                if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1401                        return -EFAULT;
1402                ufdset += 2;
1403                nr -= 2;
1404        }
1405        if (odd && __put_user(*fdset, ufdset))
1406                return -EFAULT;
1407        return 0;
1408}
1409
1410
1411/*
1412 * This is a virtual copy of sys_select from fs/select.c and probably
1413 * should be compared to it from time to time
1414 */
1415
1416/*
1417 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1418 * like to be certain this leads to no problems. So I return
1419 * EINTR just for safety.
1420 *
1421 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1422 * I'm trying ERESTARTNOHAND which restart only when you want to.
1423 */
1424int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1425        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1426        struct timespec *end_time)
1427{
1428        fd_set_bits fds;
1429        void *bits;
1430        int size, max_fds, ret = -EINVAL;
1431        struct fdtable *fdt;
1432        long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1433
1434        if (n < 0)
1435                goto out_nofds;
1436
1437        /* max_fds can increase, so grab it once to avoid race */
1438        rcu_read_lock();
1439        fdt = files_fdtable(current->files);
1440        max_fds = fdt->max_fds;
1441        rcu_read_unlock();
1442        if (n > max_fds)
1443                n = max_fds;
1444
1445        /*
1446         * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1447         * since we used fdset we need to allocate memory in units of
1448         * long-words.
1449         */
1450        size = FDS_BYTES(n);
1451        bits = stack_fds;
1452        if (size > sizeof(stack_fds) / 6) {
1453                bits = kmalloc(6 * size, GFP_KERNEL);
1454                ret = -ENOMEM;
1455                if (!bits)
1456                        goto out_nofds;
1457        }
1458        fds.in      = (unsigned long *)  bits;
1459        fds.out     = (unsigned long *) (bits +   size);
1460        fds.ex      = (unsigned long *) (bits + 2*size);
1461        fds.res_in  = (unsigned long *) (bits + 3*size);
1462        fds.res_out = (unsigned long *) (bits + 4*size);
1463        fds.res_ex  = (unsigned long *) (bits + 5*size);
1464
1465        if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1466            (ret = compat_get_fd_set(n, outp, fds.out)) ||
1467            (ret = compat_get_fd_set(n, exp, fds.ex)))
1468                goto out;
1469        zero_fd_set(n, fds.res_in);
1470        zero_fd_set(n, fds.res_out);
1471        zero_fd_set(n, fds.res_ex);
1472
1473        ret = do_select(n, &fds, end_time);
1474
1475        if (ret < 0)
1476                goto out;
1477        if (!ret) {
1478                ret = -ERESTARTNOHAND;
1479                if (signal_pending(current))
1480                        goto out;
1481                ret = 0;
1482        }
1483
1484        if (compat_set_fd_set(n, inp, fds.res_in) ||
1485            compat_set_fd_set(n, outp, fds.res_out) ||
1486            compat_set_fd_set(n, exp, fds.res_ex))
1487                ret = -EFAULT;
1488out:
1489        if (bits != stack_fds)
1490                kfree(bits);
1491out_nofds:
1492        return ret;
1493}
1494
1495asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1496        compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1497        struct compat_timeval __user *tvp)
1498{
1499        struct timespec end_time, *to = NULL;
1500        struct compat_timeval tv;
1501        int ret;
1502
1503        if (tvp) {
1504                if (copy_from_user(&tv, tvp, sizeof(tv)))
1505                        return -EFAULT;
1506
1507                to = &end_time;
1508                if (poll_select_set_timeout(to,
1509                                tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1510                                (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1511                        return -EINVAL;
1512        }
1513
1514        ret = compat_core_sys_select(n, inp, outp, exp, to);
1515        ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1516
1517        return ret;
1518}
1519
1520struct compat_sel_arg_struct {
1521        compat_ulong_t n;
1522        compat_uptr_t inp;
1523        compat_uptr_t outp;
1524        compat_uptr_t exp;
1525        compat_uptr_t tvp;
1526};
1527
1528asmlinkage long compat_sys_old_select(struct compat_sel_arg_struct __user *arg)
1529{
1530        struct compat_sel_arg_struct a;
1531
1532        if (copy_from_user(&a, arg, sizeof(a)))
1533                return -EFAULT;
1534        return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1535                                 compat_ptr(a.exp), compat_ptr(a.tvp));
1536}
1537
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
1661#ifdef CONFIG_EPOLL
1662
1663asmlinkage long compat_sys_epoll_pwait(int epfd,
1664                        struct compat_epoll_event __user *events,
1665                        int maxevents, int timeout,
1666                        const compat_sigset_t __user *sigmask,
1667                        compat_size_t sigsetsize)
1668{
1669        long err;
1670        compat_sigset_t csigmask;
1671        sigset_t ksigmask, sigsaved;
1672
1673        /*
1674         * If the caller wants a certain signal mask to be set during the wait,
1675         * we apply it here.
1676         */
1677        if (sigmask) {
1678                if (sigsetsize != sizeof(compat_sigset_t))
1679                        return -EINVAL;
1680                if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
1681                        return -EFAULT;
1682                sigset_from_compat(&ksigmask, &csigmask);
1683                sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1684                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1685        }
1686
1687        err = sys_epoll_wait(epfd, events, maxevents, timeout);
1688
1689        /*
1690         * If we changed the signal mask, we need to restore the original one.
1691         * In case we've got a signal while waiting, we do not restore the
1692         * signal mask yet, and we allow do_signal() to deliver the signal on
1693         * the way back to userspace, before the signal mask is restored.
1694         */
1695        if (sigmask) {
1696                if (err == -EINTR) {
1697                        memcpy(&current->saved_sigmask, &sigsaved,
1698                               sizeof(sigsaved));
1699                        set_restore_sigmask();
1700                } else
1701                        sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1702        }
1703
1704        return err;
1705}
1706
1707#endif /* CONFIG_EPOLL */
1708
1709#ifdef CONFIG_SIGNALFD
1710
1711asmlinkage long compat_sys_signalfd4(int ufd,
1712                                     const compat_sigset_t __user *sigmask,
1713                                     compat_size_t sigsetsize, int flags)
1714{
1715        compat_sigset_t ss32;
1716        sigset_t tmp;
1717        sigset_t __user *ksigmask;
1718
1719        if (sigsetsize != sizeof(compat_sigset_t))
1720                return -EINVAL;
1721        if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1722                return -EFAULT;
1723        sigset_from_compat(&tmp, &ss32);
1724        ksigmask = compat_alloc_user_space(sizeof(sigset_t));
1725        if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
1726                return -EFAULT;
1727
1728        return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
1729}
1730
1731asmlinkage long compat_sys_signalfd(int ufd,
1732                                    const compat_sigset_t __user *sigmask,
1733                                    compat_size_t sigsetsize)
1734{
1735        return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
1736}
1737#endif /* CONFIG_SIGNALFD */
1738
1739#ifdef CONFIG_FHANDLE
1740/*
1741 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1742 * doesn't set the O_LARGEFILE flag.
1743 */
1744COMPAT_SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
1745                             struct file_handle __user *, handle, int, flags)
1746{
1747        return do_handle_open(mountdirfd, handle, flags);
1748}
1749#endif
1750
1751#ifdef __ARCH_WANT_COMPAT_SYS_SENDFILE
1752asmlinkage long compat_sys_sendfile(int out_fd, int in_fd,
1753                                    compat_off_t __user *offset, compat_size_t count)
1754{
1755        loff_t pos;
1756        off_t off;
1757        ssize_t ret;
1758
1759        if (offset) {
1760                if (unlikely(get_user(off, offset)))
1761                        return -EFAULT;
1762                pos = off;
1763                ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1764                if (unlikely(put_user(pos, offset)))
1765                        return -EFAULT;
1766                return ret;
1767        }
1768
1769        return do_sendfile(out_fd, in_fd, NULL, count, 0);
1770}
1771#endif /* __ARCH_WANT_COMPAT_SYS_SENDFILE */
1772
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.