fs/read_write.c

   1 /*
   2  *  linux/fs/read_write.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/slab.h>
   8 #include <linux/stat.h>
   9 #include <linux/fcntl.h>
  10 #include <linux/file.h>
  11 #include <linux/uio.h>
  12 #include <linux/smp_lock.h>
  13 #include <linux/fsnotify.h>
  14 #include <linux/security.h>
  15 #include <linux/module.h>
  16 #include <linux/syscalls.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/splice.h>
  19 #include "read_write.h"
  20
  21 #include <asm/uaccess.h>
  22 #include <asm/unistd.h>
  23
  24 const struct file_operations generic_ro_fops = {
  25         .llseek         = generic_file_llseek,
  26         .read           = do_sync_read,
  27         .aio_read       = generic_file_aio_read,
  28         .mmap           = generic_file_readonly_mmap,
  29         .splice_read    = generic_file_splice_read,
  30 };
  31
  32 EXPORT_SYMBOL(generic_ro_fops);
  33
  34 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
  35 {
  36         long long retval;
  37         struct inode *inode = file->f_mapping->host;
  38
  39         mutex_lock(&inode->i_mutex);
  40         switch (origin) {
  41                 case SEEK_END:
  42                         offset += inode->i_size;
  43                         break;
  44                 case SEEK_CUR:
  45                         offset += file->f_pos;
  46         }
  47         retval = -EINVAL;
  48         if (offset>=0 && offset<=inode->i_sb->s_maxbytes) {
  49                 if (offset != file->f_pos) {
  50                         file->f_pos = offset;
  51                         file->f_version = 0;
  52                 }
  53                 retval = offset;
  54         }
  55         mutex_unlock(&inode->i_mutex);
  56         return retval;
  57 }
  58
  59 EXPORT_SYMBOL(generic_file_llseek);
  60
  61 loff_t remote_llseek(struct file *file, loff_t offset, int origin)
  62 {
  63         long long retval;
  64
  65         lock_kernel();
  66         switch (origin) {
  67                 case SEEK_END:
  68                         offset += i_size_read(file->f_path.dentry->d_inode);
  69                         break;
  70                 case SEEK_CUR:
  71                         offset += file->f_pos;
  72         }
  73         retval = -EINVAL;
  74         if (offset>=0 && offset<=file->f_path.dentry->d_inode->i_sb->s_maxbytes) {
  75                 if (offset != file->f_pos) {
  76                         file->f_pos = offset;
  77                         file->f_version = 0;
  78                 }
  79                 retval = offset;
  80         }
  81         unlock_kernel();
  82         return retval;
  83 }
  84 EXPORT_SYMBOL(remote_llseek);
  85
  86 loff_t no_llseek(struct file *file, loff_t offset, int origin)
  87 {
  88         return -ESPIPE;
  89 }
  90 EXPORT_SYMBOL(no_llseek);
  91
  92 loff_t default_llseek(struct file *file, loff_t offset, int origin)
  93 {
  94         long long retval;
  95
  96         lock_kernel();
  97         switch (origin) {
  98                 case SEEK_END:
  99                         offset += i_size_read(file->f_path.dentry->d_inode);
 100                         break;
 101                 case SEEK_CUR:
 102                         offset += file->f_pos;
 103         }
 104         retval = -EINVAL;
 105         if (offset >= 0) {
 106                 if (offset != file->f_pos) {
 107                         file->f_pos = offset;
 108                         file->f_version = 0;
 109                 }
 110                 retval = offset;
 111         }
 112         unlock_kernel();
 113         return retval;
 114 }
 115 EXPORT_SYMBOL(default_llseek);
 116
 117 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
 118 {
 119         loff_t (*fn)(struct file *, loff_t, int);
 120
 121         fn = no_llseek;
 122         if (file->f_mode & FMODE_LSEEK) {
 123                 fn = default_llseek;
 124                 if (file->f_op && file->f_op->llseek)
 125                         fn = file->f_op->llseek;
 126         }
 127         return fn(file, offset, origin);
 128 }
 129 EXPORT_SYMBOL(vfs_llseek);
 130
 131 asmlinkage off_t sys_lseek(unsigned int fd, off_t offset, unsigned int origin)
 132 {
 133         off_t retval;
 134         struct file * file;
 135         int fput_needed;
 136
 137         retval = -EBADF;
 138         file = fget_light(fd, &fput_needed);
 139         if (!file)
 140                 goto bad;
 141
 142         retval = -EINVAL;
 143         if (origin <= SEEK_MAX) {
 144                 loff_t res = vfs_llseek(file, offset, origin);
 145                 retval = res;
 146                 if (res != (loff_t)retval)
 147                         retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
 148         }
 149         fput_light(file, fput_needed);
 150 bad:
 151         return retval;
 152 }
 153
 154 #ifdef __ARCH_WANT_SYS_LLSEEK
 155 asmlinkage long sys_llseek(unsigned int fd, unsigned long offset_high,
 156                            unsigned long offset_low, loff_t __user * result,
 157                            unsigned int origin)
 158 {
 159         int retval;
 160         struct file * file;
 161         loff_t offset;
 162         int fput_needed;
 163
 164         retval = -EBADF;
 165         file = fget_light(fd, &fput_needed);
 166         if (!file)
 167                 goto bad;
 168
 169         retval = -EINVAL;
 170         if (origin > SEEK_MAX)
 171                 goto out_putf;
 172
 173         offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
 174                         origin);
 175
 176         retval = (int)offset;
 177         if (offset >= 0) {
 178                 retval = -EFAULT;
 179                 if (!copy_to_user(result, &offset, sizeof(offset)))
 180                         retval = 0;
 181         }
 182 out_putf:
 183         fput_light(file, fput_needed);
 184 bad:
 185         return retval;
 186 }
 187 #endif
 188
 189 /*
 190  * rw_verify_area doesn't like huge counts. We limit
 191  * them to something that fits in "int" so that others
 192  * won't have to do range checks all the time.
 193  */
 194 #define MAX_RW_COUNT (INT_MAX & PAGE_CACHE_MASK)
 195
 196 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
 197 {
 198         struct inode *inode;
 199         loff_t pos;
 200         int retval = -EINVAL;
 201
 202         inode = file->f_path.dentry->d_inode;
 203         if (unlikely((ssize_t) count < 0))
 204                 return retval;
 205         pos = *ppos;
 206         if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))
 207                 return retval;
 208
 209         if (unlikely(inode->i_flock && mandatory_lock(inode))) {
 210                 retval = locks_mandatory_area(
 211                         read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
 212                         inode, file, pos, count);
 213                 if (retval < 0)
 214                         return retval;
 215         }
 216         retval = security_file_permission(file,
 217                                 read_write == READ ? MAY_READ : MAY_WRITE);
 218         if (retval)
 219                 return retval;
 220         return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
 221 }
 222
 223 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
 224 {
 225         set_current_state(TASK_UNINTERRUPTIBLE);
 226         if (!kiocbIsKicked(iocb))
 227                 schedule();
 228         else
 229                 kiocbClearKicked(iocb);
 230         __set_current_state(TASK_RUNNING);
 231 }
 232
 233 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 234 {
 235         struct iovec iov = { .iov_base = buf, .iov_len = len };
 236         struct kiocb kiocb;
 237         ssize_t ret;
 238
 239         init_sync_kiocb(&kiocb, filp);
 240         kiocb.ki_pos = *ppos;
 241         kiocb.ki_left = len;
 242
 243         for (;;) {
 244                 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
 245                 if (ret != -EIOCBRETRY)
 246                         break;
 247                 wait_on_retry_sync_kiocb(&kiocb);
 248         }
 249
 250         if (-EIOCBQUEUED == ret)
 251                 ret = wait_on_sync_kiocb(&kiocb);
 252         *ppos = kiocb.ki_pos;
 253         return ret;
 254 }
 255
 256 EXPORT_SYMBOL(do_sync_read);
 257
 258 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 259 {
 260         ssize_t ret;
 261
 262         if (!(file->f_mode & FMODE_READ))
 263                 return -EBADF;
 264         if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
 265                 return -EINVAL;
 266         if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
 267                 return -EFAULT;
 268
 269         ret = rw_verify_area(READ, file, pos, count);
 270         if (ret >= 0) {
 271                 count = ret;
 272                 if (file->f_op->read)
 273                         ret = file->f_op->read(file, buf, count, pos);
 274                 else
 275                         ret = do_sync_read(file, buf, count, pos);
 276                 if (ret > 0) {
 277                         fsnotify_access(file->f_path.dentry);
 278                         add_rchar(current, ret);
 279                 }
 280                 inc_syscr(current);
 281         }
 282
 283         return ret;
 284 }
 285
 286 EXPORT_SYMBOL(vfs_read);
 287
 288 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 289 {
 290         struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
 291         struct kiocb kiocb;
 292         ssize_t ret;
 293
 294         init_sync_kiocb(&kiocb, filp);
 295         kiocb.ki_pos = *ppos;
 296         kiocb.ki_left = len;
 297
 298         for (;;) {
 299                 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
 300                 if (ret != -EIOCBRETRY)
 301                         break;
 302                 wait_on_retry_sync_kiocb(&kiocb);
 303         }
 304
 305         if (-EIOCBQUEUED == ret)
 306                 ret = wait_on_sync_kiocb(&kiocb);
 307         *ppos = kiocb.ki_pos;
 308         return ret;
 309 }
 310
 311 EXPORT_SYMBOL(do_sync_write);
 312
 313 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 314 {
 315         ssize_t ret;
 316
 317         if (!(file->f_mode & FMODE_WRITE))
 318                 return -EBADF;
 319         if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
 320                 return -EINVAL;
 321         if (unlikely(!access_ok(VERIFY_READ, buf, count)))
 322                 return -EFAULT;
 323
 324         ret = rw_verify_area(WRITE, file, pos, count);
 325         if (ret >= 0) {
 326                 count = ret;
 327                 if (file->f_op->write)
 328                         ret = file->f_op->write(file, buf, count, pos);
 329                 else
 330                         ret = do_sync_write(file, buf, count, pos);
 331                 if (ret > 0) {
 332                         fsnotify_modify(file->f_path.dentry);
 333                         add_wchar(current, ret);
 334                 }
 335                 inc_syscw(current);
 336         }
 337
 338         return ret;
 339 }
 340
 341 EXPORT_SYMBOL(vfs_write);
 342
 343 static inline loff_t file_pos_read(struct file *file)
 344 {
 345         return file->f_pos;
 346 }
 347
 348 static inline void file_pos_write(struct file *file, loff_t pos)
 349 {
 350         file->f_pos = pos;
 351 }
 352
 353 asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count)
 354 {
 355         struct file *file;
 356         ssize_t ret = -EBADF;
 357         int fput_needed;
 358
 359         file = fget_light(fd, &fput_needed);
 360         if (file) {
 361                 loff_t pos = file_pos_read(file);
 362                 ret = vfs_read(file, buf, count, &pos);
 363                 file_pos_write(file, pos);
 364                 fput_light(file, fput_needed);
 365         }
 366
 367         return ret;
 368 }
 369
 370 asmlinkage ssize_t sys_write(unsigned int fd, const char __user * buf, size_t count)
 371 {
 372         struct file *file;
 373         ssize_t ret = -EBADF;
 374         int fput_needed;
 375
 376         file = fget_light(fd, &fput_needed);
 377         if (file) {
 378                 loff_t pos = file_pos_read(file);
 379                 ret = vfs_write(file, buf, count, &pos);
 380                 file_pos_write(file, pos);
 381                 fput_light(file, fput_needed);
 382         }
 383
 384         return ret;
 385 }
 386
 387 asmlinkage ssize_t sys_pread64(unsigned int fd, char __user *buf,
 388                              size_t count, loff_t pos)
 389 {
 390         struct file *file;
 391         ssize_t ret = -EBADF;
 392         int fput_needed;
 393
 394         if (pos < 0)
 395                 return -EINVAL;
 396
 397         file = fget_light(fd, &fput_needed);
 398         if (file) {
 399                 ret = -ESPIPE;
 400                 if (file->f_mode & FMODE_PREAD)
 401                         ret = vfs_read(file, buf, count, &pos);
 402                 fput_light(file, fput_needed);
 403         }
 404
 405         return ret;
 406 }
 407
 408 asmlinkage ssize_t sys_pwrite64(unsigned int fd, const char __user *buf,
 409                               size_t count, loff_t pos)
 410 {
 411         struct file *file;
 412         ssize_t ret = -EBADF;
 413         int fput_needed;
 414
 415         if (pos < 0)
 416                 return -EINVAL;
 417
 418         file = fget_light(fd, &fput_needed);
 419         if (file) {
 420                 ret = -ESPIPE;
 421                 if (file->f_mode & FMODE_PWRITE)
 422                         ret = vfs_write(file, buf, count, &pos);
 423                 fput_light(file, fput_needed);
 424         }
 425
 426         return ret;
 427 }
 428
 429 /*
 430  * Reduce an iovec's length in-place.  Return the resulting number of segments
 431  */
 432 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
 433 {
 434         unsigned long seg = 0;
 435         size_t len = 0;
 436
 437         while (seg < nr_segs) {
 438                 seg++;
 439                 if (len + iov->iov_len >= to) {
 440                         iov->iov_len = to - len;
 441                         break;
 442                 }
 443                 len += iov->iov_len;
 444                 iov++;
 445         }
 446         return seg;
 447 }
 448 EXPORT_SYMBOL(iov_shorten);
 449
 450 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
 451                 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
 452 {
 453         struct kiocb kiocb;
 454         ssize_t ret;
 455
 456         init_sync_kiocb(&kiocb, filp);
 457         kiocb.ki_pos = *ppos;
 458         kiocb.ki_left = len;
 459         kiocb.ki_nbytes = len;
 460
 461         for (;;) {
 462                 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
 463                 if (ret != -EIOCBRETRY)
 464                         break;
 465                 wait_on_retry_sync_kiocb(&kiocb);
 466         }
 467
 468         if (ret == -EIOCBQUEUED)
 469                 ret = wait_on_sync_kiocb(&kiocb);
 470         *ppos = kiocb.ki_pos;
 471         return ret;
 472 }
 473
 474 /* Do it by hand, with file-ops */
 475 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
 476                 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
 477 {
 478         struct iovec *vector = iov;
 479         ssize_t ret = 0;
 480
 481         while (nr_segs > 0) {
 482                 void __user *base;
 483                 size_t len;
 484                 ssize_t nr;
 485
 486                 base = vector->iov_base;
 487                 len = vector->iov_len;
 488                 vector++;
 489                 nr_segs--;
 490
 491                 nr = fn(filp, base, len, ppos);
 492
 493                 if (nr < 0) {
 494                         if (!ret)
 495                                 ret = nr;
 496                         break;
 497                 }
 498                 ret += nr;
 499                 if (nr != len)
 500                         break;
 501         }
 502
 503         return ret;
 504 }
 505
 506 /* A write operation does a read from user space and vice versa */
 507 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 508
 509 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
 510                               unsigned long nr_segs, unsigned long fast_segs,
 511                               struct iovec *fast_pointer,
 512                               struct iovec **ret_pointer)
 513   {
 514         unsigned long seg;
 515         ssize_t ret;
 516         struct iovec *iov = fast_pointer;
 517
 518         /*
 519          * SuS says "The readv() function *may* fail if the iovcnt argument
 520          * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 521          * traditionally returned zero for zero segments, so...
 522          */
 523         if (nr_segs == 0) {
 524                 ret = 0;
 525                 goto out;
 526         }
 527
 528         /*
 529          * First get the "struct iovec" from user memory and
 530          * verify all the pointers
 531          */
 532         if (nr_segs > UIO_MAXIOV) {
 533                 ret = -EINVAL;
 534                 goto out;
 535         }
 536         if (nr_segs > fast_segs) {
 537                 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 538                 if (iov == NULL) {
 539                         ret = -ENOMEM;
 540                         goto out;
 541                 }
 542         }
 543         if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
 544                 ret = -EFAULT;
 545                 goto out;
 546         }
 547
 548         /*
 549          * According to the Single Unix Specification we should return EINVAL
 550          * if an element length is < 0 when cast to ssize_t or if the
 551          * total length would overflow the ssize_t return value of the
 552          * system call.
 553          */
 554         ret = 0;
 555         for (seg = 0; seg < nr_segs; seg++) {
 556                 void __user *buf = iov[seg].iov_base;
 557                 ssize_t len = (ssize_t)iov[seg].iov_len;
 558
 559                 /* see if we we're about to use an invalid len or if
 560                  * it's about to overflow ssize_t */
 561                 if (len < 0 || (ret + len < ret)) {
 562                         ret = -EINVAL;
 563                         goto out;
 564                 }
 565                 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
 566                         ret = -EFAULT;
 567                         goto out;
 568                 }
 569
 570                 ret += len;
 571         }
 572 out:
 573         *ret_pointer = iov;
 574         return ret;
 575 }
 576
 577 static ssize_t do_readv_writev(int type, struct file *file,
 578                                const struct iovec __user * uvector,
 579                                unsigned long nr_segs, loff_t *pos)
 580 {
 581         size_t tot_len;
 582         struct iovec iovstack[UIO_FASTIOV];
 583         struct iovec *iov = iovstack;
 584         ssize_t ret;
 585         io_fn_t fn;
 586         iov_fn_t fnv;
 587
 588         if (!file->f_op) {
 589                 ret = -EINVAL;
 590                 goto out;
 591         }
 592
 593         ret = rw_copy_check_uvector(type, uvector, nr_segs,
 594                         ARRAY_SIZE(iovstack), iovstack, &iov);
 595         if (ret <= 0)
 596                 goto out;
 597
 598         tot_len = ret;
 599         ret = rw_verify_area(type, file, pos, tot_len);
 600         if (ret < 0)
 601                 goto out;
 602
 603         fnv = NULL;
 604         if (type == READ) {
 605                 fn = file->f_op->read;
 606                 fnv = file->f_op->aio_read;
 607         } else {
 608                 fn = (io_fn_t)file->f_op->write;
 609                 fnv = file->f_op->aio_write;
 610         }
 611
 612         if (fnv)
 613                 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
 614                                                 pos, fnv);
 615         else
 616                 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
 617
 618 out:
 619         if (iov != iovstack)
 620                 kfree(iov);
 621         if ((ret + (type == READ)) > 0) {
 622                 if (type == READ)
 623                         fsnotify_access(file->f_path.dentry);
 624                 else
 625                         fsnotify_modify(file->f_path.dentry);
 626         }
 627         return ret;
 628 }
 629
 630 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
 631                   unsigned long vlen, loff_t *pos)
 632 {
 633         if (!(file->f_mode & FMODE_READ))
 634                 return -EBADF;
 635         if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
 636                 return -EINVAL;
 637
 638         return do_readv_writev(READ, file, vec, vlen, pos);
 639 }
 640
 641 EXPORT_SYMBOL(vfs_readv);
 642
 643 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 644                    unsigned long vlen, loff_t *pos)
 645 {
 646         if (!(file->f_mode & FMODE_WRITE))
 647                 return -EBADF;
 648         if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
 649                 return -EINVAL;
 650
 651         return do_readv_writev(WRITE, file, vec, vlen, pos);
 652 }
 653
 654 EXPORT_SYMBOL(vfs_writev);
 655
 656 asmlinkage ssize_t
 657 sys_readv(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
 658 {
 659         struct file *file;
 660         ssize_t ret = -EBADF;
 661         int fput_needed;
 662
 663         file = fget_light(fd, &fput_needed);
 664         if (file) {
 665                 loff_t pos = file_pos_read(file);
 666                 ret = vfs_readv(file, vec, vlen, &pos);
 667                 file_pos_write(file, pos);
 668                 fput_light(file, fput_needed);
 669         }
 670
 671         if (ret > 0)
 672                 add_rchar(current, ret);
 673         inc_syscr(current);
 674         return ret;
 675 }
 676
 677 asmlinkage ssize_t
 678 sys_writev(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
 679 {
 680         struct file *file;
 681         ssize_t ret = -EBADF;
 682         int fput_needed;
 683
 684         file = fget_light(fd, &fput_needed);
 685         if (file) {
 686                 loff_t pos = file_pos_read(file);
 687                 ret = vfs_writev(file, vec, vlen, &pos);
 688                 file_pos_write(file, pos);
 689                 fput_light(file, fput_needed);
 690         }
 691
 692         if (ret > 0)
 693                 add_wchar(current, ret);
 694         inc_syscw(current);
 695         return ret;
 696 }
 697
 698 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
 699                            size_t count, loff_t max)
 700 {
 701         struct file * in_file, * out_file;
 702         struct inode * in_inode, * out_inode;
 703         loff_t pos;
 704         ssize_t retval;
 705         int fput_needed_in, fput_needed_out, fl;
 706
 707         /*
 708          * Get input file, and verify that it is ok..
 709          */
 710         retval = -EBADF;
 711         in_file = fget_light(in_fd, &fput_needed_in);
 712         if (!in_file)
 713                 goto out;
 714         if (!(in_file->f_mode & FMODE_READ))
 715                 goto fput_in;
 716         retval = -EINVAL;
 717         in_inode = in_file->f_path.dentry->d_inode;
 718         if (!in_inode)
 719                 goto fput_in;
 720         if (!in_file->f_op || !in_file->f_op->splice_read)
 721                 goto fput_in;
 722         retval = -ESPIPE;
 723         if (!ppos)
 724                 ppos = &in_file->f_pos;
 725         else
 726                 if (!(in_file->f_mode & FMODE_PREAD))
 727                         goto fput_in;
 728         retval = rw_verify_area(READ, in_file, ppos, count);
 729         if (retval < 0)
 730                 goto fput_in;
 731         count = retval;
 732
 733         /*
 734          * Get output file, and verify that it is ok..
 735          */
 736         retval = -EBADF;
 737         out_file = fget_light(out_fd, &fput_needed_out);
 738         if (!out_file)
 739                 goto fput_in;
 740         if (!(out_file->f_mode & FMODE_WRITE))
 741                 goto fput_out;
 742         retval = -EINVAL;
 743         if (!out_file->f_op || !out_file->f_op->sendpage)
 744                 goto fput_out;
 745         out_inode = out_file->f_path.dentry->d_inode;
 746         retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
 747         if (retval < 0)
 748                 goto fput_out;
 749         count = retval;
 750
 751         if (!max)
 752                 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
 753
 754         pos = *ppos;
 755         retval = -EINVAL;
 756         if (unlikely(pos < 0))
 757                 goto fput_out;
 758         if (unlikely(pos + count > max)) {
 759                 retval = -EOVERFLOW;
 760                 if (pos >= max)
 761                         goto fput_out;
 762                 count = max - pos;
 763         }
 764
 765         fl = 0;
 766 #if 0
 767         /*
 768          * We need to debate whether we can enable this or not. The
 769          * man page documents EAGAIN return for the output at least,
 770          * and the application is arguably buggy if it doesn't expect
 771          * EAGAIN on a non-blocking file descriptor.
 772          */
 773         if (in_file->f_flags & O_NONBLOCK)
 774                 fl = SPLICE_F_NONBLOCK;
 775 #endif
 776         retval = do_splice_direct(in_file, ppos, out_file, count, fl);
 777
 778         if (retval > 0) {
 779                 add_rchar(current, retval);
 780                 add_wchar(current, retval);
 781         }
 782
 783         inc_syscr(current);
 784         inc_syscw(current);
 785         if (*ppos > max)
 786                 retval = -EOVERFLOW;
 787
 788 fput_out:
 789         fput_light(out_file, fput_needed_out);
 790 fput_in:
 791         fput_light(in_file, fput_needed_in);
 792 out:
 793         return retval;
 794 }
 795
 796 asmlinkage ssize_t sys_sendfile(int out_fd, int in_fd, off_t __user *offset, size_t count)
 797 {
 798         loff_t pos;
 799         off_t off;
 800         ssize_t ret;
 801
 802         if (offset) {
 803                 if (unlikely(get_user(off, offset)))
 804                         return -EFAULT;
 805                 pos = off;
 806                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
 807                 if (unlikely(put_user(pos, offset)))
 808                         return -EFAULT;
 809                 return ret;
 810         }
 811
 812         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 813 }
 814
 815 asmlinkage ssize_t sys_sendfile64(int out_fd, int in_fd, loff_t __user *offset, size_t count)
 816 {
 817         loff_t pos;
 818         ssize_t ret;
 819
 820         if (offset) {
 821                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
 822                         return -EFAULT;
 823                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
 824                 if (unlikely(put_user(pos, offset)))
 825                         return -EFAULT;
 826                 return ret;
 827         }
 828
 829         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 830 }