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 static int
  35 __negative_fpos_check(struct file *file, loff_t pos, size_t count)
  36 {
  37         /*
  38          * pos or pos+count is negative here, check overflow.
  39          * too big "count" will be caught in rw_verify_area().
  40          */
  41         if ((pos < 0) && (pos + count < pos))
  42                 return -EOVERFLOW;
  43         if (file->f_mode & FMODE_UNSIGNED_OFFSET)
  44                 return 0;
  45         return -EINVAL;
  46 }
  47
  48 /**
  49  * generic_file_llseek_unlocked - lockless generic llseek implementation
  50  * @file:       file structure to seek on
  51  * @offset:     file offset to seek to
  52  * @origin:     type of seek
  53  *
  54  * Updates the file offset to the value specified by @offset and @origin.
  55  * Locking must be provided by the caller.
  56  */
  57 loff_t
  58 generic_file_llseek_unlocked(struct file *file, loff_t offset, int origin)
  59 {
  60         struct inode *inode = file->f_mapping->host;
  61
  62         switch (origin) {
  63         case SEEK_END:
  64                 offset += inode->i_size;
  65                 break;
  66         case SEEK_CUR:
  67                 /*
  68                  * Here we special-case the lseek(fd, 0, SEEK_CUR)
  69                  * position-querying operation.  Avoid rewriting the "same"
  70                  * f_pos value back to the file because a concurrent read(),
  71                  * write() or lseek() might have altered it
  72                  */
  73                 if (offset == 0)
  74                         return file->f_pos;
  75                 offset += file->f_pos;
  76                 break;
  77         }
  78
  79         if (offset < 0 && __negative_fpos_check(file, offset, 0))
  80                 return -EINVAL;
  81         if (offset > inode->i_sb->s_maxbytes)
  82                 return -EINVAL;
  83
  84         /* Special lock needed here? */
  85         if (offset != file->f_pos) {
  86                 file->f_pos = offset;
  87                 file->f_version = 0;
  88         }
  89
  90         return offset;
  91 }
  92 EXPORT_SYMBOL(generic_file_llseek_unlocked);
  93
  94 /**
  95  * generic_file_llseek - generic llseek implementation for regular files
  96  * @file:       file structure to seek on
  97  * @offset:     file offset to seek to
  98  * @origin:     type of seek
  99  *
 100  * This is a generic implemenation of ->llseek useable for all normal local
 101  * filesystems.  It just updates the file offset to the value specified by
 102  * @offset and @origin under i_mutex.
 103  */
 104 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
 105 {
 106         loff_t rval;
 107
 108         mutex_lock(&file->f_dentry->d_inode->i_mutex);
 109         rval = generic_file_llseek_unlocked(file, offset, origin);
 110         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 111
 112         return rval;
 113 }
 114 EXPORT_SYMBOL(generic_file_llseek);
 115
 116 /**
 117  * noop_llseek - No Operation Performed llseek implementation
 118  * @file:       file structure to seek on
 119  * @offset:     file offset to seek to
 120  * @origin:     type of seek
 121  *
 122  * This is an implementation of ->llseek useable for the rare special case when
 123  * userspace expects the seek to succeed but the (device) file is actually not
 124  * able to perform the seek. In this case you use noop_llseek() instead of
 125  * falling back to the default implementation of ->llseek.
 126  */
 127 loff_t noop_llseek(struct file *file, loff_t offset, int origin)
 128 {
 129         return file->f_pos;
 130 }
 131 EXPORT_SYMBOL(noop_llseek);
 132
 133 loff_t no_llseek(struct file *file, loff_t offset, int origin)
 134 {
 135         return -ESPIPE;
 136 }
 137 EXPORT_SYMBOL(no_llseek);
 138
 139 loff_t default_llseek(struct file *file, loff_t offset, int origin)
 140 {
 141         loff_t retval;
 142
 143         mutex_lock(&file->f_dentry->d_inode->i_mutex);
 144         switch (origin) {
 145                 case SEEK_END:
 146                         offset += i_size_read(file->f_path.dentry->d_inode);
 147                         break;
 148                 case SEEK_CUR:
 149                         if (offset == 0) {
 150                                 retval = file->f_pos;
 151                                 goto out;
 152                         }
 153                         offset += file->f_pos;
 154         }
 155         retval = -EINVAL;
 156         if (offset >= 0 || !__negative_fpos_check(file, offset, 0)) {
 157                 if (offset != file->f_pos) {
 158                         file->f_pos = offset;
 159                         file->f_version = 0;
 160                 }
 161                 retval = offset;
 162         }
 163 out:
 164         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 165         return retval;
 166 }
 167 EXPORT_SYMBOL(default_llseek);
 168
 169 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
 170 {
 171         loff_t (*fn)(struct file *, loff_t, int);
 172
 173         fn = no_llseek;
 174         if (file->f_mode & FMODE_LSEEK) {
 175                 if (file->f_op && file->f_op->llseek)
 176                         fn = file->f_op->llseek;
 177         }
 178         return fn(file, offset, origin);
 179 }
 180 EXPORT_SYMBOL(vfs_llseek);
 181
 182 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
 183 {
 184         off_t retval;
 185         struct file * file;
 186         int fput_needed;
 187
 188         retval = -EBADF;
 189         file = fget_light(fd, &fput_needed);
 190         if (!file)
 191                 goto bad;
 192
 193         retval = -EINVAL;
 194         if (origin <= SEEK_MAX) {
 195                 loff_t res = vfs_llseek(file, offset, origin);
 196                 retval = res;
 197                 if (res != (loff_t)retval)
 198                         retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
 199         }
 200         fput_light(file, fput_needed);
 201 bad:
 202         return retval;
 203 }
 204
 205 #ifdef __ARCH_WANT_SYS_LLSEEK
 206 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
 207                 unsigned long, offset_low, loff_t __user *, result,
 208                 unsigned int, origin)
 209 {
 210         int retval;
 211         struct file * file;
 212         loff_t offset;
 213         int fput_needed;
 214
 215         retval = -EBADF;
 216         file = fget_light(fd, &fput_needed);
 217         if (!file)
 218                 goto bad;
 219
 220         retval = -EINVAL;
 221         if (origin > SEEK_MAX)
 222                 goto out_putf;
 223
 224         offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
 225                         origin);
 226
 227         retval = (int)offset;
 228         if (offset >= 0) {
 229                 retval = -EFAULT;
 230                 if (!copy_to_user(result, &offset, sizeof(offset)))
 231                         retval = 0;
 232         }
 233 out_putf:
 234         fput_light(file, fput_needed);
 235 bad:
 236         return retval;
 237 }
 238 #endif
 239
 240
 241 /*
 242  * rw_verify_area doesn't like huge counts. We limit
 243  * them to something that fits in "int" so that others
 244  * won't have to do range checks all the time.
 245  */
 246 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
 247 {
 248         struct inode *inode;
 249         loff_t pos;
 250         int retval = -EINVAL;
 251
 252         inode = file->f_path.dentry->d_inode;
 253         if (unlikely((ssize_t) count < 0))
 254                 return retval;
 255         pos = *ppos;
 256         if (unlikely((pos < 0) || (loff_t) (pos + count) < 0)) {
 257                 retval = __negative_fpos_check(file, pos, count);
 258                 if (retval)
 259                         return retval;
 260         }
 261
 262         if (unlikely(inode->i_flock && mandatory_lock(inode))) {
 263                 retval = locks_mandatory_area(
 264                         read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
 265                         inode, file, pos, count);
 266                 if (retval < 0)
 267                         return retval;
 268         }
 269         retval = security_file_permission(file,
 270                                 read_write == READ ? MAY_READ : MAY_WRITE);
 271         if (retval)
 272                 return retval;
 273         return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
 274 }
 275
 276 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
 277 {
 278         set_current_state(TASK_UNINTERRUPTIBLE);
 279         if (!kiocbIsKicked(iocb))
 280                 schedule();
 281         else
 282                 kiocbClearKicked(iocb);
 283         __set_current_state(TASK_RUNNING);
 284 }
 285
 286 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 287 {
 288         struct iovec iov = { .iov_base = buf, .iov_len = len };
 289         struct kiocb kiocb;
 290         ssize_t ret;
 291
 292         init_sync_kiocb(&kiocb, filp);
 293         kiocb.ki_pos = *ppos;
 294         kiocb.ki_left = len;
 295         kiocb.ki_nbytes = len;
 296
 297         for (;;) {
 298                 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
 299                 if (ret != -EIOCBRETRY)
 300                         break;
 301                 wait_on_retry_sync_kiocb(&kiocb);
 302         }
 303
 304         if (-EIOCBQUEUED == ret)
 305                 ret = wait_on_sync_kiocb(&kiocb);
 306         *ppos = kiocb.ki_pos;
 307         return ret;
 308 }
 309
 310 EXPORT_SYMBOL(do_sync_read);
 311
 312 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 313 {
 314         ssize_t ret;
 315
 316         if (!(file->f_mode & FMODE_READ))
 317                 return -EBADF;
 318         if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
 319                 return -EINVAL;
 320         if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
 321                 return -EFAULT;
 322
 323         ret = rw_verify_area(READ, file, pos, count);
 324         if (ret >= 0) {
 325                 count = ret;
 326                 if (file->f_op->read)
 327                         ret = file->f_op->read(file, buf, count, pos);
 328                 else
 329                         ret = do_sync_read(file, buf, count, pos);
 330                 if (ret > 0) {
 331                         fsnotify_access(file);
 332                         add_rchar(current, ret);
 333                 }
 334                 inc_syscr(current);
 335         }
 336
 337         return ret;
 338 }
 339
 340 EXPORT_SYMBOL(vfs_read);
 341
 342 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 343 {
 344         struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
 345         struct kiocb kiocb;
 346         ssize_t ret;
 347
 348         init_sync_kiocb(&kiocb, filp);
 349         kiocb.ki_pos = *ppos;
 350         kiocb.ki_left = len;
 351         kiocb.ki_nbytes = len;
 352
 353         for (;;) {
 354                 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
 355                 if (ret != -EIOCBRETRY)
 356                         break;
 357                 wait_on_retry_sync_kiocb(&kiocb);
 358         }
 359
 360         if (-EIOCBQUEUED == ret)
 361                 ret = wait_on_sync_kiocb(&kiocb);
 362         *ppos = kiocb.ki_pos;
 363         return ret;
 364 }
 365
 366 EXPORT_SYMBOL(do_sync_write);
 367
 368 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 369 {
 370         ssize_t ret;
 371
 372         if (!(file->f_mode & FMODE_WRITE))
 373                 return -EBADF;
 374         if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
 375                 return -EINVAL;
 376         if (unlikely(!access_ok(VERIFY_READ, buf, count)))
 377                 return -EFAULT;
 378
 379         ret = rw_verify_area(WRITE, file, pos, count);
 380         if (ret >= 0) {
 381                 count = ret;
 382                 if (file->f_op->write)
 383                         ret = file->f_op->write(file, buf, count, pos);
 384                 else
 385                         ret = do_sync_write(file, buf, count, pos);
 386                 if (ret > 0) {
 387                         fsnotify_modify(file);
 388                         add_wchar(current, ret);
 389                 }
 390                 inc_syscw(current);
 391         }
 392
 393         return ret;
 394 }
 395
 396 EXPORT_SYMBOL(vfs_write);
 397
 398 static inline loff_t file_pos_read(struct file *file)
 399 {
 400         return file->f_pos;
 401 }
 402
 403 static inline void file_pos_write(struct file *file, loff_t pos)
 404 {
 405         file->f_pos = pos;
 406 }
 407
 408 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
 409 {
 410         struct file *file;
 411         ssize_t ret = -EBADF;
 412         int fput_needed;
 413
 414         file = fget_light(fd, &fput_needed);
 415         if (file) {
 416                 loff_t pos = file_pos_read(file);
 417                 ret = vfs_read(file, buf, count, &pos);
 418                 file_pos_write(file, pos);
 419                 fput_light(file, fput_needed);
 420         }
 421
 422         return ret;
 423 }
 424
 425 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
 426                 size_t, count)
 427 {
 428         struct file *file;
 429         ssize_t ret = -EBADF;
 430         int fput_needed;
 431
 432         file = fget_light(fd, &fput_needed);
 433         if (file) {
 434                 loff_t pos = file_pos_read(file);
 435                 ret = vfs_write(file, buf, count, &pos);
 436                 file_pos_write(file, pos);
 437                 fput_light(file, fput_needed);
 438         }
 439
 440         return ret;
 441 }
 442
 443 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
 444                         size_t count, loff_t pos)
 445 {
 446         struct file *file;
 447         ssize_t ret = -EBADF;
 448         int fput_needed;
 449
 450         if (pos < 0)
 451                 return -EINVAL;
 452
 453         file = fget_light(fd, &fput_needed);
 454         if (file) {
 455                 ret = -ESPIPE;
 456                 if (file->f_mode & FMODE_PREAD)
 457                         ret = vfs_read(file, buf, count, &pos);
 458                 fput_light(file, fput_needed);
 459         }
 460
 461         return ret;
 462 }
 463 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 464 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
 465 {
 466         return SYSC_pread64((unsigned int) fd, (char __user *) buf,
 467                             (size_t) count, pos);
 468 }
 469 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
 470 #endif
 471
 472 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
 473                          size_t count, loff_t pos)
 474 {
 475         struct file *file;
 476         ssize_t ret = -EBADF;
 477         int fput_needed;
 478
 479         if (pos < 0)
 480                 return -EINVAL;
 481
 482         file = fget_light(fd, &fput_needed);
 483         if (file) {
 484                 ret = -ESPIPE;
 485                 if (file->f_mode & FMODE_PWRITE)
 486                         ret = vfs_write(file, buf, count, &pos);
 487                 fput_light(file, fput_needed);
 488         }
 489
 490         return ret;
 491 }
 492 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 493 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
 494 {
 495         return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
 496                              (size_t) count, pos);
 497 }
 498 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
 499 #endif
 500
 501 /*
 502  * Reduce an iovec's length in-place.  Return the resulting number of segments
 503  */
 504 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
 505 {
 506         unsigned long seg = 0;
 507         size_t len = 0;
 508
 509         while (seg < nr_segs) {
 510                 seg++;
 511                 if (len + iov->iov_len >= to) {
 512                         iov->iov_len = to - len;
 513                         break;
 514                 }
 515                 len += iov->iov_len;
 516                 iov++;
 517         }
 518         return seg;
 519 }
 520 EXPORT_SYMBOL(iov_shorten);
 521
 522 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
 523                 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
 524 {
 525         struct kiocb kiocb;
 526         ssize_t ret;
 527
 528         init_sync_kiocb(&kiocb, filp);
 529         kiocb.ki_pos = *ppos;
 530         kiocb.ki_left = len;
 531         kiocb.ki_nbytes = len;
 532
 533         for (;;) {
 534                 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
 535                 if (ret != -EIOCBRETRY)
 536                         break;
 537                 wait_on_retry_sync_kiocb(&kiocb);
 538         }
 539
 540         if (ret == -EIOCBQUEUED)
 541                 ret = wait_on_sync_kiocb(&kiocb);
 542         *ppos = kiocb.ki_pos;
 543         return ret;
 544 }
 545
 546 /* Do it by hand, with file-ops */
 547 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
 548                 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
 549 {
 550         struct iovec *vector = iov;
 551         ssize_t ret = 0;
 552
 553         while (nr_segs > 0) {
 554                 void __user *base;
 555                 size_t len;
 556                 ssize_t nr;
 557
 558                 base = vector->iov_base;
 559                 len = vector->iov_len;
 560                 vector++;
 561                 nr_segs--;
 562
 563                 nr = fn(filp, base, len, ppos);
 564
 565                 if (nr < 0) {
 566                         if (!ret)
 567                                 ret = nr;
 568                         break;
 569                 }
 570                 ret += nr;
 571                 if (nr != len)
 572                         break;
 573         }
 574
 575         return ret;
 576 }
 577
 578 /* A write operation does a read from user space and vice versa */
 579 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 580
 581 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
 582                               unsigned long nr_segs, unsigned long fast_segs,
 583                               struct iovec *fast_pointer,
 584                               struct iovec **ret_pointer)
 585 {
 586         unsigned long seg;
 587         ssize_t ret;
 588         struct iovec *iov = fast_pointer;
 589
 590         /*
 591          * SuS says "The readv() function *may* fail if the iovcnt argument
 592          * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 593          * traditionally returned zero for zero segments, so...
 594          */
 595         if (nr_segs == 0) {
 596                 ret = 0;
 597                 goto out;
 598         }
 599
 600         /*
 601          * First get the "struct iovec" from user memory and
 602          * verify all the pointers
 603          */
 604         if (nr_segs > UIO_MAXIOV) {
 605                 ret = -EINVAL;
 606                 goto out;
 607         }
 608         if (nr_segs > fast_segs) {
 609                 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 610                 if (iov == NULL) {
 611                         ret = -ENOMEM;
 612                         goto out;
 613                 }
 614         }
 615         if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
 616                 ret = -EFAULT;
 617                 goto out;
 618         }
 619
 620         /*
 621          * According to the Single Unix Specification we should return EINVAL
 622          * if an element length is < 0 when cast to ssize_t or if the
 623          * total length would overflow the ssize_t return value of the
 624          * system call.
 625          *
 626          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
 627          * overflow case.
 628          */
 629         ret = 0;
 630         for (seg = 0; seg < nr_segs; seg++) {
 631                 void __user *buf = iov[seg].iov_base;
 632                 ssize_t len = (ssize_t)iov[seg].iov_len;
 633
 634                 /* see if we we're about to use an invalid len or if
 635                  * it's about to overflow ssize_t */
 636                 if (len < 0) {
 637                         ret = -EINVAL;
 638                         goto out;
 639                 }
 640                 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
 641                         ret = -EFAULT;
 642                         goto out;
 643                 }
 644                 if (len > MAX_RW_COUNT - ret) {
 645                         len = MAX_RW_COUNT - ret;
 646                         iov[seg].iov_len = len;
 647                 }
 648                 ret += len;
 649         }
 650 out:
 651         *ret_pointer = iov;
 652         return ret;
 653 }
 654
 655 static ssize_t do_readv_writev(int type, struct file *file,
 656                                const struct iovec __user * uvector,
 657                                unsigned long nr_segs, loff_t *pos)
 658 {
 659         size_t tot_len;
 660         struct iovec iovstack[UIO_FASTIOV];
 661         struct iovec *iov = iovstack;
 662         ssize_t ret;
 663         io_fn_t fn;
 664         iov_fn_t fnv;
 665
 666         if (!file->f_op) {
 667                 ret = -EINVAL;
 668                 goto out;
 669         }
 670
 671         ret = rw_copy_check_uvector(type, uvector, nr_segs,
 672                         ARRAY_SIZE(iovstack), iovstack, &iov);
 673         if (ret <= 0)
 674                 goto out;
 675
 676         tot_len = ret;
 677         ret = rw_verify_area(type, file, pos, tot_len);
 678         if (ret < 0)
 679                 goto out;
 680
 681         fnv = NULL;
 682         if (type == READ) {
 683                 fn = file->f_op->read;
 684                 fnv = file->f_op->aio_read;
 685         } else {
 686                 fn = (io_fn_t)file->f_op->write;
 687                 fnv = file->f_op->aio_write;
 688         }
 689
 690         if (fnv)
 691                 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
 692                                                 pos, fnv);
 693         else
 694                 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
 695
 696 out:
 697         if (iov != iovstack)
 698                 kfree(iov);
 699         if ((ret + (type == READ)) > 0) {
 700                 if (type == READ)
 701                         fsnotify_access(file);
 702                 else
 703                         fsnotify_modify(file);
 704         }
 705         return ret;
 706 }
 707
 708 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
 709                   unsigned long vlen, loff_t *pos)
 710 {
 711         if (!(file->f_mode & FMODE_READ))
 712                 return -EBADF;
 713         if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
 714                 return -EINVAL;
 715
 716         return do_readv_writev(READ, file, vec, vlen, pos);
 717 }
 718
 719 EXPORT_SYMBOL(vfs_readv);
 720
 721 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 722                    unsigned long vlen, loff_t *pos)
 723 {
 724         if (!(file->f_mode & FMODE_WRITE))
 725                 return -EBADF;
 726         if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
 727                 return -EINVAL;
 728
 729         return do_readv_writev(WRITE, file, vec, vlen, pos);
 730 }
 731
 732 EXPORT_SYMBOL(vfs_writev);
 733
 734 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
 735                 unsigned long, vlen)
 736 {
 737         struct file *file;
 738         ssize_t ret = -EBADF;
 739         int fput_needed;
 740
 741         file = fget_light(fd, &fput_needed);
 742         if (file) {
 743                 loff_t pos = file_pos_read(file);
 744                 ret = vfs_readv(file, vec, vlen, &pos);
 745                 file_pos_write(file, pos);
 746                 fput_light(file, fput_needed);
 747         }
 748
 749         if (ret > 0)
 750                 add_rchar(current, ret);
 751         inc_syscr(current);
 752         return ret;
 753 }
 754
 755 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
 756                 unsigned long, vlen)
 757 {
 758         struct file *file;
 759         ssize_t ret = -EBADF;
 760         int fput_needed;
 761
 762         file = fget_light(fd, &fput_needed);
 763         if (file) {
 764                 loff_t pos = file_pos_read(file);
 765                 ret = vfs_writev(file, vec, vlen, &pos);
 766                 file_pos_write(file, pos);
 767                 fput_light(file, fput_needed);
 768         }
 769
 770         if (ret > 0)
 771                 add_wchar(current, ret);
 772         inc_syscw(current);
 773         return ret;
 774 }
 775
 776 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
 777 {
 778 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
 779         return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
 780 }
 781
 782 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
 783                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
 784 {
 785         loff_t pos = pos_from_hilo(pos_h, pos_l);
 786         struct file *file;
 787         ssize_t ret = -EBADF;
 788         int fput_needed;
 789
 790         if (pos < 0)
 791                 return -EINVAL;
 792
 793         file = fget_light(fd, &fput_needed);
 794         if (file) {
 795                 ret = -ESPIPE;
 796                 if (file->f_mode & FMODE_PREAD)
 797                         ret = vfs_readv(file, vec, vlen, &pos);
 798                 fput_light(file, fput_needed);
 799         }
 800
 801         if (ret > 0)
 802                 add_rchar(current, ret);
 803         inc_syscr(current);
 804         return ret;
 805 }
 806
 807 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
 808                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
 809 {
 810         loff_t pos = pos_from_hilo(pos_h, pos_l);
 811         struct file *file;
 812         ssize_t ret = -EBADF;
 813         int fput_needed;
 814
 815         if (pos < 0)
 816                 return -EINVAL;
 817
 818         file = fget_light(fd, &fput_needed);
 819         if (file) {
 820                 ret = -ESPIPE;
 821                 if (file->f_mode & FMODE_PWRITE)
 822                         ret = vfs_writev(file, vec, vlen, &pos);
 823                 fput_light(file, fput_needed);
 824         }
 825
 826         if (ret > 0)
 827                 add_wchar(current, ret);
 828         inc_syscw(current);
 829         return ret;
 830 }
 831
 832 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
 833                            size_t count, loff_t max)
 834 {
 835         struct file * in_file, * out_file;
 836         struct inode * in_inode, * out_inode;
 837         loff_t pos;
 838         ssize_t retval;
 839         int fput_needed_in, fput_needed_out, fl;
 840
 841         /*
 842          * Get input file, and verify that it is ok..
 843          */
 844         retval = -EBADF;
 845         in_file = fget_light(in_fd, &fput_needed_in);
 846         if (!in_file)
 847                 goto out;
 848         if (!(in_file->f_mode & FMODE_READ))
 849                 goto fput_in;
 850         retval = -ESPIPE;
 851         if (!ppos)
 852                 ppos = &in_file->f_pos;
 853         else
 854                 if (!(in_file->f_mode & FMODE_PREAD))
 855                         goto fput_in;
 856         retval = rw_verify_area(READ, in_file, ppos, count);
 857         if (retval < 0)
 858                 goto fput_in;
 859         count = retval;
 860
 861         /*
 862          * Get output file, and verify that it is ok..
 863          */
 864         retval = -EBADF;
 865         out_file = fget_light(out_fd, &fput_needed_out);
 866         if (!out_file)
 867                 goto fput_in;
 868         if (!(out_file->f_mode & FMODE_WRITE))
 869                 goto fput_out;
 870         retval = -EINVAL;
 871         in_inode = in_file->f_path.dentry->d_inode;
 872         out_inode = out_file->f_path.dentry->d_inode;
 873         retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
 874         if (retval < 0)
 875                 goto fput_out;
 876         count = retval;
 877
 878         if (!max)
 879                 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
 880
 881         pos = *ppos;
 882         if (unlikely(pos + count > max)) {
 883                 retval = -EOVERFLOW;
 884                 if (pos >= max)
 885                         goto fput_out;
 886                 count = max - pos;
 887         }
 888
 889         fl = 0;
 890 #if 0
 891         /*
 892          * We need to debate whether we can enable this or not. The
 893          * man page documents EAGAIN return for the output at least,
 894          * and the application is arguably buggy if it doesn't expect
 895          * EAGAIN on a non-blocking file descriptor.
 896          */
 897         if (in_file->f_flags & O_NONBLOCK)
 898                 fl = SPLICE_F_NONBLOCK;
 899 #endif
 900         retval = do_splice_direct(in_file, ppos, out_file, count, fl);
 901
 902         if (retval > 0) {
 903                 add_rchar(current, retval);
 904                 add_wchar(current, retval);
 905         }
 906
 907         inc_syscr(current);
 908         inc_syscw(current);
 909         if (*ppos > max)
 910                 retval = -EOVERFLOW;
 911
 912 fput_out:
 913         fput_light(out_file, fput_needed_out);
 914 fput_in:
 915         fput_light(in_file, fput_needed_in);
 916 out:
 917         return retval;
 918 }
 919
 920 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
 921 {
 922         loff_t pos;
 923         off_t off;
 924         ssize_t ret;
 925
 926         if (offset) {
 927                 if (unlikely(get_user(off, offset)))
 928                         return -EFAULT;
 929                 pos = off;
 930                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
 931                 if (unlikely(put_user(pos, offset)))
 932                         return -EFAULT;
 933                 return ret;
 934         }
 935
 936         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 937 }
 938
 939 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
 940 {
 941         loff_t pos;
 942         ssize_t ret;
 943
 944         if (offset) {
 945                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
 946                         return -EFAULT;
 947                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
 948                 if (unlikely(put_user(pos, offset)))
 949                         return -EFAULT;
 950                 return ret;
 951         }
 952
 953         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 954 }