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/fsnotify.h>
  13 #include <linux/security.h>
  14 #include <linux/module.h>
  15 #include <linux/syscalls.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/splice.h>
  18 #include "read_write.h"
  19
  20 #include <asm/uaccess.h>
  21 #include <asm/unistd.h>
  22
  23 const struct file_operations generic_ro_fops = {
  24         .llseek         = generic_file_llseek,
  25         .read           = do_sync_read,
  26         .aio_read       = generic_file_aio_read,
  27         .mmap           = generic_file_readonly_mmap,
  28         .splice_read    = generic_file_splice_read,
  29 };
  30
  31 EXPORT_SYMBOL(generic_ro_fops);
  32
  33 static inline int unsigned_offsets(struct file *file)
  34 {
  35         return file->f_mode & FMODE_UNSIGNED_OFFSET;
  36 }
  37
  38 /**
  39  * generic_file_llseek_unlocked - lockless generic llseek implementation
  40  * @file:       file structure to seek on
  41  * @offset:     file offset to seek to
  42  * @origin:     type of seek
  43  *
  44  * Updates the file offset to the value specified by @offset and @origin.
  45  * Locking must be provided by the caller.
  46  */
  47 loff_t
  48 generic_file_llseek_unlocked(struct file *file, loff_t offset, int origin)
  49 {
  50         struct inode *inode = file->f_mapping->host;
  51
  52         switch (origin) {
  53         case SEEK_END:
  54                 offset += inode->i_size;
  55                 break;
  56         case SEEK_CUR:
  57                 /*
  58                  * Here we special-case the lseek(fd, 0, SEEK_CUR)
  59                  * position-querying operation.  Avoid rewriting the "same"
  60                  * f_pos value back to the file because a concurrent read(),
  61                  * write() or lseek() might have altered it
  62                  */
  63                 if (offset == 0)
  64                         return file->f_pos;
  65                 offset += file->f_pos;
  66                 break;
  67         case SEEK_DATA:
  68                 /*
  69                  * In the generic case the entire file is data, so as long as
  70                  * offset isn't at the end of the file then the offset is data.
  71                  */
  72                 if (offset >= inode->i_size)
  73                         return -ENXIO;
  74                 break;
  75         case SEEK_HOLE:
  76                 /*
  77                  * There is a virtual hole at the end of the file, so as long as
  78                  * offset isn't i_size or larger, return i_size.
  79                  */
  80                 if (offset >= inode->i_size)
  81                         return -ENXIO;
  82                 offset = inode->i_size;
  83                 break;
  84         }
  85
  86         if (offset < 0 && !unsigned_offsets(file))
  87                 return -EINVAL;
  88         if (offset > inode->i_sb->s_maxbytes)
  89                 return -EINVAL;
  90
  91         /* Special lock needed here? */
  92         if (offset != file->f_pos) {
  93                 file->f_pos = offset;
  94                 file->f_version = 0;
  95         }
  96
  97         return offset;
  98 }
  99 EXPORT_SYMBOL(generic_file_llseek_unlocked);
 100
 101 /**
 102  * generic_file_llseek - generic llseek implementation for regular files
 103  * @file:       file structure to seek on
 104  * @offset:     file offset to seek to
 105  * @origin:     type of seek
 106  *
 107  * This is a generic implemenation of ->llseek useable for all normal local
 108  * filesystems.  It just updates the file offset to the value specified by
 109  * @offset and @origin under i_mutex.
 110  */
 111 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
 112 {
 113         loff_t rval;
 114
 115         mutex_lock(&file->f_dentry->d_inode->i_mutex);
 116         rval = generic_file_llseek_unlocked(file, offset, origin);
 117         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 118
 119         return rval;
 120 }
 121 EXPORT_SYMBOL(generic_file_llseek);
 122
 123 /**
 124  * noop_llseek - No Operation Performed llseek implementation
 125  * @file:       file structure to seek on
 126  * @offset:     file offset to seek to
 127  * @origin:     type of seek
 128  *
 129  * This is an implementation of ->llseek useable for the rare special case when
 130  * userspace expects the seek to succeed but the (device) file is actually not
 131  * able to perform the seek. In this case you use noop_llseek() instead of
 132  * falling back to the default implementation of ->llseek.
 133  */
 134 loff_t noop_llseek(struct file *file, loff_t offset, int origin)
 135 {
 136         return file->f_pos;
 137 }
 138 EXPORT_SYMBOL(noop_llseek);
 139
 140 loff_t no_llseek(struct file *file, loff_t offset, int origin)
 141 {
 142         return -ESPIPE;
 143 }
 144 EXPORT_SYMBOL(no_llseek);
 145
 146 loff_t default_llseek(struct file *file, loff_t offset, int origin)
 147 {
 148         struct inode *inode = file->f_path.dentry->d_inode;
 149         loff_t retval;
 150
 151         mutex_lock(&inode->i_mutex);
 152         switch (origin) {
 153                 case SEEK_END:
 154                         offset += i_size_read(inode);
 155                         break;
 156                 case SEEK_CUR:
 157                         if (offset == 0) {
 158                                 retval = file->f_pos;
 159                                 goto out;
 160                         }
 161                         offset += file->f_pos;
 162                         break;
 163                 case SEEK_DATA:
 164                         /*
 165                          * In the generic case the entire file is data, so as
 166                          * long as offset isn't at the end of the file then the
 167                          * offset is data.
 168                          */
 169                         if (offset >= inode->i_size) {
 170                                 retval = -ENXIO;
 171                                 goto out;
 172                         }
 173                         break;
 174                 case SEEK_HOLE:
 175                         /*
 176                          * There is a virtual hole at the end of the file, so
 177                          * as long as offset isn't i_size or larger, return
 178                          * i_size.
 179                          */
 180                         if (offset >= inode->i_size) {
 181                                 retval = -ENXIO;
 182                                 goto out;
 183                         }
 184                         offset = inode->i_size;
 185                         break;
 186         }
 187         retval = -EINVAL;
 188         if (offset >= 0 || unsigned_offsets(file)) {
 189                 if (offset != file->f_pos) {
 190                         file->f_pos = offset;
 191                         file->f_version = 0;
 192                 }
 193                 retval = offset;
 194         }
 195 out:
 196         mutex_unlock(&inode->i_mutex);
 197         return retval;
 198 }
 199 EXPORT_SYMBOL(default_llseek);
 200
 201 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
 202 {
 203         loff_t (*fn)(struct file *, loff_t, int);
 204
 205         fn = no_llseek;
 206         if (file->f_mode & FMODE_LSEEK) {
 207                 if (file->f_op && file->f_op->llseek)
 208                         fn = file->f_op->llseek;
 209         }
 210         return fn(file, offset, origin);
 211 }
 212 EXPORT_SYMBOL(vfs_llseek);
 213
 214 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
 215 {
 216         off_t retval;
 217         struct file * file;
 218         int fput_needed;
 219
 220         retval = -EBADF;
 221         file = fget_light(fd, &fput_needed);
 222         if (!file)
 223                 goto bad;
 224
 225         retval = -EINVAL;
 226         if (origin <= SEEK_MAX) {
 227                 loff_t res = vfs_llseek(file, offset, origin);
 228                 retval = res;
 229                 if (res != (loff_t)retval)
 230                         retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
 231         }
 232         fput_light(file, fput_needed);
 233 bad:
 234         return retval;
 235 }
 236
 237 #ifdef __ARCH_WANT_SYS_LLSEEK
 238 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
 239                 unsigned long, offset_low, loff_t __user *, result,
 240                 unsigned int, origin)
 241 {
 242         int retval;
 243         struct file * file;
 244         loff_t offset;
 245         int fput_needed;
 246
 247         retval = -EBADF;
 248         file = fget_light(fd, &fput_needed);
 249         if (!file)
 250                 goto bad;
 251
 252         retval = -EINVAL;
 253         if (origin > SEEK_MAX)
 254                 goto out_putf;
 255
 256         offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
 257                         origin);
 258
 259         retval = (int)offset;
 260         if (offset >= 0) {
 261                 retval = -EFAULT;
 262                 if (!copy_to_user(result, &offset, sizeof(offset)))
 263                         retval = 0;
 264         }
 265 out_putf:
 266         fput_light(file, fput_needed);
 267 bad:
 268         return retval;
 269 }
 270 #endif
 271
 272
 273 /*
 274  * rw_verify_area doesn't like huge counts. We limit
 275  * them to something that fits in "int" so that others
 276  * won't have to do range checks all the time.
 277  */
 278 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
 279 {
 280         struct inode *inode;
 281         loff_t pos;
 282         int retval = -EINVAL;
 283
 284         inode = file->f_path.dentry->d_inode;
 285         if (unlikely((ssize_t) count < 0))
 286                 return retval;
 287         pos = *ppos;
 288         if (unlikely(pos < 0)) {
 289                 if (!unsigned_offsets(file))
 290                         return retval;
 291                 if (count >= -pos) /* both values are in 0..LLONG_MAX */
 292                         return -EOVERFLOW;
 293         } else if (unlikely((loff_t) (pos + count) < 0)) {
 294                 if (!unsigned_offsets(file))
 295                         return retval;
 296         }
 297
 298         if (unlikely(inode->i_flock && mandatory_lock(inode))) {
 299                 retval = locks_mandatory_area(
 300                         read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
 301                         inode, file, pos, count);
 302                 if (retval < 0)
 303                         return retval;
 304         }
 305         retval = security_file_permission(file,
 306                                 read_write == READ ? MAY_READ : MAY_WRITE);
 307         if (retval)
 308                 return retval;
 309         return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
 310 }
 311
 312 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
 313 {
 314         set_current_state(TASK_UNINTERRUPTIBLE);
 315         if (!kiocbIsKicked(iocb))
 316                 schedule();
 317         else
 318                 kiocbClearKicked(iocb);
 319         __set_current_state(TASK_RUNNING);
 320 }
 321
 322 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 323 {
 324         struct iovec iov = { .iov_base = buf, .iov_len = len };
 325         struct kiocb kiocb;
 326         ssize_t ret;
 327
 328         init_sync_kiocb(&kiocb, filp);
 329         kiocb.ki_pos = *ppos;
 330         kiocb.ki_left = len;
 331         kiocb.ki_nbytes = len;
 332
 333         for (;;) {
 334                 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
 335                 if (ret != -EIOCBRETRY)
 336                         break;
 337                 wait_on_retry_sync_kiocb(&kiocb);
 338         }
 339
 340         if (-EIOCBQUEUED == ret)
 341                 ret = wait_on_sync_kiocb(&kiocb);
 342         *ppos = kiocb.ki_pos;
 343         return ret;
 344 }
 345
 346 EXPORT_SYMBOL(do_sync_read);
 347
 348 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 349 {
 350         ssize_t ret;
 351
 352         if (!(file->f_mode & FMODE_READ))
 353                 return -EBADF;
 354         if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
 355                 return -EINVAL;
 356         if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
 357                 return -EFAULT;
 358
 359         ret = rw_verify_area(READ, file, pos, count);
 360         if (ret >= 0) {
 361                 count = ret;
 362                 if (file->f_op->read)
 363                         ret = file->f_op->read(file, buf, count, pos);
 364                 else
 365                         ret = do_sync_read(file, buf, count, pos);
 366                 if (ret > 0) {
 367                         fsnotify_access(file);
 368                         add_rchar(current, ret);
 369                 }
 370                 inc_syscr(current);
 371         }
 372
 373         return ret;
 374 }
 375
 376 EXPORT_SYMBOL(vfs_read);
 377
 378 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 379 {
 380         struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
 381         struct kiocb kiocb;
 382         ssize_t ret;
 383
 384         init_sync_kiocb(&kiocb, filp);
 385         kiocb.ki_pos = *ppos;
 386         kiocb.ki_left = len;
 387         kiocb.ki_nbytes = len;
 388
 389         for (;;) {
 390                 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
 391                 if (ret != -EIOCBRETRY)
 392                         break;
 393                 wait_on_retry_sync_kiocb(&kiocb);
 394         }
 395
 396         if (-EIOCBQUEUED == ret)
 397                 ret = wait_on_sync_kiocb(&kiocb);
 398         *ppos = kiocb.ki_pos;
 399         return ret;
 400 }
 401
 402 EXPORT_SYMBOL(do_sync_write);
 403
 404 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 405 {
 406         ssize_t ret;
 407
 408         if (!(file->f_mode & FMODE_WRITE))
 409                 return -EBADF;
 410         if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
 411                 return -EINVAL;
 412         if (unlikely(!access_ok(VERIFY_READ, buf, count)))
 413                 return -EFAULT;
 414
 415         ret = rw_verify_area(WRITE, file, pos, count);
 416         if (ret >= 0) {
 417                 count = ret;
 418                 if (file->f_op->write)
 419                         ret = file->f_op->write(file, buf, count, pos);
 420                 else
 421                         ret = do_sync_write(file, buf, count, pos);
 422                 if (ret > 0) {
 423                         fsnotify_modify(file);
 424                         add_wchar(current, ret);
 425                 }
 426                 inc_syscw(current);
 427         }
 428
 429         return ret;
 430 }
 431
 432 EXPORT_SYMBOL(vfs_write);
 433
 434 static inline loff_t file_pos_read(struct file *file)
 435 {
 436         return file->f_pos;
 437 }
 438
 439 static inline void file_pos_write(struct file *file, loff_t pos)
 440 {
 441         file->f_pos = pos;
 442 }
 443
 444 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
 445 {
 446         struct file *file;
 447         ssize_t ret = -EBADF;
 448         int fput_needed;
 449
 450         file = fget_light(fd, &fput_needed);
 451         if (file) {
 452                 loff_t pos = file_pos_read(file);
 453                 ret = vfs_read(file, buf, count, &pos);
 454                 file_pos_write(file, pos);
 455                 fput_light(file, fput_needed);
 456         }
 457
 458         return ret;
 459 }
 460
 461 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
 462                 size_t, count)
 463 {
 464         struct file *file;
 465         ssize_t ret = -EBADF;
 466         int fput_needed;
 467
 468         file = fget_light(fd, &fput_needed);
 469         if (file) {
 470                 loff_t pos = file_pos_read(file);
 471                 ret = vfs_write(file, buf, count, &pos);
 472                 file_pos_write(file, pos);
 473                 fput_light(file, fput_needed);
 474         }
 475
 476         return ret;
 477 }
 478
 479 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
 480                         size_t count, loff_t pos)
 481 {
 482         struct file *file;
 483         ssize_t ret = -EBADF;
 484         int fput_needed;
 485
 486         if (pos < 0)
 487                 return -EINVAL;
 488
 489         file = fget_light(fd, &fput_needed);
 490         if (file) {
 491                 ret = -ESPIPE;
 492                 if (file->f_mode & FMODE_PREAD)
 493                         ret = vfs_read(file, buf, count, &pos);
 494                 fput_light(file, fput_needed);
 495         }
 496
 497         return ret;
 498 }
 499 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 500 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
 501 {
 502         return SYSC_pread64((unsigned int) fd, (char __user *) buf,
 503                             (size_t) count, pos);
 504 }
 505 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
 506 #endif
 507
 508 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
 509                          size_t count, loff_t pos)
 510 {
 511         struct file *file;
 512         ssize_t ret = -EBADF;
 513         int fput_needed;
 514
 515         if (pos < 0)
 516                 return -EINVAL;
 517
 518         file = fget_light(fd, &fput_needed);
 519         if (file) {
 520                 ret = -ESPIPE;
 521                 if (file->f_mode & FMODE_PWRITE)
 522                         ret = vfs_write(file, buf, count, &pos);
 523                 fput_light(file, fput_needed);
 524         }
 525
 526         return ret;
 527 }
 528 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 529 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
 530 {
 531         return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
 532                              (size_t) count, pos);
 533 }
 534 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
 535 #endif
 536
 537 /*
 538  * Reduce an iovec's length in-place.  Return the resulting number of segments
 539  */
 540 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
 541 {
 542         unsigned long seg = 0;
 543         size_t len = 0;
 544
 545         while (seg < nr_segs) {
 546                 seg++;
 547                 if (len + iov->iov_len >= to) {
 548                         iov->iov_len = to - len;
 549                         break;
 550                 }
 551                 len += iov->iov_len;
 552                 iov++;
 553         }
 554         return seg;
 555 }
 556 EXPORT_SYMBOL(iov_shorten);
 557
 558 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
 559                 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
 560 {
 561         struct kiocb kiocb;
 562         ssize_t ret;
 563
 564         init_sync_kiocb(&kiocb, filp);
 565         kiocb.ki_pos = *ppos;
 566         kiocb.ki_left = len;
 567         kiocb.ki_nbytes = len;
 568
 569         for (;;) {
 570                 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
 571                 if (ret != -EIOCBRETRY)
 572                         break;
 573                 wait_on_retry_sync_kiocb(&kiocb);
 574         }
 575
 576         if (ret == -EIOCBQUEUED)
 577                 ret = wait_on_sync_kiocb(&kiocb);
 578         *ppos = kiocb.ki_pos;
 579         return ret;
 580 }
 581
 582 /* Do it by hand, with file-ops */
 583 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
 584                 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
 585 {
 586         struct iovec *vector = iov;
 587         ssize_t ret = 0;
 588
 589         while (nr_segs > 0) {
 590                 void __user *base;
 591                 size_t len;
 592                 ssize_t nr;
 593
 594                 base = vector->iov_base;
 595                 len = vector->iov_len;
 596                 vector++;
 597                 nr_segs--;
 598
 599                 nr = fn(filp, base, len, ppos);
 600
 601                 if (nr < 0) {
 602                         if (!ret)
 603                                 ret = nr;
 604                         break;
 605                 }
 606                 ret += nr;
 607                 if (nr != len)
 608                         break;
 609         }
 610
 611         return ret;
 612 }
 613
 614 /* A write operation does a read from user space and vice versa */
 615 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 616
 617 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
 618                               unsigned long nr_segs, unsigned long fast_segs,
 619                               struct iovec *fast_pointer,
 620                               struct iovec **ret_pointer)
 621 {
 622         unsigned long seg;
 623         ssize_t ret;
 624         struct iovec *iov = fast_pointer;
 625
 626         /*
 627          * SuS says "The readv() function *may* fail if the iovcnt argument
 628          * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 629          * traditionally returned zero for zero segments, so...
 630          */
 631         if (nr_segs == 0) {
 632                 ret = 0;
 633                 goto out;
 634         }
 635
 636         /*
 637          * First get the "struct iovec" from user memory and
 638          * verify all the pointers
 639          */
 640         if (nr_segs > UIO_MAXIOV) {
 641                 ret = -EINVAL;
 642                 goto out;
 643         }
 644         if (nr_segs > fast_segs) {
 645                 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 646                 if (iov == NULL) {
 647                         ret = -ENOMEM;
 648                         goto out;
 649                 }
 650         }
 651         if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
 652                 ret = -EFAULT;
 653                 goto out;
 654         }
 655
 656         /*
 657          * According to the Single Unix Specification we should return EINVAL
 658          * if an element length is < 0 when cast to ssize_t or if the
 659          * total length would overflow the ssize_t return value of the
 660          * system call.
 661          *
 662          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
 663          * overflow case.
 664          */
 665         ret = 0;
 666         for (seg = 0; seg < nr_segs; seg++) {
 667                 void __user *buf = iov[seg].iov_base;
 668                 ssize_t len = (ssize_t)iov[seg].iov_len;
 669
 670                 /* see if we we're about to use an invalid len or if
 671                  * it's about to overflow ssize_t */
 672                 if (len < 0) {
 673                         ret = -EINVAL;
 674                         goto out;
 675                 }
 676                 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
 677                         ret = -EFAULT;
 678                         goto out;
 679                 }
 680                 if (len > MAX_RW_COUNT - ret) {
 681                         len = MAX_RW_COUNT - ret;
 682                         iov[seg].iov_len = len;
 683                 }
 684                 ret += len;
 685         }
 686 out:
 687         *ret_pointer = iov;
 688         return ret;
 689 }
 690
 691 static ssize_t do_readv_writev(int type, struct file *file,
 692                                const struct iovec __user * uvector,
 693                                unsigned long nr_segs, loff_t *pos)
 694 {
 695         size_t tot_len;
 696         struct iovec iovstack[UIO_FASTIOV];
 697         struct iovec *iov = iovstack;
 698         ssize_t ret;
 699         io_fn_t fn;
 700         iov_fn_t fnv;
 701
 702         if (!file->f_op) {
 703                 ret = -EINVAL;
 704                 goto out;
 705         }
 706
 707         ret = rw_copy_check_uvector(type, uvector, nr_segs,
 708                         ARRAY_SIZE(iovstack), iovstack, &iov);
 709         if (ret <= 0)
 710                 goto out;
 711
 712         tot_len = ret;
 713         ret = rw_verify_area(type, file, pos, tot_len);
 714         if (ret < 0)
 715                 goto out;
 716
 717         fnv = NULL;
 718         if (type == READ) {
 719                 fn = file->f_op->read;
 720                 fnv = file->f_op->aio_read;
 721         } else {
 722                 fn = (io_fn_t)file->f_op->write;
 723                 fnv = file->f_op->aio_write;
 724         }
 725
 726         if (fnv)
 727                 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
 728                                                 pos, fnv);
 729         else
 730                 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
 731
 732 out:
 733         if (iov != iovstack)
 734                 kfree(iov);
 735         if ((ret + (type == READ)) > 0) {
 736                 if (type == READ)
 737                         fsnotify_access(file);
 738                 else
 739                         fsnotify_modify(file);
 740         }
 741         return ret;
 742 }
 743
 744 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
 745                   unsigned long vlen, loff_t *pos)
 746 {
 747         if (!(file->f_mode & FMODE_READ))
 748                 return -EBADF;
 749         if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
 750                 return -EINVAL;
 751
 752         return do_readv_writev(READ, file, vec, vlen, pos);
 753 }
 754
 755 EXPORT_SYMBOL(vfs_readv);
 756
 757 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 758                    unsigned long vlen, loff_t *pos)
 759 {
 760         if (!(file->f_mode & FMODE_WRITE))
 761                 return -EBADF;
 762         if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
 763                 return -EINVAL;
 764
 765         return do_readv_writev(WRITE, file, vec, vlen, pos);
 766 }
 767
 768 EXPORT_SYMBOL(vfs_writev);
 769
 770 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
 771                 unsigned long, vlen)
 772 {
 773         struct file *file;
 774         ssize_t ret = -EBADF;
 775         int fput_needed;
 776
 777         file = fget_light(fd, &fput_needed);
 778         if (file) {
 779                 loff_t pos = file_pos_read(file);
 780                 ret = vfs_readv(file, vec, vlen, &pos);
 781                 file_pos_write(file, pos);
 782                 fput_light(file, fput_needed);
 783         }
 784
 785         if (ret > 0)
 786                 add_rchar(current, ret);
 787         inc_syscr(current);
 788         return ret;
 789 }
 790
 791 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
 792                 unsigned long, vlen)
 793 {
 794         struct file *file;
 795         ssize_t ret = -EBADF;
 796         int fput_needed;
 797
 798         file = fget_light(fd, &fput_needed);
 799         if (file) {
 800                 loff_t pos = file_pos_read(file);
 801                 ret = vfs_writev(file, vec, vlen, &pos);
 802                 file_pos_write(file, pos);
 803                 fput_light(file, fput_needed);
 804         }
 805
 806         if (ret > 0)
 807                 add_wchar(current, ret);
 808         inc_syscw(current);
 809         return ret;
 810 }
 811
 812 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
 813 {
 814 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
 815         return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
 816 }
 817
 818 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
 819                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
 820 {
 821         loff_t pos = pos_from_hilo(pos_h, pos_l);
 822         struct file *file;
 823         ssize_t ret = -EBADF;
 824         int fput_needed;
 825
 826         if (pos < 0)
 827                 return -EINVAL;
 828
 829         file = fget_light(fd, &fput_needed);
 830         if (file) {
 831                 ret = -ESPIPE;
 832                 if (file->f_mode & FMODE_PREAD)
 833                         ret = vfs_readv(file, vec, vlen, &pos);
 834                 fput_light(file, fput_needed);
 835         }
 836
 837         if (ret > 0)
 838                 add_rchar(current, ret);
 839         inc_syscr(current);
 840         return ret;
 841 }
 842
 843 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
 844                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
 845 {
 846         loff_t pos = pos_from_hilo(pos_h, pos_l);
 847         struct file *file;
 848         ssize_t ret = -EBADF;
 849         int fput_needed;
 850
 851         if (pos < 0)
 852                 return -EINVAL;
 853
 854         file = fget_light(fd, &fput_needed);
 855         if (file) {
 856                 ret = -ESPIPE;
 857                 if (file->f_mode & FMODE_PWRITE)
 858                         ret = vfs_writev(file, vec, vlen, &pos);
 859                 fput_light(file, fput_needed);
 860         }
 861
 862         if (ret > 0)
 863                 add_wchar(current, ret);
 864         inc_syscw(current);
 865         return ret;
 866 }
 867
 868 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
 869                            size_t count, loff_t max)
 870 {
 871         struct file * in_file, * out_file;
 872         struct inode * in_inode, * out_inode;
 873         loff_t pos;
 874         ssize_t retval;
 875         int fput_needed_in, fput_needed_out, fl;
 876
 877         /*
 878          * Get input file, and verify that it is ok..
 879          */
 880         retval = -EBADF;
 881         in_file = fget_light(in_fd, &fput_needed_in);
 882         if (!in_file)
 883                 goto out;
 884         if (!(in_file->f_mode & FMODE_READ))
 885                 goto fput_in;
 886         retval = -ESPIPE;
 887         if (!ppos)
 888                 ppos = &in_file->f_pos;
 889         else
 890                 if (!(in_file->f_mode & FMODE_PREAD))
 891                         goto fput_in;
 892         retval = rw_verify_area(READ, in_file, ppos, count);
 893         if (retval < 0)
 894                 goto fput_in;
 895         count = retval;
 896
 897         /*
 898          * Get output file, and verify that it is ok..
 899          */
 900         retval = -EBADF;
 901         out_file = fget_light(out_fd, &fput_needed_out);
 902         if (!out_file)
 903                 goto fput_in;
 904         if (!(out_file->f_mode & FMODE_WRITE))
 905                 goto fput_out;
 906         retval = -EINVAL;
 907         in_inode = in_file->f_path.dentry->d_inode;
 908         out_inode = out_file->f_path.dentry->d_inode;
 909         retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
 910         if (retval < 0)
 911                 goto fput_out;
 912         count = retval;
 913
 914         if (!max)
 915                 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
 916
 917         pos = *ppos;
 918         if (unlikely(pos + count > max)) {
 919                 retval = -EOVERFLOW;
 920                 if (pos >= max)
 921                         goto fput_out;
 922                 count = max - pos;
 923         }
 924
 925         fl = 0;
 926 #if 0
 927         /*
 928          * We need to debate whether we can enable this or not. The
 929          * man page documents EAGAIN return for the output at least,
 930          * and the application is arguably buggy if it doesn't expect
 931          * EAGAIN on a non-blocking file descriptor.
 932          */
 933         if (in_file->f_flags & O_NONBLOCK)
 934                 fl = SPLICE_F_NONBLOCK;
 935 #endif
 936         retval = do_splice_direct(in_file, ppos, out_file, count, fl);
 937
 938         if (retval > 0) {
 939                 add_rchar(current, retval);
 940                 add_wchar(current, retval);
 941         }
 942
 943         inc_syscr(current);
 944         inc_syscw(current);
 945         if (*ppos > max)
 946                 retval = -EOVERFLOW;
 947
 948 fput_out:
 949         fput_light(out_file, fput_needed_out);
 950 fput_in:
 951         fput_light(in_file, fput_needed_in);
 952 out:
 953         return retval;
 954 }
 955
 956 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
 957 {
 958         loff_t pos;
 959         off_t off;
 960         ssize_t ret;
 961
 962         if (offset) {
 963                 if (unlikely(get_user(off, offset)))
 964                         return -EFAULT;
 965                 pos = off;
 966                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
 967                 if (unlikely(put_user(pos, offset)))
 968                         return -EFAULT;
 969                 return ret;
 970         }
 971
 972         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 973 }
 974
 975 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
 976 {
 977         loff_t pos;
 978         ssize_t ret;
 979
 980         if (offset) {
 981                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
 982                         return -EFAULT;
 983                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
 984                 if (unlikely(put_user(pos, offset)))
 985                         return -EFAULT;
 986                 return ret;
 987         }
 988
 989         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 990 }