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/export.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 static loff_t lseek_execute(struct file *file, struct inode *inode,
  39                 loff_t offset, loff_t maxsize)
  40 {
  41         if (offset < 0 && !unsigned_offsets(file))
  42                 return -EINVAL;
  43         if (offset > maxsize)
  44                 return -EINVAL;
  45
  46         if (offset != file->f_pos) {
  47                 file->f_pos = offset;
  48                 file->f_version = 0;
  49         }
  50         return offset;
  51 }
  52
  53 /**
  54  * generic_file_llseek_size - generic llseek implementation for regular files
  55  * @file:       file structure to seek on
  56  * @offset:     file offset to seek to
  57  * @origin:     type of seek
  58  * @size:       max size of file system
  59  *
  60  * This is a variant of generic_file_llseek that allows passing in a custom
  61  * file size.
  62  *
  63  * Synchronization:
  64  * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
  65  * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
  66  * read/writes behave like SEEK_SET against seeks.
  67  */
  68 loff_t
  69 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
  70                 loff_t maxsize)
  71 {
  72         struct inode *inode = file->f_mapping->host;
  73
  74         switch (origin) {
  75         case SEEK_END:
  76                 offset += i_size_read(inode);
  77                 break;
  78         case SEEK_CUR:
  79                 /*
  80                  * Here we special-case the lseek(fd, 0, SEEK_CUR)
  81                  * position-querying operation.  Avoid rewriting the "same"
  82                  * f_pos value back to the file because a concurrent read(),
  83                  * write() or lseek() might have altered it
  84                  */
  85                 if (offset == 0)
  86                         return file->f_pos;
  87                 /*
  88                  * f_lock protects against read/modify/write race with other
  89                  * SEEK_CURs. Note that parallel writes and reads behave
  90                  * like SEEK_SET.
  91                  */
  92                 spin_lock(&file->f_lock);
  93                 offset = lseek_execute(file, inode, file->f_pos + offset,
  94                                        maxsize);
  95                 spin_unlock(&file->f_lock);
  96                 return offset;
  97         case SEEK_DATA:
  98                 /*
  99                  * In the generic case the entire file is data, so as long as
 100                  * offset isn't at the end of the file then the offset is data.
 101                  */
 102                 if (offset >= i_size_read(inode))
 103                         return -ENXIO;
 104                 break;
 105         case SEEK_HOLE:
 106                 /*
 107                  * There is a virtual hole at the end of the file, so as long as
 108                  * offset isn't i_size or larger, return i_size.
 109                  */
 110                 if (offset >= i_size_read(inode))
 111                         return -ENXIO;
 112                 offset = i_size_read(inode);
 113                 break;
 114         }
 115
 116         return lseek_execute(file, inode, offset, maxsize);
 117 }
 118 EXPORT_SYMBOL(generic_file_llseek_size);
 119
 120 /**
 121  * generic_file_llseek - generic llseek implementation for regular files
 122  * @file:       file structure to seek on
 123  * @offset:     file offset to seek to
 124  * @origin:     type of seek
 125  *
 126  * This is a generic implemenation of ->llseek useable for all normal local
 127  * filesystems.  It just updates the file offset to the value specified by
 128  * @offset and @origin under i_mutex.
 129  */
 130 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
 131 {
 132         struct inode *inode = file->f_mapping->host;
 133
 134         return generic_file_llseek_size(file, offset, origin,
 135                                         inode->i_sb->s_maxbytes);
 136 }
 137 EXPORT_SYMBOL(generic_file_llseek);
 138
 139 /**
 140  * noop_llseek - No Operation Performed llseek implementation
 141  * @file:       file structure to seek on
 142  * @offset:     file offset to seek to
 143  * @origin:     type of seek
 144  *
 145  * This is an implementation of ->llseek useable for the rare special case when
 146  * userspace expects the seek to succeed but the (device) file is actually not
 147  * able to perform the seek. In this case you use noop_llseek() instead of
 148  * falling back to the default implementation of ->llseek.
 149  */
 150 loff_t noop_llseek(struct file *file, loff_t offset, int origin)
 151 {
 152         return file->f_pos;
 153 }
 154 EXPORT_SYMBOL(noop_llseek);
 155
 156 loff_t no_llseek(struct file *file, loff_t offset, int origin)
 157 {
 158         return -ESPIPE;
 159 }
 160 EXPORT_SYMBOL(no_llseek);
 161
 162 loff_t default_llseek(struct file *file, loff_t offset, int origin)
 163 {
 164         struct inode *inode = file->f_path.dentry->d_inode;
 165         loff_t retval;
 166
 167         mutex_lock(&inode->i_mutex);
 168         switch (origin) {
 169                 case SEEK_END:
 170                         offset += i_size_read(inode);
 171                         break;
 172                 case SEEK_CUR:
 173                         if (offset == 0) {
 174                                 retval = file->f_pos;
 175                                 goto out;
 176                         }
 177                         offset += file->f_pos;
 178                         break;
 179                 case SEEK_DATA:
 180                         /*
 181                          * In the generic case the entire file is data, so as
 182                          * long as offset isn't at the end of the file then the
 183                          * offset is data.
 184                          */
 185                         if (offset >= inode->i_size) {
 186                                 retval = -ENXIO;
 187                                 goto out;
 188                         }
 189                         break;
 190                 case SEEK_HOLE:
 191                         /*
 192                          * There is a virtual hole at the end of the file, so
 193                          * as long as offset isn't i_size or larger, return
 194                          * i_size.
 195                          */
 196                         if (offset >= inode->i_size) {
 197                                 retval = -ENXIO;
 198                                 goto out;
 199                         }
 200                         offset = inode->i_size;
 201                         break;
 202         }
 203         retval = -EINVAL;
 204         if (offset >= 0 || unsigned_offsets(file)) {
 205                 if (offset != file->f_pos) {
 206                         file->f_pos = offset;
 207                         file->f_version = 0;
 208                 }
 209                 retval = offset;
 210         }
 211 out:
 212         mutex_unlock(&inode->i_mutex);
 213         return retval;
 214 }
 215 EXPORT_SYMBOL(default_llseek);
 216
 217 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
 218 {
 219         loff_t (*fn)(struct file *, loff_t, int);
 220
 221         fn = no_llseek;
 222         if (file->f_mode & FMODE_LSEEK) {
 223                 if (file->f_op && file->f_op->llseek)
 224                         fn = file->f_op->llseek;
 225         }
 226         return fn(file, offset, origin);
 227 }
 228 EXPORT_SYMBOL(vfs_llseek);
 229
 230 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
 231 {
 232         off_t retval;
 233         struct file * file;
 234         int fput_needed;
 235
 236         retval = -EBADF;
 237         file = fget_light(fd, &fput_needed);
 238         if (!file)
 239                 goto bad;
 240
 241         retval = -EINVAL;
 242         if (origin <= SEEK_MAX) {
 243                 loff_t res = vfs_llseek(file, offset, origin);
 244                 retval = res;
 245                 if (res != (loff_t)retval)
 246                         retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
 247         }
 248         fput_light(file, fput_needed);
 249 bad:
 250         return retval;
 251 }
 252
 253 #ifdef __ARCH_WANT_SYS_LLSEEK
 254 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
 255                 unsigned long, offset_low, loff_t __user *, result,
 256                 unsigned int, origin)
 257 {
 258         int retval;
 259         struct file * file;
 260         loff_t offset;
 261         int fput_needed;
 262
 263         retval = -EBADF;
 264         file = fget_light(fd, &fput_needed);
 265         if (!file)
 266                 goto bad;
 267
 268         retval = -EINVAL;
 269         if (origin > SEEK_MAX)
 270                 goto out_putf;
 271
 272         offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
 273                         origin);
 274
 275         retval = (int)offset;
 276         if (offset >= 0) {
 277                 retval = -EFAULT;
 278                 if (!copy_to_user(result, &offset, sizeof(offset)))
 279                         retval = 0;
 280         }
 281 out_putf:
 282         fput_light(file, fput_needed);
 283 bad:
 284         return retval;
 285 }
 286 #endif
 287
 288
 289 /*
 290  * rw_verify_area doesn't like huge counts. We limit
 291  * them to something that fits in "int" so that others
 292  * won't have to do range checks all the time.
 293  */
 294 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
 295 {
 296         struct inode *inode;
 297         loff_t pos;
 298         int retval = -EINVAL;
 299
 300         inode = file->f_path.dentry->d_inode;
 301         if (unlikely((ssize_t) count < 0))
 302                 return retval;
 303         pos = *ppos;
 304         if (unlikely(pos < 0)) {
 305                 if (!unsigned_offsets(file))
 306                         return retval;
 307                 if (count >= -pos) /* both values are in 0..LLONG_MAX */
 308                         return -EOVERFLOW;
 309         } else if (unlikely((loff_t) (pos + count) < 0)) {
 310                 if (!unsigned_offsets(file))
 311                         return retval;
 312         }
 313
 314         if (unlikely(inode->i_flock && mandatory_lock(inode))) {
 315                 retval = locks_mandatory_area(
 316                         read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
 317                         inode, file, pos, count);
 318                 if (retval < 0)
 319                         return retval;
 320         }
 321         retval = security_file_permission(file,
 322                                 read_write == READ ? MAY_READ : MAY_WRITE);
 323         if (retval)
 324                 return retval;
 325         return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
 326 }
 327
 328 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
 329 {
 330         set_current_state(TASK_UNINTERRUPTIBLE);
 331         if (!kiocbIsKicked(iocb))
 332                 schedule();
 333         else
 334                 kiocbClearKicked(iocb);
 335         __set_current_state(TASK_RUNNING);
 336 }
 337
 338 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 339 {
 340         struct iovec iov = { .iov_base = buf, .iov_len = len };
 341         struct kiocb kiocb;
 342         ssize_t ret;
 343
 344         init_sync_kiocb(&kiocb, filp);
 345         kiocb.ki_pos = *ppos;
 346         kiocb.ki_left = len;
 347         kiocb.ki_nbytes = len;
 348
 349         for (;;) {
 350                 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
 351                 if (ret != -EIOCBRETRY)
 352                         break;
 353                 wait_on_retry_sync_kiocb(&kiocb);
 354         }
 355
 356         if (-EIOCBQUEUED == ret)
 357                 ret = wait_on_sync_kiocb(&kiocb);
 358         *ppos = kiocb.ki_pos;
 359         return ret;
 360 }
 361
 362 EXPORT_SYMBOL(do_sync_read);
 363
 364 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 365 {
 366         ssize_t ret;
 367
 368         if (!(file->f_mode & FMODE_READ))
 369                 return -EBADF;
 370         if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
 371                 return -EINVAL;
 372         if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
 373                 return -EFAULT;
 374
 375         ret = rw_verify_area(READ, file, pos, count);
 376         if (ret >= 0) {
 377                 count = ret;
 378                 if (file->f_op->read)
 379                         ret = file->f_op->read(file, buf, count, pos);
 380                 else
 381                         ret = do_sync_read(file, buf, count, pos);
 382                 if (ret > 0) {
 383                         fsnotify_access(file);
 384                         add_rchar(current, ret);
 385                 }
 386                 inc_syscr(current);
 387         }
 388
 389         return ret;
 390 }
 391
 392 EXPORT_SYMBOL(vfs_read);
 393
 394 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 395 {
 396         struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
 397         struct kiocb kiocb;
 398         ssize_t ret;
 399
 400         init_sync_kiocb(&kiocb, filp);
 401         kiocb.ki_pos = *ppos;
 402         kiocb.ki_left = len;
 403         kiocb.ki_nbytes = len;
 404
 405         for (;;) {
 406                 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
 407                 if (ret != -EIOCBRETRY)
 408                         break;
 409                 wait_on_retry_sync_kiocb(&kiocb);
 410         }
 411
 412         if (-EIOCBQUEUED == ret)
 413                 ret = wait_on_sync_kiocb(&kiocb);
 414         *ppos = kiocb.ki_pos;
 415         return ret;
 416 }
 417
 418 EXPORT_SYMBOL(do_sync_write);
 419
 420 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 421 {
 422         ssize_t ret;
 423
 424         if (!(file->f_mode & FMODE_WRITE))
 425                 return -EBADF;
 426         if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
 427                 return -EINVAL;
 428         if (unlikely(!access_ok(VERIFY_READ, buf, count)))
 429                 return -EFAULT;
 430
 431         ret = rw_verify_area(WRITE, file, pos, count);
 432         if (ret >= 0) {
 433                 count = ret;
 434                 if (file->f_op->write)
 435                         ret = file->f_op->write(file, buf, count, pos);
 436                 else
 437                         ret = do_sync_write(file, buf, count, pos);
 438                 if (ret > 0) {
 439                         fsnotify_modify(file);
 440                         add_wchar(current, ret);
 441                 }
 442                 inc_syscw(current);
 443         }
 444
 445         return ret;
 446 }
 447
 448 EXPORT_SYMBOL(vfs_write);
 449
 450 static inline loff_t file_pos_read(struct file *file)
 451 {
 452         return file->f_pos;
 453 }
 454
 455 static inline void file_pos_write(struct file *file, loff_t pos)
 456 {
 457         file->f_pos = pos;
 458 }
 459
 460 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
 461 {
 462         struct file *file;
 463         ssize_t ret = -EBADF;
 464         int fput_needed;
 465
 466         file = fget_light(fd, &fput_needed);
 467         if (file) {
 468                 loff_t pos = file_pos_read(file);
 469                 ret = vfs_read(file, buf, count, &pos);
 470                 file_pos_write(file, pos);
 471                 fput_light(file, fput_needed);
 472         }
 473
 474         return ret;
 475 }
 476
 477 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
 478                 size_t, count)
 479 {
 480         struct file *file;
 481         ssize_t ret = -EBADF;
 482         int fput_needed;
 483
 484         file = fget_light(fd, &fput_needed);
 485         if (file) {
 486                 loff_t pos = file_pos_read(file);
 487                 ret = vfs_write(file, buf, count, &pos);
 488                 file_pos_write(file, pos);
 489                 fput_light(file, fput_needed);
 490         }
 491
 492         return ret;
 493 }
 494
 495 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
 496                         size_t count, loff_t pos)
 497 {
 498         struct file *file;
 499         ssize_t ret = -EBADF;
 500         int fput_needed;
 501
 502         if (pos < 0)
 503                 return -EINVAL;
 504
 505         file = fget_light(fd, &fput_needed);
 506         if (file) {
 507                 ret = -ESPIPE;
 508                 if (file->f_mode & FMODE_PREAD)
 509                         ret = vfs_read(file, buf, count, &pos);
 510                 fput_light(file, fput_needed);
 511         }
 512
 513         return ret;
 514 }
 515 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 516 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
 517 {
 518         return SYSC_pread64((unsigned int) fd, (char __user *) buf,
 519                             (size_t) count, pos);
 520 }
 521 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
 522 #endif
 523
 524 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
 525                          size_t count, loff_t pos)
 526 {
 527         struct file *file;
 528         ssize_t ret = -EBADF;
 529         int fput_needed;
 530
 531         if (pos < 0)
 532                 return -EINVAL;
 533
 534         file = fget_light(fd, &fput_needed);
 535         if (file) {
 536                 ret = -ESPIPE;
 537                 if (file->f_mode & FMODE_PWRITE)
 538                         ret = vfs_write(file, buf, count, &pos);
 539                 fput_light(file, fput_needed);
 540         }
 541
 542         return ret;
 543 }
 544 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 545 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
 546 {
 547         return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
 548                              (size_t) count, pos);
 549 }
 550 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
 551 #endif
 552
 553 /*
 554  * Reduce an iovec's length in-place.  Return the resulting number of segments
 555  */
 556 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
 557 {
 558         unsigned long seg = 0;
 559         size_t len = 0;
 560
 561         while (seg < nr_segs) {
 562                 seg++;
 563                 if (len + iov->iov_len >= to) {
 564                         iov->iov_len = to - len;
 565                         break;
 566                 }
 567                 len += iov->iov_len;
 568                 iov++;
 569         }
 570         return seg;
 571 }
 572 EXPORT_SYMBOL(iov_shorten);
 573
 574 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
 575                 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
 576 {
 577         struct kiocb kiocb;
 578         ssize_t ret;
 579
 580         init_sync_kiocb(&kiocb, filp);
 581         kiocb.ki_pos = *ppos;
 582         kiocb.ki_left = len;
 583         kiocb.ki_nbytes = len;
 584
 585         for (;;) {
 586                 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
 587                 if (ret != -EIOCBRETRY)
 588                         break;
 589                 wait_on_retry_sync_kiocb(&kiocb);
 590         }
 591
 592         if (ret == -EIOCBQUEUED)
 593                 ret = wait_on_sync_kiocb(&kiocb);
 594         *ppos = kiocb.ki_pos;
 595         return ret;
 596 }
 597
 598 /* Do it by hand, with file-ops */
 599 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
 600                 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
 601 {
 602         struct iovec *vector = iov;
 603         ssize_t ret = 0;
 604
 605         while (nr_segs > 0) {
 606                 void __user *base;
 607                 size_t len;
 608                 ssize_t nr;
 609
 610                 base = vector->iov_base;
 611                 len = vector->iov_len;
 612                 vector++;
 613                 nr_segs--;
 614
 615                 nr = fn(filp, base, len, ppos);
 616
 617                 if (nr < 0) {
 618                         if (!ret)
 619                                 ret = nr;
 620                         break;
 621                 }
 622                 ret += nr;
 623                 if (nr != len)
 624                         break;
 625         }
 626
 627         return ret;
 628 }
 629
 630 /* A write operation does a read from user space and vice versa */
 631 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 632
 633 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
 634                               unsigned long nr_segs, unsigned long fast_segs,
 635                               struct iovec *fast_pointer,
 636                               struct iovec **ret_pointer,
 637                               int check_access)
 638 {
 639         unsigned long seg;
 640         ssize_t ret;
 641         struct iovec *iov = fast_pointer;
 642
 643         /*
 644          * SuS says "The readv() function *may* fail if the iovcnt argument
 645          * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 646          * traditionally returned zero for zero segments, so...
 647          */
 648         if (nr_segs == 0) {
 649                 ret = 0;
 650                 goto out;
 651         }
 652
 653         /*
 654          * First get the "struct iovec" from user memory and
 655          * verify all the pointers
 656          */
 657         if (nr_segs > UIO_MAXIOV) {
 658                 ret = -EINVAL;
 659                 goto out;
 660         }
 661         if (nr_segs > fast_segs) {
 662                 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 663                 if (iov == NULL) {
 664                         ret = -ENOMEM;
 665                         goto out;
 666                 }
 667         }
 668         if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
 669                 ret = -EFAULT;
 670                 goto out;
 671         }
 672
 673         /*
 674          * According to the Single Unix Specification we should return EINVAL
 675          * if an element length is < 0 when cast to ssize_t or if the
 676          * total length would overflow the ssize_t return value of the
 677          * system call.
 678          *
 679          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
 680          * overflow case.
 681          */
 682         ret = 0;
 683         for (seg = 0; seg < nr_segs; seg++) {
 684                 void __user *buf = iov[seg].iov_base;
 685                 ssize_t len = (ssize_t)iov[seg].iov_len;
 686
 687                 /* see if we we're about to use an invalid len or if
 688                  * it's about to overflow ssize_t */
 689                 if (len < 0) {
 690                         ret = -EINVAL;
 691                         goto out;
 692                 }
 693                 if (check_access
 694                     && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
 695                         ret = -EFAULT;
 696                         goto out;
 697                 }
 698                 if (len > MAX_RW_COUNT - ret) {
 699                         len = MAX_RW_COUNT - ret;
 700                         iov[seg].iov_len = len;
 701                 }
 702                 ret += len;
 703         }
 704 out:
 705         *ret_pointer = iov;
 706         return ret;
 707 }
 708
 709 static ssize_t do_readv_writev(int type, struct file *file,
 710                                const struct iovec __user * uvector,
 711                                unsigned long nr_segs, loff_t *pos)
 712 {
 713         size_t tot_len;
 714         struct iovec iovstack[UIO_FASTIOV];
 715         struct iovec *iov = iovstack;
 716         ssize_t ret;
 717         io_fn_t fn;
 718         iov_fn_t fnv;
 719
 720         if (!file->f_op) {
 721                 ret = -EINVAL;
 722                 goto out;
 723         }
 724
 725         ret = rw_copy_check_uvector(type, uvector, nr_segs,
 726                                     ARRAY_SIZE(iovstack), iovstack, &iov, 1);
 727         if (ret <= 0)
 728                 goto out;
 729
 730         tot_len = ret;
 731         ret = rw_verify_area(type, file, pos, tot_len);
 732         if (ret < 0)
 733                 goto out;
 734
 735         fnv = NULL;
 736         if (type == READ) {
 737                 fn = file->f_op->read;
 738                 fnv = file->f_op->aio_read;
 739         } else {
 740                 fn = (io_fn_t)file->f_op->write;
 741                 fnv = file->f_op->aio_write;
 742         }
 743
 744         if (fnv)
 745                 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
 746                                                 pos, fnv);
 747         else
 748                 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
 749
 750 out:
 751         if (iov != iovstack)
 752                 kfree(iov);
 753         if ((ret + (type == READ)) > 0) {
 754                 if (type == READ)
 755                         fsnotify_access(file);
 756                 else
 757                         fsnotify_modify(file);
 758         }
 759         return ret;
 760 }
 761
 762 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
 763                   unsigned long vlen, loff_t *pos)
 764 {
 765         if (!(file->f_mode & FMODE_READ))
 766                 return -EBADF;
 767         if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
 768                 return -EINVAL;
 769
 770         return do_readv_writev(READ, file, vec, vlen, pos);
 771 }
 772
 773 EXPORT_SYMBOL(vfs_readv);
 774
 775 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 776                    unsigned long vlen, loff_t *pos)
 777 {
 778         if (!(file->f_mode & FMODE_WRITE))
 779                 return -EBADF;
 780         if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
 781                 return -EINVAL;
 782
 783         return do_readv_writev(WRITE, file, vec, vlen, pos);
 784 }
 785
 786 EXPORT_SYMBOL(vfs_writev);
 787
 788 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
 789                 unsigned long, vlen)
 790 {
 791         struct file *file;
 792         ssize_t ret = -EBADF;
 793         int fput_needed;
 794
 795         file = fget_light(fd, &fput_needed);
 796         if (file) {
 797                 loff_t pos = file_pos_read(file);
 798                 ret = vfs_readv(file, vec, vlen, &pos);
 799                 file_pos_write(file, pos);
 800                 fput_light(file, fput_needed);
 801         }
 802
 803         if (ret > 0)
 804                 add_rchar(current, ret);
 805         inc_syscr(current);
 806         return ret;
 807 }
 808
 809 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
 810                 unsigned long, vlen)
 811 {
 812         struct file *file;
 813         ssize_t ret = -EBADF;
 814         int fput_needed;
 815
 816         file = fget_light(fd, &fput_needed);
 817         if (file) {
 818                 loff_t pos = file_pos_read(file);
 819                 ret = vfs_writev(file, vec, vlen, &pos);
 820                 file_pos_write(file, pos);
 821                 fput_light(file, fput_needed);
 822         }
 823
 824         if (ret > 0)
 825                 add_wchar(current, ret);
 826         inc_syscw(current);
 827         return ret;
 828 }
 829
 830 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
 831 {
 832 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
 833         return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
 834 }
 835
 836 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
 837                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
 838 {
 839         loff_t pos = pos_from_hilo(pos_h, pos_l);
 840         struct file *file;
 841         ssize_t ret = -EBADF;
 842         int fput_needed;
 843
 844         if (pos < 0)
 845                 return -EINVAL;
 846
 847         file = fget_light(fd, &fput_needed);
 848         if (file) {
 849                 ret = -ESPIPE;
 850                 if (file->f_mode & FMODE_PREAD)
 851                         ret = vfs_readv(file, vec, vlen, &pos);
 852                 fput_light(file, fput_needed);
 853         }
 854
 855         if (ret > 0)
 856                 add_rchar(current, ret);
 857         inc_syscr(current);
 858         return ret;
 859 }
 860
 861 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
 862                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
 863 {
 864         loff_t pos = pos_from_hilo(pos_h, pos_l);
 865         struct file *file;
 866         ssize_t ret = -EBADF;
 867         int fput_needed;
 868
 869         if (pos < 0)
 870                 return -EINVAL;
 871
 872         file = fget_light(fd, &fput_needed);
 873         if (file) {
 874                 ret = -ESPIPE;
 875                 if (file->f_mode & FMODE_PWRITE)
 876                         ret = vfs_writev(file, vec, vlen, &pos);
 877                 fput_light(file, fput_needed);
 878         }
 879
 880         if (ret > 0)
 881                 add_wchar(current, ret);
 882         inc_syscw(current);
 883         return ret;
 884 }
 885
 886 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
 887                            size_t count, loff_t max)
 888 {
 889         struct file * in_file, * out_file;
 890         struct inode * in_inode, * out_inode;
 891         loff_t pos;
 892         ssize_t retval;
 893         int fput_needed_in, fput_needed_out, fl;
 894
 895         /*
 896          * Get input file, and verify that it is ok..
 897          */
 898         retval = -EBADF;
 899         in_file = fget_light(in_fd, &fput_needed_in);
 900         if (!in_file)
 901                 goto out;
 902         if (!(in_file->f_mode & FMODE_READ))
 903                 goto fput_in;
 904         retval = -ESPIPE;
 905         if (!ppos)
 906                 ppos = &in_file->f_pos;
 907         else
 908                 if (!(in_file->f_mode & FMODE_PREAD))
 909                         goto fput_in;
 910         retval = rw_verify_area(READ, in_file, ppos, count);
 911         if (retval < 0)
 912                 goto fput_in;
 913         count = retval;
 914
 915         /*
 916          * Get output file, and verify that it is ok..
 917          */
 918         retval = -EBADF;
 919         out_file = fget_light(out_fd, &fput_needed_out);
 920         if (!out_file)
 921                 goto fput_in;
 922         if (!(out_file->f_mode & FMODE_WRITE))
 923                 goto fput_out;
 924         retval = -EINVAL;
 925         in_inode = in_file->f_path.dentry->d_inode;
 926         out_inode = out_file->f_path.dentry->d_inode;
 927         retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
 928         if (retval < 0)
 929                 goto fput_out;
 930         count = retval;
 931
 932         if (!max)
 933                 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
 934
 935         pos = *ppos;
 936         if (unlikely(pos + count > max)) {
 937                 retval = -EOVERFLOW;
 938                 if (pos >= max)
 939                         goto fput_out;
 940                 count = max - pos;
 941         }
 942
 943         fl = 0;
 944 #if 0
 945         /*
 946          * We need to debate whether we can enable this or not. The
 947          * man page documents EAGAIN return for the output at least,
 948          * and the application is arguably buggy if it doesn't expect
 949          * EAGAIN on a non-blocking file descriptor.
 950          */
 951         if (in_file->f_flags & O_NONBLOCK)
 952                 fl = SPLICE_F_NONBLOCK;
 953 #endif
 954         retval = do_splice_direct(in_file, ppos, out_file, count, fl);
 955
 956         if (retval > 0) {
 957                 add_rchar(current, retval);
 958                 add_wchar(current, retval);
 959         }
 960
 961         inc_syscr(current);
 962         inc_syscw(current);
 963         if (*ppos > max)
 964                 retval = -EOVERFLOW;
 965
 966 fput_out:
 967         fput_light(out_file, fput_needed_out);
 968 fput_in:
 969         fput_light(in_file, fput_needed_in);
 970 out:
 971         return retval;
 972 }
 973
 974 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
 975 {
 976         loff_t pos;
 977         off_t off;
 978         ssize_t ret;
 979
 980         if (offset) {
 981                 if (unlikely(get_user(off, offset)))
 982                         return -EFAULT;
 983                 pos = off;
 984                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
 985                 if (unlikely(put_user(pos, offset)))
 986                         return -EFAULT;
 987                 return ret;
 988         }
 989
 990         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 991 }
 992
 993 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
 994 {
 995         loff_t pos;
 996         ssize_t ret;
 997
 998         if (offset) {
 999                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1000                         return -EFAULT;
1001                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1002                 if (unlikely(put_user(pos, offset)))
1003                         return -EFAULT;
1004                 return ret;
1005         }
1006
1007         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1008 }