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 <linux/compat.h>
  19 #include "internal.h"
  20
  21 #include <asm/uaccess.h>
  22 #include <asm/unistd.h>
  23
  24 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
  25 typedef ssize_t (*iov_fn_t)(struct kiocb *, const struct iovec *,
  26                 unsigned long, loff_t);
  27 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
  28
  29 const struct file_operations generic_ro_fops = {
  30         .llseek         = generic_file_llseek,
  31         .read           = new_sync_read,
  32         .read_iter      = generic_file_read_iter,
  33         .mmap           = generic_file_readonly_mmap,
  34         .splice_read    = generic_file_splice_read,
  35 };
  36
  37 EXPORT_SYMBOL(generic_ro_fops);
  38
  39 static inline int unsigned_offsets(struct file *file)
  40 {
  41         return file->f_mode & FMODE_UNSIGNED_OFFSET;
  42 }
  43
  44 /**
  45  * vfs_setpos - update the file offset for lseek
  46  * @file:       file structure in question
  47  * @offset:     file offset to seek to
  48  * @maxsize:    maximum file size
  49  *
  50  * This is a low-level filesystem helper for updating the file offset to
  51  * the value specified by @offset if the given offset is valid and it is
  52  * not equal to the current file offset.
  53  *
  54  * Return the specified offset on success and -EINVAL on invalid offset.
  55  */
  56 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
  57 {
  58         if (offset < 0 && !unsigned_offsets(file))
  59                 return -EINVAL;
  60         if (offset > maxsize)
  61                 return -EINVAL;
  62
  63         if (offset != file->f_pos) {
  64                 file->f_pos = offset;
  65                 file->f_version = 0;
  66         }
  67         return offset;
  68 }
  69 EXPORT_SYMBOL(vfs_setpos);
  70
  71 /**
  72  * generic_file_llseek_size - generic llseek implementation for regular files
  73  * @file:       file structure to seek on
  74  * @offset:     file offset to seek to
  75  * @whence:     type of seek
  76  * @size:       max size of this file in file system
  77  * @eof:        offset used for SEEK_END position
  78  *
  79  * This is a variant of generic_file_llseek that allows passing in a custom
  80  * maximum file size and a custom EOF position, for e.g. hashed directories
  81  *
  82  * Synchronization:
  83  * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
  84  * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
  85  * read/writes behave like SEEK_SET against seeks.
  86  */
  87 loff_t
  88 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
  89                 loff_t maxsize, loff_t eof)
  90 {
  91         switch (whence) {
  92         case SEEK_END:
  93                 offset += eof;
  94                 break;
  95         case SEEK_CUR:
  96                 /*
  97                  * Here we special-case the lseek(fd, 0, SEEK_CUR)
  98                  * position-querying operation.  Avoid rewriting the "same"
  99                  * f_pos value back to the file because a concurrent read(),
 100                  * write() or lseek() might have altered it
 101                  */
 102                 if (offset == 0)
 103                         return file->f_pos;
 104                 /*
 105                  * f_lock protects against read/modify/write race with other
 106                  * SEEK_CURs. Note that parallel writes and reads behave
 107                  * like SEEK_SET.
 108                  */
 109                 spin_lock(&file->f_lock);
 110                 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
 111                 spin_unlock(&file->f_lock);
 112                 return offset;
 113         case SEEK_DATA:
 114                 /*
 115                  * In the generic case the entire file is data, so as long as
 116                  * offset isn't at the end of the file then the offset is data.
 117                  */
 118                 if (offset >= eof)
 119                         return -ENXIO;
 120                 break;
 121         case SEEK_HOLE:
 122                 /*
 123                  * There is a virtual hole at the end of the file, so as long as
 124                  * offset isn't i_size or larger, return i_size.
 125                  */
 126                 if (offset >= eof)
 127                         return -ENXIO;
 128                 offset = eof;
 129                 break;
 130         }
 131
 132         return vfs_setpos(file, offset, maxsize);
 133 }
 134 EXPORT_SYMBOL(generic_file_llseek_size);
 135
 136 /**
 137  * generic_file_llseek - generic llseek implementation for regular files
 138  * @file:       file structure to seek on
 139  * @offset:     file offset to seek to
 140  * @whence:     type of seek
 141  *
 142  * This is a generic implemenation of ->llseek useable for all normal local
 143  * filesystems.  It just updates the file offset to the value specified by
 144  * @offset and @whence.
 145  */
 146 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
 147 {
 148         struct inode *inode = file->f_mapping->host;
 149
 150         return generic_file_llseek_size(file, offset, whence,
 151                                         inode->i_sb->s_maxbytes,
 152                                         i_size_read(inode));
 153 }
 154 EXPORT_SYMBOL(generic_file_llseek);
 155
 156 /**
 157  * fixed_size_llseek - llseek implementation for fixed-sized devices
 158  * @file:       file structure to seek on
 159  * @offset:     file offset to seek to
 160  * @whence:     type of seek
 161  * @size:       size of the file
 162  *
 163  */
 164 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
 165 {
 166         switch (whence) {
 167         case SEEK_SET: case SEEK_CUR: case SEEK_END:
 168                 return generic_file_llseek_size(file, offset, whence,
 169                                                 size, size);
 170         default:
 171                 return -EINVAL;
 172         }
 173 }
 174 EXPORT_SYMBOL(fixed_size_llseek);
 175
 176 /**
 177  * noop_llseek - No Operation Performed llseek implementation
 178  * @file:       file structure to seek on
 179  * @offset:     file offset to seek to
 180  * @whence:     type of seek
 181  *
 182  * This is an implementation of ->llseek useable for the rare special case when
 183  * userspace expects the seek to succeed but the (device) file is actually not
 184  * able to perform the seek. In this case you use noop_llseek() instead of
 185  * falling back to the default implementation of ->llseek.
 186  */
 187 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
 188 {
 189         return file->f_pos;
 190 }
 191 EXPORT_SYMBOL(noop_llseek);
 192
 193 loff_t no_llseek(struct file *file, loff_t offset, int whence)
 194 {
 195         return -ESPIPE;
 196 }
 197 EXPORT_SYMBOL(no_llseek);
 198
 199 loff_t default_llseek(struct file *file, loff_t offset, int whence)
 200 {
 201         struct inode *inode = file_inode(file);
 202         loff_t retval;
 203
 204         mutex_lock(&inode->i_mutex);
 205         switch (whence) {
 206                 case SEEK_END:
 207                         offset += i_size_read(inode);
 208                         break;
 209                 case SEEK_CUR:
 210                         if (offset == 0) {
 211                                 retval = file->f_pos;
 212                                 goto out;
 213                         }
 214                         offset += file->f_pos;
 215                         break;
 216                 case SEEK_DATA:
 217                         /*
 218                          * In the generic case the entire file is data, so as
 219                          * long as offset isn't at the end of the file then the
 220                          * offset is data.
 221                          */
 222                         if (offset >= inode->i_size) {
 223                                 retval = -ENXIO;
 224                                 goto out;
 225                         }
 226                         break;
 227                 case SEEK_HOLE:
 228                         /*
 229                          * There is a virtual hole at the end of the file, so
 230                          * as long as offset isn't i_size or larger, return
 231                          * i_size.
 232                          */
 233                         if (offset >= inode->i_size) {
 234                                 retval = -ENXIO;
 235                                 goto out;
 236                         }
 237                         offset = inode->i_size;
 238                         break;
 239         }
 240         retval = -EINVAL;
 241         if (offset >= 0 || unsigned_offsets(file)) {
 242                 if (offset != file->f_pos) {
 243                         file->f_pos = offset;
 244                         file->f_version = 0;
 245                 }
 246                 retval = offset;
 247         }
 248 out:
 249         mutex_unlock(&inode->i_mutex);
 250         return retval;
 251 }
 252 EXPORT_SYMBOL(default_llseek);
 253
 254 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
 255 {
 256         loff_t (*fn)(struct file *, loff_t, int);
 257
 258         fn = no_llseek;
 259         if (file->f_mode & FMODE_LSEEK) {
 260                 if (file->f_op->llseek)
 261                         fn = file->f_op->llseek;
 262         }
 263         return fn(file, offset, whence);
 264 }
 265 EXPORT_SYMBOL(vfs_llseek);
 266
 267 static inline struct fd fdget_pos(int fd)
 268 {
 269         return __to_fd(__fdget_pos(fd));
 270 }
 271
 272 static inline void fdput_pos(struct fd f)
 273 {
 274         if (f.flags & FDPUT_POS_UNLOCK)
 275                 mutex_unlock(&f.file->f_pos_lock);
 276         fdput(f);
 277 }
 278
 279 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
 280 {
 281         off_t retval;
 282         struct fd f = fdget_pos(fd);
 283         if (!f.file)
 284                 return -EBADF;
 285
 286         retval = -EINVAL;
 287         if (whence <= SEEK_MAX) {
 288                 loff_t res = vfs_llseek(f.file, offset, whence);
 289                 retval = res;
 290                 if (res != (loff_t)retval)
 291                         retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
 292         }
 293         fdput_pos(f);
 294         return retval;
 295 }
 296
 297 #ifdef CONFIG_COMPAT
 298 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
 299 {
 300         return sys_lseek(fd, offset, whence);
 301 }
 302 #endif
 303
 304 #ifdef __ARCH_WANT_SYS_LLSEEK
 305 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
 306                 unsigned long, offset_low, loff_t __user *, result,
 307                 unsigned int, whence)
 308 {
 309         int retval;
 310         struct fd f = fdget_pos(fd);
 311         loff_t offset;
 312
 313         if (!f.file)
 314                 return -EBADF;
 315
 316         retval = -EINVAL;
 317         if (whence > SEEK_MAX)
 318                 goto out_putf;
 319
 320         offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
 321                         whence);
 322
 323         retval = (int)offset;
 324         if (offset >= 0) {
 325                 retval = -EFAULT;
 326                 if (!copy_to_user(result, &offset, sizeof(offset)))
 327                         retval = 0;
 328         }
 329 out_putf:
 330         fdput_pos(f);
 331         return retval;
 332 }
 333 #endif
 334
 335 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
 336 {
 337         struct kiocb kiocb;
 338         ssize_t ret;
 339
 340         if (!file->f_op->read_iter)
 341                 return -EINVAL;
 342
 343         init_sync_kiocb(&kiocb, file);
 344         kiocb.ki_pos = *ppos;
 345
 346         iter->type |= READ;
 347         ret = file->f_op->read_iter(&kiocb, iter);
 348         BUG_ON(ret == -EIOCBQUEUED);
 349         if (ret > 0)
 350                 *ppos = kiocb.ki_pos;
 351         return ret;
 352 }
 353 EXPORT_SYMBOL(vfs_iter_read);
 354
 355 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
 356 {
 357         struct kiocb kiocb;
 358         ssize_t ret;
 359
 360         if (!file->f_op->write_iter)
 361                 return -EINVAL;
 362
 363         init_sync_kiocb(&kiocb, file);
 364         kiocb.ki_pos = *ppos;
 365
 366         iter->type |= WRITE;
 367         ret = file->f_op->write_iter(&kiocb, iter);
 368         BUG_ON(ret == -EIOCBQUEUED);
 369         if (ret > 0)
 370                 *ppos = kiocb.ki_pos;
 371         return ret;
 372 }
 373 EXPORT_SYMBOL(vfs_iter_write);
 374
 375 /*
 376  * rw_verify_area doesn't like huge counts. We limit
 377  * them to something that fits in "int" so that others
 378  * won't have to do range checks all the time.
 379  */
 380 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
 381 {
 382         struct inode *inode;
 383         loff_t pos;
 384         int retval = -EINVAL;
 385
 386         inode = file_inode(file);
 387         if (unlikely((ssize_t) count < 0))
 388                 return retval;
 389         pos = *ppos;
 390         if (unlikely(pos < 0)) {
 391                 if (!unsigned_offsets(file))
 392                         return retval;
 393                 if (count >= -pos) /* both values are in 0..LLONG_MAX */
 394                         return -EOVERFLOW;
 395         } else if (unlikely((loff_t) (pos + count) < 0)) {
 396                 if (!unsigned_offsets(file))
 397                         return retval;
 398         }
 399
 400         if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
 401                 retval = locks_mandatory_area(
 402                         read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
 403                         inode, file, pos, count);
 404                 if (retval < 0)
 405                         return retval;
 406         }
 407         retval = security_file_permission(file,
 408                                 read_write == READ ? MAY_READ : MAY_WRITE);
 409         if (retval)
 410                 return retval;
 411         return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
 412 }
 413
 414 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 415 {
 416         struct iovec iov = { .iov_base = buf, .iov_len = len };
 417         struct kiocb kiocb;
 418         ssize_t ret;
 419
 420         init_sync_kiocb(&kiocb, filp);
 421         kiocb.ki_pos = *ppos;
 422
 423         ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
 424         BUG_ON(ret == -EIOCBQUEUED);
 425         *ppos = kiocb.ki_pos;
 426         return ret;
 427 }
 428
 429 EXPORT_SYMBOL(do_sync_read);
 430
 431 ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 432 {
 433         struct iovec iov = { .iov_base = buf, .iov_len = len };
 434         struct kiocb kiocb;
 435         struct iov_iter iter;
 436         ssize_t ret;
 437
 438         init_sync_kiocb(&kiocb, filp);
 439         kiocb.ki_pos = *ppos;
 440         iov_iter_init(&iter, READ, &iov, 1, len);
 441
 442         ret = filp->f_op->read_iter(&kiocb, &iter);
 443         BUG_ON(ret == -EIOCBQUEUED);
 444         *ppos = kiocb.ki_pos;
 445         return ret;
 446 }
 447
 448 EXPORT_SYMBOL(new_sync_read);
 449
 450 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
 451                    loff_t *pos)
 452 {
 453         ssize_t ret;
 454
 455         if (file->f_op->read)
 456                 ret = file->f_op->read(file, buf, count, pos);
 457         else if (file->f_op->aio_read)
 458                 ret = do_sync_read(file, buf, count, pos);
 459         else if (file->f_op->read_iter)
 460                 ret = new_sync_read(file, buf, count, pos);
 461         else
 462                 ret = -EINVAL;
 463
 464         return ret;
 465 }
 466
 467 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 468 {
 469         ssize_t ret;
 470
 471         if (!(file->f_mode & FMODE_READ))
 472                 return -EBADF;
 473         if (!(file->f_mode & FMODE_CAN_READ))
 474                 return -EINVAL;
 475         if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
 476                 return -EFAULT;
 477
 478         ret = rw_verify_area(READ, file, pos, count);
 479         if (ret >= 0) {
 480                 count = ret;
 481                 ret = __vfs_read(file, buf, count, pos);
 482                 if (ret > 0) {
 483                         fsnotify_access(file);
 484                         add_rchar(current, ret);
 485                 }
 486                 inc_syscr(current);
 487         }
 488
 489         return ret;
 490 }
 491
 492 EXPORT_SYMBOL(vfs_read);
 493
 494 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 495 {
 496         struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
 497         struct kiocb kiocb;
 498         ssize_t ret;
 499
 500         init_sync_kiocb(&kiocb, filp);
 501         kiocb.ki_pos = *ppos;
 502
 503         ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
 504         BUG_ON(ret == -EIOCBQUEUED);
 505         *ppos = kiocb.ki_pos;
 506         return ret;
 507 }
 508
 509 EXPORT_SYMBOL(do_sync_write);
 510
 511 ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 512 {
 513         struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
 514         struct kiocb kiocb;
 515         struct iov_iter iter;
 516         ssize_t ret;
 517
 518         init_sync_kiocb(&kiocb, filp);
 519         kiocb.ki_pos = *ppos;
 520         iov_iter_init(&iter, WRITE, &iov, 1, len);
 521
 522         ret = filp->f_op->write_iter(&kiocb, &iter);
 523         BUG_ON(ret == -EIOCBQUEUED);
 524         *ppos = kiocb.ki_pos;
 525         return ret;
 526 }
 527
 528 EXPORT_SYMBOL(new_sync_write);
 529
 530 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
 531 {
 532         mm_segment_t old_fs;
 533         const char __user *p;
 534         ssize_t ret;
 535
 536         if (!(file->f_mode & FMODE_CAN_WRITE))
 537                 return -EINVAL;
 538
 539         old_fs = get_fs();
 540         set_fs(get_ds());
 541         p = (__force const char __user *)buf;
 542         if (count > MAX_RW_COUNT)
 543                 count =  MAX_RW_COUNT;
 544         if (file->f_op->write)
 545                 ret = file->f_op->write(file, p, count, pos);
 546         else if (file->f_op->aio_write)
 547                 ret = do_sync_write(file, p, count, pos);
 548         else
 549                 ret = new_sync_write(file, p, count, pos);
 550         set_fs(old_fs);
 551         if (ret > 0) {
 552                 fsnotify_modify(file);
 553                 add_wchar(current, ret);
 554         }
 555         inc_syscw(current);
 556         return ret;
 557 }
 558
 559 EXPORT_SYMBOL(__kernel_write);
 560
 561 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 562 {
 563         ssize_t ret;
 564
 565         if (!(file->f_mode & FMODE_WRITE))
 566                 return -EBADF;
 567         if (!(file->f_mode & FMODE_CAN_WRITE))
 568                 return -EINVAL;
 569         if (unlikely(!access_ok(VERIFY_READ, buf, count)))
 570                 return -EFAULT;
 571
 572         ret = rw_verify_area(WRITE, file, pos, count);
 573         if (ret >= 0) {
 574                 count = ret;
 575                 file_start_write(file);
 576                 if (file->f_op->write)
 577                         ret = file->f_op->write(file, buf, count, pos);
 578                 else if (file->f_op->aio_write)
 579                         ret = do_sync_write(file, buf, count, pos);
 580                 else
 581                         ret = new_sync_write(file, buf, count, pos);
 582                 if (ret > 0) {
 583                         fsnotify_modify(file);
 584                         add_wchar(current, ret);
 585                 }
 586                 inc_syscw(current);
 587                 file_end_write(file);
 588         }
 589
 590         return ret;
 591 }
 592
 593 EXPORT_SYMBOL(vfs_write);
 594
 595 static inline loff_t file_pos_read(struct file *file)
 596 {
 597         return file->f_pos;
 598 }
 599
 600 static inline void file_pos_write(struct file *file, loff_t pos)
 601 {
 602         file->f_pos = pos;
 603 }
 604
 605 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
 606 {
 607         struct fd f = fdget_pos(fd);
 608         ssize_t ret = -EBADF;
 609
 610         if (f.file) {
 611                 loff_t pos = file_pos_read(f.file);
 612                 ret = vfs_read(f.file, buf, count, &pos);
 613                 if (ret >= 0)
 614                         file_pos_write(f.file, pos);
 615                 fdput_pos(f);
 616         }
 617         return ret;
 618 }
 619
 620 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
 621                 size_t, count)
 622 {
 623         struct fd f = fdget_pos(fd);
 624         ssize_t ret = -EBADF;
 625
 626         if (f.file) {
 627                 loff_t pos = file_pos_read(f.file);
 628                 ret = vfs_write(f.file, buf, count, &pos);
 629                 if (ret >= 0)
 630                         file_pos_write(f.file, pos);
 631                 fdput_pos(f);
 632         }
 633
 634         return ret;
 635 }
 636
 637 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
 638                         size_t, count, loff_t, pos)
 639 {
 640         struct fd f;
 641         ssize_t ret = -EBADF;
 642
 643         if (pos < 0)
 644                 return -EINVAL;
 645
 646         f = fdget(fd);
 647         if (f.file) {
 648                 ret = -ESPIPE;
 649                 if (f.file->f_mode & FMODE_PREAD)
 650                         ret = vfs_read(f.file, buf, count, &pos);
 651                 fdput(f);
 652         }
 653
 654         return ret;
 655 }
 656
 657 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
 658                          size_t, count, loff_t, pos)
 659 {
 660         struct fd f;
 661         ssize_t ret = -EBADF;
 662
 663         if (pos < 0)
 664                 return -EINVAL;
 665
 666         f = fdget(fd);
 667         if (f.file) {
 668                 ret = -ESPIPE;
 669                 if (f.file->f_mode & FMODE_PWRITE)
 670                         ret = vfs_write(f.file, buf, count, &pos);
 671                 fdput(f);
 672         }
 673
 674         return ret;
 675 }
 676
 677 /*
 678  * Reduce an iovec's length in-place.  Return the resulting number of segments
 679  */
 680 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
 681 {
 682         unsigned long seg = 0;
 683         size_t len = 0;
 684
 685         while (seg < nr_segs) {
 686                 seg++;
 687                 if (len + iov->iov_len >= to) {
 688                         iov->iov_len = to - len;
 689                         break;
 690                 }
 691                 len += iov->iov_len;
 692                 iov++;
 693         }
 694         return seg;
 695 }
 696 EXPORT_SYMBOL(iov_shorten);
 697
 698 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
 699                 loff_t *ppos, iter_fn_t fn)
 700 {
 701         struct kiocb kiocb;
 702         ssize_t ret;
 703
 704         init_sync_kiocb(&kiocb, filp);
 705         kiocb.ki_pos = *ppos;
 706
 707         ret = fn(&kiocb, iter);
 708         BUG_ON(ret == -EIOCBQUEUED);
 709         *ppos = kiocb.ki_pos;
 710         return ret;
 711 }
 712
 713 static ssize_t do_sync_readv_writev(struct file *filp, struct iov_iter *iter,
 714                 loff_t *ppos, iov_fn_t fn)
 715 {
 716         struct kiocb kiocb;
 717         ssize_t ret;
 718
 719         init_sync_kiocb(&kiocb, filp);
 720         kiocb.ki_pos = *ppos;
 721
 722         ret = fn(&kiocb, iter->iov, iter->nr_segs, kiocb.ki_pos);
 723         BUG_ON(ret == -EIOCBQUEUED);
 724         *ppos = kiocb.ki_pos;
 725         return ret;
 726 }
 727
 728 /* Do it by hand, with file-ops */
 729 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
 730                 loff_t *ppos, io_fn_t fn)
 731 {
 732         ssize_t ret = 0;
 733
 734         while (iov_iter_count(iter)) {
 735                 struct iovec iovec = iov_iter_iovec(iter);
 736                 ssize_t nr;
 737
 738                 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
 739
 740                 if (nr < 0) {
 741                         if (!ret)
 742                                 ret = nr;
 743                         break;
 744                 }
 745                 ret += nr;
 746                 if (nr != iovec.iov_len)
 747                         break;
 748                 iov_iter_advance(iter, nr);
 749         }
 750
 751         return ret;
 752 }
 753
 754 /* A write operation does a read from user space and vice versa */
 755 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 756
 757 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
 758                               unsigned long nr_segs, unsigned long fast_segs,
 759                               struct iovec *fast_pointer,
 760                               struct iovec **ret_pointer)
 761 {
 762         unsigned long seg;
 763         ssize_t ret;
 764         struct iovec *iov = fast_pointer;
 765
 766         /*
 767          * SuS says "The readv() function *may* fail if the iovcnt argument
 768          * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 769          * traditionally returned zero for zero segments, so...
 770          */
 771         if (nr_segs == 0) {
 772                 ret = 0;
 773                 goto out;
 774         }
 775
 776         /*
 777          * First get the "struct iovec" from user memory and
 778          * verify all the pointers
 779          */
 780         if (nr_segs > UIO_MAXIOV) {
 781                 ret = -EINVAL;
 782                 goto out;
 783         }
 784         if (nr_segs > fast_segs) {
 785                 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 786                 if (iov == NULL) {
 787                         ret = -ENOMEM;
 788                         goto out;
 789                 }
 790         }
 791         if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
 792                 ret = -EFAULT;
 793                 goto out;
 794         }
 795
 796         /*
 797          * According to the Single Unix Specification we should return EINVAL
 798          * if an element length is < 0 when cast to ssize_t or if the
 799          * total length would overflow the ssize_t return value of the
 800          * system call.
 801          *
 802          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
 803          * overflow case.
 804          */
 805         ret = 0;
 806         for (seg = 0; seg < nr_segs; seg++) {
 807                 void __user *buf = iov[seg].iov_base;
 808                 ssize_t len = (ssize_t)iov[seg].iov_len;
 809
 810                 /* see if we we're about to use an invalid len or if
 811                  * it's about to overflow ssize_t */
 812                 if (len < 0) {
 813                         ret = -EINVAL;
 814                         goto out;
 815                 }
 816                 if (type >= 0
 817                     && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
 818                         ret = -EFAULT;
 819                         goto out;
 820                 }
 821                 if (len > MAX_RW_COUNT - ret) {
 822                         len = MAX_RW_COUNT - ret;
 823                         iov[seg].iov_len = len;
 824                 }
 825                 ret += len;
 826         }
 827 out:
 828         *ret_pointer = iov;
 829         return ret;
 830 }
 831
 832 static ssize_t do_readv_writev(int type, struct file *file,
 833                                const struct iovec __user * uvector,
 834                                unsigned long nr_segs, loff_t *pos)
 835 {
 836         size_t tot_len;
 837         struct iovec iovstack[UIO_FASTIOV];
 838         struct iovec *iov = iovstack;
 839         struct iov_iter iter;
 840         ssize_t ret;
 841         io_fn_t fn;
 842         iov_fn_t fnv;
 843         iter_fn_t iter_fn;
 844
 845         ret = import_iovec(type, uvector, nr_segs,
 846                            ARRAY_SIZE(iovstack), &iov, &iter);
 847         if (ret < 0)
 848                 return ret;
 849
 850         tot_len = iov_iter_count(&iter);
 851         if (!tot_len)
 852                 goto out;
 853         ret = rw_verify_area(type, file, pos, tot_len);
 854         if (ret < 0)
 855                 goto out;
 856
 857         fnv = NULL;
 858         if (type == READ) {
 859                 fn = file->f_op->read;
 860                 fnv = file->f_op->aio_read;
 861                 iter_fn = file->f_op->read_iter;
 862         } else {
 863                 fn = (io_fn_t)file->f_op->write;
 864                 fnv = file->f_op->aio_write;
 865                 iter_fn = file->f_op->write_iter;
 866                 file_start_write(file);
 867         }
 868
 869         if (iter_fn)
 870                 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
 871         else if (fnv)
 872                 ret = do_sync_readv_writev(file, &iter, pos, fnv);
 873         else
 874                 ret = do_loop_readv_writev(file, &iter, pos, fn);
 875
 876         if (type != READ)
 877                 file_end_write(file);
 878
 879 out:
 880         kfree(iov);
 881         if ((ret + (type == READ)) > 0) {
 882                 if (type == READ)
 883                         fsnotify_access(file);
 884                 else
 885                         fsnotify_modify(file);
 886         }
 887         return ret;
 888 }
 889
 890 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
 891                   unsigned long vlen, loff_t *pos)
 892 {
 893         if (!(file->f_mode & FMODE_READ))
 894                 return -EBADF;
 895         if (!(file->f_mode & FMODE_CAN_READ))
 896                 return -EINVAL;
 897
 898         return do_readv_writev(READ, file, vec, vlen, pos);
 899 }
 900
 901 EXPORT_SYMBOL(vfs_readv);
 902
 903 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 904                    unsigned long vlen, loff_t *pos)
 905 {
 906         if (!(file->f_mode & FMODE_WRITE))
 907                 return -EBADF;
 908         if (!(file->f_mode & FMODE_CAN_WRITE))
 909                 return -EINVAL;
 910
 911         return do_readv_writev(WRITE, file, vec, vlen, pos);
 912 }
 913
 914 EXPORT_SYMBOL(vfs_writev);
 915
 916 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
 917                 unsigned long, vlen)
 918 {
 919         struct fd f = fdget_pos(fd);
 920         ssize_t ret = -EBADF;
 921
 922         if (f.file) {
 923                 loff_t pos = file_pos_read(f.file);
 924                 ret = vfs_readv(f.file, vec, vlen, &pos);
 925                 if (ret >= 0)
 926                         file_pos_write(f.file, pos);
 927                 fdput_pos(f);
 928         }
 929
 930         if (ret > 0)
 931                 add_rchar(current, ret);
 932         inc_syscr(current);
 933         return ret;
 934 }
 935
 936 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
 937                 unsigned long, vlen)
 938 {
 939         struct fd f = fdget_pos(fd);
 940         ssize_t ret = -EBADF;
 941
 942         if (f.file) {
 943                 loff_t pos = file_pos_read(f.file);
 944                 ret = vfs_writev(f.file, vec, vlen, &pos);
 945                 if (ret >= 0)
 946                         file_pos_write(f.file, pos);
 947                 fdput_pos(f);
 948         }
 949
 950         if (ret > 0)
 951                 add_wchar(current, ret);
 952         inc_syscw(current);
 953         return ret;
 954 }
 955
 956 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
 957 {
 958 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
 959         return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
 960 }
 961
 962 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
 963                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
 964 {
 965         loff_t pos = pos_from_hilo(pos_h, pos_l);
 966         struct fd f;
 967         ssize_t ret = -EBADF;
 968
 969         if (pos < 0)
 970                 return -EINVAL;
 971
 972         f = fdget(fd);
 973         if (f.file) {
 974                 ret = -ESPIPE;
 975                 if (f.file->f_mode & FMODE_PREAD)
 976                         ret = vfs_readv(f.file, vec, vlen, &pos);
 977                 fdput(f);
 978         }
 979
 980         if (ret > 0)
 981                 add_rchar(current, ret);
 982         inc_syscr(current);
 983         return ret;
 984 }
 985
 986 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
 987                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
 988 {
 989         loff_t pos = pos_from_hilo(pos_h, pos_l);
 990         struct fd f;
 991         ssize_t ret = -EBADF;
 992
 993         if (pos < 0)
 994                 return -EINVAL;
 995
 996         f = fdget(fd);
 997         if (f.file) {
 998                 ret = -ESPIPE;
 999                 if (f.file->f_mode & FMODE_PWRITE)
1000                         ret = vfs_writev(f.file, vec, vlen, &pos);
1001                 fdput(f);
1002         }
1003
1004         if (ret > 0)
1005                 add_wchar(current, ret);
1006         inc_syscw(current);
1007         return ret;
1008 }
1009
1010 #ifdef CONFIG_COMPAT
1011
1012 static ssize_t compat_do_readv_writev(int type, struct file *file,
1013                                const struct compat_iovec __user *uvector,
1014                                unsigned long nr_segs, loff_t *pos)
1015 {
1016         compat_ssize_t tot_len;
1017         struct iovec iovstack[UIO_FASTIOV];
1018         struct iovec *iov = iovstack;
1019         struct iov_iter iter;
1020         ssize_t ret;
1021         io_fn_t fn;
1022         iov_fn_t fnv;
1023         iter_fn_t iter_fn;
1024
1025         ret = compat_import_iovec(type, uvector, nr_segs,
1026                                   UIO_FASTIOV, &iov, &iter);
1027         if (ret < 0)
1028                 return ret;
1029
1030         tot_len = iov_iter_count(&iter);
1031         if (!tot_len)
1032                 goto out;
1033         ret = rw_verify_area(type, file, pos, tot_len);
1034         if (ret < 0)
1035                 goto out;
1036
1037         fnv = NULL;
1038         if (type == READ) {
1039                 fn = file->f_op->read;
1040                 fnv = file->f_op->aio_read;
1041                 iter_fn = file->f_op->read_iter;
1042         } else {
1043                 fn = (io_fn_t)file->f_op->write;
1044                 fnv = file->f_op->aio_write;
1045                 iter_fn = file->f_op->write_iter;
1046                 file_start_write(file);
1047         }
1048
1049         if (iter_fn)
1050                 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
1051         else if (fnv)
1052                 ret = do_sync_readv_writev(file, &iter, pos, fnv);
1053         else
1054                 ret = do_loop_readv_writev(file, &iter, pos, fn);
1055
1056         if (type != READ)
1057                 file_end_write(file);
1058
1059 out:
1060         kfree(iov);
1061         if ((ret + (type == READ)) > 0) {
1062                 if (type == READ)
1063                         fsnotify_access(file);
1064                 else
1065                         fsnotify_modify(file);
1066         }
1067         return ret;
1068 }
1069
1070 static size_t compat_readv(struct file *file,
1071                            const struct compat_iovec __user *vec,
1072                            unsigned long vlen, loff_t *pos)
1073 {
1074         ssize_t ret = -EBADF;
1075
1076         if (!(file->f_mode & FMODE_READ))
1077                 goto out;
1078
1079         ret = -EINVAL;
1080         if (!(file->f_mode & FMODE_CAN_READ))
1081                 goto out;
1082
1083         ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1084
1085 out:
1086         if (ret > 0)
1087                 add_rchar(current, ret);
1088         inc_syscr(current);
1089         return ret;
1090 }
1091
1092 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1093                 const struct compat_iovec __user *,vec,
1094                 compat_ulong_t, vlen)
1095 {
1096         struct fd f = fdget_pos(fd);
1097         ssize_t ret;
1098         loff_t pos;
1099
1100         if (!f.file)
1101                 return -EBADF;
1102         pos = f.file->f_pos;
1103         ret = compat_readv(f.file, vec, vlen, &pos);
1104         if (ret >= 0)
1105                 f.file->f_pos = pos;
1106         fdput_pos(f);
1107         return ret;
1108 }
1109
1110 static long __compat_sys_preadv64(unsigned long fd,
1111                                   const struct compat_iovec __user *vec,
1112                                   unsigned long vlen, loff_t pos)
1113 {
1114         struct fd f;
1115         ssize_t ret;
1116
1117         if (pos < 0)
1118                 return -EINVAL;
1119         f = fdget(fd);
1120         if (!f.file)
1121                 return -EBADF;
1122         ret = -ESPIPE;
1123         if (f.file->f_mode & FMODE_PREAD)
1124                 ret = compat_readv(f.file, vec, vlen, &pos);
1125         fdput(f);
1126         return ret;
1127 }
1128
1129 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1130 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1131                 const struct compat_iovec __user *,vec,
1132                 unsigned long, vlen, loff_t, pos)
1133 {
1134         return __compat_sys_preadv64(fd, vec, vlen, pos);
1135 }
1136 #endif
1137
1138 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1139                 const struct compat_iovec __user *,vec,
1140                 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1141 {
1142         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1143
1144         return __compat_sys_preadv64(fd, vec, vlen, pos);
1145 }
1146
1147 static size_t compat_writev(struct file *file,
1148                             const struct compat_iovec __user *vec,
1149                             unsigned long vlen, loff_t *pos)
1150 {
1151         ssize_t ret = -EBADF;
1152
1153         if (!(file->f_mode & FMODE_WRITE))
1154                 goto out;
1155
1156         ret = -EINVAL;
1157         if (!(file->f_mode & FMODE_CAN_WRITE))
1158                 goto out;
1159
1160         ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1161
1162 out:
1163         if (ret > 0)
1164                 add_wchar(current, ret);
1165         inc_syscw(current);
1166         return ret;
1167 }
1168
1169 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1170                 const struct compat_iovec __user *, vec,
1171                 compat_ulong_t, vlen)
1172 {
1173         struct fd f = fdget_pos(fd);
1174         ssize_t ret;
1175         loff_t pos;
1176
1177         if (!f.file)
1178                 return -EBADF;
1179         pos = f.file->f_pos;
1180         ret = compat_writev(f.file, vec, vlen, &pos);
1181         if (ret >= 0)
1182                 f.file->f_pos = pos;
1183         fdput_pos(f);
1184         return ret;
1185 }
1186
1187 static long __compat_sys_pwritev64(unsigned long fd,
1188                                    const struct compat_iovec __user *vec,
1189                                    unsigned long vlen, loff_t pos)
1190 {
1191         struct fd f;
1192         ssize_t ret;
1193
1194         if (pos < 0)
1195                 return -EINVAL;
1196         f = fdget(fd);
1197         if (!f.file)
1198                 return -EBADF;
1199         ret = -ESPIPE;
1200         if (f.file->f_mode & FMODE_PWRITE)
1201                 ret = compat_writev(f.file, vec, vlen, &pos);
1202         fdput(f);
1203         return ret;
1204 }
1205
1206 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1207 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1208                 const struct compat_iovec __user *,vec,
1209                 unsigned long, vlen, loff_t, pos)
1210 {
1211         return __compat_sys_pwritev64(fd, vec, vlen, pos);
1212 }
1213 #endif
1214
1215 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1216                 const struct compat_iovec __user *,vec,
1217                 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1218 {
1219         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1220
1221         return __compat_sys_pwritev64(fd, vec, vlen, pos);
1222 }
1223 #endif
1224
1225 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1226                            size_t count, loff_t max)
1227 {
1228         struct fd in, out;
1229         struct inode *in_inode, *out_inode;
1230         loff_t pos;
1231         loff_t out_pos;
1232         ssize_t retval;
1233         int fl;
1234
1235         /*
1236          * Get input file, and verify that it is ok..
1237          */
1238         retval = -EBADF;
1239         in = fdget(in_fd);
1240         if (!in.file)
1241                 goto out;
1242         if (!(in.file->f_mode & FMODE_READ))
1243                 goto fput_in;
1244         retval = -ESPIPE;
1245         if (!ppos) {
1246                 pos = in.file->f_pos;
1247         } else {
1248                 pos = *ppos;
1249                 if (!(in.file->f_mode & FMODE_PREAD))
1250                         goto fput_in;
1251         }
1252         retval = rw_verify_area(READ, in.file, &pos, count);
1253         if (retval < 0)
1254                 goto fput_in;
1255         count = retval;
1256
1257         /*
1258          * Get output file, and verify that it is ok..
1259          */
1260         retval = -EBADF;
1261         out = fdget(out_fd);
1262         if (!out.file)
1263                 goto fput_in;
1264         if (!(out.file->f_mode & FMODE_WRITE))
1265                 goto fput_out;
1266         retval = -EINVAL;
1267         in_inode = file_inode(in.file);
1268         out_inode = file_inode(out.file);
1269         out_pos = out.file->f_pos;
1270         retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1271         if (retval < 0)
1272                 goto fput_out;
1273         count = retval;
1274
1275         if (!max)
1276                 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1277
1278         if (unlikely(pos + count > max)) {
1279                 retval = -EOVERFLOW;
1280                 if (pos >= max)
1281                         goto fput_out;
1282                 count = max - pos;
1283         }
1284
1285         fl = 0;
1286 #if 0
1287         /*
1288          * We need to debate whether we can enable this or not. The
1289          * man page documents EAGAIN return for the output at least,
1290          * and the application is arguably buggy if it doesn't expect
1291          * EAGAIN on a non-blocking file descriptor.
1292          */
1293         if (in.file->f_flags & O_NONBLOCK)
1294                 fl = SPLICE_F_NONBLOCK;
1295 #endif
1296         file_start_write(out.file);
1297         retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1298         file_end_write(out.file);
1299
1300         if (retval > 0) {
1301                 add_rchar(current, retval);
1302                 add_wchar(current, retval);
1303                 fsnotify_access(in.file);
1304                 fsnotify_modify(out.file);
1305                 out.file->f_pos = out_pos;
1306                 if (ppos)
1307                         *ppos = pos;
1308                 else
1309                         in.file->f_pos = pos;
1310         }
1311
1312         inc_syscr(current);
1313         inc_syscw(current);
1314         if (pos > max)
1315                 retval = -EOVERFLOW;
1316
1317 fput_out:
1318         fdput(out);
1319 fput_in:
1320         fdput(in);
1321 out:
1322         return retval;
1323 }
1324
1325 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1326 {
1327         loff_t pos;
1328         off_t off;
1329         ssize_t ret;
1330
1331         if (offset) {
1332                 if (unlikely(get_user(off, offset)))
1333                         return -EFAULT;
1334                 pos = off;
1335                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1336                 if (unlikely(put_user(pos, offset)))
1337                         return -EFAULT;
1338                 return ret;
1339         }
1340
1341         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1342 }
1343
1344 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1345 {
1346         loff_t pos;
1347         ssize_t ret;
1348
1349         if (offset) {
1350                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1351                         return -EFAULT;
1352                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1353                 if (unlikely(put_user(pos, offset)))
1354                         return -EFAULT;
1355                 return ret;
1356         }
1357
1358         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1359 }
1360
1361 #ifdef CONFIG_COMPAT
1362 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1363                 compat_off_t __user *, offset, compat_size_t, count)
1364 {
1365         loff_t pos;
1366         off_t off;
1367         ssize_t ret;
1368
1369         if (offset) {
1370                 if (unlikely(get_user(off, offset)))
1371                         return -EFAULT;
1372                 pos = off;
1373                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1374                 if (unlikely(put_user(pos, offset)))
1375                         return -EFAULT;
1376                 return ret;
1377         }
1378
1379         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1380 }
1381
1382 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1383                 compat_loff_t __user *, offset, compat_size_t, count)
1384 {
1385         loff_t pos;
1386         ssize_t ret;
1387
1388         if (offset) {
1389                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1390                         return -EFAULT;
1391                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1392                 if (unlikely(put_user(pos, offset)))
1393                         return -EFAULT;
1394                 return ret;
1395         }
1396
1397         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1398 }
1399 #endif