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