fs/read_write.c

   1 /*
   2  *  linux/fs/read_write.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/slab.h>
   8 #include <linux/stat.h>
   9 #include <linux/fcntl.h>
  10 #include <linux/file.h>
  11 #include <linux/uio.h>
  12 #include <linux/smp_lock.h>
  13 #include <linux/dnotify.h>
  14 #include <linux/security.h>
  15 #include <linux/module.h>
  16
  17 #include <asm/uaccess.h>
  18 #include <asm/unistd.h>
  19
  20 struct file_operations generic_ro_fops = {
  21         .llseek         = generic_file_llseek,
  22         .read           = generic_file_read,
  23         .mmap           = generic_file_readonly_mmap,
  24         .sendfile       = generic_file_sendfile,
  25 };
  26
  27 EXPORT_SYMBOL(generic_ro_fops);
  28
  29 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
  30 {
  31         long long retval;
  32         struct inode *inode = file->f_mapping->host;
  33
  34         down(&inode->i_sem);
  35         switch (origin) {
  36                 case 2:
  37                         offset += inode->i_size;
  38                         break;
  39                 case 1:
  40                         offset += file->f_pos;
  41         }
  42         retval = -EINVAL;
  43         if (offset>=0 && offset<=inode->i_sb->s_maxbytes) {
  44                 if (offset != file->f_pos) {
  45                         file->f_pos = offset;
  46                         file->f_version = 0;
  47                 }
  48                 retval = offset;
  49         }
  50         up(&inode->i_sem);
  51         return retval;
  52 }
  53
  54 EXPORT_SYMBOL(generic_file_llseek);
  55
  56 loff_t remote_llseek(struct file *file, loff_t offset, int origin)
  57 {
  58         long long retval;
  59
  60         lock_kernel();
  61         switch (origin) {
  62                 case 2:
  63                         offset += i_size_read(file->f_dentry->d_inode);
  64                         break;
  65                 case 1:
  66                         offset += file->f_pos;
  67         }
  68         retval = -EINVAL;
  69         if (offset>=0 && offset<=file->f_dentry->d_inode->i_sb->s_maxbytes) {
  70                 if (offset != file->f_pos) {
  71                         file->f_pos = offset;
  72                         file->f_version = 0;
  73                 }
  74                 retval = offset;
  75         }
  76         unlock_kernel();
  77         return retval;
  78 }
  79 EXPORT_SYMBOL(remote_llseek);
  80
  81 loff_t no_llseek(struct file *file, loff_t offset, int origin)
  82 {
  83         return -ESPIPE;
  84 }
  85 EXPORT_SYMBOL(no_llseek);
  86
  87 loff_t default_llseek(struct file *file, loff_t offset, int origin)
  88 {
  89         long long retval;
  90
  91         lock_kernel();
  92         switch (origin) {
  93                 case 2:
  94                         offset += i_size_read(file->f_dentry->d_inode);
  95                         break;
  96                 case 1:
  97                         offset += file->f_pos;
  98         }
  99         retval = -EINVAL;
 100         if (offset >= 0) {
 101                 if (offset != file->f_pos) {
 102                         file->f_pos = offset;
 103                         file->f_version = 0;
 104                 }
 105                 retval = offset;
 106         }
 107         unlock_kernel();
 108         return retval;
 109 }
 110 EXPORT_SYMBOL(default_llseek);
 111
 112 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
 113 {
 114         loff_t (*fn)(struct file *, loff_t, int);
 115
 116         fn = no_llseek;
 117         if (file->f_mode & FMODE_LSEEK) {
 118                 fn = default_llseek;
 119                 if (file->f_op && file->f_op->llseek)
 120                         fn = file->f_op->llseek;
 121         }
 122         return fn(file, offset, origin);
 123 }
 124 EXPORT_SYMBOL(vfs_llseek);
 125
 126 asmlinkage off_t sys_lseek(unsigned int fd, off_t offset, unsigned int origin)
 127 {
 128         off_t retval;
 129         struct file * file;
 130         int fput_needed;
 131
 132         retval = -EBADF;
 133         file = fget_light(fd, &fput_needed);
 134         if (!file)
 135                 goto bad;
 136
 137         retval = -EINVAL;
 138         if (origin <= 2) {
 139                 loff_t res = vfs_llseek(file, offset, origin);
 140                 retval = res;
 141                 if (res != (loff_t)retval)
 142                         retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
 143         }
 144         fput_light(file, fput_needed);
 145 bad:
 146         return retval;
 147 }
 148 EXPORT_SYMBOL_GPL(sys_lseek);
 149
 150 #ifdef __ARCH_WANT_SYS_LLSEEK
 151 asmlinkage long sys_llseek(unsigned int fd, unsigned long offset_high,
 152                            unsigned long offset_low, loff_t __user * result,
 153                            unsigned int origin)
 154 {
 155         int retval;
 156         struct file * file;
 157         loff_t offset;
 158         int fput_needed;
 159
 160         retval = -EBADF;
 161         file = fget_light(fd, &fput_needed);
 162         if (!file)
 163                 goto bad;
 164
 165         retval = -EINVAL;
 166         if (origin > 2)
 167                 goto out_putf;
 168
 169         offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
 170                         origin);
 171
 172         retval = (int)offset;
 173         if (offset >= 0) {
 174                 retval = -EFAULT;
 175                 if (!copy_to_user(result, &offset, sizeof(offset)))
 176                         retval = 0;
 177         }
 178 out_putf:
 179         fput_light(file, fput_needed);
 180 bad:
 181         return retval;
 182 }
 183 #endif
 184
 185 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 186 {
 187         struct kiocb kiocb;
 188         ssize_t ret;
 189
 190         init_sync_kiocb(&kiocb, filp);
 191         kiocb.ki_pos = *ppos;
 192         ret = filp->f_op->aio_read(&kiocb, buf, len, kiocb.ki_pos);
 193         if (-EIOCBQUEUED == ret)
 194                 ret = wait_on_sync_kiocb(&kiocb);
 195         *ppos = kiocb.ki_pos;
 196         return ret;
 197 }
 198
 199 EXPORT_SYMBOL(do_sync_read);
 200
 201 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 202 {
 203         struct inode *inode = file->f_dentry->d_inode;
 204         ssize_t ret;
 205
 206         if (!(file->f_mode & FMODE_READ))
 207                 return -EBADF;
 208         if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
 209                 return -EINVAL;
 210
 211         ret = locks_verify_area(FLOCK_VERIFY_READ, inode, file, *pos, count);
 212         if (!ret) {
 213                 ret = security_file_permission (file, MAY_READ);
 214                 if (!ret) {
 215                         if (file->f_op->read)
 216                                 ret = file->f_op->read(file, buf, count, pos);
 217                         else
 218                                 ret = do_sync_read(file, buf, count, pos);
 219                         if (ret > 0)
 220                                 dnotify_parent(file->f_dentry, DN_ACCESS);
 221                 }
 222         }
 223
 224         return ret;
 225 }
 226
 227 EXPORT_SYMBOL(vfs_read);
 228
 229 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 230 {
 231         struct kiocb kiocb;
 232         ssize_t ret;
 233
 234         init_sync_kiocb(&kiocb, filp);
 235         kiocb.ki_pos = *ppos;
 236         ret = filp->f_op->aio_write(&kiocb, buf, len, kiocb.ki_pos);
 237         if (-EIOCBQUEUED == ret)
 238                 ret = wait_on_sync_kiocb(&kiocb);
 239         *ppos = kiocb.ki_pos;
 240         return ret;
 241 }
 242
 243 EXPORT_SYMBOL(do_sync_write);
 244
 245 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 246 {
 247         struct inode *inode = file->f_dentry->d_inode;
 248         ssize_t ret;
 249
 250         if (!(file->f_mode & FMODE_WRITE))
 251                 return -EBADF;
 252         if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
 253                 return -EINVAL;
 254
 255         ret = locks_verify_area(FLOCK_VERIFY_WRITE, inode, file, *pos, count);
 256         if (!ret) {
 257                 ret = security_file_permission (file, MAY_WRITE);
 258                 if (!ret) {
 259                         if (file->f_op->write)
 260                                 ret = file->f_op->write(file, buf, count, pos);
 261                         else
 262                                 ret = do_sync_write(file, buf, count, pos);
 263                         if (ret > 0)
 264                                 dnotify_parent(file->f_dentry, DN_MODIFY);
 265                 }
 266         }
 267
 268         return ret;
 269 }
 270
 271 EXPORT_SYMBOL(vfs_write);
 272
 273 static inline loff_t file_pos_read(struct file *file)
 274 {
 275         return file->f_pos;
 276 }
 277
 278 static inline void file_pos_write(struct file *file, loff_t pos)
 279 {
 280         file->f_pos = pos;
 281 }
 282
 283 asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count)
 284 {
 285         struct file *file;
 286         ssize_t ret = -EBADF;
 287         int fput_needed;
 288
 289         file = fget_light(fd, &fput_needed);
 290         if (file) {
 291                 loff_t pos = file_pos_read(file);
 292                 ret = vfs_read(file, buf, count, &pos);
 293                 file_pos_write(file, pos);
 294                 fput_light(file, fput_needed);
 295         }
 296
 297         return ret;
 298 }
 299 EXPORT_SYMBOL_GPL(sys_read);
 300
 301 asmlinkage ssize_t sys_write(unsigned int fd, const char __user * buf, size_t count)
 302 {
 303         struct file *file;
 304         ssize_t ret = -EBADF;
 305         int fput_needed;
 306
 307         file = fget_light(fd, &fput_needed);
 308         if (file) {
 309                 loff_t pos = file_pos_read(file);
 310                 ret = vfs_write(file, buf, count, &pos);
 311                 file_pos_write(file, pos);
 312                 fput_light(file, fput_needed);
 313         }
 314
 315         return ret;
 316 }
 317
 318 asmlinkage ssize_t sys_pread64(unsigned int fd, char __user *buf,
 319                              size_t count, loff_t pos)
 320 {
 321         struct file *file;
 322         ssize_t ret = -EBADF;
 323         int fput_needed;
 324
 325         if (pos < 0)
 326                 return -EINVAL;
 327
 328         file = fget_light(fd, &fput_needed);
 329         if (file) {
 330                 ret = -ESPIPE;
 331                 if (file->f_mode & FMODE_PREAD)
 332                         ret = vfs_read(file, buf, count, &pos);
 333                 fput_light(file, fput_needed);
 334         }
 335
 336         return ret;
 337 }
 338
 339 asmlinkage ssize_t sys_pwrite64(unsigned int fd, const char __user *buf,
 340                               size_t count, loff_t pos)
 341 {
 342         struct file *file;
 343         ssize_t ret = -EBADF;
 344         int fput_needed;
 345
 346         if (pos < 0)
 347                 return -EINVAL;
 348
 349         file = fget_light(fd, &fput_needed);
 350         if (file) {
 351                 ret = -ESPIPE;
 352                 if (file->f_mode & FMODE_PWRITE)
 353                         ret = vfs_write(file, buf, count, &pos);
 354                 fput_light(file, fput_needed);
 355         }
 356
 357         return ret;
 358 }
 359
 360 /*
 361  * Reduce an iovec's length in-place.  Return the resulting number of segments
 362  */
 363 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
 364 {
 365         unsigned long seg = 0;
 366         size_t len = 0;
 367
 368         while (seg < nr_segs) {
 369                 seg++;
 370                 if (len + iov->iov_len >= to) {
 371                         iov->iov_len = to - len;
 372                         break;
 373                 }
 374                 len += iov->iov_len;
 375                 iov++;
 376         }
 377         return seg;
 378 }
 379
 380 EXPORT_SYMBOL(iov_shorten);
 381
 382 static ssize_t do_readv_writev(int type, struct file *file,
 383                                const struct iovec __user * uvector,
 384                                unsigned long nr_segs, loff_t *pos)
 385 {
 386         typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
 387         typedef ssize_t (*iov_fn_t)(struct file *, const struct iovec *, unsigned long, loff_t *);
 388
 389         size_t tot_len;
 390         struct iovec iovstack[UIO_FASTIOV];
 391         struct iovec *iov=iovstack, *vector;
 392         ssize_t ret;
 393         int seg;
 394         io_fn_t fn;
 395         iov_fn_t fnv;
 396         struct inode *inode;
 397
 398         /*
 399          * SuS says "The readv() function *may* fail if the iovcnt argument
 400          * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 401          * traditionally returned zero for zero segments, so...
 402          */
 403         ret = 0;
 404         if (nr_segs == 0)
 405                 goto out;
 406
 407         /*
 408          * First get the "struct iovec" from user memory and
 409          * verify all the pointers
 410          */
 411         ret = -EINVAL;
 412         if ((nr_segs > UIO_MAXIOV) || (nr_segs <= 0))
 413                 goto out;
 414         if (!file->f_op)
 415                 goto out;
 416         if (nr_segs > UIO_FASTIOV) {
 417                 ret = -ENOMEM;
 418                 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 419                 if (!iov)
 420                         goto out;
 421         }
 422         ret = -EFAULT;
 423         if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector)))
 424                 goto out;
 425
 426         /*
 427          * Single unix specification:
 428          * We should -EINVAL if an element length is not >= 0 and fitting an
 429          * ssize_t.  The total length is fitting an ssize_t
 430          *
 431          * Be careful here because iov_len is a size_t not an ssize_t
 432          */
 433         tot_len = 0;
 434         ret = -EINVAL;
 435         for (seg = 0; seg < nr_segs; seg++) {
 436                 ssize_t len = (ssize_t)iov[seg].iov_len;
 437
 438                 if (len < 0)    /* size_t not fitting an ssize_t .. */
 439                         goto out;
 440                 tot_len += len;
 441                 if ((ssize_t)tot_len < 0) /* maths overflow on the ssize_t */
 442                         goto out;
 443         }
 444         if (tot_len == 0) {
 445                 ret = 0;
 446                 goto out;
 447         }
 448
 449         inode = file->f_dentry->d_inode;
 450         /* VERIFY_WRITE actually means a read, as we write to user space */
 451         ret = locks_verify_area((type == READ
 452                                  ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE),
 453                                 inode, file, *pos, tot_len);
 454         if (ret)
 455                 goto out;
 456
 457         fnv = NULL;
 458         if (type == READ) {
 459                 fn = file->f_op->read;
 460                 fnv = file->f_op->readv;
 461         } else {
 462                 fn = (io_fn_t)file->f_op->write;
 463                 fnv = file->f_op->writev;
 464         }
 465         if (fnv) {
 466                 ret = fnv(file, iov, nr_segs, pos);
 467                 goto out;
 468         }
 469
 470         /* Do it by hand, with file-ops */
 471         ret = 0;
 472         vector = iov;
 473         while (nr_segs > 0) {
 474                 void __user * base;
 475                 size_t len;
 476                 ssize_t nr;
 477
 478                 base = vector->iov_base;
 479                 len = vector->iov_len;
 480                 vector++;
 481                 nr_segs--;
 482
 483                 nr = fn(file, base, len, pos);
 484
 485                 if (nr < 0) {
 486                         if (!ret) ret = nr;
 487                         break;
 488                 }
 489                 ret += nr;
 490                 if (nr != len)
 491                         break;
 492         }
 493 out:
 494         if (iov != iovstack)
 495                 kfree(iov);
 496         if ((ret + (type == READ)) > 0)
 497                 dnotify_parent(file->f_dentry,
 498                                 (type == READ) ? DN_ACCESS : DN_MODIFY);
 499         return ret;
 500 }
 501
 502 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
 503                   unsigned long vlen, loff_t *pos)
 504 {
 505         if (!(file->f_mode & FMODE_READ))
 506                 return -EBADF;
 507         if (!file->f_op || (!file->f_op->readv && !file->f_op->read))
 508                 return -EINVAL;
 509
 510         return do_readv_writev(READ, file, vec, vlen, pos);
 511 }
 512
 513 EXPORT_SYMBOL(vfs_readv);
 514
 515 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 516                    unsigned long vlen, loff_t *pos)
 517 {
 518         if (!(file->f_mode & FMODE_WRITE))
 519                 return -EBADF;
 520         if (!file->f_op || (!file->f_op->writev && !file->f_op->write))
 521                 return -EINVAL;
 522
 523         return do_readv_writev(WRITE, file, vec, vlen, pos);
 524 }
 525
 526 EXPORT_SYMBOL(vfs_writev);
 527
 528 asmlinkage ssize_t
 529 sys_readv(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
 530 {
 531         struct file *file;
 532         ssize_t ret = -EBADF;
 533         int fput_needed;
 534
 535         file = fget_light(fd, &fput_needed);
 536         if (file) {
 537                 loff_t pos = file_pos_read(file);
 538                 ret = vfs_readv(file, vec, vlen, &pos);
 539                 file_pos_write(file, pos);
 540                 fput_light(file, fput_needed);
 541         }
 542
 543         return ret;
 544 }
 545
 546 asmlinkage ssize_t
 547 sys_writev(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
 548 {
 549         struct file *file;
 550         ssize_t ret = -EBADF;
 551         int fput_needed;
 552
 553         file = fget_light(fd, &fput_needed);
 554         if (file) {
 555                 loff_t pos = file_pos_read(file);
 556                 ret = vfs_writev(file, vec, vlen, &pos);
 557                 file_pos_write(file, pos);
 558                 fput_light(file, fput_needed);
 559         }
 560
 561         return ret;
 562 }
 563
 564 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
 565                            size_t count, loff_t max)
 566 {
 567         struct file * in_file, * out_file;
 568         struct inode * in_inode, * out_inode;
 569         loff_t pos;
 570         ssize_t retval;
 571         int fput_needed_in, fput_needed_out;
 572
 573         /*
 574          * Get input file, and verify that it is ok..
 575          */
 576         retval = -EBADF;
 577         in_file = fget_light(in_fd, &fput_needed_in);
 578         if (!in_file)
 579                 goto out;
 580         if (!(in_file->f_mode & FMODE_READ))
 581                 goto fput_in;
 582         retval = -EINVAL;
 583         in_inode = in_file->f_dentry->d_inode;
 584         if (!in_inode)
 585                 goto fput_in;
 586         if (!in_file->f_op || !in_file->f_op->sendfile)
 587                 goto fput_in;
 588         retval = -ESPIPE;
 589         if (!ppos)
 590                 ppos = &in_file->f_pos;
 591         else
 592                 if (!(in_file->f_mode & FMODE_PREAD))
 593                         goto fput_in;
 594         retval = locks_verify_area(FLOCK_VERIFY_READ, in_inode, in_file, *ppos, count);
 595         if (retval)
 596                 goto fput_in;
 597
 598         retval = security_file_permission (in_file, MAY_READ);
 599         if (retval)
 600                 goto fput_in;
 601
 602         /*
 603          * Get output file, and verify that it is ok..
 604          */
 605         retval = -EBADF;
 606         out_file = fget_light(out_fd, &fput_needed_out);
 607         if (!out_file)
 608                 goto fput_in;
 609         if (!(out_file->f_mode & FMODE_WRITE))
 610                 goto fput_out;
 611         retval = -EINVAL;
 612         if (!out_file->f_op || !out_file->f_op->sendpage)
 613                 goto fput_out;
 614         out_inode = out_file->f_dentry->d_inode;
 615         retval = locks_verify_area(FLOCK_VERIFY_WRITE, out_inode, out_file, out_file->f_pos, count);
 616         if (retval)
 617                 goto fput_out;
 618
 619         retval = security_file_permission (out_file, MAY_WRITE);
 620         if (retval)
 621                 goto fput_out;
 622
 623         if (!max)
 624                 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
 625
 626         pos = *ppos;
 627         retval = -EINVAL;
 628         if (unlikely(pos < 0))
 629                 goto fput_out;
 630         if (unlikely(pos + count > max)) {
 631                 retval = -EOVERFLOW;
 632                 if (pos >= max)
 633                         goto fput_out;
 634                 count = max - pos;
 635         }
 636
 637         retval = in_file->f_op->sendfile(in_file, ppos, count, file_send_actor, out_file);
 638
 639         if (*ppos > max)
 640                 retval = -EOVERFLOW;
 641
 642 fput_out:
 643         fput_light(out_file, fput_needed_out);
 644 fput_in:
 645         fput_light(in_file, fput_needed_in);
 646 out:
 647         return retval;
 648 }
 649
 650 asmlinkage ssize_t sys_sendfile(int out_fd, int in_fd, off_t __user *offset, size_t count)
 651 {
 652         loff_t pos;
 653         off_t off;
 654         ssize_t ret;
 655
 656         if (offset) {
 657                 if (unlikely(get_user(off, offset)))
 658                         return -EFAULT;
 659                 pos = off;
 660                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
 661                 if (unlikely(put_user(pos, offset)))
 662                         return -EFAULT;
 663                 return ret;
 664         }
 665
 666         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 667 }
 668
 669 asmlinkage ssize_t sys_sendfile64(int out_fd, int in_fd, loff_t __user *offset, size_t count)
 670 {
 671         loff_t pos;
 672         ssize_t ret;
 673
 674         if (offset) {
 675                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
 676                         return -EFAULT;
 677                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
 678                 if (unlikely(put_user(pos, offset)))
 679                         return -EFAULT;
 680                 return ret;
 681         }
 682
 683         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 684 }