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 #include <linux/syscalls.h>
  17
  18 #include <asm/uaccess.h>
  19 #include <asm/unistd.h>
  20
  21 struct file_operations generic_ro_fops = {
  22         .llseek         = generic_file_llseek,
  23         .read           = generic_file_read,
  24         .mmap           = generic_file_readonly_mmap,
  25         .sendfile       = generic_file_sendfile,
  26 };
  27
  28 EXPORT_SYMBOL(generic_ro_fops);
  29
  30 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
  31 {
  32         long long retval;
  33         struct inode *inode = file->f_mapping->host;
  34
  35         down(&inode->i_sem);
  36         switch (origin) {
  37                 case 2:
  38                         offset += inode->i_size;
  39                         break;
  40                 case 1:
  41                         offset += file->f_pos;
  42         }
  43         retval = -EINVAL;
  44         if (offset>=0 && offset<=inode->i_sb->s_maxbytes) {
  45                 if (offset != file->f_pos) {
  46                         file->f_pos = offset;
  47                         file->f_version = 0;
  48                 }
  49                 retval = offset;
  50         }
  51         up(&inode->i_sem);
  52         return retval;
  53 }
  54
  55 EXPORT_SYMBOL(generic_file_llseek);
  56
  57 loff_t remote_llseek(struct file *file, loff_t offset, int origin)
  58 {
  59         long long retval;
  60
  61         lock_kernel();
  62         switch (origin) {
  63                 case 2:
  64                         offset += i_size_read(file->f_dentry->d_inode);
  65                         break;
  66                 case 1:
  67                         offset += file->f_pos;
  68         }
  69         retval = -EINVAL;
  70         if (offset>=0 && offset<=file->f_dentry->d_inode->i_sb->s_maxbytes) {
  71                 if (offset != file->f_pos) {
  72                         file->f_pos = offset;
  73                         file->f_version = 0;
  74                 }
  75                 retval = offset;
  76         }
  77         unlock_kernel();
  78         return retval;
  79 }
  80 EXPORT_SYMBOL(remote_llseek);
  81
  82 loff_t no_llseek(struct file *file, loff_t offset, int origin)
  83 {
  84         return -ESPIPE;
  85 }
  86 EXPORT_SYMBOL(no_llseek);
  87
  88 loff_t default_llseek(struct file *file, loff_t offset, int origin)
  89 {
  90         long long retval;
  91
  92         lock_kernel();
  93         switch (origin) {
  94                 case 2:
  95                         offset += i_size_read(file->f_dentry->d_inode);
  96                         break;
  97                 case 1:
  98                         offset += file->f_pos;
  99         }
 100         retval = -EINVAL;
 101         if (offset >= 0) {
 102                 if (offset != file->f_pos) {
 103                         file->f_pos = offset;
 104                         file->f_version = 0;
 105                 }
 106                 retval = offset;
 107         }
 108         unlock_kernel();
 109         return retval;
 110 }
 111 EXPORT_SYMBOL(default_llseek);
 112
 113 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
 114 {
 115         loff_t (*fn)(struct file *, loff_t, int);
 116
 117         fn = no_llseek;
 118         if (file->f_mode & FMODE_LSEEK) {
 119                 fn = default_llseek;
 120                 if (file->f_op && file->f_op->llseek)
 121                         fn = file->f_op->llseek;
 122         }
 123         return fn(file, offset, origin);
 124 }
 125 EXPORT_SYMBOL(vfs_llseek);
 126
 127 asmlinkage off_t sys_lseek(unsigned int fd, off_t offset, unsigned int origin)
 128 {
 129         off_t retval;
 130         struct file * file;
 131         int fput_needed;
 132
 133         retval = -EBADF;
 134         file = fget_light(fd, &fput_needed);
 135         if (!file)
 136                 goto bad;
 137
 138         retval = -EINVAL;
 139         if (origin <= 2) {
 140                 loff_t res = vfs_llseek(file, offset, origin);
 141                 retval = res;
 142                 if (res != (loff_t)retval)
 143                         retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
 144         }
 145         fput_light(file, fput_needed);
 146 bad:
 147         return retval;
 148 }
 149 EXPORT_SYMBOL_GPL(sys_lseek);
 150
 151 #ifdef __ARCH_WANT_SYS_LLSEEK
 152 asmlinkage long sys_llseek(unsigned int fd, unsigned long offset_high,
 153                            unsigned long offset_low, loff_t __user * result,
 154                            unsigned int origin)
 155 {
 156         int retval;
 157         struct file * file;
 158         loff_t offset;
 159         int fput_needed;
 160
 161         retval = -EBADF;
 162         file = fget_light(fd, &fput_needed);
 163         if (!file)
 164                 goto bad;
 165
 166         retval = -EINVAL;
 167         if (origin > 2)
 168                 goto out_putf;
 169
 170         offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
 171                         origin);
 172
 173         retval = (int)offset;
 174         if (offset >= 0) {
 175                 retval = -EFAULT;
 176                 if (!copy_to_user(result, &offset, sizeof(offset)))
 177                         retval = 0;
 178         }
 179 out_putf:
 180         fput_light(file, fput_needed);
 181 bad:
 182         return retval;
 183 }
 184 #endif
 185
 186 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 187 {
 188         struct kiocb kiocb;
 189         ssize_t ret;
 190
 191         init_sync_kiocb(&kiocb, filp);
 192         kiocb.ki_pos = *ppos;
 193         ret = filp->f_op->aio_read(&kiocb, buf, len, kiocb.ki_pos);
 194         if (-EIOCBQUEUED == ret)
 195                 ret = wait_on_sync_kiocb(&kiocb);
 196         *ppos = kiocb.ki_pos;
 197         return ret;
 198 }
 199
 200 EXPORT_SYMBOL(do_sync_read);
 201
 202 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 203 {
 204         struct inode *inode = file->f_dentry->d_inode;
 205         ssize_t ret;
 206
 207         if (!(file->f_mode & FMODE_READ))
 208                 return -EBADF;
 209         if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
 210                 return -EINVAL;
 211
 212         ret = locks_verify_area(FLOCK_VERIFY_READ, inode, file, *pos, count);
 213         if (!ret) {
 214                 ret = security_file_permission (file, MAY_READ);
 215                 if (!ret) {
 216                         if (file->f_op->read)
 217                                 ret = file->f_op->read(file, buf, count, pos);
 218                         else
 219                                 ret = do_sync_read(file, buf, count, pos);
 220                         if (ret > 0)
 221                                 dnotify_parent(file->f_dentry, DN_ACCESS);
 222                 }
 223         }
 224
 225         return ret;
 226 }
 227
 228 EXPORT_SYMBOL(vfs_read);
 229
 230 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 231 {
 232         struct kiocb kiocb;
 233         ssize_t ret;
 234
 235         init_sync_kiocb(&kiocb, filp);
 236         kiocb.ki_pos = *ppos;
 237         ret = filp->f_op->aio_write(&kiocb, buf, len, kiocb.ki_pos);
 238         if (-EIOCBQUEUED == ret)
 239                 ret = wait_on_sync_kiocb(&kiocb);
 240         *ppos = kiocb.ki_pos;
 241         return ret;
 242 }
 243
 244 EXPORT_SYMBOL(do_sync_write);
 245
 246 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 247 {
 248         struct inode *inode = file->f_dentry->d_inode;
 249         ssize_t ret;
 250
 251         if (!(file->f_mode & FMODE_WRITE))
 252                 return -EBADF;
 253         if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
 254                 return -EINVAL;
 255
 256         ret = locks_verify_area(FLOCK_VERIFY_WRITE, inode, file, *pos, count);
 257         if (!ret) {
 258                 ret = security_file_permission (file, MAY_WRITE);
 259                 if (!ret) {
 260                         if (file->f_op->write)
 261                                 ret = file->f_op->write(file, buf, count, pos);
 262                         else
 263                                 ret = do_sync_write(file, buf, count, pos);
 264                         if (ret > 0)
 265                                 dnotify_parent(file->f_dentry, DN_MODIFY);
 266                 }
 267         }
 268
 269         return ret;
 270 }
 271
 272 EXPORT_SYMBOL(vfs_write);
 273
 274 static inline loff_t file_pos_read(struct file *file)
 275 {
 276         return file->f_pos;
 277 }
 278
 279 static inline void file_pos_write(struct file *file, loff_t pos)
 280 {
 281         file->f_pos = pos;
 282 }
 283
 284 asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count)
 285 {
 286         struct file *file;
 287         ssize_t ret = -EBADF;
 288         int fput_needed;
 289
 290         file = fget_light(fd, &fput_needed);
 291         if (file) {
 292                 loff_t pos = file_pos_read(file);
 293                 ret = vfs_read(file, buf, count, &pos);
 294                 file_pos_write(file, pos);
 295                 fput_light(file, fput_needed);
 296         }
 297
 298         return ret;
 299 }
 300 EXPORT_SYMBOL_GPL(sys_read);
 301
 302 asmlinkage ssize_t sys_write(unsigned int fd, const char __user * buf, size_t count)
 303 {
 304         struct file *file;
 305         ssize_t ret = -EBADF;
 306         int fput_needed;
 307
 308         file = fget_light(fd, &fput_needed);
 309         if (file) {
 310                 loff_t pos = file_pos_read(file);
 311                 ret = vfs_write(file, buf, count, &pos);
 312                 file_pos_write(file, pos);
 313                 fput_light(file, fput_needed);
 314         }
 315
 316         return ret;
 317 }
 318
 319 asmlinkage ssize_t sys_pread64(unsigned int fd, char __user *buf,
 320                              size_t count, loff_t pos)
 321 {
 322         struct file *file;
 323         ssize_t ret = -EBADF;
 324         int fput_needed;
 325
 326         if (pos < 0)
 327                 return -EINVAL;
 328
 329         file = fget_light(fd, &fput_needed);
 330         if (file) {
 331                 ret = -ESPIPE;
 332                 if (file->f_mode & FMODE_PREAD)
 333                         ret = vfs_read(file, buf, count, &pos);
 334                 fput_light(file, fput_needed);
 335         }
 336
 337         return ret;
 338 }
 339
 340 asmlinkage ssize_t sys_pwrite64(unsigned int fd, const char __user *buf,
 341                               size_t count, loff_t pos)
 342 {
 343         struct file *file;
 344         ssize_t ret = -EBADF;
 345         int fput_needed;
 346
 347         if (pos < 0)
 348                 return -EINVAL;
 349
 350         file = fget_light(fd, &fput_needed);
 351         if (file) {
 352                 ret = -ESPIPE;
 353                 if (file->f_mode & FMODE_PWRITE)
 354                         ret = vfs_write(file, buf, count, &pos);
 355                 fput_light(file, fput_needed);
 356         }
 357
 358         return ret;
 359 }
 360
 361 /*
 362  * Reduce an iovec's length in-place.  Return the resulting number of segments
 363  */
 364 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
 365 {
 366         unsigned long seg = 0;
 367         size_t len = 0;
 368
 369         while (seg < nr_segs) {
 370                 seg++;
 371                 if (len + iov->iov_len >= to) {
 372                         iov->iov_len = to - len;
 373                         break;
 374                 }
 375                 len += iov->iov_len;
 376                 iov++;
 377         }
 378         return seg;
 379 }
 380
 381 EXPORT_SYMBOL(iov_shorten);
 382
 383 static ssize_t do_readv_writev(int type, struct file *file,
 384                                const struct iovec __user * uvector,
 385                                unsigned long nr_segs, loff_t *pos)
 386 {
 387         typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
 388         typedef ssize_t (*iov_fn_t)(struct file *, const struct iovec *, unsigned long, loff_t *);
 389
 390         size_t tot_len;
 391         struct iovec iovstack[UIO_FASTIOV];
 392         struct iovec *iov=iovstack, *vector;
 393         ssize_t ret;
 394         int seg;
 395         io_fn_t fn;
 396         iov_fn_t fnv;
 397         struct inode *inode;
 398
 399         /*
 400          * SuS says "The readv() function *may* fail if the iovcnt argument
 401          * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 402          * traditionally returned zero for zero segments, so...
 403          */
 404         ret = 0;
 405         if (nr_segs == 0)
 406                 goto out;
 407
 408         /*
 409          * First get the "struct iovec" from user memory and
 410          * verify all the pointers
 411          */
 412         ret = -EINVAL;
 413         if ((nr_segs > UIO_MAXIOV) || (nr_segs <= 0))
 414                 goto out;
 415         if (!file->f_op)
 416                 goto out;
 417         if (nr_segs > UIO_FASTIOV) {
 418                 ret = -ENOMEM;
 419                 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 420                 if (!iov)
 421                         goto out;
 422         }
 423         ret = -EFAULT;
 424         if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector)))
 425                 goto out;
 426
 427         /*
 428          * Single unix specification:
 429          * We should -EINVAL if an element length is not >= 0 and fitting an
 430          * ssize_t.  The total length is fitting an ssize_t
 431          *
 432          * Be careful here because iov_len is a size_t not an ssize_t
 433          */
 434         tot_len = 0;
 435         ret = -EINVAL;
 436         for (seg = 0; seg < nr_segs; seg++) {
 437                 ssize_t len = (ssize_t)iov[seg].iov_len;
 438
 439                 if (len < 0)    /* size_t not fitting an ssize_t .. */
 440                         goto out;
 441                 tot_len += len;
 442                 if ((ssize_t)tot_len < 0) /* maths overflow on the ssize_t */
 443                         goto out;
 444         }
 445         if (tot_len == 0) {
 446                 ret = 0;
 447                 goto out;
 448         }
 449
 450         inode = file->f_dentry->d_inode;
 451         /* VERIFY_WRITE actually means a read, as we write to user space */
 452         ret = locks_verify_area((type == READ
 453                                  ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE),
 454                                 inode, file, *pos, tot_len);
 455         if (ret)
 456                 goto out;
 457
 458         fnv = NULL;
 459         if (type == READ) {
 460                 fn = file->f_op->read;
 461                 fnv = file->f_op->readv;
 462         } else {
 463                 fn = (io_fn_t)file->f_op->write;
 464                 fnv = file->f_op->writev;
 465         }
 466         if (fnv) {
 467                 ret = fnv(file, iov, nr_segs, pos);
 468                 goto out;
 469         }
 470
 471         /* Do it by hand, with file-ops */
 472         ret = 0;
 473         vector = iov;
 474         while (nr_segs > 0) {
 475                 void __user * base;
 476                 size_t len;
 477                 ssize_t nr;
 478
 479                 base = vector->iov_base;
 480                 len = vector->iov_len;
 481                 vector++;
 482                 nr_segs--;
 483
 484                 nr = fn(file, base, len, pos);
 485
 486                 if (nr < 0) {
 487                         if (!ret) ret = nr;
 488                         break;
 489                 }
 490                 ret += nr;
 491                 if (nr != len)
 492                         break;
 493         }
 494 out:
 495         if (iov != iovstack)
 496                 kfree(iov);
 497         if ((ret + (type == READ)) > 0)
 498                 dnotify_parent(file->f_dentry,
 499                                 (type == READ) ? DN_ACCESS : DN_MODIFY);
 500         return ret;
 501 }
 502
 503 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
 504                   unsigned long vlen, loff_t *pos)
 505 {
 506         if (!(file->f_mode & FMODE_READ))
 507                 return -EBADF;
 508         if (!file->f_op || (!file->f_op->readv && !file->f_op->read))
 509                 return -EINVAL;
 510
 511         return do_readv_writev(READ, file, vec, vlen, pos);
 512 }
 513
 514 EXPORT_SYMBOL(vfs_readv);
 515
 516 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 517                    unsigned long vlen, loff_t *pos)
 518 {
 519         if (!(file->f_mode & FMODE_WRITE))
 520                 return -EBADF;
 521         if (!file->f_op || (!file->f_op->writev && !file->f_op->write))
 522                 return -EINVAL;
 523
 524         return do_readv_writev(WRITE, file, vec, vlen, pos);
 525 }
 526
 527 EXPORT_SYMBOL(vfs_writev);
 528
 529 asmlinkage ssize_t
 530 sys_readv(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
 531 {
 532         struct file *file;
 533         ssize_t ret = -EBADF;
 534         int fput_needed;
 535
 536         file = fget_light(fd, &fput_needed);
 537         if (file) {
 538                 loff_t pos = file_pos_read(file);
 539                 ret = vfs_readv(file, vec, vlen, &pos);
 540                 file_pos_write(file, pos);
 541                 fput_light(file, fput_needed);
 542         }
 543
 544         return ret;
 545 }
 546
 547 asmlinkage ssize_t
 548 sys_writev(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
 549 {
 550         struct file *file;
 551         ssize_t ret = -EBADF;
 552         int fput_needed;
 553
 554         file = fget_light(fd, &fput_needed);
 555         if (file) {
 556                 loff_t pos = file_pos_read(file);
 557                 ret = vfs_writev(file, vec, vlen, &pos);
 558                 file_pos_write(file, pos);
 559                 fput_light(file, fput_needed);
 560         }
 561
 562         return ret;
 563 }
 564
 565 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
 566                            size_t count, loff_t max)
 567 {
 568         struct file * in_file, * out_file;
 569         struct inode * in_inode, * out_inode;
 570         loff_t pos;
 571         ssize_t retval;
 572         int fput_needed_in, fput_needed_out;
 573
 574         /*
 575          * Get input file, and verify that it is ok..
 576          */
 577         retval = -EBADF;
 578         in_file = fget_light(in_fd, &fput_needed_in);
 579         if (!in_file)
 580                 goto out;
 581         if (!(in_file->f_mode & FMODE_READ))
 582                 goto fput_in;
 583         retval = -EINVAL;
 584         in_inode = in_file->f_dentry->d_inode;
 585         if (!in_inode)
 586                 goto fput_in;
 587         if (!in_file->f_op || !in_file->f_op->sendfile)
 588                 goto fput_in;
 589         retval = -ESPIPE;
 590         if (!ppos)
 591                 ppos = &in_file->f_pos;
 592         else
 593                 if (!(in_file->f_mode & FMODE_PREAD))
 594                         goto fput_in;
 595         retval = locks_verify_area(FLOCK_VERIFY_READ, in_inode, in_file, *ppos, count);
 596         if (retval)
 597                 goto fput_in;
 598
 599         retval = security_file_permission (in_file, MAY_READ);
 600         if (retval)
 601                 goto fput_in;
 602
 603         /*
 604          * Get output file, and verify that it is ok..
 605          */
 606         retval = -EBADF;
 607         out_file = fget_light(out_fd, &fput_needed_out);
 608         if (!out_file)
 609                 goto fput_in;
 610         if (!(out_file->f_mode & FMODE_WRITE))
 611                 goto fput_out;
 612         retval = -EINVAL;
 613         if (!out_file->f_op || !out_file->f_op->sendpage)
 614                 goto fput_out;
 615         out_inode = out_file->f_dentry->d_inode;
 616         retval = locks_verify_area(FLOCK_VERIFY_WRITE, out_inode, out_file, out_file->f_pos, count);
 617         if (retval)
 618                 goto fput_out;
 619
 620         retval = security_file_permission (out_file, MAY_WRITE);
 621         if (retval)
 622                 goto fput_out;
 623
 624         if (!max)
 625                 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
 626
 627         pos = *ppos;
 628         retval = -EINVAL;
 629         if (unlikely(pos < 0))
 630                 goto fput_out;
 631         if (unlikely(pos + count > max)) {
 632                 retval = -EOVERFLOW;
 633                 if (pos >= max)
 634                         goto fput_out;
 635                 count = max - pos;
 636         }
 637
 638         retval = in_file->f_op->sendfile(in_file, ppos, count, file_send_actor, out_file);
 639
 640         if (*ppos > max)
 641                 retval = -EOVERFLOW;
 642
 643 fput_out:
 644         fput_light(out_file, fput_needed_out);
 645 fput_in:
 646         fput_light(in_file, fput_needed_in);
 647 out:
 648         return retval;
 649 }
 650
 651 asmlinkage ssize_t sys_sendfile(int out_fd, int in_fd, off_t __user *offset, size_t count)
 652 {
 653         loff_t pos;
 654         off_t off;
 655         ssize_t ret;
 656
 657         if (offset) {
 658                 if (unlikely(get_user(off, offset)))
 659                         return -EFAULT;
 660                 pos = off;
 661                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
 662                 if (unlikely(put_user(pos, offset)))
 663                         return -EFAULT;
 664                 return ret;
 665         }
 666
 667         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 668 }
 669
 670 asmlinkage ssize_t sys_sendfile64(int out_fd, int in_fd, loff_t __user *offset, size_t count)
 671 {
 672         loff_t pos;
 673         ssize_t ret;
 674
 675         if (offset) {
 676                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
 677                         return -EFAULT;
 678                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
 679                 if (unlikely(put_user(pos, offset)))
 680                         return -EFAULT;
 681                 return ret;
 682         }
 683
 684         return do_sendfile(out_fd, in_fd, NULL, count, 0);
 685 }