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