1 /* 2 * linux/fs/read_write.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 */ 6 7 #include <linux/slab.h> 8 #include <linux/stat.h> 9 #include <linux/fcntl.h> 10 #include <linux/file.h> 11 #include <linux/uio.h> 12 #include <linux/fsnotify.h> 13 #include <linux/security.h> 14 #include <linux/export.h> 15 #include <linux/syscalls.h> 16 #include <linux/pagemap.h> 17 #include <linux/splice.h> 18 #include <linux/compat.h> 19 #include <linux/mount.h> 20 #include <linux/fs.h> 21 #include "internal.h" 22 23 #include <asm/uaccess.h> 24 #include <asm/unistd.h> 25 26 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *); 27 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *); 28 29 const struct file_operations generic_ro_fops = { 30 .llseek = generic_file_llseek, 31 .read_iter = generic_file_read_iter, 32 .mmap = generic_file_readonly_mmap, 33 .splice_read = generic_file_splice_read, 34 }; 35 36 EXPORT_SYMBOL(generic_ro_fops); 37 38 static inline int unsigned_offsets(struct file *file) 39 { 40 return file->f_mode & FMODE_UNSIGNED_OFFSET; 41 } 42 43 /** 44 * vfs_setpos - update the file offset for lseek 45 * @file: file structure in question 46 * @offset: file offset to seek to 47 * @maxsize: maximum file size 48 * 49 * This is a low-level filesystem helper for updating the file offset to 50 * the value specified by @offset if the given offset is valid and it is 51 * not equal to the current file offset. 52 * 53 * Return the specified offset on success and -EINVAL on invalid offset. 54 */ 55 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize) 56 { 57 if (offset < 0 && !unsigned_offsets(file)) 58 return -EINVAL; 59 if (offset > maxsize) 60 return -EINVAL; 61 62 if (offset != file->f_pos) { 63 file->f_pos = offset; 64 file->f_version = 0; 65 } 66 return offset; 67 } 68 EXPORT_SYMBOL(vfs_setpos); 69 70 /** 71 * generic_file_llseek_size - generic llseek implementation for regular files 72 * @file: file structure to seek on 73 * @offset: file offset to seek to 74 * @whence: type of seek 75 * @size: max size of this file in file system 76 * @eof: offset used for SEEK_END position 77 * 78 * This is a variant of generic_file_llseek that allows passing in a custom 79 * maximum file size and a custom EOF position, for e.g. hashed directories 80 * 81 * Synchronization: 82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms) 83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes. 84 * read/writes behave like SEEK_SET against seeks. 85 */ 86 loff_t 87 generic_file_llseek_size(struct file *file, loff_t offset, int whence, 88 loff_t maxsize, loff_t eof) 89 { 90 switch (whence) { 91 case SEEK_END: 92 offset += eof; 93 break; 94 case SEEK_CUR: 95 /* 96 * Here we special-case the lseek(fd, 0, SEEK_CUR) 97 * position-querying operation. Avoid rewriting the "same" 98 * f_pos value back to the file because a concurrent read(), 99 * write() or lseek() might have altered it 100 */ 101 if (offset == 0) 102 return file->f_pos; 103 /* 104 * f_lock protects against read/modify/write race with other 105 * SEEK_CURs. Note that parallel writes and reads behave 106 * like SEEK_SET. 107 */ 108 spin_lock(&file->f_lock); 109 offset = vfs_setpos(file, file->f_pos + offset, maxsize); 110 spin_unlock(&file->f_lock); 111 return offset; 112 case SEEK_DATA: 113 /* 114 * In the generic case the entire file is data, so as long as 115 * offset isn't at the end of the file then the offset is data. 116 */ 117 if (offset >= eof) 118 return -ENXIO; 119 break; 120 case SEEK_HOLE: 121 /* 122 * There is a virtual hole at the end of the file, so as long as 123 * offset isn't i_size or larger, return i_size. 124 */ 125 if (offset >= eof) 126 return -ENXIO; 127 offset = eof; 128 break; 129 } 130 131 return vfs_setpos(file, offset, maxsize); 132 } 133 EXPORT_SYMBOL(generic_file_llseek_size); 134 135 /** 136 * generic_file_llseek - generic llseek implementation for regular files 137 * @file: file structure to seek on 138 * @offset: file offset to seek to 139 * @whence: type of seek 140 * 141 * This is a generic implemenation of ->llseek useable for all normal local 142 * filesystems. It just updates the file offset to the value specified by 143 * @offset and @whence. 144 */ 145 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence) 146 { 147 struct inode *inode = file->f_mapping->host; 148 149 return generic_file_llseek_size(file, offset, whence, 150 inode->i_sb->s_maxbytes, 151 i_size_read(inode)); 152 } 153 EXPORT_SYMBOL(generic_file_llseek); 154 155 /** 156 * fixed_size_llseek - llseek implementation for fixed-sized devices 157 * @file: file structure to seek on 158 * @offset: file offset to seek to 159 * @whence: type of seek 160 * @size: size of the file 161 * 162 */ 163 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size) 164 { 165 switch (whence) { 166 case SEEK_SET: case SEEK_CUR: case SEEK_END: 167 return generic_file_llseek_size(file, offset, whence, 168 size, size); 169 default: 170 return -EINVAL; 171 } 172 } 173 EXPORT_SYMBOL(fixed_size_llseek); 174 175 /** 176 * no_seek_end_llseek - llseek implementation for fixed-sized devices 177 * @file: file structure to seek on 178 * @offset: file offset to seek to 179 * @whence: type of seek 180 * 181 */ 182 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence) 183 { 184 switch (whence) { 185 case SEEK_SET: case SEEK_CUR: 186 return generic_file_llseek_size(file, offset, whence, 187 OFFSET_MAX, 0); 188 default: 189 return -EINVAL; 190 } 191 } 192 EXPORT_SYMBOL(no_seek_end_llseek); 193 194 /** 195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices 196 * @file: file structure to seek on 197 * @offset: file offset to seek to 198 * @whence: type of seek 199 * @size: maximal offset allowed 200 * 201 */ 202 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size) 203 { 204 switch (whence) { 205 case SEEK_SET: case SEEK_CUR: 206 return generic_file_llseek_size(file, offset, whence, 207 size, 0); 208 default: 209 return -EINVAL; 210 } 211 } 212 EXPORT_SYMBOL(no_seek_end_llseek_size); 213 214 /** 215 * noop_llseek - No Operation Performed llseek implementation 216 * @file: file structure to seek on 217 * @offset: file offset to seek to 218 * @whence: type of seek 219 * 220 * This is an implementation of ->llseek useable for the rare special case when 221 * userspace expects the seek to succeed but the (device) file is actually not 222 * able to perform the seek. In this case you use noop_llseek() instead of 223 * falling back to the default implementation of ->llseek. 224 */ 225 loff_t noop_llseek(struct file *file, loff_t offset, int whence) 226 { 227 return file->f_pos; 228 } 229 EXPORT_SYMBOL(noop_llseek); 230 231 loff_t no_llseek(struct file *file, loff_t offset, int whence) 232 { 233 return -ESPIPE; 234 } 235 EXPORT_SYMBOL(no_llseek); 236 237 loff_t default_llseek(struct file *file, loff_t offset, int whence) 238 { 239 struct inode *inode = file_inode(file); 240 loff_t retval; 241 242 inode_lock(inode); 243 switch (whence) { 244 case SEEK_END: 245 offset += i_size_read(inode); 246 break; 247 case SEEK_CUR: 248 if (offset == 0) { 249 retval = file->f_pos; 250 goto out; 251 } 252 offset += file->f_pos; 253 break; 254 case SEEK_DATA: 255 /* 256 * In the generic case the entire file is data, so as 257 * long as offset isn't at the end of the file then the 258 * offset is data. 259 */ 260 if (offset >= inode->i_size) { 261 retval = -ENXIO; 262 goto out; 263 } 264 break; 265 case SEEK_HOLE: 266 /* 267 * There is a virtual hole at the end of the file, so 268 * as long as offset isn't i_size or larger, return 269 * i_size. 270 */ 271 if (offset >= inode->i_size) { 272 retval = -ENXIO; 273 goto out; 274 } 275 offset = inode->i_size; 276 break; 277 } 278 retval = -EINVAL; 279 if (offset >= 0 || unsigned_offsets(file)) { 280 if (offset != file->f_pos) { 281 file->f_pos = offset; 282 file->f_version = 0; 283 } 284 retval = offset; 285 } 286 out: 287 inode_unlock(inode); 288 return retval; 289 } 290 EXPORT_SYMBOL(default_llseek); 291 292 loff_t vfs_llseek(struct file *file, loff_t offset, int whence) 293 { 294 loff_t (*fn)(struct file *, loff_t, int); 295 296 fn = no_llseek; 297 if (file->f_mode & FMODE_LSEEK) { 298 if (file->f_op->llseek) 299 fn = file->f_op->llseek; 300 } 301 return fn(file, offset, whence); 302 } 303 EXPORT_SYMBOL(vfs_llseek); 304 305 static inline struct fd fdget_pos(int fd) 306 { 307 return __to_fd(__fdget_pos(fd)); 308 } 309 310 static inline void fdput_pos(struct fd f) 311 { 312 if (f.flags & FDPUT_POS_UNLOCK) 313 mutex_unlock(&f.file->f_pos_lock); 314 fdput(f); 315 } 316 317 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence) 318 { 319 off_t retval; 320 struct fd f = fdget_pos(fd); 321 if (!f.file) 322 return -EBADF; 323 324 retval = -EINVAL; 325 if (whence <= SEEK_MAX) { 326 loff_t res = vfs_llseek(f.file, offset, whence); 327 retval = res; 328 if (res != (loff_t)retval) 329 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */ 330 } 331 fdput_pos(f); 332 return retval; 333 } 334 335 #ifdef CONFIG_COMPAT 336 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence) 337 { 338 return sys_lseek(fd, offset, whence); 339 } 340 #endif 341 342 #ifdef __ARCH_WANT_SYS_LLSEEK 343 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high, 344 unsigned long, offset_low, loff_t __user *, result, 345 unsigned int, whence) 346 { 347 int retval; 348 struct fd f = fdget_pos(fd); 349 loff_t offset; 350 351 if (!f.file) 352 return -EBADF; 353 354 retval = -EINVAL; 355 if (whence > SEEK_MAX) 356 goto out_putf; 357 358 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low, 359 whence); 360 361 retval = (int)offset; 362 if (offset >= 0) { 363 retval = -EFAULT; 364 if (!copy_to_user(result, &offset, sizeof(offset))) 365 retval = 0; 366 } 367 out_putf: 368 fdput_pos(f); 369 return retval; 370 } 371 #endif 372 373 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos) 374 { 375 struct kiocb kiocb; 376 ssize_t ret; 377 378 if (!file->f_op->read_iter) 379 return -EINVAL; 380 381 init_sync_kiocb(&kiocb, file); 382 kiocb.ki_pos = *ppos; 383 384 iter->type |= READ; 385 ret = file->f_op->read_iter(&kiocb, iter); 386 BUG_ON(ret == -EIOCBQUEUED); 387 if (ret > 0) 388 *ppos = kiocb.ki_pos; 389 return ret; 390 } 391 EXPORT_SYMBOL(vfs_iter_read); 392 393 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos) 394 { 395 struct kiocb kiocb; 396 ssize_t ret; 397 398 if (!file->f_op->write_iter) 399 return -EINVAL; 400 401 init_sync_kiocb(&kiocb, file); 402 kiocb.ki_pos = *ppos; 403 404 iter->type |= WRITE; 405 ret = file->f_op->write_iter(&kiocb, iter); 406 BUG_ON(ret == -EIOCBQUEUED); 407 if (ret > 0) 408 *ppos = kiocb.ki_pos; 409 return ret; 410 } 411 EXPORT_SYMBOL(vfs_iter_write); 412 413 /* 414 * rw_verify_area doesn't like huge counts. We limit 415 * them to something that fits in "int" so that others 416 * won't have to do range checks all the time. 417 */ 418 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count) 419 { 420 struct inode *inode; 421 loff_t pos; 422 int retval = -EINVAL; 423 424 inode = file_inode(file); 425 if (unlikely((ssize_t) count < 0)) 426 return retval; 427 pos = *ppos; 428 if (unlikely(pos < 0)) { 429 if (!unsigned_offsets(file)) 430 return retval; 431 if (count >= -pos) /* both values are in 0..LLONG_MAX */ 432 return -EOVERFLOW; 433 } else if (unlikely((loff_t) (pos + count) < 0)) { 434 if (!unsigned_offsets(file)) 435 return retval; 436 } 437 438 if (unlikely(inode->i_flctx && mandatory_lock(inode))) { 439 retval = locks_mandatory_area(inode, file, pos, pos + count - 1, 440 read_write == READ ? F_RDLCK : F_WRLCK); 441 if (retval < 0) 442 return retval; 443 } 444 retval = security_file_permission(file, 445 read_write == READ ? MAY_READ : MAY_WRITE); 446 if (retval) 447 return retval; 448 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count; 449 } 450 451 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos) 452 { 453 struct iovec iov = { .iov_base = buf, .iov_len = len }; 454 struct kiocb kiocb; 455 struct iov_iter iter; 456 ssize_t ret; 457 458 init_sync_kiocb(&kiocb, filp); 459 kiocb.ki_pos = *ppos; 460 iov_iter_init(&iter, READ, &iov, 1, len); 461 462 ret = filp->f_op->read_iter(&kiocb, &iter); 463 BUG_ON(ret == -EIOCBQUEUED); 464 *ppos = kiocb.ki_pos; 465 return ret; 466 } 467 468 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count, 469 loff_t *pos) 470 { 471 if (file->f_op->read) 472 return file->f_op->read(file, buf, count, pos); 473 else if (file->f_op->read_iter) 474 return new_sync_read(file, buf, count, pos); 475 else 476 return -EINVAL; 477 } 478 EXPORT_SYMBOL(__vfs_read); 479 480 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos) 481 { 482 ssize_t ret; 483 484 if (!(file->f_mode & FMODE_READ)) 485 return -EBADF; 486 if (!(file->f_mode & FMODE_CAN_READ)) 487 return -EINVAL; 488 if (unlikely(!access_ok(VERIFY_WRITE, buf, count))) 489 return -EFAULT; 490 491 ret = rw_verify_area(READ, file, pos, count); 492 if (ret >= 0) { 493 count = ret; 494 ret = __vfs_read(file, buf, count, pos); 495 if (ret > 0) { 496 fsnotify_access(file); 497 add_rchar(current, ret); 498 } 499 inc_syscr(current); 500 } 501 502 return ret; 503 } 504 505 EXPORT_SYMBOL(vfs_read); 506 507 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos) 508 { 509 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len }; 510 struct kiocb kiocb; 511 struct iov_iter iter; 512 ssize_t ret; 513 514 init_sync_kiocb(&kiocb, filp); 515 kiocb.ki_pos = *ppos; 516 iov_iter_init(&iter, WRITE, &iov, 1, len); 517 518 ret = filp->f_op->write_iter(&kiocb, &iter); 519 BUG_ON(ret == -EIOCBQUEUED); 520 if (ret > 0) 521 *ppos = kiocb.ki_pos; 522 return ret; 523 } 524 525 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count, 526 loff_t *pos) 527 { 528 if (file->f_op->write) 529 return file->f_op->write(file, p, count, pos); 530 else if (file->f_op->write_iter) 531 return new_sync_write(file, p, count, pos); 532 else 533 return -EINVAL; 534 } 535 EXPORT_SYMBOL(__vfs_write); 536 537 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos) 538 { 539 mm_segment_t old_fs; 540 const char __user *p; 541 ssize_t ret; 542 543 if (!(file->f_mode & FMODE_CAN_WRITE)) 544 return -EINVAL; 545 546 old_fs = get_fs(); 547 set_fs(get_ds()); 548 p = (__force const char __user *)buf; 549 if (count > MAX_RW_COUNT) 550 count = MAX_RW_COUNT; 551 ret = __vfs_write(file, p, count, pos); 552 set_fs(old_fs); 553 if (ret > 0) { 554 fsnotify_modify(file); 555 add_wchar(current, ret); 556 } 557 inc_syscw(current); 558 return ret; 559 } 560 561 EXPORT_SYMBOL(__kernel_write); 562 563 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) 564 { 565 ssize_t ret; 566 567 if (!(file->f_mode & FMODE_WRITE)) 568 return -EBADF; 569 if (!(file->f_mode & FMODE_CAN_WRITE)) 570 return -EINVAL; 571 if (unlikely(!access_ok(VERIFY_READ, buf, count))) 572 return -EFAULT; 573 574 ret = rw_verify_area(WRITE, file, pos, count); 575 if (ret >= 0) { 576 count = ret; 577 file_start_write(file); 578 ret = __vfs_write(file, buf, count, pos); 579 if (ret > 0) { 580 fsnotify_modify(file); 581 add_wchar(current, ret); 582 } 583 inc_syscw(current); 584 file_end_write(file); 585 } 586 587 return ret; 588 } 589 590 EXPORT_SYMBOL(vfs_write); 591 592 static inline loff_t file_pos_read(struct file *file) 593 { 594 return file->f_pos; 595 } 596 597 static inline void file_pos_write(struct file *file, loff_t pos) 598 { 599 file->f_pos = pos; 600 } 601 602 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count) 603 { 604 struct fd f = fdget_pos(fd); 605 ssize_t ret = -EBADF; 606 607 if (f.file) { 608 loff_t pos = file_pos_read(f.file); 609 ret = vfs_read(f.file, buf, count, &pos); 610 if (ret >= 0) 611 file_pos_write(f.file, pos); 612 fdput_pos(f); 613 } 614 return ret; 615 } 616 617 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf, 618 size_t, count) 619 { 620 struct fd f = fdget_pos(fd); 621 ssize_t ret = -EBADF; 622 623 if (f.file) { 624 loff_t pos = file_pos_read(f.file); 625 ret = vfs_write(f.file, buf, count, &pos); 626 if (ret >= 0) 627 file_pos_write(f.file, pos); 628 fdput_pos(f); 629 } 630 631 return ret; 632 } 633 634 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf, 635 size_t, count, loff_t, pos) 636 { 637 struct fd f; 638 ssize_t ret = -EBADF; 639 640 if (pos < 0) 641 return -EINVAL; 642 643 f = fdget(fd); 644 if (f.file) { 645 ret = -ESPIPE; 646 if (f.file->f_mode & FMODE_PREAD) 647 ret = vfs_read(f.file, buf, count, &pos); 648 fdput(f); 649 } 650 651 return ret; 652 } 653 654 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf, 655 size_t, count, loff_t, pos) 656 { 657 struct fd f; 658 ssize_t ret = -EBADF; 659 660 if (pos < 0) 661 return -EINVAL; 662 663 f = fdget(fd); 664 if (f.file) { 665 ret = -ESPIPE; 666 if (f.file->f_mode & FMODE_PWRITE) 667 ret = vfs_write(f.file, buf, count, &pos); 668 fdput(f); 669 } 670 671 return ret; 672 } 673 674 /* 675 * Reduce an iovec's length in-place. Return the resulting number of segments 676 */ 677 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to) 678 { 679 unsigned long seg = 0; 680 size_t len = 0; 681 682 while (seg < nr_segs) { 683 seg++; 684 if (len + iov->iov_len >= to) { 685 iov->iov_len = to - len; 686 break; 687 } 688 len += iov->iov_len; 689 iov++; 690 } 691 return seg; 692 } 693 EXPORT_SYMBOL(iov_shorten); 694 695 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter, 696 loff_t *ppos, iter_fn_t fn, int flags) 697 { 698 struct kiocb kiocb; 699 ssize_t ret; 700 701 if (flags & ~RWF_HIPRI) 702 return -EOPNOTSUPP; 703 704 init_sync_kiocb(&kiocb, filp); 705 if (flags & RWF_HIPRI) 706 kiocb.ki_flags |= IOCB_HIPRI; 707 kiocb.ki_pos = *ppos; 708 709 ret = fn(&kiocb, iter); 710 BUG_ON(ret == -EIOCBQUEUED); 711 *ppos = kiocb.ki_pos; 712 return ret; 713 } 714 715 /* Do it by hand, with file-ops */ 716 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter, 717 loff_t *ppos, io_fn_t fn, int flags) 718 { 719 ssize_t ret = 0; 720 721 if (flags & ~RWF_HIPRI) 722 return -EOPNOTSUPP; 723 724 while (iov_iter_count(iter)) { 725 struct iovec iovec = iov_iter_iovec(iter); 726 ssize_t nr; 727 728 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos); 729 730 if (nr < 0) { 731 if (!ret) 732 ret = nr; 733 break; 734 } 735 ret += nr; 736 if (nr != iovec.iov_len) 737 break; 738 iov_iter_advance(iter, nr); 739 } 740 741 return ret; 742 } 743 744 /* A write operation does a read from user space and vice versa */ 745 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ) 746 747 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector, 748 unsigned long nr_segs, unsigned long fast_segs, 749 struct iovec *fast_pointer, 750 struct iovec **ret_pointer) 751 { 752 unsigned long seg; 753 ssize_t ret; 754 struct iovec *iov = fast_pointer; 755 756 /* 757 * SuS says "The readv() function *may* fail if the iovcnt argument 758 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has 759 * traditionally returned zero for zero segments, so... 760 */ 761 if (nr_segs == 0) { 762 ret = 0; 763 goto out; 764 } 765 766 /* 767 * First get the "struct iovec" from user memory and 768 * verify all the pointers 769 */ 770 if (nr_segs > UIO_MAXIOV) { 771 ret = -EINVAL; 772 goto out; 773 } 774 if (nr_segs > fast_segs) { 775 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL); 776 if (iov == NULL) { 777 ret = -ENOMEM; 778 goto out; 779 } 780 } 781 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) { 782 ret = -EFAULT; 783 goto out; 784 } 785 786 /* 787 * According to the Single Unix Specification we should return EINVAL 788 * if an element length is < 0 when cast to ssize_t or if the 789 * total length would overflow the ssize_t return value of the 790 * system call. 791 * 792 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the 793 * overflow case. 794 */ 795 ret = 0; 796 for (seg = 0; seg < nr_segs; seg++) { 797 void __user *buf = iov[seg].iov_base; 798 ssize_t len = (ssize_t)iov[seg].iov_len; 799 800 /* see if we we're about to use an invalid len or if 801 * it's about to overflow ssize_t */ 802 if (len < 0) { 803 ret = -EINVAL; 804 goto out; 805 } 806 if (type >= 0 807 && unlikely(!access_ok(vrfy_dir(type), buf, len))) { 808 ret = -EFAULT; 809 goto out; 810 } 811 if (len > MAX_RW_COUNT - ret) { 812 len = MAX_RW_COUNT - ret; 813 iov[seg].iov_len = len; 814 } 815 ret += len; 816 } 817 out: 818 *ret_pointer = iov; 819 return ret; 820 } 821 822 static ssize_t do_readv_writev(int type, struct file *file, 823 const struct iovec __user * uvector, 824 unsigned long nr_segs, loff_t *pos, 825 int flags) 826 { 827 size_t tot_len; 828 struct iovec iovstack[UIO_FASTIOV]; 829 struct iovec *iov = iovstack; 830 struct iov_iter iter; 831 ssize_t ret; 832 io_fn_t fn; 833 iter_fn_t iter_fn; 834 835 ret = import_iovec(type, uvector, nr_segs, 836 ARRAY_SIZE(iovstack), &iov, &iter); 837 if (ret < 0) 838 return ret; 839 840 tot_len = iov_iter_count(&iter); 841 if (!tot_len) 842 goto out; 843 ret = rw_verify_area(type, file, pos, tot_len); 844 if (ret < 0) 845 goto out; 846 847 if (type == READ) { 848 fn = file->f_op->read; 849 iter_fn = file->f_op->read_iter; 850 } else { 851 fn = (io_fn_t)file->f_op->write; 852 iter_fn = file->f_op->write_iter; 853 file_start_write(file); 854 } 855 856 if (iter_fn) 857 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags); 858 else 859 ret = do_loop_readv_writev(file, &iter, pos, fn, flags); 860 861 if (type != READ) 862 file_end_write(file); 863 864 out: 865 kfree(iov); 866 if ((ret + (type == READ)) > 0) { 867 if (type == READ) 868 fsnotify_access(file); 869 else 870 fsnotify_modify(file); 871 } 872 return ret; 873 } 874 875 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec, 876 unsigned long vlen, loff_t *pos, int flags) 877 { 878 if (!(file->f_mode & FMODE_READ)) 879 return -EBADF; 880 if (!(file->f_mode & FMODE_CAN_READ)) 881 return -EINVAL; 882 883 return do_readv_writev(READ, file, vec, vlen, pos, flags); 884 } 885 886 EXPORT_SYMBOL(vfs_readv); 887 888 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec, 889 unsigned long vlen, loff_t *pos, int flags) 890 { 891 if (!(file->f_mode & FMODE_WRITE)) 892 return -EBADF; 893 if (!(file->f_mode & FMODE_CAN_WRITE)) 894 return -EINVAL; 895 896 return do_readv_writev(WRITE, file, vec, vlen, pos, flags); 897 } 898 899 EXPORT_SYMBOL(vfs_writev); 900 901 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec, 902 unsigned long vlen, int flags) 903 { 904 struct fd f = fdget_pos(fd); 905 ssize_t ret = -EBADF; 906 907 if (f.file) { 908 loff_t pos = file_pos_read(f.file); 909 ret = vfs_readv(f.file, vec, vlen, &pos, flags); 910 if (ret >= 0) 911 file_pos_write(f.file, pos); 912 fdput_pos(f); 913 } 914 915 if (ret > 0) 916 add_rchar(current, ret); 917 inc_syscr(current); 918 return ret; 919 } 920 921 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec, 922 unsigned long vlen, int flags) 923 { 924 struct fd f = fdget_pos(fd); 925 ssize_t ret = -EBADF; 926 927 if (f.file) { 928 loff_t pos = file_pos_read(f.file); 929 ret = vfs_writev(f.file, vec, vlen, &pos, flags); 930 if (ret >= 0) 931 file_pos_write(f.file, pos); 932 fdput_pos(f); 933 } 934 935 if (ret > 0) 936 add_wchar(current, ret); 937 inc_syscw(current); 938 return ret; 939 } 940 941 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low) 942 { 943 #define HALF_LONG_BITS (BITS_PER_LONG / 2) 944 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low; 945 } 946 947 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec, 948 unsigned long vlen, loff_t pos, int flags) 949 { 950 struct fd f; 951 ssize_t ret = -EBADF; 952 953 if (pos < 0) 954 return -EINVAL; 955 956 f = fdget(fd); 957 if (f.file) { 958 ret = -ESPIPE; 959 if (f.file->f_mode & FMODE_PREAD) 960 ret = vfs_readv(f.file, vec, vlen, &pos, flags); 961 fdput(f); 962 } 963 964 if (ret > 0) 965 add_rchar(current, ret); 966 inc_syscr(current); 967 return ret; 968 } 969 970 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec, 971 unsigned long vlen, loff_t pos, int flags) 972 { 973 struct fd f; 974 ssize_t ret = -EBADF; 975 976 if (pos < 0) 977 return -EINVAL; 978 979 f = fdget(fd); 980 if (f.file) { 981 ret = -ESPIPE; 982 if (f.file->f_mode & FMODE_PWRITE) 983 ret = vfs_writev(f.file, vec, vlen, &pos, flags); 984 fdput(f); 985 } 986 987 if (ret > 0) 988 add_wchar(current, ret); 989 inc_syscw(current); 990 return ret; 991 } 992 993 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec, 994 unsigned long, vlen) 995 { 996 return do_readv(fd, vec, vlen, 0); 997 } 998 999 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec, 1000 unsigned long, vlen) 1001 { 1002 return do_writev(fd, vec, vlen, 0); 1003 } 1004 1005 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec, 1006 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) 1007 { 1008 loff_t pos = pos_from_hilo(pos_h, pos_l); 1009 1010 return do_preadv(fd, vec, vlen, pos, 0); 1011 } 1012 1013 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec, 1014 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h, 1015 int, flags) 1016 { 1017 loff_t pos = pos_from_hilo(pos_h, pos_l); 1018 1019 if (pos == -1) 1020 return do_readv(fd, vec, vlen, flags); 1021 1022 return do_preadv(fd, vec, vlen, pos, flags); 1023 } 1024 1025 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec, 1026 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h) 1027 { 1028 loff_t pos = pos_from_hilo(pos_h, pos_l); 1029 1030 return do_pwritev(fd, vec, vlen, pos, 0); 1031 } 1032 1033 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec, 1034 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h, 1035 int, flags) 1036 { 1037 loff_t pos = pos_from_hilo(pos_h, pos_l); 1038 1039 if (pos == -1) 1040 return do_writev(fd, vec, vlen, flags); 1041 1042 return do_pwritev(fd, vec, vlen, pos, flags); 1043 } 1044 1045 #ifdef CONFIG_COMPAT 1046 1047 static ssize_t compat_do_readv_writev(int type, struct file *file, 1048 const struct compat_iovec __user *uvector, 1049 unsigned long nr_segs, loff_t *pos, 1050 int flags) 1051 { 1052 compat_ssize_t tot_len; 1053 struct iovec iovstack[UIO_FASTIOV]; 1054 struct iovec *iov = iovstack; 1055 struct iov_iter iter; 1056 ssize_t ret; 1057 io_fn_t fn; 1058 iter_fn_t iter_fn; 1059 1060 ret = compat_import_iovec(type, uvector, nr_segs, 1061 UIO_FASTIOV, &iov, &iter); 1062 if (ret < 0) 1063 return ret; 1064 1065 tot_len = iov_iter_count(&iter); 1066 if (!tot_len) 1067 goto out; 1068 ret = rw_verify_area(type, file, pos, tot_len); 1069 if (ret < 0) 1070 goto out; 1071 1072 if (type == READ) { 1073 fn = file->f_op->read; 1074 iter_fn = file->f_op->read_iter; 1075 } else { 1076 fn = (io_fn_t)file->f_op->write; 1077 iter_fn = file->f_op->write_iter; 1078 file_start_write(file); 1079 } 1080 1081 if (iter_fn) 1082 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags); 1083 else 1084 ret = do_loop_readv_writev(file, &iter, pos, fn, flags); 1085 1086 if (type != READ) 1087 file_end_write(file); 1088 1089 out: 1090 kfree(iov); 1091 if ((ret + (type == READ)) > 0) { 1092 if (type == READ) 1093 fsnotify_access(file); 1094 else 1095 fsnotify_modify(file); 1096 } 1097 return ret; 1098 } 1099 1100 static size_t compat_readv(struct file *file, 1101 const struct compat_iovec __user *vec, 1102 unsigned long vlen, loff_t *pos, int flags) 1103 { 1104 ssize_t ret = -EBADF; 1105 1106 if (!(file->f_mode & FMODE_READ)) 1107 goto out; 1108 1109 ret = -EINVAL; 1110 if (!(file->f_mode & FMODE_CAN_READ)) 1111 goto out; 1112 1113 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags); 1114 1115 out: 1116 if (ret > 0) 1117 add_rchar(current, ret); 1118 inc_syscr(current); 1119 return ret; 1120 } 1121 1122 static size_t do_compat_readv(compat_ulong_t fd, 1123 const struct compat_iovec __user *vec, 1124 compat_ulong_t vlen, int flags) 1125 { 1126 struct fd f = fdget_pos(fd); 1127 ssize_t ret; 1128 loff_t pos; 1129 1130 if (!f.file) 1131 return -EBADF; 1132 pos = f.file->f_pos; 1133 ret = compat_readv(f.file, vec, vlen, &pos, flags); 1134 if (ret >= 0) 1135 f.file->f_pos = pos; 1136 fdput_pos(f); 1137 return ret; 1138 1139 } 1140 1141 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd, 1142 const struct compat_iovec __user *,vec, 1143 compat_ulong_t, vlen) 1144 { 1145 return do_compat_readv(fd, vec, vlen, 0); 1146 } 1147 1148 static long do_compat_preadv64(unsigned long fd, 1149 const struct compat_iovec __user *vec, 1150 unsigned long vlen, loff_t pos, int flags) 1151 { 1152 struct fd f; 1153 ssize_t ret; 1154 1155 if (pos < 0) 1156 return -EINVAL; 1157 f = fdget(fd); 1158 if (!f.file) 1159 return -EBADF; 1160 ret = -ESPIPE; 1161 if (f.file->f_mode & FMODE_PREAD) 1162 ret = compat_readv(f.file, vec, vlen, &pos, flags); 1163 fdput(f); 1164 return ret; 1165 } 1166 1167 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64 1168 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd, 1169 const struct compat_iovec __user *,vec, 1170 unsigned long, vlen, loff_t, pos) 1171 { 1172 return do_compat_preadv64(fd, vec, vlen, pos, 0); 1173 } 1174 #endif 1175 1176 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd, 1177 const struct compat_iovec __user *,vec, 1178 compat_ulong_t, vlen, u32, pos_low, u32, pos_high) 1179 { 1180 loff_t pos = ((loff_t)pos_high << 32) | pos_low; 1181 1182 return do_compat_preadv64(fd, vec, vlen, pos, 0); 1183 } 1184 1185 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd, 1186 const struct compat_iovec __user *,vec, 1187 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, 1188 int, flags) 1189 { 1190 loff_t pos = ((loff_t)pos_high << 32) | pos_low; 1191 1192 if (pos == -1) 1193 return do_compat_readv(fd, vec, vlen, flags); 1194 1195 return do_compat_preadv64(fd, vec, vlen, pos, flags); 1196 } 1197 1198 static size_t compat_writev(struct file *file, 1199 const struct compat_iovec __user *vec, 1200 unsigned long vlen, loff_t *pos, int flags) 1201 { 1202 ssize_t ret = -EBADF; 1203 1204 if (!(file->f_mode & FMODE_WRITE)) 1205 goto out; 1206 1207 ret = -EINVAL; 1208 if (!(file->f_mode & FMODE_CAN_WRITE)) 1209 goto out; 1210 1211 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0); 1212 1213 out: 1214 if (ret > 0) 1215 add_wchar(current, ret); 1216 inc_syscw(current); 1217 return ret; 1218 } 1219 1220 static size_t do_compat_writev(compat_ulong_t fd, 1221 const struct compat_iovec __user* vec, 1222 compat_ulong_t vlen, int flags) 1223 { 1224 struct fd f = fdget_pos(fd); 1225 ssize_t ret; 1226 loff_t pos; 1227 1228 if (!f.file) 1229 return -EBADF; 1230 pos = f.file->f_pos; 1231 ret = compat_writev(f.file, vec, vlen, &pos, flags); 1232 if (ret >= 0) 1233 f.file->f_pos = pos; 1234 fdput_pos(f); 1235 return ret; 1236 } 1237 1238 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd, 1239 const struct compat_iovec __user *, vec, 1240 compat_ulong_t, vlen) 1241 { 1242 return do_compat_writev(fd, vec, vlen, 0); 1243 } 1244 1245 static long do_compat_pwritev64(unsigned long fd, 1246 const struct compat_iovec __user *vec, 1247 unsigned long vlen, loff_t pos, int flags) 1248 { 1249 struct fd f; 1250 ssize_t ret; 1251 1252 if (pos < 0) 1253 return -EINVAL; 1254 f = fdget(fd); 1255 if (!f.file) 1256 return -EBADF; 1257 ret = -ESPIPE; 1258 if (f.file->f_mode & FMODE_PWRITE) 1259 ret = compat_writev(f.file, vec, vlen, &pos, flags); 1260 fdput(f); 1261 return ret; 1262 } 1263 1264 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64 1265 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd, 1266 const struct compat_iovec __user *,vec, 1267 unsigned long, vlen, loff_t, pos) 1268 { 1269 return do_compat_pwritev64(fd, vec, vlen, pos, 0); 1270 } 1271 #endif 1272 1273 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd, 1274 const struct compat_iovec __user *,vec, 1275 compat_ulong_t, vlen, u32, pos_low, u32, pos_high) 1276 { 1277 loff_t pos = ((loff_t)pos_high << 32) | pos_low; 1278 1279 return do_compat_pwritev64(fd, vec, vlen, pos, 0); 1280 } 1281 1282 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd, 1283 const struct compat_iovec __user *,vec, 1284 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags) 1285 { 1286 loff_t pos = ((loff_t)pos_high << 32) | pos_low; 1287 1288 if (pos == -1) 1289 return do_compat_writev(fd, vec, vlen, flags); 1290 1291 return do_compat_pwritev64(fd, vec, vlen, pos, flags); 1292 } 1293 1294 #endif 1295 1296 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, 1297 size_t count, loff_t max) 1298 { 1299 struct fd in, out; 1300 struct inode *in_inode, *out_inode; 1301 loff_t pos; 1302 loff_t out_pos; 1303 ssize_t retval; 1304 int fl; 1305 1306 /* 1307 * Get input file, and verify that it is ok.. 1308 */ 1309 retval = -EBADF; 1310 in = fdget(in_fd); 1311 if (!in.file) 1312 goto out; 1313 if (!(in.file->f_mode & FMODE_READ)) 1314 goto fput_in; 1315 retval = -ESPIPE; 1316 if (!ppos) { 1317 pos = in.file->f_pos; 1318 } else { 1319 pos = *ppos; 1320 if (!(in.file->f_mode & FMODE_PREAD)) 1321 goto fput_in; 1322 } 1323 retval = rw_verify_area(READ, in.file, &pos, count); 1324 if (retval < 0) 1325 goto fput_in; 1326 count = retval; 1327 1328 /* 1329 * Get output file, and verify that it is ok.. 1330 */ 1331 retval = -EBADF; 1332 out = fdget(out_fd); 1333 if (!out.file) 1334 goto fput_in; 1335 if (!(out.file->f_mode & FMODE_WRITE)) 1336 goto fput_out; 1337 retval = -EINVAL; 1338 in_inode = file_inode(in.file); 1339 out_inode = file_inode(out.file); 1340 out_pos = out.file->f_pos; 1341 retval = rw_verify_area(WRITE, out.file, &out_pos, count); 1342 if (retval < 0) 1343 goto fput_out; 1344 count = retval; 1345 1346 if (!max) 1347 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes); 1348 1349 if (unlikely(pos + count > max)) { 1350 retval = -EOVERFLOW; 1351 if (pos >= max) 1352 goto fput_out; 1353 count = max - pos; 1354 } 1355 1356 fl = 0; 1357 #if 0 1358 /* 1359 * We need to debate whether we can enable this or not. The 1360 * man page documents EAGAIN return for the output at least, 1361 * and the application is arguably buggy if it doesn't expect 1362 * EAGAIN on a non-blocking file descriptor. 1363 */ 1364 if (in.file->f_flags & O_NONBLOCK) 1365 fl = SPLICE_F_NONBLOCK; 1366 #endif 1367 file_start_write(out.file); 1368 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl); 1369 file_end_write(out.file); 1370 1371 if (retval > 0) { 1372 add_rchar(current, retval); 1373 add_wchar(current, retval); 1374 fsnotify_access(in.file); 1375 fsnotify_modify(out.file); 1376 out.file->f_pos = out_pos; 1377 if (ppos) 1378 *ppos = pos; 1379 else 1380 in.file->f_pos = pos; 1381 } 1382 1383 inc_syscr(current); 1384 inc_syscw(current); 1385 if (pos > max) 1386 retval = -EOVERFLOW; 1387 1388 fput_out: 1389 fdput(out); 1390 fput_in: 1391 fdput(in); 1392 out: 1393 return retval; 1394 } 1395 1396 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count) 1397 { 1398 loff_t pos; 1399 off_t off; 1400 ssize_t ret; 1401 1402 if (offset) { 1403 if (unlikely(get_user(off, offset))) 1404 return -EFAULT; 1405 pos = off; 1406 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); 1407 if (unlikely(put_user(pos, offset))) 1408 return -EFAULT; 1409 return ret; 1410 } 1411 1412 return do_sendfile(out_fd, in_fd, NULL, count, 0); 1413 } 1414 1415 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count) 1416 { 1417 loff_t pos; 1418 ssize_t ret; 1419 1420 if (offset) { 1421 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) 1422 return -EFAULT; 1423 ret = do_sendfile(out_fd, in_fd, &pos, count, 0); 1424 if (unlikely(put_user(pos, offset))) 1425 return -EFAULT; 1426 return ret; 1427 } 1428 1429 return do_sendfile(out_fd, in_fd, NULL, count, 0); 1430 } 1431 1432 #ifdef CONFIG_COMPAT 1433 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, 1434 compat_off_t __user *, offset, compat_size_t, count) 1435 { 1436 loff_t pos; 1437 off_t off; 1438 ssize_t ret; 1439 1440 if (offset) { 1441 if (unlikely(get_user(off, offset))) 1442 return -EFAULT; 1443 pos = off; 1444 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS); 1445 if (unlikely(put_user(pos, offset))) 1446 return -EFAULT; 1447 return ret; 1448 } 1449 1450 return do_sendfile(out_fd, in_fd, NULL, count, 0); 1451 } 1452 1453 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, 1454 compat_loff_t __user *, offset, compat_size_t, count) 1455 { 1456 loff_t pos; 1457 ssize_t ret; 1458 1459 if (offset) { 1460 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t)))) 1461 return -EFAULT; 1462 ret = do_sendfile(out_fd, in_fd, &pos, count, 0); 1463 if (unlikely(put_user(pos, offset))) 1464 return -EFAULT; 1465 return ret; 1466 } 1467 1468 return do_sendfile(out_fd, in_fd, NULL, count, 0); 1469 } 1470 #endif 1471 1472 /* 1473 * copy_file_range() differs from regular file read and write in that it 1474 * specifically allows return partial success. When it does so is up to 1475 * the copy_file_range method. 1476 */ 1477 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, 1478 struct file *file_out, loff_t pos_out, 1479 size_t len, unsigned int flags) 1480 { 1481 struct inode *inode_in = file_inode(file_in); 1482 struct inode *inode_out = file_inode(file_out); 1483 ssize_t ret; 1484 1485 if (flags != 0) 1486 return -EINVAL; 1487 1488 /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */ 1489 ret = rw_verify_area(READ, file_in, &pos_in, len); 1490 if (ret >= 0) 1491 ret = rw_verify_area(WRITE, file_out, &pos_out, len); 1492 if (ret < 0) 1493 return ret; 1494 1495 if (!(file_in->f_mode & FMODE_READ) || 1496 !(file_out->f_mode & FMODE_WRITE) || 1497 (file_out->f_flags & O_APPEND)) 1498 return -EBADF; 1499 1500 /* this could be relaxed once a method supports cross-fs copies */ 1501 if (inode_in->i_sb != inode_out->i_sb) 1502 return -EXDEV; 1503 1504 if (len == 0) 1505 return 0; 1506 1507 ret = mnt_want_write_file(file_out); 1508 if (ret) 1509 return ret; 1510 1511 ret = -EOPNOTSUPP; 1512 if (file_out->f_op->copy_file_range) 1513 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out, 1514 pos_out, len, flags); 1515 if (ret == -EOPNOTSUPP) 1516 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out, 1517 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0); 1518 1519 if (ret > 0) { 1520 fsnotify_access(file_in); 1521 add_rchar(current, ret); 1522 fsnotify_modify(file_out); 1523 add_wchar(current, ret); 1524 } 1525 inc_syscr(current); 1526 inc_syscw(current); 1527 1528 mnt_drop_write_file(file_out); 1529 1530 return ret; 1531 } 1532 EXPORT_SYMBOL(vfs_copy_file_range); 1533 1534 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in, 1535 int, fd_out, loff_t __user *, off_out, 1536 size_t, len, unsigned int, flags) 1537 { 1538 loff_t pos_in; 1539 loff_t pos_out; 1540 struct fd f_in; 1541 struct fd f_out; 1542 ssize_t ret = -EBADF; 1543 1544 f_in = fdget(fd_in); 1545 if (!f_in.file) 1546 goto out2; 1547 1548 f_out = fdget(fd_out); 1549 if (!f_out.file) 1550 goto out1; 1551 1552 ret = -EFAULT; 1553 if (off_in) { 1554 if (copy_from_user(&pos_in, off_in, sizeof(loff_t))) 1555 goto out; 1556 } else { 1557 pos_in = f_in.file->f_pos; 1558 } 1559 1560 if (off_out) { 1561 if (copy_from_user(&pos_out, off_out, sizeof(loff_t))) 1562 goto out; 1563 } else { 1564 pos_out = f_out.file->f_pos; 1565 } 1566 1567 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len, 1568 flags); 1569 if (ret > 0) { 1570 pos_in += ret; 1571 pos_out += ret; 1572 1573 if (off_in) { 1574 if (copy_to_user(off_in, &pos_in, sizeof(loff_t))) 1575 ret = -EFAULT; 1576 } else { 1577 f_in.file->f_pos = pos_in; 1578 } 1579 1580 if (off_out) { 1581 if (copy_to_user(off_out, &pos_out, sizeof(loff_t))) 1582 ret = -EFAULT; 1583 } else { 1584 f_out.file->f_pos = pos_out; 1585 } 1586 } 1587 1588 out: 1589 fdput(f_out); 1590 out1: 1591 fdput(f_in); 1592 out2: 1593 return ret; 1594 } 1595 1596 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write) 1597 { 1598 struct inode *inode = file_inode(file); 1599 1600 if (unlikely(pos < 0)) 1601 return -EINVAL; 1602 1603 if (unlikely((loff_t) (pos + len) < 0)) 1604 return -EINVAL; 1605 1606 if (unlikely(inode->i_flctx && mandatory_lock(inode))) { 1607 loff_t end = len ? pos + len - 1 : OFFSET_MAX; 1608 int retval; 1609 1610 retval = locks_mandatory_area(inode, file, pos, end, 1611 write ? F_WRLCK : F_RDLCK); 1612 if (retval < 0) 1613 return retval; 1614 } 1615 1616 return security_file_permission(file, write ? MAY_WRITE : MAY_READ); 1617 } 1618 1619 int vfs_clone_file_range(struct file *file_in, loff_t pos_in, 1620 struct file *file_out, loff_t pos_out, u64 len) 1621 { 1622 struct inode *inode_in = file_inode(file_in); 1623 struct inode *inode_out = file_inode(file_out); 1624 int ret; 1625 1626 if (inode_in->i_sb != inode_out->i_sb || 1627 file_in->f_path.mnt != file_out->f_path.mnt) 1628 return -EXDEV; 1629 1630 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) 1631 return -EISDIR; 1632 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) 1633 return -EINVAL; 1634 1635 if (!(file_in->f_mode & FMODE_READ) || 1636 !(file_out->f_mode & FMODE_WRITE) || 1637 (file_out->f_flags & O_APPEND)) 1638 return -EBADF; 1639 1640 if (!file_in->f_op->clone_file_range) 1641 return -EOPNOTSUPP; 1642 1643 ret = clone_verify_area(file_in, pos_in, len, false); 1644 if (ret) 1645 return ret; 1646 1647 ret = clone_verify_area(file_out, pos_out, len, true); 1648 if (ret) 1649 return ret; 1650 1651 if (pos_in + len > i_size_read(inode_in)) 1652 return -EINVAL; 1653 1654 ret = mnt_want_write_file(file_out); 1655 if (ret) 1656 return ret; 1657 1658 ret = file_in->f_op->clone_file_range(file_in, pos_in, 1659 file_out, pos_out, len); 1660 if (!ret) { 1661 fsnotify_access(file_in); 1662 fsnotify_modify(file_out); 1663 } 1664 1665 mnt_drop_write_file(file_out); 1666 return ret; 1667 } 1668 EXPORT_SYMBOL(vfs_clone_file_range); 1669 1670 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same) 1671 { 1672 struct file_dedupe_range_info *info; 1673 struct inode *src = file_inode(file); 1674 u64 off; 1675 u64 len; 1676 int i; 1677 int ret; 1678 bool is_admin = capable(CAP_SYS_ADMIN); 1679 u16 count = same->dest_count; 1680 struct file *dst_file; 1681 loff_t dst_off; 1682 ssize_t deduped; 1683 1684 if (!(file->f_mode & FMODE_READ)) 1685 return -EINVAL; 1686 1687 if (same->reserved1 || same->reserved2) 1688 return -EINVAL; 1689 1690 off = same->src_offset; 1691 len = same->src_length; 1692 1693 ret = -EISDIR; 1694 if (S_ISDIR(src->i_mode)) 1695 goto out; 1696 1697 ret = -EINVAL; 1698 if (!S_ISREG(src->i_mode)) 1699 goto out; 1700 1701 ret = clone_verify_area(file, off, len, false); 1702 if (ret < 0) 1703 goto out; 1704 ret = 0; 1705 1706 /* pre-format output fields to sane values */ 1707 for (i = 0; i < count; i++) { 1708 same->info[i].bytes_deduped = 0ULL; 1709 same->info[i].status = FILE_DEDUPE_RANGE_SAME; 1710 } 1711 1712 for (i = 0, info = same->info; i < count; i++, info++) { 1713 struct inode *dst; 1714 struct fd dst_fd = fdget(info->dest_fd); 1715 1716 dst_file = dst_fd.file; 1717 if (!dst_file) { 1718 info->status = -EBADF; 1719 goto next_loop; 1720 } 1721 dst = file_inode(dst_file); 1722 1723 ret = mnt_want_write_file(dst_file); 1724 if (ret) { 1725 info->status = ret; 1726 goto next_loop; 1727 } 1728 1729 dst_off = info->dest_offset; 1730 ret = clone_verify_area(dst_file, dst_off, len, true); 1731 if (ret < 0) { 1732 info->status = ret; 1733 goto next_file; 1734 } 1735 ret = 0; 1736 1737 if (info->reserved) { 1738 info->status = -EINVAL; 1739 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) { 1740 info->status = -EINVAL; 1741 } else if (file->f_path.mnt != dst_file->f_path.mnt) { 1742 info->status = -EXDEV; 1743 } else if (S_ISDIR(dst->i_mode)) { 1744 info->status = -EISDIR; 1745 } else if (dst_file->f_op->dedupe_file_range == NULL) { 1746 info->status = -EINVAL; 1747 } else { 1748 deduped = dst_file->f_op->dedupe_file_range(file, off, 1749 len, dst_file, 1750 info->dest_offset); 1751 if (deduped == -EBADE) 1752 info->status = FILE_DEDUPE_RANGE_DIFFERS; 1753 else if (deduped < 0) 1754 info->status = deduped; 1755 else 1756 info->bytes_deduped += deduped; 1757 } 1758 1759 next_file: 1760 mnt_drop_write_file(dst_file); 1761 next_loop: 1762 fdput(dst_fd); 1763 1764 if (fatal_signal_pending(current)) 1765 goto out; 1766 } 1767 1768 out: 1769 return ret; 1770 } 1771 EXPORT_SYMBOL(vfs_dedupe_file_range); 1772