fs/read_write.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/read_write.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 #include <linux/slab.h>
   9 #include <linux/stat.h>
  10 #include <linux/sched/xacct.h>
  11 #include <linux/fcntl.h>
  12 #include <linux/file.h>
  13 #include <linux/uio.h>
  14 #include <linux/fsnotify.h>
  15 #include <linux/security.h>
  16 #include <linux/export.h>
  17 #include <linux/syscalls.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/splice.h>
  20 #include <linux/compat.h>
  21 #include <linux/mount.h>
  22 #include <linux/fs.h>
  23 #include "internal.h"
  24
  25 #include <linux/uaccess.h>
  26 #include <asm/unistd.h>
  27
  28 const struct file_operations generic_ro_fops = {
  29         .llseek         = generic_file_llseek,
  30         .read_iter      = generic_file_read_iter,
  31         .mmap           = generic_file_readonly_mmap,
  32         .splice_read    = generic_file_splice_read,
  33 };
  34
  35 EXPORT_SYMBOL(generic_ro_fops);
  36
  37 static inline bool unsigned_offsets(struct file *file)
  38 {
  39         return file->f_mode & FMODE_UNSIGNED_OFFSET;
  40 }
  41
  42 /**
  43  * vfs_setpos - update the file offset for lseek
  44  * @file:       file structure in question
  45  * @offset:     file offset to seek to
  46  * @maxsize:    maximum file size
  47  *
  48  * This is a low-level filesystem helper for updating the file offset to
  49  * the value specified by @offset if the given offset is valid and it is
  50  * not equal to the current file offset.
  51  *
  52  * Return the specified offset on success and -EINVAL on invalid offset.
  53  */
  54 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
  55 {
  56         if (offset < 0 && !unsigned_offsets(file))
  57                 return -EINVAL;
  58         if (offset > maxsize)
  59                 return -EINVAL;
  60
  61         if (offset != file->f_pos) {
  62                 file->f_pos = offset;
  63                 file->f_version = 0;
  64         }
  65         return offset;
  66 }
  67 EXPORT_SYMBOL(vfs_setpos);
  68
  69 /**
  70  * generic_file_llseek_size - generic llseek implementation for regular files
  71  * @file:       file structure to seek on
  72  * @offset:     file offset to seek to
  73  * @whence:     type of seek
  74  * @size:       max size of this file in file system
  75  * @eof:        offset used for SEEK_END position
  76  *
  77  * This is a variant of generic_file_llseek that allows passing in a custom
  78  * maximum file size and a custom EOF position, for e.g. hashed directories
  79  *
  80  * Synchronization:
  81  * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
  82  * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
  83  * read/writes behave like SEEK_SET against seeks.
  84  */
  85 loff_t
  86 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
  87                 loff_t maxsize, loff_t eof)
  88 {
  89         switch (whence) {
  90         case SEEK_END:
  91                 offset += eof;
  92                 break;
  93         case SEEK_CUR:
  94                 /*
  95                  * Here we special-case the lseek(fd, 0, SEEK_CUR)
  96                  * position-querying operation.  Avoid rewriting the "same"
  97                  * f_pos value back to the file because a concurrent read(),
  98                  * write() or lseek() might have altered it
  99                  */
 100                 if (offset == 0)
 101                         return file->f_pos;
 102                 /*
 103                  * f_lock protects against read/modify/write race with other
 104                  * SEEK_CURs. Note that parallel writes and reads behave
 105                  * like SEEK_SET.
 106                  */
 107                 spin_lock(&file->f_lock);
 108                 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
 109                 spin_unlock(&file->f_lock);
 110                 return offset;
 111         case SEEK_DATA:
 112                 /*
 113                  * In the generic case the entire file is data, so as long as
 114                  * offset isn't at the end of the file then the offset is data.
 115                  */
 116                 if ((unsigned long long)offset >= eof)
 117                         return -ENXIO;
 118                 break;
 119         case SEEK_HOLE:
 120                 /*
 121                  * There is a virtual hole at the end of the file, so as long as
 122                  * offset isn't i_size or larger, return i_size.
 123                  */
 124                 if ((unsigned long long)offset >= eof)
 125                         return -ENXIO;
 126                 offset = eof;
 127                 break;
 128         }
 129
 130         return vfs_setpos(file, offset, maxsize);
 131 }
 132 EXPORT_SYMBOL(generic_file_llseek_size);
 133
 134 /**
 135  * generic_file_llseek - generic llseek implementation for regular files
 136  * @file:       file structure to seek on
 137  * @offset:     file offset to seek to
 138  * @whence:     type of seek
 139  *
 140  * This is a generic implemenation of ->llseek useable for all normal local
 141  * filesystems.  It just updates the file offset to the value specified by
 142  * @offset and @whence.
 143  */
 144 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
 145 {
 146         struct inode *inode = file->f_mapping->host;
 147
 148         return generic_file_llseek_size(file, offset, whence,
 149                                         inode->i_sb->s_maxbytes,
 150                                         i_size_read(inode));
 151 }
 152 EXPORT_SYMBOL(generic_file_llseek);
 153
 154 /**
 155  * fixed_size_llseek - llseek implementation for fixed-sized devices
 156  * @file:       file structure to seek on
 157  * @offset:     file offset to seek to
 158  * @whence:     type of seek
 159  * @size:       size of the file
 160  *
 161  */
 162 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
 163 {
 164         switch (whence) {
 165         case SEEK_SET: case SEEK_CUR: case SEEK_END:
 166                 return generic_file_llseek_size(file, offset, whence,
 167                                                 size, size);
 168         default:
 169                 return -EINVAL;
 170         }
 171 }
 172 EXPORT_SYMBOL(fixed_size_llseek);
 173
 174 /**
 175  * no_seek_end_llseek - llseek implementation for fixed-sized devices
 176  * @file:       file structure to seek on
 177  * @offset:     file offset to seek to
 178  * @whence:     type of seek
 179  *
 180  */
 181 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
 182 {
 183         switch (whence) {
 184         case SEEK_SET: case SEEK_CUR:
 185                 return generic_file_llseek_size(file, offset, whence,
 186                                                 OFFSET_MAX, 0);
 187         default:
 188                 return -EINVAL;
 189         }
 190 }
 191 EXPORT_SYMBOL(no_seek_end_llseek);
 192
 193 /**
 194  * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
 195  * @file:       file structure to seek on
 196  * @offset:     file offset to seek to
 197  * @whence:     type of seek
 198  * @size:       maximal offset allowed
 199  *
 200  */
 201 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
 202 {
 203         switch (whence) {
 204         case SEEK_SET: case SEEK_CUR:
 205                 return generic_file_llseek_size(file, offset, whence,
 206                                                 size, 0);
 207         default:
 208                 return -EINVAL;
 209         }
 210 }
 211 EXPORT_SYMBOL(no_seek_end_llseek_size);
 212
 213 /**
 214  * noop_llseek - No Operation Performed llseek implementation
 215  * @file:       file structure to seek on
 216  * @offset:     file offset to seek to
 217  * @whence:     type of seek
 218  *
 219  * This is an implementation of ->llseek useable for the rare special case when
 220  * userspace expects the seek to succeed but the (device) file is actually not
 221  * able to perform the seek. In this case you use noop_llseek() instead of
 222  * falling back to the default implementation of ->llseek.
 223  */
 224 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
 225 {
 226         return file->f_pos;
 227 }
 228 EXPORT_SYMBOL(noop_llseek);
 229
 230 loff_t no_llseek(struct file *file, loff_t offset, int whence)
 231 {
 232         return -ESPIPE;
 233 }
 234 EXPORT_SYMBOL(no_llseek);
 235
 236 loff_t default_llseek(struct file *file, loff_t offset, int whence)
 237 {
 238         struct inode *inode = file_inode(file);
 239         loff_t retval;
 240
 241         inode_lock(inode);
 242         switch (whence) {
 243                 case SEEK_END:
 244                         offset += i_size_read(inode);
 245                         break;
 246                 case SEEK_CUR:
 247                         if (offset == 0) {
 248                                 retval = file->f_pos;
 249                                 goto out;
 250                         }
 251                         offset += file->f_pos;
 252                         break;
 253                 case SEEK_DATA:
 254                         /*
 255                          * In the generic case the entire file is data, so as
 256                          * long as offset isn't at the end of the file then the
 257                          * offset is data.
 258                          */
 259                         if (offset >= inode->i_size) {
 260                                 retval = -ENXIO;
 261                                 goto out;
 262                         }
 263                         break;
 264                 case SEEK_HOLE:
 265                         /*
 266                          * There is a virtual hole at the end of the file, so
 267                          * as long as offset isn't i_size or larger, return
 268                          * i_size.
 269                          */
 270                         if (offset >= inode->i_size) {
 271                                 retval = -ENXIO;
 272                                 goto out;
 273                         }
 274                         offset = inode->i_size;
 275                         break;
 276         }
 277         retval = -EINVAL;
 278         if (offset >= 0 || unsigned_offsets(file)) {
 279                 if (offset != file->f_pos) {
 280                         file->f_pos = offset;
 281                         file->f_version = 0;
 282                 }
 283                 retval = offset;
 284         }
 285 out:
 286         inode_unlock(inode);
 287         return retval;
 288 }
 289 EXPORT_SYMBOL(default_llseek);
 290
 291 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
 292 {
 293         loff_t (*fn)(struct file *, loff_t, int);
 294
 295         fn = no_llseek;
 296         if (file->f_mode & FMODE_LSEEK) {
 297                 if (file->f_op->llseek)
 298                         fn = file->f_op->llseek;
 299         }
 300         return fn(file, offset, whence);
 301 }
 302 EXPORT_SYMBOL(vfs_llseek);
 303
 304 off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
 305 {
 306         off_t retval;
 307         struct fd f = fdget_pos(fd);
 308         if (!f.file)
 309                 return -EBADF;
 310
 311         retval = -EINVAL;
 312         if (whence <= SEEK_MAX) {
 313                 loff_t res = vfs_llseek(f.file, offset, whence);
 314                 retval = res;
 315                 if (res != (loff_t)retval)
 316                         retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
 317         }
 318         fdput_pos(f);
 319         return retval;
 320 }
 321
 322 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
 323 {
 324         return ksys_lseek(fd, offset, whence);
 325 }
 326
 327 #ifdef CONFIG_COMPAT
 328 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
 329 {
 330         return ksys_lseek(fd, offset, whence);
 331 }
 332 #endif
 333
 334 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
 335         defined(__ARCH_WANT_SYS_LLSEEK)
 336 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
 337                 unsigned long, offset_low, loff_t __user *, result,
 338                 unsigned int, whence)
 339 {
 340         int retval;
 341         struct fd f = fdget_pos(fd);
 342         loff_t offset;
 343
 344         if (!f.file)
 345                 return -EBADF;
 346
 347         retval = -EINVAL;
 348         if (whence > SEEK_MAX)
 349                 goto out_putf;
 350
 351         offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
 352                         whence);
 353
 354         retval = (int)offset;
 355         if (offset >= 0) {
 356                 retval = -EFAULT;
 357                 if (!copy_to_user(result, &offset, sizeof(offset)))
 358                         retval = 0;
 359         }
 360 out_putf:
 361         fdput_pos(f);
 362         return retval;
 363 }
 364 #endif
 365
 366 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
 367 {
 368         struct inode *inode;
 369         int retval = -EINVAL;
 370
 371         inode = file_inode(file);
 372         if (unlikely((ssize_t) count < 0))
 373                 return retval;
 374
 375         /*
 376          * ranged mandatory locking does not apply to streams - it makes sense
 377          * only for files where position has a meaning.
 378          */
 379         if (ppos) {
 380                 loff_t pos = *ppos;
 381
 382                 if (unlikely(pos < 0)) {
 383                         if (!unsigned_offsets(file))
 384                                 return retval;
 385                         if (count >= -pos) /* both values are in 0..LLONG_MAX */
 386                                 return -EOVERFLOW;
 387                 } else if (unlikely((loff_t) (pos + count) < 0)) {
 388                         if (!unsigned_offsets(file))
 389                                 return retval;
 390                 }
 391
 392                 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
 393                         retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
 394                                         read_write == READ ? F_RDLCK : F_WRLCK);
 395                         if (retval < 0)
 396                                 return retval;
 397                 }
 398         }
 399
 400         return security_file_permission(file,
 401                                 read_write == READ ? MAY_READ : MAY_WRITE);
 402 }
 403
 404 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 405 {
 406         struct iovec iov = { .iov_base = buf, .iov_len = len };
 407         struct kiocb kiocb;
 408         struct iov_iter iter;
 409         ssize_t ret;
 410
 411         init_sync_kiocb(&kiocb, filp);
 412         kiocb.ki_pos = (ppos ? *ppos : 0);
 413         iov_iter_init(&iter, READ, &iov, 1, len);
 414
 415         ret = call_read_iter(filp, &kiocb, &iter);
 416         BUG_ON(ret == -EIOCBQUEUED);
 417         if (ppos)
 418                 *ppos = kiocb.ki_pos;
 419         return ret;
 420 }
 421
 422 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
 423                    loff_t *pos)
 424 {
 425         if (file->f_op->read)
 426                 return file->f_op->read(file, buf, count, pos);
 427         else if (file->f_op->read_iter)
 428                 return new_sync_read(file, buf, count, pos);
 429         else
 430                 return -EINVAL;
 431 }
 432
 433 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
 434 {
 435         mm_segment_t old_fs;
 436         ssize_t result;
 437
 438         old_fs = get_fs();
 439         set_fs(KERNEL_DS);
 440         /* The cast to a user pointer is valid due to the set_fs() */
 441         result = vfs_read(file, (void __user *)buf, count, pos);
 442         set_fs(old_fs);
 443         return result;
 444 }
 445 EXPORT_SYMBOL(kernel_read);
 446
 447 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 448 {
 449         ssize_t ret;
 450
 451         if (!(file->f_mode & FMODE_READ))
 452                 return -EBADF;
 453         if (!(file->f_mode & FMODE_CAN_READ))
 454                 return -EINVAL;
 455         if (unlikely(!access_ok(buf, count)))
 456                 return -EFAULT;
 457
 458         ret = rw_verify_area(READ, file, pos, count);
 459         if (!ret) {
 460                 if (count > MAX_RW_COUNT)
 461                         count =  MAX_RW_COUNT;
 462                 ret = __vfs_read(file, buf, count, pos);
 463                 if (ret > 0) {
 464                         fsnotify_access(file);
 465                         add_rchar(current, ret);
 466                 }
 467                 inc_syscr(current);
 468         }
 469
 470         return ret;
 471 }
 472
 473 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 474 {
 475         struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
 476         struct kiocb kiocb;
 477         struct iov_iter iter;
 478         ssize_t ret;
 479
 480         init_sync_kiocb(&kiocb, filp);
 481         kiocb.ki_pos = (ppos ? *ppos : 0);
 482         iov_iter_init(&iter, WRITE, &iov, 1, len);
 483
 484         ret = call_write_iter(filp, &kiocb, &iter);
 485         BUG_ON(ret == -EIOCBQUEUED);
 486         if (ret > 0 && ppos)
 487                 *ppos = kiocb.ki_pos;
 488         return ret;
 489 }
 490
 491 static ssize_t __vfs_write(struct file *file, const char __user *p,
 492                            size_t count, loff_t *pos)
 493 {
 494         if (file->f_op->write)
 495                 return file->f_op->write(file, p, count, pos);
 496         else if (file->f_op->write_iter)
 497                 return new_sync_write(file, p, count, pos);
 498         else
 499                 return -EINVAL;
 500 }
 501
 502 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
 503 {
 504         mm_segment_t old_fs;
 505         const char __user *p;
 506         ssize_t ret;
 507
 508         if (!(file->f_mode & FMODE_CAN_WRITE))
 509                 return -EINVAL;
 510
 511         old_fs = get_fs();
 512         set_fs(KERNEL_DS);
 513         p = (__force const char __user *)buf;
 514         if (count > MAX_RW_COUNT)
 515                 count =  MAX_RW_COUNT;
 516         ret = __vfs_write(file, p, count, pos);
 517         set_fs(old_fs);
 518         if (ret > 0) {
 519                 fsnotify_modify(file);
 520                 add_wchar(current, ret);
 521         }
 522         inc_syscw(current);
 523         return ret;
 524 }
 525 EXPORT_SYMBOL(__kernel_write);
 526
 527 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
 528                             loff_t *pos)
 529 {
 530         mm_segment_t old_fs;
 531         ssize_t res;
 532
 533         old_fs = get_fs();
 534         set_fs(KERNEL_DS);
 535         /* The cast to a user pointer is valid due to the set_fs() */
 536         res = vfs_write(file, (__force const char __user *)buf, count, pos);
 537         set_fs(old_fs);
 538
 539         return res;
 540 }
 541 EXPORT_SYMBOL(kernel_write);
 542
 543 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 544 {
 545         ssize_t ret;
 546
 547         if (!(file->f_mode & FMODE_WRITE))
 548                 return -EBADF;
 549         if (!(file->f_mode & FMODE_CAN_WRITE))
 550                 return -EINVAL;
 551         if (unlikely(!access_ok(buf, count)))
 552                 return -EFAULT;
 553
 554         ret = rw_verify_area(WRITE, file, pos, count);
 555         if (!ret) {
 556                 if (count > MAX_RW_COUNT)
 557                         count =  MAX_RW_COUNT;
 558                 file_start_write(file);
 559                 ret = __vfs_write(file, buf, count, pos);
 560                 if (ret > 0) {
 561                         fsnotify_modify(file);
 562                         add_wchar(current, ret);
 563                 }
 564                 inc_syscw(current);
 565                 file_end_write(file);
 566         }
 567
 568         return ret;
 569 }
 570
 571 /* file_ppos returns &file->f_pos or NULL if file is stream */
 572 static inline loff_t *file_ppos(struct file *file)
 573 {
 574         return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
 575 }
 576
 577 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
 578 {
 579         struct fd f = fdget_pos(fd);
 580         ssize_t ret = -EBADF;
 581
 582         if (f.file) {
 583                 loff_t pos, *ppos = file_ppos(f.file);
 584                 if (ppos) {
 585                         pos = *ppos;
 586                         ppos = &pos;
 587                 }
 588                 ret = vfs_read(f.file, buf, count, ppos);
 589                 if (ret >= 0 && ppos)
 590                         f.file->f_pos = pos;
 591                 fdput_pos(f);
 592         }
 593         return ret;
 594 }
 595
 596 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
 597 {
 598         return ksys_read(fd, buf, count);
 599 }
 600
 601 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
 602 {
 603         struct fd f = fdget_pos(fd);
 604         ssize_t ret = -EBADF;
 605
 606         if (f.file) {
 607                 loff_t pos, *ppos = file_ppos(f.file);
 608                 if (ppos) {
 609                         pos = *ppos;
 610                         ppos = &pos;
 611                 }
 612                 ret = vfs_write(f.file, buf, count, ppos);
 613                 if (ret >= 0 && ppos)
 614                         f.file->f_pos = pos;
 615                 fdput_pos(f);
 616         }
 617
 618         return ret;
 619 }
 620
 621 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
 622                 size_t, count)
 623 {
 624         return ksys_write(fd, buf, count);
 625 }
 626
 627 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
 628                      loff_t pos)
 629 {
 630         struct fd f;
 631         ssize_t ret = -EBADF;
 632
 633         if (pos < 0)
 634                 return -EINVAL;
 635
 636         f = fdget(fd);
 637         if (f.file) {
 638                 ret = -ESPIPE;
 639                 if (f.file->f_mode & FMODE_PREAD)
 640                         ret = vfs_read(f.file, buf, count, &pos);
 641                 fdput(f);
 642         }
 643
 644         return ret;
 645 }
 646
 647 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
 648                         size_t, count, loff_t, pos)
 649 {
 650         return ksys_pread64(fd, buf, count, pos);
 651 }
 652
 653 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
 654                       size_t count, loff_t pos)
 655 {
 656         struct fd f;
 657         ssize_t ret = -EBADF;
 658
 659         if (pos < 0)
 660                 return -EINVAL;
 661
 662         f = fdget(fd);
 663         if (f.file) {
 664                 ret = -ESPIPE;
 665                 if (f.file->f_mode & FMODE_PWRITE)
 666                         ret = vfs_write(f.file, buf, count, &pos);
 667                 fdput(f);
 668         }
 669
 670         return ret;
 671 }
 672
 673 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
 674                          size_t, count, loff_t, pos)
 675 {
 676         return ksys_pwrite64(fd, buf, count, pos);
 677 }
 678
 679 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
 680                 loff_t *ppos, int type, rwf_t flags)
 681 {
 682         struct kiocb kiocb;
 683         ssize_t ret;
 684
 685         init_sync_kiocb(&kiocb, filp);
 686         ret = kiocb_set_rw_flags(&kiocb, flags);
 687         if (ret)
 688                 return ret;
 689         kiocb.ki_pos = (ppos ? *ppos : 0);
 690
 691         if (type == READ)
 692                 ret = call_read_iter(filp, &kiocb, iter);
 693         else
 694                 ret = call_write_iter(filp, &kiocb, iter);
 695         BUG_ON(ret == -EIOCBQUEUED);
 696         if (ppos)
 697                 *ppos = kiocb.ki_pos;
 698         return ret;
 699 }
 700
 701 /* Do it by hand, with file-ops */
 702 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
 703                 loff_t *ppos, int type, rwf_t flags)
 704 {
 705         ssize_t ret = 0;
 706
 707         if (flags & ~RWF_HIPRI)
 708                 return -EOPNOTSUPP;
 709
 710         while (iov_iter_count(iter)) {
 711                 struct iovec iovec = iov_iter_iovec(iter);
 712                 ssize_t nr;
 713
 714                 if (type == READ) {
 715                         nr = filp->f_op->read(filp, iovec.iov_base,
 716                                               iovec.iov_len, ppos);
 717                 } else {
 718                         nr = filp->f_op->write(filp, iovec.iov_base,
 719                                                iovec.iov_len, ppos);
 720                 }
 721
 722                 if (nr < 0) {
 723                         if (!ret)
 724                                 ret = nr;
 725                         break;
 726                 }
 727                 ret += nr;
 728                 if (nr != iovec.iov_len)
 729                         break;
 730                 iov_iter_advance(iter, nr);
 731         }
 732
 733         return ret;
 734 }
 735
 736 /**
 737  * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
 738  *     into the kernel and check that it is valid.
 739  *
 740  * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
 741  * @uvector: Pointer to the userspace array.
 742  * @nr_segs: Number of elements in userspace array.
 743  * @fast_segs: Number of elements in @fast_pointer.
 744  * @fast_pointer: Pointer to (usually small on-stack) kernel array.
 745  * @ret_pointer: (output parameter) Pointer to a variable that will point to
 746  *     either @fast_pointer, a newly allocated kernel array, or NULL,
 747  *     depending on which array was used.
 748  *
 749  * This function copies an array of &struct iovec of @nr_segs from
 750  * userspace into the kernel and checks that each element is valid (e.g.
 751  * it does not point to a kernel address or cause overflow by being too
 752  * large, etc.).
 753  *
 754  * As an optimization, the caller may provide a pointer to a small
 755  * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
 756  * (the size of this array, or 0 if unused, should be given in @fast_segs).
 757  *
 758  * @ret_pointer will always point to the array that was used, so the
 759  * caller must take care not to call kfree() on it e.g. in case the
 760  * @fast_pointer array was used and it was allocated on the stack.
 761  *
 762  * Return: The total number of bytes covered by the iovec array on success
 763  *   or a negative error code on error.
 764  */
 765 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
 766                               unsigned long nr_segs, unsigned long fast_segs,
 767                               struct iovec *fast_pointer,
 768                               struct iovec **ret_pointer)
 769 {
 770         unsigned long seg;
 771         ssize_t ret;
 772         struct iovec *iov = fast_pointer;
 773
 774         /*
 775          * SuS says "The readv() function *may* fail if the iovcnt argument
 776          * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 777          * traditionally returned zero for zero segments, so...
 778          */
 779         if (nr_segs == 0) {
 780                 ret = 0;
 781                 goto out;
 782         }
 783
 784         /*
 785          * First get the "struct iovec" from user memory and
 786          * verify all the pointers
 787          */
 788         if (nr_segs > UIO_MAXIOV) {
 789                 ret = -EINVAL;
 790                 goto out;
 791         }
 792         if (nr_segs > fast_segs) {
 793                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
 794                 if (iov == NULL) {
 795                         ret = -ENOMEM;
 796                         goto out;
 797                 }
 798         }
 799         if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
 800                 ret = -EFAULT;
 801                 goto out;
 802         }
 803
 804         /*
 805          * According to the Single Unix Specification we should return EINVAL
 806          * if an element length is < 0 when cast to ssize_t or if the
 807          * total length would overflow the ssize_t return value of the
 808          * system call.
 809          *
 810          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
 811          * overflow case.
 812          */
 813         ret = 0;
 814         for (seg = 0; seg < nr_segs; seg++) {
 815                 void __user *buf = iov[seg].iov_base;
 816                 ssize_t len = (ssize_t)iov[seg].iov_len;
 817
 818                 /* see if we we're about to use an invalid len or if
 819                  * it's about to overflow ssize_t */
 820                 if (len < 0) {
 821                         ret = -EINVAL;
 822                         goto out;
 823                 }
 824                 if (type >= 0
 825                     && unlikely(!access_ok(buf, len))) {
 826                         ret = -EFAULT;
 827                         goto out;
 828                 }
 829                 if (len > MAX_RW_COUNT - ret) {
 830                         len = MAX_RW_COUNT - ret;
 831                         iov[seg].iov_len = len;
 832                 }
 833                 ret += len;
 834         }
 835 out:
 836         *ret_pointer = iov;
 837         return ret;
 838 }
 839
 840 #ifdef CONFIG_COMPAT
 841 ssize_t compat_rw_copy_check_uvector(int type,
 842                 const struct compat_iovec __user *uvector, unsigned long nr_segs,
 843                 unsigned long fast_segs, struct iovec *fast_pointer,
 844                 struct iovec **ret_pointer)
 845 {
 846         compat_ssize_t tot_len;
 847         struct iovec *iov = *ret_pointer = fast_pointer;
 848         ssize_t ret = 0;
 849         int seg;
 850
 851         /*
 852          * SuS says "The readv() function *may* fail if the iovcnt argument
 853          * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 854          * traditionally returned zero for zero segments, so...
 855          */
 856         if (nr_segs == 0)
 857                 goto out;
 858
 859         ret = -EINVAL;
 860         if (nr_segs > UIO_MAXIOV)
 861                 goto out;
 862         if (nr_segs > fast_segs) {
 863                 ret = -ENOMEM;
 864                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
 865                 if (iov == NULL)
 866                         goto out;
 867         }
 868         *ret_pointer = iov;
 869
 870         ret = -EFAULT;
 871         if (!access_ok(uvector, nr_segs*sizeof(*uvector)))
 872                 goto out;
 873
 874         /*
 875          * Single unix specification:
 876          * We should -EINVAL if an element length is not >= 0 and fitting an
 877          * ssize_t.
 878          *
 879          * In Linux, the total length is limited to MAX_RW_COUNT, there is
 880          * no overflow possibility.
 881          */
 882         tot_len = 0;
 883         ret = -EINVAL;
 884         for (seg = 0; seg < nr_segs; seg++) {
 885                 compat_uptr_t buf;
 886                 compat_ssize_t len;
 887
 888                 if (__get_user(len, &uvector->iov_len) ||
 889                    __get_user(buf, &uvector->iov_base)) {
 890                         ret = -EFAULT;
 891                         goto out;
 892                 }
 893                 if (len < 0)    /* size_t not fitting in compat_ssize_t .. */
 894                         goto out;
 895                 if (type >= 0 &&
 896                     !access_ok(compat_ptr(buf), len)) {
 897                         ret = -EFAULT;
 898                         goto out;
 899                 }
 900                 if (len > MAX_RW_COUNT - tot_len)
 901                         len = MAX_RW_COUNT - tot_len;
 902                 tot_len += len;
 903                 iov->iov_base = compat_ptr(buf);
 904                 iov->iov_len = (compat_size_t) len;
 905                 uvector++;
 906                 iov++;
 907         }
 908         ret = tot_len;
 909
 910 out:
 911         return ret;
 912 }
 913 #endif
 914
 915 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
 916                 loff_t *pos, rwf_t flags)
 917 {
 918         size_t tot_len;
 919         ssize_t ret = 0;
 920
 921         if (!(file->f_mode & FMODE_READ))
 922                 return -EBADF;
 923         if (!(file->f_mode & FMODE_CAN_READ))
 924                 return -EINVAL;
 925
 926         tot_len = iov_iter_count(iter);
 927         if (!tot_len)
 928                 goto out;
 929         ret = rw_verify_area(READ, file, pos, tot_len);
 930         if (ret < 0)
 931                 return ret;
 932
 933         if (file->f_op->read_iter)
 934                 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
 935         else
 936                 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
 937 out:
 938         if (ret >= 0)
 939                 fsnotify_access(file);
 940         return ret;
 941 }
 942
 943 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
 944                            struct iov_iter *iter)
 945 {
 946         size_t tot_len;
 947         ssize_t ret = 0;
 948
 949         if (!file->f_op->read_iter)
 950                 return -EINVAL;
 951         if (!(file->f_mode & FMODE_READ))
 952                 return -EBADF;
 953         if (!(file->f_mode & FMODE_CAN_READ))
 954                 return -EINVAL;
 955
 956         tot_len = iov_iter_count(iter);
 957         if (!tot_len)
 958                 goto out;
 959         ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
 960         if (ret < 0)
 961                 return ret;
 962
 963         ret = call_read_iter(file, iocb, iter);
 964 out:
 965         if (ret >= 0)
 966                 fsnotify_access(file);
 967         return ret;
 968 }
 969 EXPORT_SYMBOL(vfs_iocb_iter_read);
 970
 971 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
 972                 rwf_t flags)
 973 {
 974         if (!file->f_op->read_iter)
 975                 return -EINVAL;
 976         return do_iter_read(file, iter, ppos, flags);
 977 }
 978 EXPORT_SYMBOL(vfs_iter_read);
 979
 980 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
 981                 loff_t *pos, rwf_t flags)
 982 {
 983         size_t tot_len;
 984         ssize_t ret = 0;
 985
 986         if (!(file->f_mode & FMODE_WRITE))
 987                 return -EBADF;
 988         if (!(file->f_mode & FMODE_CAN_WRITE))
 989                 return -EINVAL;
 990
 991         tot_len = iov_iter_count(iter);
 992         if (!tot_len)
 993                 return 0;
 994         ret = rw_verify_area(WRITE, file, pos, tot_len);
 995         if (ret < 0)
 996                 return ret;
 997
 998         if (file->f_op->write_iter)
 999                 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
1000         else
1001                 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
1002         if (ret > 0)
1003                 fsnotify_modify(file);
1004         return ret;
1005 }
1006
1007 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
1008                             struct iov_iter *iter)
1009 {
1010         size_t tot_len;
1011         ssize_t ret = 0;
1012
1013         if (!file->f_op->write_iter)
1014                 return -EINVAL;
1015         if (!(file->f_mode & FMODE_WRITE))
1016                 return -EBADF;
1017         if (!(file->f_mode & FMODE_CAN_WRITE))
1018                 return -EINVAL;
1019
1020         tot_len = iov_iter_count(iter);
1021         if (!tot_len)
1022                 return 0;
1023         ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
1024         if (ret < 0)
1025                 return ret;
1026
1027         ret = call_write_iter(file, iocb, iter);
1028         if (ret > 0)
1029                 fsnotify_modify(file);
1030
1031         return ret;
1032 }
1033 EXPORT_SYMBOL(vfs_iocb_iter_write);
1034
1035 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
1036                 rwf_t flags)
1037 {
1038         if (!file->f_op->write_iter)
1039                 return -EINVAL;
1040         return do_iter_write(file, iter, ppos, flags);
1041 }
1042 EXPORT_SYMBOL(vfs_iter_write);
1043
1044 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
1045                   unsigned long vlen, loff_t *pos, rwf_t flags)
1046 {
1047         struct iovec iovstack[UIO_FASTIOV];
1048         struct iovec *iov = iovstack;
1049         struct iov_iter iter;
1050         ssize_t ret;
1051
1052         ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
1053         if (ret >= 0) {
1054                 ret = do_iter_read(file, &iter, pos, flags);
1055                 kfree(iov);
1056         }
1057
1058         return ret;
1059 }
1060
1061 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
1062                    unsigned long vlen, loff_t *pos, rwf_t flags)
1063 {
1064         struct iovec iovstack[UIO_FASTIOV];
1065         struct iovec *iov = iovstack;
1066         struct iov_iter iter;
1067         ssize_t ret;
1068
1069         ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
1070         if (ret >= 0) {
1071                 file_start_write(file);
1072                 ret = do_iter_write(file, &iter, pos, flags);
1073                 file_end_write(file);
1074                 kfree(iov);
1075         }
1076         return ret;
1077 }
1078
1079 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1080                         unsigned long vlen, rwf_t flags)
1081 {
1082         struct fd f = fdget_pos(fd);
1083         ssize_t ret = -EBADF;
1084
1085         if (f.file) {
1086                 loff_t pos, *ppos = file_ppos(f.file);
1087                 if (ppos) {
1088                         pos = *ppos;
1089                         ppos = &pos;
1090                 }
1091                 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
1092                 if (ret >= 0 && ppos)
1093                         f.file->f_pos = pos;
1094                 fdput_pos(f);
1095         }
1096
1097         if (ret > 0)
1098                 add_rchar(current, ret);
1099         inc_syscr(current);
1100         return ret;
1101 }
1102
1103 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1104                          unsigned long vlen, rwf_t flags)
1105 {
1106         struct fd f = fdget_pos(fd);
1107         ssize_t ret = -EBADF;
1108
1109         if (f.file) {
1110                 loff_t pos, *ppos = file_ppos(f.file);
1111                 if (ppos) {
1112                         pos = *ppos;
1113                         ppos = &pos;
1114                 }
1115                 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
1116                 if (ret >= 0 && ppos)
1117                         f.file->f_pos = pos;
1118                 fdput_pos(f);
1119         }
1120
1121         if (ret > 0)
1122                 add_wchar(current, ret);
1123         inc_syscw(current);
1124         return ret;
1125 }
1126
1127 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1128 {
1129 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1130         return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1131 }
1132
1133 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1134                          unsigned long vlen, loff_t pos, rwf_t flags)
1135 {
1136         struct fd f;
1137         ssize_t ret = -EBADF;
1138
1139         if (pos < 0)
1140                 return -EINVAL;
1141
1142         f = fdget(fd);
1143         if (f.file) {
1144                 ret = -ESPIPE;
1145                 if (f.file->f_mode & FMODE_PREAD)
1146                         ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1147                 fdput(f);
1148         }
1149
1150         if (ret > 0)
1151                 add_rchar(current, ret);
1152         inc_syscr(current);
1153         return ret;
1154 }
1155
1156 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1157                           unsigned long vlen, loff_t pos, rwf_t flags)
1158 {
1159         struct fd f;
1160         ssize_t ret = -EBADF;
1161
1162         if (pos < 0)
1163                 return -EINVAL;
1164
1165         f = fdget(fd);
1166         if (f.file) {
1167                 ret = -ESPIPE;
1168                 if (f.file->f_mode & FMODE_PWRITE)
1169                         ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1170                 fdput(f);
1171         }
1172
1173         if (ret > 0)
1174                 add_wchar(current, ret);
1175         inc_syscw(current);
1176         return ret;
1177 }
1178
1179 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1180                 unsigned long, vlen)
1181 {
1182         return do_readv(fd, vec, vlen, 0);
1183 }
1184
1185 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1186                 unsigned long, vlen)
1187 {
1188         return do_writev(fd, vec, vlen, 0);
1189 }
1190
1191 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1192                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1193 {
1194         loff_t pos = pos_from_hilo(pos_h, pos_l);
1195
1196         return do_preadv(fd, vec, vlen, pos, 0);
1197 }
1198
1199 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1200                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1201                 rwf_t, flags)
1202 {
1203         loff_t pos = pos_from_hilo(pos_h, pos_l);
1204
1205         if (pos == -1)
1206                 return do_readv(fd, vec, vlen, flags);
1207
1208         return do_preadv(fd, vec, vlen, pos, flags);
1209 }
1210
1211 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1212                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1213 {
1214         loff_t pos = pos_from_hilo(pos_h, pos_l);
1215
1216         return do_pwritev(fd, vec, vlen, pos, 0);
1217 }
1218
1219 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1220                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1221                 rwf_t, flags)
1222 {
1223         loff_t pos = pos_from_hilo(pos_h, pos_l);
1224
1225         if (pos == -1)
1226                 return do_writev(fd, vec, vlen, flags);
1227
1228         return do_pwritev(fd, vec, vlen, pos, flags);
1229 }
1230
1231 #ifdef CONFIG_COMPAT
1232 static size_t compat_readv(struct file *file,
1233                            const struct compat_iovec __user *vec,
1234                            unsigned long vlen, loff_t *pos, rwf_t flags)
1235 {
1236         struct iovec iovstack[UIO_FASTIOV];
1237         struct iovec *iov = iovstack;
1238         struct iov_iter iter;
1239         ssize_t ret;
1240
1241         ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1242         if (ret >= 0) {
1243                 ret = do_iter_read(file, &iter, pos, flags);
1244                 kfree(iov);
1245         }
1246         if (ret > 0)
1247                 add_rchar(current, ret);
1248         inc_syscr(current);
1249         return ret;
1250 }
1251
1252 static size_t do_compat_readv(compat_ulong_t fd,
1253                                  const struct compat_iovec __user *vec,
1254                                  compat_ulong_t vlen, rwf_t flags)
1255 {
1256         struct fd f = fdget_pos(fd);
1257         ssize_t ret;
1258         loff_t pos;
1259
1260         if (!f.file)
1261                 return -EBADF;
1262         pos = f.file->f_pos;
1263         ret = compat_readv(f.file, vec, vlen, &pos, flags);
1264         if (ret >= 0)
1265                 f.file->f_pos = pos;
1266         fdput_pos(f);
1267         return ret;
1268
1269 }
1270
1271 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1272                 const struct compat_iovec __user *,vec,
1273                 compat_ulong_t, vlen)
1274 {
1275         return do_compat_readv(fd, vec, vlen, 0);
1276 }
1277
1278 static long do_compat_preadv64(unsigned long fd,
1279                                   const struct compat_iovec __user *vec,
1280                                   unsigned long vlen, loff_t pos, rwf_t flags)
1281 {
1282         struct fd f;
1283         ssize_t ret;
1284
1285         if (pos < 0)
1286                 return -EINVAL;
1287         f = fdget(fd);
1288         if (!f.file)
1289                 return -EBADF;
1290         ret = -ESPIPE;
1291         if (f.file->f_mode & FMODE_PREAD)
1292                 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1293         fdput(f);
1294         return ret;
1295 }
1296
1297 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1298 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1299                 const struct compat_iovec __user *,vec,
1300                 unsigned long, vlen, loff_t, pos)
1301 {
1302         return do_compat_preadv64(fd, vec, vlen, pos, 0);
1303 }
1304 #endif
1305
1306 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1307                 const struct compat_iovec __user *,vec,
1308                 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1309 {
1310         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1311
1312         return do_compat_preadv64(fd, vec, vlen, pos, 0);
1313 }
1314
1315 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1316 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1317                 const struct compat_iovec __user *,vec,
1318                 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1319 {
1320         if (pos == -1)
1321                 return do_compat_readv(fd, vec, vlen, flags);
1322
1323         return do_compat_preadv64(fd, vec, vlen, pos, flags);
1324 }
1325 #endif
1326
1327 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1328                 const struct compat_iovec __user *,vec,
1329                 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1330                 rwf_t, flags)
1331 {
1332         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1333
1334         if (pos == -1)
1335                 return do_compat_readv(fd, vec, vlen, flags);
1336
1337         return do_compat_preadv64(fd, vec, vlen, pos, flags);
1338 }
1339
1340 static size_t compat_writev(struct file *file,
1341                             const struct compat_iovec __user *vec,
1342                             unsigned long vlen, loff_t *pos, rwf_t flags)
1343 {
1344         struct iovec iovstack[UIO_FASTIOV];
1345         struct iovec *iov = iovstack;
1346         struct iov_iter iter;
1347         ssize_t ret;
1348
1349         ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1350         if (ret >= 0) {
1351                 file_start_write(file);
1352                 ret = do_iter_write(file, &iter, pos, flags);
1353                 file_end_write(file);
1354                 kfree(iov);
1355         }
1356         if (ret > 0)
1357                 add_wchar(current, ret);
1358         inc_syscw(current);
1359         return ret;
1360 }
1361
1362 static size_t do_compat_writev(compat_ulong_t fd,
1363                                   const struct compat_iovec __user* vec,
1364                                   compat_ulong_t vlen, rwf_t flags)
1365 {
1366         struct fd f = fdget_pos(fd);
1367         ssize_t ret;
1368         loff_t pos;
1369
1370         if (!f.file)
1371                 return -EBADF;
1372         pos = f.file->f_pos;
1373         ret = compat_writev(f.file, vec, vlen, &pos, flags);
1374         if (ret >= 0)
1375                 f.file->f_pos = pos;
1376         fdput_pos(f);
1377         return ret;
1378 }
1379
1380 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1381                 const struct compat_iovec __user *, vec,
1382                 compat_ulong_t, vlen)
1383 {
1384         return do_compat_writev(fd, vec, vlen, 0);
1385 }
1386
1387 static long do_compat_pwritev64(unsigned long fd,
1388                                    const struct compat_iovec __user *vec,
1389                                    unsigned long vlen, loff_t pos, rwf_t flags)
1390 {
1391         struct fd f;
1392         ssize_t ret;
1393
1394         if (pos < 0)
1395                 return -EINVAL;
1396         f = fdget(fd);
1397         if (!f.file)
1398                 return -EBADF;
1399         ret = -ESPIPE;
1400         if (f.file->f_mode & FMODE_PWRITE)
1401                 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1402         fdput(f);
1403         return ret;
1404 }
1405
1406 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1407 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1408                 const struct compat_iovec __user *,vec,
1409                 unsigned long, vlen, loff_t, pos)
1410 {
1411         return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1412 }
1413 #endif
1414
1415 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1416                 const struct compat_iovec __user *,vec,
1417                 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1418 {
1419         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1420
1421         return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1422 }
1423
1424 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1425 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1426                 const struct compat_iovec __user *,vec,
1427                 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1428 {
1429         if (pos == -1)
1430                 return do_compat_writev(fd, vec, vlen, flags);
1431
1432         return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1433 }
1434 #endif
1435
1436 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1437                 const struct compat_iovec __user *,vec,
1438                 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1439 {
1440         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1441
1442         if (pos == -1)
1443                 return do_compat_writev(fd, vec, vlen, flags);
1444
1445         return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1446 }
1447
1448 #endif
1449
1450 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1451                            size_t count, loff_t max)
1452 {
1453         struct fd in, out;
1454         struct inode *in_inode, *out_inode;
1455         loff_t pos;
1456         loff_t out_pos;
1457         ssize_t retval;
1458         int fl;
1459
1460         /*
1461          * Get input file, and verify that it is ok..
1462          */
1463         retval = -EBADF;
1464         in = fdget(in_fd);
1465         if (!in.file)
1466                 goto out;
1467         if (!(in.file->f_mode & FMODE_READ))
1468                 goto fput_in;
1469         retval = -ESPIPE;
1470         if (!ppos) {
1471                 pos = in.file->f_pos;
1472         } else {
1473                 pos = *ppos;
1474                 if (!(in.file->f_mode & FMODE_PREAD))
1475                         goto fput_in;
1476         }
1477         retval = rw_verify_area(READ, in.file, &pos, count);
1478         if (retval < 0)
1479                 goto fput_in;
1480         if (count > MAX_RW_COUNT)
1481                 count =  MAX_RW_COUNT;
1482
1483         /*
1484          * Get output file, and verify that it is ok..
1485          */
1486         retval = -EBADF;
1487         out = fdget(out_fd);
1488         if (!out.file)
1489                 goto fput_in;
1490         if (!(out.file->f_mode & FMODE_WRITE))
1491                 goto fput_out;
1492         in_inode = file_inode(in.file);
1493         out_inode = file_inode(out.file);
1494         out_pos = out.file->f_pos;
1495         retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1496         if (retval < 0)
1497                 goto fput_out;
1498
1499         if (!max)
1500                 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1501
1502         if (unlikely(pos + count > max)) {
1503                 retval = -EOVERFLOW;
1504                 if (pos >= max)
1505                         goto fput_out;
1506                 count = max - pos;
1507         }
1508
1509         fl = 0;
1510 #if 0
1511         /*
1512          * We need to debate whether we can enable this or not. The
1513          * man page documents EAGAIN return for the output at least,
1514          * and the application is arguably buggy if it doesn't expect
1515          * EAGAIN on a non-blocking file descriptor.
1516          */
1517         if (in.file->f_flags & O_NONBLOCK)
1518                 fl = SPLICE_F_NONBLOCK;
1519 #endif
1520         file_start_write(out.file);
1521         retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1522         file_end_write(out.file);
1523
1524         if (retval > 0) {
1525                 add_rchar(current, retval);
1526                 add_wchar(current, retval);
1527                 fsnotify_access(in.file);
1528                 fsnotify_modify(out.file);
1529                 out.file->f_pos = out_pos;
1530                 if (ppos)
1531                         *ppos = pos;
1532                 else
1533                         in.file->f_pos = pos;
1534         }
1535
1536         inc_syscr(current);
1537         inc_syscw(current);
1538         if (pos > max)
1539                 retval = -EOVERFLOW;
1540
1541 fput_out:
1542         fdput(out);
1543 fput_in:
1544         fdput(in);
1545 out:
1546         return retval;
1547 }
1548
1549 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1550 {
1551         loff_t pos;
1552         off_t off;
1553         ssize_t ret;
1554
1555         if (offset) {
1556                 if (unlikely(get_user(off, offset)))
1557                         return -EFAULT;
1558                 pos = off;
1559                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1560                 if (unlikely(put_user(pos, offset)))
1561                         return -EFAULT;
1562                 return ret;
1563         }
1564
1565         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1566 }
1567
1568 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1569 {
1570         loff_t pos;
1571         ssize_t ret;
1572
1573         if (offset) {
1574                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1575                         return -EFAULT;
1576                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1577                 if (unlikely(put_user(pos, offset)))
1578                         return -EFAULT;
1579                 return ret;
1580         }
1581
1582         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1583 }
1584
1585 #ifdef CONFIG_COMPAT
1586 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1587                 compat_off_t __user *, offset, compat_size_t, count)
1588 {
1589         loff_t pos;
1590         off_t off;
1591         ssize_t ret;
1592
1593         if (offset) {
1594                 if (unlikely(get_user(off, offset)))
1595                         return -EFAULT;
1596                 pos = off;
1597                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1598                 if (unlikely(put_user(pos, offset)))
1599                         return -EFAULT;
1600                 return ret;
1601         }
1602
1603         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1604 }
1605
1606 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1607                 compat_loff_t __user *, offset, compat_size_t, count)
1608 {
1609         loff_t pos;
1610         ssize_t ret;
1611
1612         if (offset) {
1613                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1614                         return -EFAULT;
1615                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1616                 if (unlikely(put_user(pos, offset)))
1617                         return -EFAULT;
1618                 return ret;
1619         }
1620
1621         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1622 }
1623 #endif
1624
1625 /**
1626  * generic_copy_file_range - copy data between two files
1627  * @file_in:    file structure to read from
1628  * @pos_in:     file offset to read from
1629  * @file_out:   file structure to write data to
1630  * @pos_out:    file offset to write data to
1631  * @len:        amount of data to copy
1632  * @flags:      copy flags
1633  *
1634  * This is a generic filesystem helper to copy data from one file to another.
1635  * It has no constraints on the source or destination file owners - the files
1636  * can belong to different superblocks and different filesystem types. Short
1637  * copies are allowed.
1638  *
1639  * This should be called from the @file_out filesystem, as per the
1640  * ->copy_file_range() method.
1641  *
1642  * Returns the number of bytes copied or a negative error indicating the
1643  * failure.
1644  */
1645
1646 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1647                                 struct file *file_out, loff_t pos_out,
1648                                 size_t len, unsigned int flags)
1649 {
1650         return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1651                                 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1652 }
1653 EXPORT_SYMBOL(generic_copy_file_range);
1654
1655 static ssize_t do_copy_file_range(struct file *file_in, loff_t pos_in,
1656                                   struct file *file_out, loff_t pos_out,
1657                                   size_t len, unsigned int flags)
1658 {
1659         /*
1660          * Although we now allow filesystems to handle cross sb copy, passing
1661          * a file of the wrong filesystem type to filesystem driver can result
1662          * in an attempt to dereference the wrong type of ->private_data, so
1663          * avoid doing that until we really have a good reason.  NFS defines
1664          * several different file_system_type structures, but they all end up
1665          * using the same ->copy_file_range() function pointer.
1666          */
1667         if (file_out->f_op->copy_file_range &&
1668             file_out->f_op->copy_file_range == file_in->f_op->copy_file_range)
1669                 return file_out->f_op->copy_file_range(file_in, pos_in,
1670                                                        file_out, pos_out,
1671                                                        len, flags);
1672
1673         return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1674                                        flags);
1675 }
1676
1677 /*
1678  * copy_file_range() differs from regular file read and write in that it
1679  * specifically allows return partial success.  When it does so is up to
1680  * the copy_file_range method.
1681  */
1682 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1683                             struct file *file_out, loff_t pos_out,
1684                             size_t len, unsigned int flags)
1685 {
1686         ssize_t ret;
1687
1688         if (flags != 0)
1689                 return -EINVAL;
1690
1691         ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1692                                        flags);
1693         if (unlikely(ret))
1694                 return ret;
1695
1696         ret = rw_verify_area(READ, file_in, &pos_in, len);
1697         if (unlikely(ret))
1698                 return ret;
1699
1700         ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1701         if (unlikely(ret))
1702                 return ret;
1703
1704         if (len == 0)
1705                 return 0;
1706
1707         file_start_write(file_out);
1708
1709         /*
1710          * Try cloning first, this is supported by more file systems, and
1711          * more efficient if both clone and copy are supported (e.g. NFS).
1712          */
1713         if (file_in->f_op->remap_file_range &&
1714             file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1715                 loff_t cloned;
1716
1717                 cloned = file_in->f_op->remap_file_range(file_in, pos_in,
1718                                 file_out, pos_out,
1719                                 min_t(loff_t, MAX_RW_COUNT, len),
1720                                 REMAP_FILE_CAN_SHORTEN);
1721                 if (cloned > 0) {
1722                         ret = cloned;
1723                         goto done;
1724                 }
1725         }
1726
1727         ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1728                                 flags);
1729         WARN_ON_ONCE(ret == -EOPNOTSUPP);
1730 done:
1731         if (ret > 0) {
1732                 fsnotify_access(file_in);
1733                 add_rchar(current, ret);
1734                 fsnotify_modify(file_out);
1735                 add_wchar(current, ret);
1736         }
1737
1738         inc_syscr(current);
1739         inc_syscw(current);
1740
1741         file_end_write(file_out);
1742
1743         return ret;
1744 }
1745 EXPORT_SYMBOL(vfs_copy_file_range);
1746
1747 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1748                 int, fd_out, loff_t __user *, off_out,
1749                 size_t, len, unsigned int, flags)
1750 {
1751         loff_t pos_in;
1752         loff_t pos_out;
1753         struct fd f_in;
1754         struct fd f_out;
1755         ssize_t ret = -EBADF;
1756
1757         f_in = fdget(fd_in);
1758         if (!f_in.file)
1759                 goto out2;
1760
1761         f_out = fdget(fd_out);
1762         if (!f_out.file)
1763                 goto out1;
1764
1765         ret = -EFAULT;
1766         if (off_in) {
1767                 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1768                         goto out;
1769         } else {
1770                 pos_in = f_in.file->f_pos;
1771         }
1772
1773         if (off_out) {
1774                 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1775                         goto out;
1776         } else {
1777                 pos_out = f_out.file->f_pos;
1778         }
1779
1780         ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1781                                   flags);
1782         if (ret > 0) {
1783                 pos_in += ret;
1784                 pos_out += ret;
1785
1786                 if (off_in) {
1787                         if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1788                                 ret = -EFAULT;
1789                 } else {
1790                         f_in.file->f_pos = pos_in;
1791                 }
1792
1793                 if (off_out) {
1794                         if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1795                                 ret = -EFAULT;
1796                 } else {
1797                         f_out.file->f_pos = pos_out;
1798                 }
1799         }
1800
1801 out:
1802         fdput(f_out);
1803 out1:
1804         fdput(f_in);
1805 out2:
1806         return ret;
1807 }
1808
1809 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
1810                              bool write)
1811 {
1812         struct inode *inode = file_inode(file);
1813
1814         if (unlikely(pos < 0 || len < 0))
1815                 return -EINVAL;
1816
1817          if (unlikely((loff_t) (pos + len) < 0))
1818                 return -EINVAL;
1819
1820         if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1821                 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1822                 int retval;
1823
1824                 retval = locks_mandatory_area(inode, file, pos, end,
1825                                 write ? F_WRLCK : F_RDLCK);
1826                 if (retval < 0)
1827                         return retval;
1828         }
1829
1830         return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1831 }
1832 /*
1833  * Ensure that we don't remap a partial EOF block in the middle of something
1834  * else.  Assume that the offsets have already been checked for block
1835  * alignment.
1836  *
1837  * For clone we only link a partial EOF block above or at the destination file's
1838  * EOF.  For deduplication we accept a partial EOF block only if it ends at the
1839  * destination file's EOF (can not link it into the middle of a file).
1840  *
1841  * Shorten the request if possible.
1842  */
1843 static int generic_remap_check_len(struct inode *inode_in,
1844                                    struct inode *inode_out,
1845                                    loff_t pos_out,
1846                                    loff_t *len,
1847                                    unsigned int remap_flags)
1848 {
1849         u64 blkmask = i_blocksize(inode_in) - 1;
1850         loff_t new_len = *len;
1851
1852         if ((*len & blkmask) == 0)
1853                 return 0;
1854
1855         if (pos_out + *len < i_size_read(inode_out))
1856                 new_len &= ~blkmask;
1857
1858         if (new_len == *len)
1859                 return 0;
1860
1861         if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
1862                 *len = new_len;
1863                 return 0;
1864         }
1865
1866         return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
1867 }
1868
1869 /* Read a page's worth of file data into the page cache. */
1870 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1871 {
1872         struct page *page;
1873
1874         page = read_mapping_page(inode->i_mapping, offset >> PAGE_SHIFT, NULL);
1875         if (IS_ERR(page))
1876                 return page;
1877         if (!PageUptodate(page)) {
1878                 put_page(page);
1879                 return ERR_PTR(-EIO);
1880         }
1881         return page;
1882 }
1883
1884 /*
1885  * Lock two pages, ensuring that we lock in offset order if the pages are from
1886  * the same file.
1887  */
1888 static void vfs_lock_two_pages(struct page *page1, struct page *page2)
1889 {
1890         /* Always lock in order of increasing index. */
1891         if (page1->index > page2->index)
1892                 swap(page1, page2);
1893
1894         lock_page(page1);
1895         if (page1 != page2)
1896                 lock_page(page2);
1897 }
1898
1899 /* Unlock two pages, being careful not to unlock the same page twice. */
1900 static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
1901 {
1902         unlock_page(page1);
1903         if (page1 != page2)
1904                 unlock_page(page2);
1905 }
1906
1907 /*
1908  * Compare extents of two files to see if they are the same.
1909  * Caller must have locked both inodes to prevent write races.
1910  */
1911 static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1912                                          struct inode *dest, loff_t destoff,
1913                                          loff_t len, bool *is_same)
1914 {
1915         loff_t src_poff;
1916         loff_t dest_poff;
1917         void *src_addr;
1918         void *dest_addr;
1919         struct page *src_page;
1920         struct page *dest_page;
1921         loff_t cmp_len;
1922         bool same;
1923         int error;
1924
1925         error = -EINVAL;
1926         same = true;
1927         while (len) {
1928                 src_poff = srcoff & (PAGE_SIZE - 1);
1929                 dest_poff = destoff & (PAGE_SIZE - 1);
1930                 cmp_len = min(PAGE_SIZE - src_poff,
1931                               PAGE_SIZE - dest_poff);
1932                 cmp_len = min(cmp_len, len);
1933                 if (cmp_len <= 0)
1934                         goto out_error;
1935
1936                 src_page = vfs_dedupe_get_page(src, srcoff);
1937                 if (IS_ERR(src_page)) {
1938                         error = PTR_ERR(src_page);
1939                         goto out_error;
1940                 }
1941                 dest_page = vfs_dedupe_get_page(dest, destoff);
1942                 if (IS_ERR(dest_page)) {
1943                         error = PTR_ERR(dest_page);
1944                         put_page(src_page);
1945                         goto out_error;
1946                 }
1947
1948                 vfs_lock_two_pages(src_page, dest_page);
1949
1950                 /*
1951                  * Now that we've locked both pages, make sure they're still
1952                  * mapped to the file data we're interested in.  If not,
1953                  * someone is invalidating pages on us and we lose.
1954                  */
1955                 if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
1956                     src_page->mapping != src->i_mapping ||
1957                     dest_page->mapping != dest->i_mapping) {
1958                         same = false;
1959                         goto unlock;
1960                 }
1961
1962                 src_addr = kmap_atomic(src_page);
1963                 dest_addr = kmap_atomic(dest_page);
1964
1965                 flush_dcache_page(src_page);
1966                 flush_dcache_page(dest_page);
1967
1968                 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1969                         same = false;
1970
1971                 kunmap_atomic(dest_addr);
1972                 kunmap_atomic(src_addr);
1973 unlock:
1974                 vfs_unlock_two_pages(src_page, dest_page);
1975                 put_page(dest_page);
1976                 put_page(src_page);
1977
1978                 if (!same)
1979                         break;
1980
1981                 srcoff += cmp_len;
1982                 destoff += cmp_len;
1983                 len -= cmp_len;
1984         }
1985
1986         *is_same = same;
1987         return 0;
1988
1989 out_error:
1990         return error;
1991 }
1992
1993 /*
1994  * Check that the two inodes are eligible for cloning, the ranges make
1995  * sense, and then flush all dirty data.  Caller must ensure that the
1996  * inodes have been locked against any other modifications.
1997  *
1998  * If there's an error, then the usual negative error code is returned.
1999  * Otherwise returns 0 with *len set to the request length.
2000  */
2001 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
2002                                   struct file *file_out, loff_t pos_out,
2003                                   loff_t *len, unsigned int remap_flags)
2004 {
2005         struct inode *inode_in = file_inode(file_in);
2006         struct inode *inode_out = file_inode(file_out);
2007         bool same_inode = (inode_in == inode_out);
2008         int ret;
2009
2010         /* Don't touch certain kinds of inodes */
2011         if (IS_IMMUTABLE(inode_out))
2012                 return -EPERM;
2013
2014         if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
2015                 return -ETXTBSY;
2016
2017         /* Don't reflink dirs, pipes, sockets... */
2018         if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
2019                 return -EISDIR;
2020         if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
2021                 return -EINVAL;
2022
2023         /* Zero length dedupe exits immediately; reflink goes to EOF. */
2024         if (*len == 0) {
2025                 loff_t isize = i_size_read(inode_in);
2026
2027                 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
2028                         return 0;
2029                 if (pos_in > isize)
2030                         return -EINVAL;
2031                 *len = isize - pos_in;
2032                 if (*len == 0)
2033                         return 0;
2034         }
2035
2036         /* Check that we don't violate system file offset limits. */
2037         ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
2038                         remap_flags);
2039         if (ret)
2040                 return ret;
2041
2042         /* Wait for the completion of any pending IOs on both files */
2043         inode_dio_wait(inode_in);
2044         if (!same_inode)
2045                 inode_dio_wait(inode_out);
2046
2047         ret = filemap_write_and_wait_range(inode_in->i_mapping,
2048                         pos_in, pos_in + *len - 1);
2049         if (ret)
2050                 return ret;
2051
2052         ret = filemap_write_and_wait_range(inode_out->i_mapping,
2053                         pos_out, pos_out + *len - 1);
2054         if (ret)
2055                 return ret;
2056
2057         /*
2058          * Check that the extents are the same.
2059          */
2060         if (remap_flags & REMAP_FILE_DEDUP) {
2061                 bool            is_same = false;
2062
2063                 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
2064                                 inode_out, pos_out, *len, &is_same);
2065                 if (ret)
2066                         return ret;
2067                 if (!is_same)
2068                         return -EBADE;
2069         }
2070
2071         ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
2072                         remap_flags);
2073         if (ret)
2074                 return ret;
2075
2076         /* If can't alter the file contents, we're done. */
2077         if (!(remap_flags & REMAP_FILE_DEDUP))
2078                 ret = file_modified(file_out);
2079
2080         return ret;
2081 }
2082 EXPORT_SYMBOL(generic_remap_file_range_prep);
2083
2084 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
2085                            struct file *file_out, loff_t pos_out,
2086                            loff_t len, unsigned int remap_flags)
2087 {
2088         loff_t ret;
2089
2090         WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
2091
2092         /*
2093          * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
2094          * the same mount. Practically, they only need to be on the same file
2095          * system.
2096          */
2097         if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
2098                 return -EXDEV;
2099
2100         ret = generic_file_rw_checks(file_in, file_out);
2101         if (ret < 0)
2102                 return ret;
2103
2104         if (!file_in->f_op->remap_file_range)
2105                 return -EOPNOTSUPP;
2106
2107         ret = remap_verify_area(file_in, pos_in, len, false);
2108         if (ret)
2109                 return ret;
2110
2111         ret = remap_verify_area(file_out, pos_out, len, true);
2112         if (ret)
2113                 return ret;
2114
2115         ret = file_in->f_op->remap_file_range(file_in, pos_in,
2116                         file_out, pos_out, len, remap_flags);
2117         if (ret < 0)
2118                 return ret;
2119
2120         fsnotify_access(file_in);
2121         fsnotify_modify(file_out);
2122         return ret;
2123 }
2124 EXPORT_SYMBOL(do_clone_file_range);
2125
2126 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
2127                             struct file *file_out, loff_t pos_out,
2128                             loff_t len, unsigned int remap_flags)
2129 {
2130         loff_t ret;
2131
2132         file_start_write(file_out);
2133         ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
2134                                   remap_flags);
2135         file_end_write(file_out);
2136
2137         return ret;
2138 }
2139 EXPORT_SYMBOL(vfs_clone_file_range);
2140
2141 /* Check whether we are allowed to dedupe the destination file */
2142 static bool allow_file_dedupe(struct file *file)
2143 {
2144         if (capable(CAP_SYS_ADMIN))
2145                 return true;
2146         if (file->f_mode & FMODE_WRITE)
2147                 return true;
2148         if (uid_eq(current_fsuid(), file_inode(file)->i_uid))
2149                 return true;
2150         if (!inode_permission(file_inode(file), MAY_WRITE))
2151                 return true;
2152         return false;
2153 }
2154
2155 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
2156                                  struct file *dst_file, loff_t dst_pos,
2157                                  loff_t len, unsigned int remap_flags)
2158 {
2159         loff_t ret;
2160
2161         WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
2162                                      REMAP_FILE_CAN_SHORTEN));
2163
2164         ret = mnt_want_write_file(dst_file);
2165         if (ret)
2166                 return ret;
2167
2168         ret = remap_verify_area(dst_file, dst_pos, len, true);
2169         if (ret < 0)
2170                 goto out_drop_write;
2171
2172         ret = -EPERM;
2173         if (!allow_file_dedupe(dst_file))
2174                 goto out_drop_write;
2175
2176         ret = -EXDEV;
2177         if (src_file->f_path.mnt != dst_file->f_path.mnt)
2178                 goto out_drop_write;
2179
2180         ret = -EISDIR;
2181         if (S_ISDIR(file_inode(dst_file)->i_mode))
2182                 goto out_drop_write;
2183
2184         ret = -EINVAL;
2185         if (!dst_file->f_op->remap_file_range)
2186                 goto out_drop_write;
2187
2188         if (len == 0) {
2189                 ret = 0;
2190                 goto out_drop_write;
2191         }
2192
2193         ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
2194                         dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
2195 out_drop_write:
2196         mnt_drop_write_file(dst_file);
2197
2198         return ret;
2199 }
2200 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
2201
2202 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
2203 {
2204         struct file_dedupe_range_info *info;
2205         struct inode *src = file_inode(file);
2206         u64 off;
2207         u64 len;
2208         int i;
2209         int ret;
2210         u16 count = same->dest_count;
2211         loff_t deduped;
2212
2213         if (!(file->f_mode & FMODE_READ))
2214                 return -EINVAL;
2215
2216         if (same->reserved1 || same->reserved2)
2217                 return -EINVAL;
2218
2219         off = same->src_offset;
2220         len = same->src_length;
2221
2222         if (S_ISDIR(src->i_mode))
2223                 return -EISDIR;
2224
2225         if (!S_ISREG(src->i_mode))
2226                 return -EINVAL;
2227
2228         if (!file->f_op->remap_file_range)
2229                 return -EOPNOTSUPP;
2230
2231         ret = remap_verify_area(file, off, len, false);
2232         if (ret < 0)
2233                 return ret;
2234         ret = 0;
2235
2236         if (off + len > i_size_read(src))
2237                 return -EINVAL;
2238
2239         /* Arbitrary 1G limit on a single dedupe request, can be raised. */
2240         len = min_t(u64, len, 1 << 30);
2241
2242         /* pre-format output fields to sane values */
2243         for (i = 0; i < count; i++) {
2244                 same->info[i].bytes_deduped = 0ULL;
2245                 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2246         }
2247
2248         for (i = 0, info = same->info; i < count; i++, info++) {
2249                 struct fd dst_fd = fdget(info->dest_fd);
2250                 struct file *dst_file = dst_fd.file;
2251
2252                 if (!dst_file) {
2253                         info->status = -EBADF;
2254                         goto next_loop;
2255                 }
2256
2257                 if (info->reserved) {
2258                         info->status = -EINVAL;
2259                         goto next_fdput;
2260                 }
2261
2262                 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
2263                                                     info->dest_offset, len,
2264                                                     REMAP_FILE_CAN_SHORTEN);
2265                 if (deduped == -EBADE)
2266                         info->status = FILE_DEDUPE_RANGE_DIFFERS;
2267                 else if (deduped < 0)
2268                         info->status = deduped;
2269                 else
2270                         info->bytes_deduped = len;
2271
2272 next_fdput:
2273                 fdput(dst_fd);
2274 next_loop:
2275                 if (fatal_signal_pending(current))
2276                         break;
2277         }
2278         return ret;
2279 }
2280 EXPORT_SYMBOL(vfs_dedupe_file_range);