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