linux/fs/read_write.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/read_write.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7#include <linux/slab.h> 
   8#include <linux/stat.h>
   9#include <linux/fcntl.h>
  10#include <linux/file.h>
  11#include <linux/uio.h>
  12#include <linux/fsnotify.h>
  13#include <linux/security.h>
  14#include <linux/export.h>
  15#include <linux/syscalls.h>
  16#include <linux/pagemap.h>
  17#include <linux/splice.h>
  18#include <linux/compat.h>
  19#include <linux/mount.h>
  20#include <linux/fs.h>
  21#include "internal.h"
  22
  23#include <asm/uaccess.h>
  24#include <asm/unistd.h>
  25
  26typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
  27typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
  28
  29const struct file_operations generic_ro_fops = {
  30        .llseek         = generic_file_llseek,
  31        .read_iter      = generic_file_read_iter,
  32        .mmap           = generic_file_readonly_mmap,
  33        .splice_read    = generic_file_splice_read,
  34};
  35
  36EXPORT_SYMBOL(generic_ro_fops);
  37
  38static inline int unsigned_offsets(struct file *file)
  39{
  40        return file->f_mode & FMODE_UNSIGNED_OFFSET;
  41}
  42
  43/**
  44 * vfs_setpos - update the file offset for lseek
  45 * @file:       file structure in question
  46 * @offset:     file offset to seek to
  47 * @maxsize:    maximum file size
  48 *
  49 * This is a low-level filesystem helper for updating the file offset to
  50 * the value specified by @offset if the given offset is valid and it is
  51 * not equal to the current file offset.
  52 *
  53 * Return the specified offset on success and -EINVAL on invalid offset.
  54 */
  55loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
  56{
  57        if (offset < 0 && !unsigned_offsets(file))
  58                return -EINVAL;
  59        if (offset > maxsize)
  60                return -EINVAL;
  61
  62        if (offset != file->f_pos) {
  63                file->f_pos = offset;
  64                file->f_version = 0;
  65        }
  66        return offset;
  67}
  68EXPORT_SYMBOL(vfs_setpos);
  69
  70/**
  71 * generic_file_llseek_size - generic llseek implementation for regular files
  72 * @file:       file structure to seek on
  73 * @offset:     file offset to seek to
  74 * @whence:     type of seek
  75 * @size:       max size of this file in file system
  76 * @eof:        offset used for SEEK_END position
  77 *
  78 * This is a variant of generic_file_llseek that allows passing in a custom
  79 * maximum file size and a custom EOF position, for e.g. hashed directories
  80 *
  81 * Synchronization:
  82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
  83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
  84 * read/writes behave like SEEK_SET against seeks.
  85 */
  86loff_t
  87generic_file_llseek_size(struct file *file, loff_t offset, int whence,
  88                loff_t maxsize, loff_t eof)
  89{
  90        switch (whence) {
  91        case SEEK_END:
  92                offset += eof;
  93                break;
  94        case SEEK_CUR:
  95                /*
  96                 * Here we special-case the lseek(fd, 0, SEEK_CUR)
  97                 * position-querying operation.  Avoid rewriting the "same"
  98                 * f_pos value back to the file because a concurrent read(),
  99                 * write() or lseek() might have altered it
 100                 */
 101                if (offset == 0)
 102                        return file->f_pos;
 103                /*
 104                 * f_lock protects against read/modify/write race with other
 105                 * SEEK_CURs. Note that parallel writes and reads behave
 106                 * like SEEK_SET.
 107                 */
 108                spin_lock(&file->f_lock);
 109                offset = vfs_setpos(file, file->f_pos + offset, maxsize);
 110                spin_unlock(&file->f_lock);
 111                return offset;
 112        case SEEK_DATA:
 113                /*
 114                 * In the generic case the entire file is data, so as long as
 115                 * offset isn't at the end of the file then the offset is data.
 116                 */
 117                if (offset >= eof)
 118                        return -ENXIO;
 119                break;
 120        case SEEK_HOLE:
 121                /*
 122                 * There is a virtual hole at the end of the file, so as long as
 123                 * offset isn't i_size or larger, return i_size.
 124                 */
 125                if (offset >= eof)
 126                        return -ENXIO;
 127                offset = eof;
 128                break;
 129        }
 130
 131        return vfs_setpos(file, offset, maxsize);
 132}
 133EXPORT_SYMBOL(generic_file_llseek_size);
 134
 135/**
 136 * generic_file_llseek - generic llseek implementation for regular files
 137 * @file:       file structure to seek on
 138 * @offset:     file offset to seek to
 139 * @whence:     type of seek
 140 *
 141 * This is a generic implemenation of ->llseek useable for all normal local
 142 * filesystems.  It just updates the file offset to the value specified by
 143 * @offset and @whence.
 144 */
 145loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
 146{
 147        struct inode *inode = file->f_mapping->host;
 148
 149        return generic_file_llseek_size(file, offset, whence,
 150                                        inode->i_sb->s_maxbytes,
 151                                        i_size_read(inode));
 152}
 153EXPORT_SYMBOL(generic_file_llseek);
 154
 155/**
 156 * fixed_size_llseek - llseek implementation for fixed-sized devices
 157 * @file:       file structure to seek on
 158 * @offset:     file offset to seek to
 159 * @whence:     type of seek
 160 * @size:       size of the file
 161 *
 162 */
 163loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
 164{
 165        switch (whence) {
 166        case SEEK_SET: case SEEK_CUR: case SEEK_END:
 167                return generic_file_llseek_size(file, offset, whence,
 168                                                size, size);
 169        default:
 170                return -EINVAL;
 171        }
 172}
 173EXPORT_SYMBOL(fixed_size_llseek);
 174
 175/**
 176 * no_seek_end_llseek - llseek implementation for fixed-sized devices
 177 * @file:       file structure to seek on
 178 * @offset:     file offset to seek to
 179 * @whence:     type of seek
 180 *
 181 */
 182loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
 183{
 184        switch (whence) {
 185        case SEEK_SET: case SEEK_CUR:
 186                return generic_file_llseek_size(file, offset, whence,
 187                                                OFFSET_MAX, 0);
 188        default:
 189                return -EINVAL;
 190        }
 191}
 192EXPORT_SYMBOL(no_seek_end_llseek);
 193
 194/**
 195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
 196 * @file:       file structure to seek on
 197 * @offset:     file offset to seek to
 198 * @whence:     type of seek
 199 * @size:       maximal offset allowed
 200 *
 201 */
 202loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
 203{
 204        switch (whence) {
 205        case SEEK_SET: case SEEK_CUR:
 206                return generic_file_llseek_size(file, offset, whence,
 207                                                size, 0);
 208        default:
 209                return -EINVAL;
 210        }
 211}
 212EXPORT_SYMBOL(no_seek_end_llseek_size);
 213
 214/**
 215 * noop_llseek - No Operation Performed llseek implementation
 216 * @file:       file structure to seek on
 217 * @offset:     file offset to seek to
 218 * @whence:     type of seek
 219 *
 220 * This is an implementation of ->llseek useable for the rare special case when
 221 * userspace expects the seek to succeed but the (device) file is actually not
 222 * able to perform the seek. In this case you use noop_llseek() instead of
 223 * falling back to the default implementation of ->llseek.
 224 */
 225loff_t noop_llseek(struct file *file, loff_t offset, int whence)
 226{
 227        return file->f_pos;
 228}
 229EXPORT_SYMBOL(noop_llseek);
 230
 231loff_t no_llseek(struct file *file, loff_t offset, int whence)
 232{
 233        return -ESPIPE;
 234}
 235EXPORT_SYMBOL(no_llseek);
 236
 237loff_t default_llseek(struct file *file, loff_t offset, int whence)
 238{
 239        struct inode *inode = file_inode(file);
 240        loff_t retval;
 241
 242        inode_lock(inode);
 243        switch (whence) {
 244                case SEEK_END:
 245                        offset += i_size_read(inode);
 246                        break;
 247                case SEEK_CUR:
 248                        if (offset == 0) {
 249                                retval = file->f_pos;
 250                                goto out;
 251                        }
 252                        offset += file->f_pos;
 253                        break;
 254                case SEEK_DATA:
 255                        /*
 256                         * In the generic case the entire file is data, so as
 257                         * long as offset isn't at the end of the file then the
 258                         * offset is data.
 259                         */
 260                        if (offset >= inode->i_size) {
 261                                retval = -ENXIO;
 262                                goto out;
 263                        }
 264                        break;
 265                case SEEK_HOLE:
 266                        /*
 267                         * There is a virtual hole at the end of the file, so
 268                         * as long as offset isn't i_size or larger, return
 269                         * i_size.
 270                         */
 271                        if (offset >= inode->i_size) {
 272                                retval = -ENXIO;
 273                                goto out;
 274                        }
 275                        offset = inode->i_size;
 276                        break;
 277        }
 278        retval = -EINVAL;
 279        if (offset >= 0 || unsigned_offsets(file)) {
 280                if (offset != file->f_pos) {
 281                        file->f_pos = offset;
 282                        file->f_version = 0;
 283                }
 284                retval = offset;
 285        }
 286out:
 287        inode_unlock(inode);
 288        return retval;
 289}
 290EXPORT_SYMBOL(default_llseek);
 291
 292loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
 293{
 294        loff_t (*fn)(struct file *, loff_t, int);
 295
 296        fn = no_llseek;
 297        if (file->f_mode & FMODE_LSEEK) {
 298                if (file->f_op->llseek)
 299                        fn = file->f_op->llseek;
 300        }
 301        return fn(file, offset, whence);
 302}
 303EXPORT_SYMBOL(vfs_llseek);
 304
 305SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
 306{
 307        off_t retval;
 308        struct fd f = fdget_pos(fd);
 309        if (!f.file)
 310                return -EBADF;
 311
 312        retval = -EINVAL;
 313        if (whence <= SEEK_MAX) {
 314                loff_t res = vfs_llseek(f.file, offset, whence);
 315                retval = res;
 316                if (res != (loff_t)retval)
 317                        retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
 318        }
 319        fdput_pos(f);
 320        return retval;
 321}
 322
 323#ifdef CONFIG_COMPAT
 324COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
 325{
 326        return sys_lseek(fd, offset, whence);
 327}
 328#endif
 329
 330#ifdef __ARCH_WANT_SYS_LLSEEK
 331SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
 332                unsigned long, offset_low, loff_t __user *, result,
 333                unsigned int, whence)
 334{
 335        int retval;
 336        struct fd f = fdget_pos(fd);
 337        loff_t offset;
 338
 339        if (!f.file)
 340                return -EBADF;
 341
 342        retval = -EINVAL;
 343        if (whence > SEEK_MAX)
 344                goto out_putf;
 345
 346        offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
 347                        whence);
 348
 349        retval = (int)offset;
 350        if (offset >= 0) {
 351                retval = -EFAULT;
 352                if (!copy_to_user(result, &offset, sizeof(offset)))
 353                        retval = 0;
 354        }
 355out_putf:
 356        fdput_pos(f);
 357        return retval;
 358}
 359#endif
 360
 361ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
 362{
 363        struct kiocb kiocb;
 364        ssize_t ret;
 365
 366        if (!file->f_op->read_iter)
 367                return -EINVAL;
 368
 369        init_sync_kiocb(&kiocb, file);
 370        kiocb.ki_pos = *ppos;
 371
 372        iter->type |= READ;
 373        ret = file->f_op->read_iter(&kiocb, iter);
 374        BUG_ON(ret == -EIOCBQUEUED);
 375        if (ret > 0)
 376                *ppos = kiocb.ki_pos;
 377        return ret;
 378}
 379EXPORT_SYMBOL(vfs_iter_read);
 380
 381ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
 382{
 383        struct kiocb kiocb;
 384        ssize_t ret;
 385
 386        if (!file->f_op->write_iter)
 387                return -EINVAL;
 388
 389        init_sync_kiocb(&kiocb, file);
 390        kiocb.ki_pos = *ppos;
 391
 392        iter->type |= WRITE;
 393        ret = file->f_op->write_iter(&kiocb, iter);
 394        BUG_ON(ret == -EIOCBQUEUED);
 395        if (ret > 0)
 396                *ppos = kiocb.ki_pos;
 397        return ret;
 398}
 399EXPORT_SYMBOL(vfs_iter_write);
 400
 401int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
 402{
 403        struct inode *inode;
 404        loff_t pos;
 405        int retval = -EINVAL;
 406
 407        inode = file_inode(file);
 408        if (unlikely((ssize_t) count < 0))
 409                return retval;
 410        pos = *ppos;
 411        if (unlikely(pos < 0)) {
 412                if (!unsigned_offsets(file))
 413                        return retval;
 414                if (count >= -pos) /* both values are in 0..LLONG_MAX */
 415                        return -EOVERFLOW;
 416        } else if (unlikely((loff_t) (pos + count) < 0)) {
 417                if (!unsigned_offsets(file))
 418                        return retval;
 419        }
 420
 421        if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
 422                retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
 423                                read_write == READ ? F_RDLCK : F_WRLCK);
 424                if (retval < 0)
 425                        return retval;
 426        }
 427        return security_file_permission(file,
 428                                read_write == READ ? MAY_READ : MAY_WRITE);
 429}
 430
 431static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 432{
 433        struct iovec iov = { .iov_base = buf, .iov_len = len };
 434        struct kiocb kiocb;
 435        struct iov_iter iter;
 436        ssize_t ret;
 437
 438        init_sync_kiocb(&kiocb, filp);
 439        kiocb.ki_pos = *ppos;
 440        iov_iter_init(&iter, READ, &iov, 1, len);
 441
 442        ret = filp->f_op->read_iter(&kiocb, &iter);
 443        BUG_ON(ret == -EIOCBQUEUED);
 444        *ppos = kiocb.ki_pos;
 445        return ret;
 446}
 447
 448ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
 449                   loff_t *pos)
 450{
 451        if (file->f_op->read)
 452                return file->f_op->read(file, buf, count, pos);
 453        else if (file->f_op->read_iter)
 454                return new_sync_read(file, buf, count, pos);
 455        else
 456                return -EINVAL;
 457}
 458EXPORT_SYMBOL(__vfs_read);
 459
 460ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 461{
 462        ssize_t ret;
 463
 464        if (!(file->f_mode & FMODE_READ))
 465                return -EBADF;
 466        if (!(file->f_mode & FMODE_CAN_READ))
 467                return -EINVAL;
 468        if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
 469                return -EFAULT;
 470
 471        ret = rw_verify_area(READ, file, pos, count);
 472        if (!ret) {
 473                if (count > MAX_RW_COUNT)
 474                        count =  MAX_RW_COUNT;
 475                ret = __vfs_read(file, buf, count, pos);
 476                if (ret > 0) {
 477                        fsnotify_access(file);
 478                        add_rchar(current, ret);
 479                }
 480                inc_syscr(current);
 481        }
 482
 483        return ret;
 484}
 485
 486EXPORT_SYMBOL(vfs_read);
 487
 488static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 489{
 490        struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
 491        struct kiocb kiocb;
 492        struct iov_iter iter;
 493        ssize_t ret;
 494
 495        init_sync_kiocb(&kiocb, filp);
 496        kiocb.ki_pos = *ppos;
 497        iov_iter_init(&iter, WRITE, &iov, 1, len);
 498
 499        ret = filp->f_op->write_iter(&kiocb, &iter);
 500        BUG_ON(ret == -EIOCBQUEUED);
 501        if (ret > 0)
 502                *ppos = kiocb.ki_pos;
 503        return ret;
 504}
 505
 506ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
 507                    loff_t *pos)
 508{
 509        if (file->f_op->write)
 510                return file->f_op->write(file, p, count, pos);
 511        else if (file->f_op->write_iter)
 512                return new_sync_write(file, p, count, pos);
 513        else
 514                return -EINVAL;
 515}
 516EXPORT_SYMBOL(__vfs_write);
 517
 518ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
 519{
 520        mm_segment_t old_fs;
 521        const char __user *p;
 522        ssize_t ret;
 523
 524        if (!(file->f_mode & FMODE_CAN_WRITE))
 525                return -EINVAL;
 526
 527        old_fs = get_fs();
 528        set_fs(get_ds());
 529        p = (__force const char __user *)buf;
 530        if (count > MAX_RW_COUNT)
 531                count =  MAX_RW_COUNT;
 532        ret = __vfs_write(file, p, count, pos);
 533        set_fs(old_fs);
 534        if (ret > 0) {
 535                fsnotify_modify(file);
 536                add_wchar(current, ret);
 537        }
 538        inc_syscw(current);
 539        return ret;
 540}
 541
 542EXPORT_SYMBOL(__kernel_write);
 543
 544ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 545{
 546        ssize_t ret;
 547
 548        if (!(file->f_mode & FMODE_WRITE))
 549                return -EBADF;
 550        if (!(file->f_mode & FMODE_CAN_WRITE))
 551                return -EINVAL;
 552        if (unlikely(!access_ok(VERIFY_READ, buf, count)))
 553                return -EFAULT;
 554
 555        ret = rw_verify_area(WRITE, file, pos, count);
 556        if (!ret) {
 557                if (count > MAX_RW_COUNT)
 558                        count =  MAX_RW_COUNT;
 559                file_start_write(file);
 560                ret = __vfs_write(file, buf, count, pos);
 561                if (ret > 0) {
 562                        fsnotify_modify(file);
 563                        add_wchar(current, ret);
 564                }
 565                inc_syscw(current);
 566                file_end_write(file);
 567        }
 568
 569        return ret;
 570}
 571
 572EXPORT_SYMBOL(vfs_write);
 573
 574static inline loff_t file_pos_read(struct file *file)
 575{
 576        return file->f_pos;
 577}
 578
 579static inline void file_pos_write(struct file *file, loff_t pos)
 580{
 581        file->f_pos = pos;
 582}
 583
 584SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
 585{
 586        struct fd f = fdget_pos(fd);
 587        ssize_t ret = -EBADF;
 588
 589        if (f.file) {
 590                loff_t pos = file_pos_read(f.file);
 591                ret = vfs_read(f.file, buf, count, &pos);
 592                if (ret >= 0)
 593                        file_pos_write(f.file, pos);
 594                fdput_pos(f);
 595        }
 596        return ret;
 597}
 598
 599SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
 600                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 = file_pos_read(f.file);
 607                ret = vfs_write(f.file, buf, count, &pos);
 608                if (ret >= 0)
 609                        file_pos_write(f.file, pos);
 610                fdput_pos(f);
 611        }
 612
 613        return ret;
 614}
 615
 616SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
 617                        size_t, count, loff_t, pos)
 618{
 619        struct fd f;
 620        ssize_t ret = -EBADF;
 621
 622        if (pos < 0)
 623                return -EINVAL;
 624
 625        f = fdget(fd);
 626        if (f.file) {
 627                ret = -ESPIPE;
 628                if (f.file->f_mode & FMODE_PREAD)
 629                        ret = vfs_read(f.file, buf, count, &pos);
 630                fdput(f);
 631        }
 632
 633        return ret;
 634}
 635
 636SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
 637                         size_t, count, loff_t, pos)
 638{
 639        struct fd f;
 640        ssize_t ret = -EBADF;
 641
 642        if (pos < 0)
 643                return -EINVAL;
 644
 645        f = fdget(fd);
 646        if (f.file) {
 647                ret = -ESPIPE;
 648                if (f.file->f_mode & FMODE_PWRITE)  
 649                        ret = vfs_write(f.file, buf, count, &pos);
 650                fdput(f);
 651        }
 652
 653        return ret;
 654}
 655
 656/*
 657 * Reduce an iovec's length in-place.  Return the resulting number of segments
 658 */
 659unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
 660{
 661        unsigned long seg = 0;
 662        size_t len = 0;
 663
 664        while (seg < nr_segs) {
 665                seg++;
 666                if (len + iov->iov_len >= to) {
 667                        iov->iov_len = to - len;
 668                        break;
 669                }
 670                len += iov->iov_len;
 671                iov++;
 672        }
 673        return seg;
 674}
 675EXPORT_SYMBOL(iov_shorten);
 676
 677static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
 678                loff_t *ppos, iter_fn_t fn, int flags)
 679{
 680        struct kiocb kiocb;
 681        ssize_t ret;
 682
 683        if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
 684                return -EOPNOTSUPP;
 685
 686        init_sync_kiocb(&kiocb, filp);
 687        if (flags & RWF_HIPRI)
 688                kiocb.ki_flags |= IOCB_HIPRI;
 689        if (flags & RWF_DSYNC)
 690                kiocb.ki_flags |= IOCB_DSYNC;
 691        if (flags & RWF_SYNC)
 692                kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
 693        kiocb.ki_pos = *ppos;
 694
 695        ret = fn(&kiocb, iter);
 696        BUG_ON(ret == -EIOCBQUEUED);
 697        *ppos = kiocb.ki_pos;
 698        return ret;
 699}
 700
 701/* Do it by hand, with file-ops */
 702static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
 703                loff_t *ppos, io_fn_t fn, int 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                nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
 715
 716                if (nr < 0) {
 717                        if (!ret)
 718                                ret = nr;
 719                        break;
 720                }
 721                ret += nr;
 722                if (nr != iovec.iov_len)
 723                        break;
 724                iov_iter_advance(iter, nr);
 725        }
 726
 727        return ret;
 728}
 729
 730/* A write operation does a read from user space and vice versa */
 731#define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 732
 733ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
 734                              unsigned long nr_segs, unsigned long fast_segs,
 735                              struct iovec *fast_pointer,
 736                              struct iovec **ret_pointer)
 737{
 738        unsigned long seg;
 739        ssize_t ret;
 740        struct iovec *iov = fast_pointer;
 741
 742        /*
 743         * SuS says "The readv() function *may* fail if the iovcnt argument
 744         * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
 745         * traditionally returned zero for zero segments, so...
 746         */
 747        if (nr_segs == 0) {
 748                ret = 0;
 749                goto out;
 750        }
 751
 752        /*
 753         * First get the "struct iovec" from user memory and
 754         * verify all the pointers
 755         */
 756        if (nr_segs > UIO_MAXIOV) {
 757                ret = -EINVAL;
 758                goto out;
 759        }
 760        if (nr_segs > fast_segs) {
 761                iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
 762                if (iov == NULL) {
 763                        ret = -ENOMEM;
 764                        goto out;
 765                }
 766        }
 767        if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
 768                ret = -EFAULT;
 769                goto out;
 770        }
 771
 772        /*
 773         * According to the Single Unix Specification we should return EINVAL
 774         * if an element length is < 0 when cast to ssize_t or if the
 775         * total length would overflow the ssize_t return value of the
 776         * system call.
 777         *
 778         * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
 779         * overflow case.
 780         */
 781        ret = 0;
 782        for (seg = 0; seg < nr_segs; seg++) {
 783                void __user *buf = iov[seg].iov_base;
 784                ssize_t len = (ssize_t)iov[seg].iov_len;
 785
 786                /* see if we we're about to use an invalid len or if
 787                 * it's about to overflow ssize_t */
 788                if (len < 0) {
 789                        ret = -EINVAL;
 790                        goto out;
 791                }
 792                if (type >= 0
 793                    && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
 794                        ret = -EFAULT;
 795                        goto out;
 796                }
 797                if (len > MAX_RW_COUNT - ret) {
 798                        len = MAX_RW_COUNT - ret;
 799                        iov[seg].iov_len = len;
 800                }
 801                ret += len;
 802        }
 803out:
 804        *ret_pointer = iov;
 805        return ret;
 806}
 807
 808static ssize_t do_readv_writev(int type, struct file *file,
 809                               const struct iovec __user * uvector,
 810                               unsigned long nr_segs, loff_t *pos,
 811                               int flags)
 812{
 813        size_t tot_len;
 814        struct iovec iovstack[UIO_FASTIOV];
 815        struct iovec *iov = iovstack;
 816        struct iov_iter iter;
 817        ssize_t ret;
 818        io_fn_t fn;
 819        iter_fn_t iter_fn;
 820
 821        ret = import_iovec(type, uvector, nr_segs,
 822                           ARRAY_SIZE(iovstack), &iov, &iter);
 823        if (ret < 0)
 824                return ret;
 825
 826        tot_len = iov_iter_count(&iter);
 827        if (!tot_len)
 828                goto out;
 829        ret = rw_verify_area(type, file, pos, tot_len);
 830        if (ret < 0)
 831                goto out;
 832
 833        if (type == READ) {
 834                fn = file->f_op->read;
 835                iter_fn = file->f_op->read_iter;
 836        } else {
 837                fn = (io_fn_t)file->f_op->write;
 838                iter_fn = file->f_op->write_iter;
 839                file_start_write(file);
 840        }
 841
 842        if (iter_fn)
 843                ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
 844        else
 845                ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
 846
 847        if (type != READ)
 848                file_end_write(file);
 849
 850out:
 851        kfree(iov);
 852        if ((ret + (type == READ)) > 0) {
 853                if (type == READ)
 854                        fsnotify_access(file);
 855                else
 856                        fsnotify_modify(file);
 857        }
 858        return ret;
 859}
 860
 861ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
 862                  unsigned long vlen, loff_t *pos, int flags)
 863{
 864        if (!(file->f_mode & FMODE_READ))
 865                return -EBADF;
 866        if (!(file->f_mode & FMODE_CAN_READ))
 867                return -EINVAL;
 868
 869        return do_readv_writev(READ, file, vec, vlen, pos, flags);
 870}
 871
 872EXPORT_SYMBOL(vfs_readv);
 873
 874ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 875                   unsigned long vlen, loff_t *pos, int flags)
 876{
 877        if (!(file->f_mode & FMODE_WRITE))
 878                return -EBADF;
 879        if (!(file->f_mode & FMODE_CAN_WRITE))
 880                return -EINVAL;
 881
 882        return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
 883}
 884
 885EXPORT_SYMBOL(vfs_writev);
 886
 887static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
 888                        unsigned long vlen, int flags)
 889{
 890        struct fd f = fdget_pos(fd);
 891        ssize_t ret = -EBADF;
 892
 893        if (f.file) {
 894                loff_t pos = file_pos_read(f.file);
 895                ret = vfs_readv(f.file, vec, vlen, &pos, flags);
 896                if (ret >= 0)
 897                        file_pos_write(f.file, pos);
 898                fdput_pos(f);
 899        }
 900
 901        if (ret > 0)
 902                add_rchar(current, ret);
 903        inc_syscr(current);
 904        return ret;
 905}
 906
 907static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
 908                         unsigned long vlen, int flags)
 909{
 910        struct fd f = fdget_pos(fd);
 911        ssize_t ret = -EBADF;
 912
 913        if (f.file) {
 914                loff_t pos = file_pos_read(f.file);
 915                ret = vfs_writev(f.file, vec, vlen, &pos, flags);
 916                if (ret >= 0)
 917                        file_pos_write(f.file, pos);
 918                fdput_pos(f);
 919        }
 920
 921        if (ret > 0)
 922                add_wchar(current, ret);
 923        inc_syscw(current);
 924        return ret;
 925}
 926
 927static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
 928{
 929#define HALF_LONG_BITS (BITS_PER_LONG / 2)
 930        return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
 931}
 932
 933static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
 934                         unsigned long vlen, loff_t pos, int flags)
 935{
 936        struct fd f;
 937        ssize_t ret = -EBADF;
 938
 939        if (pos < 0)
 940                return -EINVAL;
 941
 942        f = fdget(fd);
 943        if (f.file) {
 944                ret = -ESPIPE;
 945                if (f.file->f_mode & FMODE_PREAD)
 946                        ret = vfs_readv(f.file, vec, vlen, &pos, flags);
 947                fdput(f);
 948        }
 949
 950        if (ret > 0)
 951                add_rchar(current, ret);
 952        inc_syscr(current);
 953        return ret;
 954}
 955
 956static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
 957                          unsigned long vlen, loff_t pos, int flags)
 958{
 959        struct fd f;
 960        ssize_t ret = -EBADF;
 961
 962        if (pos < 0)
 963                return -EINVAL;
 964
 965        f = fdget(fd);
 966        if (f.file) {
 967                ret = -ESPIPE;
 968                if (f.file->f_mode & FMODE_PWRITE)
 969                        ret = vfs_writev(f.file, vec, vlen, &pos, flags);
 970                fdput(f);
 971        }
 972
 973        if (ret > 0)
 974                add_wchar(current, ret);
 975        inc_syscw(current);
 976        return ret;
 977}
 978
 979SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
 980                unsigned long, vlen)
 981{
 982        return do_readv(fd, vec, vlen, 0);
 983}
 984
 985SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
 986                unsigned long, vlen)
 987{
 988        return do_writev(fd, vec, vlen, 0);
 989}
 990
 991SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
 992                unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
 993{
 994        loff_t pos = pos_from_hilo(pos_h, pos_l);
 995
 996        return do_preadv(fd, vec, vlen, pos, 0);
 997}
 998
 999SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1000                unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1001                int, flags)
1002{
1003        loff_t pos = pos_from_hilo(pos_h, pos_l);
1004
1005        if (pos == -1)
1006                return do_readv(fd, vec, vlen, flags);
1007
1008        return do_preadv(fd, vec, vlen, pos, flags);
1009}
1010
1011SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1012                unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1013{
1014        loff_t pos = pos_from_hilo(pos_h, pos_l);
1015
1016        return do_pwritev(fd, vec, vlen, pos, 0);
1017}
1018
1019SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1020                unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1021                int, flags)
1022{
1023        loff_t pos = pos_from_hilo(pos_h, pos_l);
1024
1025        if (pos == -1)
1026                return do_writev(fd, vec, vlen, flags);
1027
1028        return do_pwritev(fd, vec, vlen, pos, flags);
1029}
1030
1031#ifdef CONFIG_COMPAT
1032
1033static ssize_t compat_do_readv_writev(int type, struct file *file,
1034                               const struct compat_iovec __user *uvector,
1035                               unsigned long nr_segs, loff_t *pos,
1036                               int flags)
1037{
1038        compat_ssize_t tot_len;
1039        struct iovec iovstack[UIO_FASTIOV];
1040        struct iovec *iov = iovstack;
1041        struct iov_iter iter;
1042        ssize_t ret;
1043        io_fn_t fn;
1044        iter_fn_t iter_fn;
1045
1046        ret = compat_import_iovec(type, uvector, nr_segs,
1047                                  UIO_FASTIOV, &iov, &iter);
1048        if (ret < 0)
1049                return ret;
1050
1051        tot_len = iov_iter_count(&iter);
1052        if (!tot_len)
1053                goto out;
1054        ret = rw_verify_area(type, file, pos, tot_len);
1055        if (ret < 0)
1056                goto out;
1057
1058        if (type == READ) {
1059                fn = file->f_op->read;
1060                iter_fn = file->f_op->read_iter;
1061        } else {
1062                fn = (io_fn_t)file->f_op->write;
1063                iter_fn = file->f_op->write_iter;
1064                file_start_write(file);
1065        }
1066
1067        if (iter_fn)
1068                ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
1069        else
1070                ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
1071
1072        if (type != READ)
1073                file_end_write(file);
1074
1075out:
1076        kfree(iov);
1077        if ((ret + (type == READ)) > 0) {
1078                if (type == READ)
1079                        fsnotify_access(file);
1080                else
1081                        fsnotify_modify(file);
1082        }
1083        return ret;
1084}
1085
1086static size_t compat_readv(struct file *file,
1087                           const struct compat_iovec __user *vec,
1088                           unsigned long vlen, loff_t *pos, int flags)
1089{
1090        ssize_t ret = -EBADF;
1091
1092        if (!(file->f_mode & FMODE_READ))
1093                goto out;
1094
1095        ret = -EINVAL;
1096        if (!(file->f_mode & FMODE_CAN_READ))
1097                goto out;
1098
1099        ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1100
1101out:
1102        if (ret > 0)
1103                add_rchar(current, ret);
1104        inc_syscr(current);
1105        return ret;
1106}
1107
1108static size_t do_compat_readv(compat_ulong_t fd,
1109                                 const struct compat_iovec __user *vec,
1110                                 compat_ulong_t vlen, int flags)
1111{
1112        struct fd f = fdget_pos(fd);
1113        ssize_t ret;
1114        loff_t pos;
1115
1116        if (!f.file)
1117                return -EBADF;
1118        pos = f.file->f_pos;
1119        ret = compat_readv(f.file, vec, vlen, &pos, flags);
1120        if (ret >= 0)
1121                f.file->f_pos = pos;
1122        fdput_pos(f);
1123        return ret;
1124
1125}
1126
1127COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1128                const struct compat_iovec __user *,vec,
1129                compat_ulong_t, vlen)
1130{
1131        return do_compat_readv(fd, vec, vlen, 0);
1132}
1133
1134static long do_compat_preadv64(unsigned long fd,
1135                                  const struct compat_iovec __user *vec,
1136                                  unsigned long vlen, loff_t pos, int flags)
1137{
1138        struct fd f;
1139        ssize_t ret;
1140
1141        if (pos < 0)
1142                return -EINVAL;
1143        f = fdget(fd);
1144        if (!f.file)
1145                return -EBADF;
1146        ret = -ESPIPE;
1147        if (f.file->f_mode & FMODE_PREAD)
1148                ret = compat_readv(f.file, vec, vlen, &pos, flags);
1149        fdput(f);
1150        return ret;
1151}
1152
1153#ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1154COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1155                const struct compat_iovec __user *,vec,
1156                unsigned long, vlen, loff_t, pos)
1157{
1158        return do_compat_preadv64(fd, vec, vlen, pos, 0);
1159}
1160#endif
1161
1162COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1163                const struct compat_iovec __user *,vec,
1164                compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1165{
1166        loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1167
1168        return do_compat_preadv64(fd, vec, vlen, pos, 0);
1169}
1170
1171#ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1172COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1173                const struct compat_iovec __user *,vec,
1174                unsigned long, vlen, loff_t, pos, int, flags)
1175{
1176        return do_compat_preadv64(fd, vec, vlen, pos, flags);
1177}
1178#endif
1179
1180COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1181                const struct compat_iovec __user *,vec,
1182                compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1183                int, flags)
1184{
1185        loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1186
1187        if (pos == -1)
1188                return do_compat_readv(fd, vec, vlen, flags);
1189
1190        return do_compat_preadv64(fd, vec, vlen, pos, flags);
1191}
1192
1193static size_t compat_writev(struct file *file,
1194                            const struct compat_iovec __user *vec,
1195                            unsigned long vlen, loff_t *pos, int flags)
1196{
1197        ssize_t ret = -EBADF;
1198
1199        if (!(file->f_mode & FMODE_WRITE))
1200                goto out;
1201
1202        ret = -EINVAL;
1203        if (!(file->f_mode & FMODE_CAN_WRITE))
1204                goto out;
1205
1206        ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0);
1207
1208out:
1209        if (ret > 0)
1210                add_wchar(current, ret);
1211        inc_syscw(current);
1212        return ret;
1213}
1214
1215static size_t do_compat_writev(compat_ulong_t fd,
1216                                  const struct compat_iovec __user* vec,
1217                                  compat_ulong_t vlen, int flags)
1218{
1219        struct fd f = fdget_pos(fd);
1220        ssize_t ret;
1221        loff_t pos;
1222
1223        if (!f.file)
1224                return -EBADF;
1225        pos = f.file->f_pos;
1226        ret = compat_writev(f.file, vec, vlen, &pos, flags);
1227        if (ret >= 0)
1228                f.file->f_pos = pos;
1229        fdput_pos(f);
1230        return ret;
1231}
1232
1233COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1234                const struct compat_iovec __user *, vec,
1235                compat_ulong_t, vlen)
1236{
1237        return do_compat_writev(fd, vec, vlen, 0);
1238}
1239
1240static long do_compat_pwritev64(unsigned long fd,
1241                                   const struct compat_iovec __user *vec,
1242                                   unsigned long vlen, loff_t pos, int flags)
1243{
1244        struct fd f;
1245        ssize_t ret;
1246
1247        if (pos < 0)
1248                return -EINVAL;
1249        f = fdget(fd);
1250        if (!f.file)
1251                return -EBADF;
1252        ret = -ESPIPE;
1253        if (f.file->f_mode & FMODE_PWRITE)
1254                ret = compat_writev(f.file, vec, vlen, &pos, flags);
1255        fdput(f);
1256        return ret;
1257}
1258
1259#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1260COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1261                const struct compat_iovec __user *,vec,
1262                unsigned long, vlen, loff_t, pos)
1263{
1264        return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1265}
1266#endif
1267
1268COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1269                const struct compat_iovec __user *,vec,
1270                compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1271{
1272        loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1273
1274        return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1275}
1276
1277#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1278COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1279                const struct compat_iovec __user *,vec,
1280                unsigned long, vlen, loff_t, pos, int, flags)
1281{
1282        return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1283}
1284#endif
1285
1286COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1287                const struct compat_iovec __user *,vec,
1288                compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1289{
1290        loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1291
1292        if (pos == -1)
1293                return do_compat_writev(fd, vec, vlen, flags);
1294
1295        return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1296}
1297
1298#endif
1299
1300static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1301                           size_t count, loff_t max)
1302{
1303        struct fd in, out;
1304        struct inode *in_inode, *out_inode;
1305        loff_t pos;
1306        loff_t out_pos;
1307        ssize_t retval;
1308        int fl;
1309
1310        /*
1311         * Get input file, and verify that it is ok..
1312         */
1313        retval = -EBADF;
1314        in = fdget(in_fd);
1315        if (!in.file)
1316                goto out;
1317        if (!(in.file->f_mode & FMODE_READ))
1318                goto fput_in;
1319        retval = -ESPIPE;
1320        if (!ppos) {
1321                pos = in.file->f_pos;
1322        } else {
1323                pos = *ppos;
1324                if (!(in.file->f_mode & FMODE_PREAD))
1325                        goto fput_in;
1326        }
1327        retval = rw_verify_area(READ, in.file, &pos, count);
1328        if (retval < 0)
1329                goto fput_in;
1330        if (count > MAX_RW_COUNT)
1331                count =  MAX_RW_COUNT;
1332
1333        /*
1334         * Get output file, and verify that it is ok..
1335         */
1336        retval = -EBADF;
1337        out = fdget(out_fd);
1338        if (!out.file)
1339                goto fput_in;
1340        if (!(out.file->f_mode & FMODE_WRITE))
1341                goto fput_out;
1342        retval = -EINVAL;
1343        in_inode = file_inode(in.file);
1344        out_inode = file_inode(out.file);
1345        out_pos = out.file->f_pos;
1346        retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1347        if (retval < 0)
1348                goto fput_out;
1349
1350        if (!max)
1351                max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1352
1353        if (unlikely(pos + count > max)) {
1354                retval = -EOVERFLOW;
1355                if (pos >= max)
1356                        goto fput_out;
1357                count = max - pos;
1358        }
1359
1360        fl = 0;
1361#if 0
1362        /*
1363         * We need to debate whether we can enable this or not. The
1364         * man page documents EAGAIN return for the output at least,
1365         * and the application is arguably buggy if it doesn't expect
1366         * EAGAIN on a non-blocking file descriptor.
1367         */
1368        if (in.file->f_flags & O_NONBLOCK)
1369                fl = SPLICE_F_NONBLOCK;
1370#endif
1371        file_start_write(out.file);
1372        retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1373        file_end_write(out.file);
1374
1375        if (retval > 0) {
1376                add_rchar(current, retval);
1377                add_wchar(current, retval);
1378                fsnotify_access(in.file);
1379                fsnotify_modify(out.file);
1380                out.file->f_pos = out_pos;
1381                if (ppos)
1382                        *ppos = pos;
1383                else
1384                        in.file->f_pos = pos;
1385        }
1386
1387        inc_syscr(current);
1388        inc_syscw(current);
1389        if (pos > max)
1390                retval = -EOVERFLOW;
1391
1392fput_out:
1393        fdput(out);
1394fput_in:
1395        fdput(in);
1396out:
1397        return retval;
1398}
1399
1400SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1401{
1402        loff_t pos;
1403        off_t off;
1404        ssize_t ret;
1405
1406        if (offset) {
1407                if (unlikely(get_user(off, offset)))
1408                        return -EFAULT;
1409                pos = off;
1410                ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1411                if (unlikely(put_user(pos, offset)))
1412                        return -EFAULT;
1413                return ret;
1414        }
1415
1416        return do_sendfile(out_fd, in_fd, NULL, count, 0);
1417}
1418
1419SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1420{
1421        loff_t pos;
1422        ssize_t ret;
1423
1424        if (offset) {
1425                if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1426                        return -EFAULT;
1427                ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1428                if (unlikely(put_user(pos, offset)))
1429                        return -EFAULT;
1430                return ret;
1431        }
1432
1433        return do_sendfile(out_fd, in_fd, NULL, count, 0);
1434}
1435
1436#ifdef CONFIG_COMPAT
1437COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1438                compat_off_t __user *, offset, compat_size_t, count)
1439{
1440        loff_t pos;
1441        off_t off;
1442        ssize_t ret;
1443
1444        if (offset) {
1445                if (unlikely(get_user(off, offset)))
1446                        return -EFAULT;
1447                pos = off;
1448                ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1449                if (unlikely(put_user(pos, offset)))
1450                        return -EFAULT;
1451                return ret;
1452        }
1453
1454        return do_sendfile(out_fd, in_fd, NULL, count, 0);
1455}
1456
1457COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1458                compat_loff_t __user *, offset, compat_size_t, count)
1459{
1460        loff_t pos;
1461        ssize_t ret;
1462
1463        if (offset) {
1464                if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1465                        return -EFAULT;
1466                ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1467                if (unlikely(put_user(pos, offset)))
1468                        return -EFAULT;
1469                return ret;
1470        }
1471
1472        return do_sendfile(out_fd, in_fd, NULL, count, 0);
1473}
1474#endif
1475
1476/*
1477 * copy_file_range() differs from regular file read and write in that it
1478 * specifically allows return partial success.  When it does so is up to
1479 * the copy_file_range method.
1480 */
1481ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1482                            struct file *file_out, loff_t pos_out,
1483                            size_t len, unsigned int flags)
1484{
1485        struct inode *inode_in = file_inode(file_in);
1486        struct inode *inode_out = file_inode(file_out);
1487        ssize_t ret;
1488
1489        if (flags != 0)
1490                return -EINVAL;
1491
1492        ret = rw_verify_area(READ, file_in, &pos_in, len);
1493        if (unlikely(ret))
1494                return ret;
1495
1496        ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1497        if (unlikely(ret))
1498                return ret;
1499
1500        if (!(file_in->f_mode & FMODE_READ) ||
1501            !(file_out->f_mode & FMODE_WRITE) ||
1502            (file_out->f_flags & O_APPEND))
1503                return -EBADF;
1504
1505        /* this could be relaxed once a method supports cross-fs copies */
1506        if (inode_in->i_sb != inode_out->i_sb)
1507                return -EXDEV;
1508
1509        if (len == 0)
1510                return 0;
1511
1512        ret = mnt_want_write_file(file_out);
1513        if (ret)
1514                return ret;
1515
1516        ret = -EOPNOTSUPP;
1517        if (file_out->f_op->copy_file_range)
1518                ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1519                                                      pos_out, len, flags);
1520        if (ret == -EOPNOTSUPP)
1521                ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1522                                len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1523
1524        if (ret > 0) {
1525                fsnotify_access(file_in);
1526                add_rchar(current, ret);
1527                fsnotify_modify(file_out);
1528                add_wchar(current, ret);
1529        }
1530        inc_syscr(current);
1531        inc_syscw(current);
1532
1533        mnt_drop_write_file(file_out);
1534
1535        return ret;
1536}
1537EXPORT_SYMBOL(vfs_copy_file_range);
1538
1539SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1540                int, fd_out, loff_t __user *, off_out,
1541                size_t, len, unsigned int, flags)
1542{
1543        loff_t pos_in;
1544        loff_t pos_out;
1545        struct fd f_in;
1546        struct fd f_out;
1547        ssize_t ret = -EBADF;
1548
1549        f_in = fdget(fd_in);
1550        if (!f_in.file)
1551                goto out2;
1552
1553        f_out = fdget(fd_out);
1554        if (!f_out.file)
1555                goto out1;
1556
1557        ret = -EFAULT;
1558        if (off_in) {
1559                if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1560                        goto out;
1561        } else {
1562                pos_in = f_in.file->f_pos;
1563        }
1564
1565        if (off_out) {
1566                if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1567                        goto out;
1568        } else {
1569                pos_out = f_out.file->f_pos;
1570        }
1571
1572        ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1573                                  flags);
1574        if (ret > 0) {
1575                pos_in += ret;
1576                pos_out += ret;
1577
1578                if (off_in) {
1579                        if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1580                                ret = -EFAULT;
1581                } else {
1582                        f_in.file->f_pos = pos_in;
1583                }
1584
1585                if (off_out) {
1586                        if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1587                                ret = -EFAULT;
1588                } else {
1589                        f_out.file->f_pos = pos_out;
1590                }
1591        }
1592
1593out:
1594        fdput(f_out);
1595out1:
1596        fdput(f_in);
1597out2:
1598        return ret;
1599}
1600
1601static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1602{
1603        struct inode *inode = file_inode(file);
1604
1605        if (unlikely(pos < 0))
1606                return -EINVAL;
1607
1608         if (unlikely((loff_t) (pos + len) < 0))
1609                return -EINVAL;
1610
1611        if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1612                loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1613                int retval;
1614
1615                retval = locks_mandatory_area(inode, file, pos, end,
1616                                write ? F_WRLCK : F_RDLCK);
1617                if (retval < 0)
1618                        return retval;
1619        }
1620
1621        return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1622}
1623
1624int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1625                struct file *file_out, loff_t pos_out, u64 len)
1626{
1627        struct inode *inode_in = file_inode(file_in);
1628        struct inode *inode_out = file_inode(file_out);
1629        int ret;
1630
1631        if (inode_in->i_sb != inode_out->i_sb ||
1632            file_in->f_path.mnt != file_out->f_path.mnt)
1633                return -EXDEV;
1634
1635        if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1636                return -EISDIR;
1637        if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1638                return -EINVAL;
1639
1640        if (!(file_in->f_mode & FMODE_READ) ||
1641            !(file_out->f_mode & FMODE_WRITE) ||
1642            (file_out->f_flags & O_APPEND))
1643                return -EBADF;
1644
1645        if (!file_in->f_op->clone_file_range)
1646                return -EOPNOTSUPP;
1647
1648        ret = clone_verify_area(file_in, pos_in, len, false);
1649        if (ret)
1650                return ret;
1651
1652        ret = clone_verify_area(file_out, pos_out, len, true);
1653        if (ret)
1654                return ret;
1655
1656        if (pos_in + len > i_size_read(inode_in))
1657                return -EINVAL;
1658
1659        ret = mnt_want_write_file(file_out);
1660        if (ret)
1661                return ret;
1662
1663        ret = file_in->f_op->clone_file_range(file_in, pos_in,
1664                        file_out, pos_out, len);
1665        if (!ret) {
1666                fsnotify_access(file_in);
1667                fsnotify_modify(file_out);
1668        }
1669
1670        mnt_drop_write_file(file_out);
1671        return ret;
1672}
1673EXPORT_SYMBOL(vfs_clone_file_range);
1674
1675int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1676{
1677        struct file_dedupe_range_info *info;
1678        struct inode *src = file_inode(file);
1679        u64 off;
1680        u64 len;
1681        int i;
1682        int ret;
1683        bool is_admin = capable(CAP_SYS_ADMIN);
1684        u16 count = same->dest_count;
1685        struct file *dst_file;
1686        loff_t dst_off;
1687        ssize_t deduped;
1688
1689        if (!(file->f_mode & FMODE_READ))
1690                return -EINVAL;
1691
1692        if (same->reserved1 || same->reserved2)
1693                return -EINVAL;
1694
1695        off = same->src_offset;
1696        len = same->src_length;
1697
1698        ret = -EISDIR;
1699        if (S_ISDIR(src->i_mode))
1700                goto out;
1701
1702        ret = -EINVAL;
1703        if (!S_ISREG(src->i_mode))
1704                goto out;
1705
1706        ret = clone_verify_area(file, off, len, false);
1707        if (ret < 0)
1708                goto out;
1709        ret = 0;
1710
1711        /* pre-format output fields to sane values */
1712        for (i = 0; i < count; i++) {
1713                same->info[i].bytes_deduped = 0ULL;
1714                same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1715        }
1716
1717        for (i = 0, info = same->info; i < count; i++, info++) {
1718                struct inode *dst;
1719                struct fd dst_fd = fdget(info->dest_fd);
1720
1721                dst_file = dst_fd.file;
1722                if (!dst_file) {
1723                        info->status = -EBADF;
1724                        goto next_loop;
1725                }
1726                dst = file_inode(dst_file);
1727
1728                ret = mnt_want_write_file(dst_file);
1729                if (ret) {
1730                        info->status = ret;
1731                        goto next_loop;
1732                }
1733
1734                dst_off = info->dest_offset;
1735                ret = clone_verify_area(dst_file, dst_off, len, true);
1736                if (ret < 0) {
1737                        info->status = ret;
1738                        goto next_file;
1739                }
1740                ret = 0;
1741
1742                if (info->reserved) {
1743                        info->status = -EINVAL;
1744                } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1745                        info->status = -EINVAL;
1746                } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1747                        info->status = -EXDEV;
1748                } else if (S_ISDIR(dst->i_mode)) {
1749                        info->status = -EISDIR;
1750                } else if (dst_file->f_op->dedupe_file_range == NULL) {
1751                        info->status = -EINVAL;
1752                } else {
1753                        deduped = dst_file->f_op->dedupe_file_range(file, off,
1754                                                        len, dst_file,
1755                                                        info->dest_offset);
1756                        if (deduped == -EBADE)
1757                                info->status = FILE_DEDUPE_RANGE_DIFFERS;
1758                        else if (deduped < 0)
1759                                info->status = deduped;
1760                        else
1761                                info->bytes_deduped += deduped;
1762                }
1763
1764next_file:
1765                mnt_drop_write_file(dst_file);
1766next_loop:
1767                fdput(dst_fd);
1768
1769                if (fatal_signal_pending(current))
1770                        goto out;
1771        }
1772
1773out:
1774        return ret;
1775}
1776EXPORT_SYMBOL(vfs_dedupe_file_range);
1777