linux/fs/xfs/xfs_file.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_log.h"
  21#include "xfs_sb.h"
  22#include "xfs_ag.h"
  23#include "xfs_trans.h"
  24#include "xfs_mount.h"
  25#include "xfs_bmap_btree.h"
  26#include "xfs_alloc.h"
  27#include "xfs_dinode.h"
  28#include "xfs_inode.h"
  29#include "xfs_inode_item.h"
  30#include "xfs_bmap.h"
  31#include "xfs_error.h"
  32#include "xfs_vnodeops.h"
  33#include "xfs_da_btree.h"
  34#include "xfs_ioctl.h"
  35#include "xfs_trace.h"
  36
  37#include <linux/dcache.h>
  38#include <linux/falloc.h>
  39
  40static const struct vm_operations_struct xfs_file_vm_ops;
  41
  42/*
  43 * Locking primitives for read and write IO paths to ensure we consistently use
  44 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
  45 */
  46static inline void
  47xfs_rw_ilock(
  48        struct xfs_inode        *ip,
  49        int                     type)
  50{
  51        if (type & XFS_IOLOCK_EXCL)
  52                mutex_lock(&VFS_I(ip)->i_mutex);
  53        xfs_ilock(ip, type);
  54}
  55
  56static inline void
  57xfs_rw_iunlock(
  58        struct xfs_inode        *ip,
  59        int                     type)
  60{
  61        xfs_iunlock(ip, type);
  62        if (type & XFS_IOLOCK_EXCL)
  63                mutex_unlock(&VFS_I(ip)->i_mutex);
  64}
  65
  66static inline void
  67xfs_rw_ilock_demote(
  68        struct xfs_inode        *ip,
  69        int                     type)
  70{
  71        xfs_ilock_demote(ip, type);
  72        if (type & XFS_IOLOCK_EXCL)
  73                mutex_unlock(&VFS_I(ip)->i_mutex);
  74}
  75
  76/*
  77 *      xfs_iozero
  78 *
  79 *      xfs_iozero clears the specified range of buffer supplied,
  80 *      and marks all the affected blocks as valid and modified.  If
  81 *      an affected block is not allocated, it will be allocated.  If
  82 *      an affected block is not completely overwritten, and is not
  83 *      valid before the operation, it will be read from disk before
  84 *      being partially zeroed.
  85 */
  86STATIC int
  87xfs_iozero(
  88        struct xfs_inode        *ip,    /* inode                        */
  89        loff_t                  pos,    /* offset in file               */
  90        size_t                  count)  /* size of data to zero         */
  91{
  92        struct page             *page;
  93        struct address_space    *mapping;
  94        int                     status;
  95
  96        mapping = VFS_I(ip)->i_mapping;
  97        do {
  98                unsigned offset, bytes;
  99                void *fsdata;
 100
 101                offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 102                bytes = PAGE_CACHE_SIZE - offset;
 103                if (bytes > count)
 104                        bytes = count;
 105
 106                status = pagecache_write_begin(NULL, mapping, pos, bytes,
 107                                        AOP_FLAG_UNINTERRUPTIBLE,
 108                                        &page, &fsdata);
 109                if (status)
 110                        break;
 111
 112                zero_user(page, offset, bytes);
 113
 114                status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
 115                                        page, fsdata);
 116                WARN_ON(status <= 0); /* can't return less than zero! */
 117                pos += bytes;
 118                count -= bytes;
 119                status = 0;
 120        } while (count);
 121
 122        return (-status);
 123}
 124
 125/*
 126 * Fsync operations on directories are much simpler than on regular files,
 127 * as there is no file data to flush, and thus also no need for explicit
 128 * cache flush operations, and there are no non-transaction metadata updates
 129 * on directories either.
 130 */
 131STATIC int
 132xfs_dir_fsync(
 133        struct file             *file,
 134        loff_t                  start,
 135        loff_t                  end,
 136        int                     datasync)
 137{
 138        struct xfs_inode        *ip = XFS_I(file->f_mapping->host);
 139        struct xfs_mount        *mp = ip->i_mount;
 140        xfs_lsn_t               lsn = 0;
 141
 142        trace_xfs_dir_fsync(ip);
 143
 144        xfs_ilock(ip, XFS_ILOCK_SHARED);
 145        if (xfs_ipincount(ip))
 146                lsn = ip->i_itemp->ili_last_lsn;
 147        xfs_iunlock(ip, XFS_ILOCK_SHARED);
 148
 149        if (!lsn)
 150                return 0;
 151        return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
 152}
 153
 154STATIC int
 155xfs_file_fsync(
 156        struct file             *file,
 157        loff_t                  start,
 158        loff_t                  end,
 159        int                     datasync)
 160{
 161        struct inode            *inode = file->f_mapping->host;
 162        struct xfs_inode        *ip = XFS_I(inode);
 163        struct xfs_mount        *mp = ip->i_mount;
 164        int                     error = 0;
 165        int                     log_flushed = 0;
 166        xfs_lsn_t               lsn = 0;
 167
 168        trace_xfs_file_fsync(ip);
 169
 170        error = filemap_write_and_wait_range(inode->i_mapping, start, end);
 171        if (error)
 172                return error;
 173
 174        if (XFS_FORCED_SHUTDOWN(mp))
 175                return -XFS_ERROR(EIO);
 176
 177        xfs_iflags_clear(ip, XFS_ITRUNCATED);
 178
 179        if (mp->m_flags & XFS_MOUNT_BARRIER) {
 180                /*
 181                 * If we have an RT and/or log subvolume we need to make sure
 182                 * to flush the write cache the device used for file data
 183                 * first.  This is to ensure newly written file data make
 184                 * it to disk before logging the new inode size in case of
 185                 * an extending write.
 186                 */
 187                if (XFS_IS_REALTIME_INODE(ip))
 188                        xfs_blkdev_issue_flush(mp->m_rtdev_targp);
 189                else if (mp->m_logdev_targp != mp->m_ddev_targp)
 190                        xfs_blkdev_issue_flush(mp->m_ddev_targp);
 191        }
 192
 193        /*
 194         * All metadata updates are logged, which means that we just have
 195         * to flush the log up to the latest LSN that touched the inode.
 196         */
 197        xfs_ilock(ip, XFS_ILOCK_SHARED);
 198        if (xfs_ipincount(ip)) {
 199                if (!datasync ||
 200                    (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
 201                        lsn = ip->i_itemp->ili_last_lsn;
 202        }
 203        xfs_iunlock(ip, XFS_ILOCK_SHARED);
 204
 205        if (lsn)
 206                error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
 207
 208        /*
 209         * If we only have a single device, and the log force about was
 210         * a no-op we might have to flush the data device cache here.
 211         * This can only happen for fdatasync/O_DSYNC if we were overwriting
 212         * an already allocated file and thus do not have any metadata to
 213         * commit.
 214         */
 215        if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
 216            mp->m_logdev_targp == mp->m_ddev_targp &&
 217            !XFS_IS_REALTIME_INODE(ip) &&
 218            !log_flushed)
 219                xfs_blkdev_issue_flush(mp->m_ddev_targp);
 220
 221        return -error;
 222}
 223
 224STATIC ssize_t
 225xfs_file_aio_read(
 226        struct kiocb            *iocb,
 227        const struct iovec      *iovp,
 228        unsigned long           nr_segs,
 229        loff_t                  pos)
 230{
 231        struct file             *file = iocb->ki_filp;
 232        struct inode            *inode = file->f_mapping->host;
 233        struct xfs_inode        *ip = XFS_I(inode);
 234        struct xfs_mount        *mp = ip->i_mount;
 235        size_t                  size = 0;
 236        ssize_t                 ret = 0;
 237        int                     ioflags = 0;
 238        xfs_fsize_t             n;
 239
 240        XFS_STATS_INC(xs_read_calls);
 241
 242        BUG_ON(iocb->ki_pos != pos);
 243
 244        if (unlikely(file->f_flags & O_DIRECT))
 245                ioflags |= IO_ISDIRECT;
 246        if (file->f_mode & FMODE_NOCMTIME)
 247                ioflags |= IO_INVIS;
 248
 249        ret = generic_segment_checks(iovp, &nr_segs, &size, VERIFY_WRITE);
 250        if (ret < 0)
 251                return ret;
 252
 253        if (unlikely(ioflags & IO_ISDIRECT)) {
 254                xfs_buftarg_t   *target =
 255                        XFS_IS_REALTIME_INODE(ip) ?
 256                                mp->m_rtdev_targp : mp->m_ddev_targp;
 257                if ((iocb->ki_pos & target->bt_smask) ||
 258                    (size & target->bt_smask)) {
 259                        if (iocb->ki_pos == i_size_read(inode))
 260                                return 0;
 261                        return -XFS_ERROR(EINVAL);
 262                }
 263        }
 264
 265        n = mp->m_super->s_maxbytes - iocb->ki_pos;
 266        if (n <= 0 || size == 0)
 267                return 0;
 268
 269        if (n < size)
 270                size = n;
 271
 272        if (XFS_FORCED_SHUTDOWN(mp))
 273                return -EIO;
 274
 275        /*
 276         * Locking is a bit tricky here. If we take an exclusive lock
 277         * for direct IO, we effectively serialise all new concurrent
 278         * read IO to this file and block it behind IO that is currently in
 279         * progress because IO in progress holds the IO lock shared. We only
 280         * need to hold the lock exclusive to blow away the page cache, so
 281         * only take lock exclusively if the page cache needs invalidation.
 282         * This allows the normal direct IO case of no page cache pages to
 283         * proceeed concurrently without serialisation.
 284         */
 285        xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
 286        if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
 287                xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
 288                xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
 289
 290                if (inode->i_mapping->nrpages) {
 291                        ret = -xfs_flushinval_pages(ip,
 292                                        (iocb->ki_pos & PAGE_CACHE_MASK),
 293                                        -1, FI_REMAPF_LOCKED);
 294                        if (ret) {
 295                                xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
 296                                return ret;
 297                        }
 298                }
 299                xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
 300        }
 301
 302        trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
 303
 304        ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
 305        if (ret > 0)
 306                XFS_STATS_ADD(xs_read_bytes, ret);
 307
 308        xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
 309        return ret;
 310}
 311
 312STATIC ssize_t
 313xfs_file_splice_read(
 314        struct file             *infilp,
 315        loff_t                  *ppos,
 316        struct pipe_inode_info  *pipe,
 317        size_t                  count,
 318        unsigned int            flags)
 319{
 320        struct xfs_inode        *ip = XFS_I(infilp->f_mapping->host);
 321        int                     ioflags = 0;
 322        ssize_t                 ret;
 323
 324        XFS_STATS_INC(xs_read_calls);
 325
 326        if (infilp->f_mode & FMODE_NOCMTIME)
 327                ioflags |= IO_INVIS;
 328
 329        if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 330                return -EIO;
 331
 332        xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
 333
 334        trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
 335
 336        ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
 337        if (ret > 0)
 338                XFS_STATS_ADD(xs_read_bytes, ret);
 339
 340        xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
 341        return ret;
 342}
 343
 344/*
 345 * xfs_file_splice_write() does not use xfs_rw_ilock() because
 346 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
 347 * couuld cause lock inversions between the aio_write path and the splice path
 348 * if someone is doing concurrent splice(2) based writes and write(2) based
 349 * writes to the same inode. The only real way to fix this is to re-implement
 350 * the generic code here with correct locking orders.
 351 */
 352STATIC ssize_t
 353xfs_file_splice_write(
 354        struct pipe_inode_info  *pipe,
 355        struct file             *outfilp,
 356        loff_t                  *ppos,
 357        size_t                  count,
 358        unsigned int            flags)
 359{
 360        struct inode            *inode = outfilp->f_mapping->host;
 361        struct xfs_inode        *ip = XFS_I(inode);
 362        int                     ioflags = 0;
 363        ssize_t                 ret;
 364
 365        XFS_STATS_INC(xs_write_calls);
 366
 367        if (outfilp->f_mode & FMODE_NOCMTIME)
 368                ioflags |= IO_INVIS;
 369
 370        if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 371                return -EIO;
 372
 373        xfs_ilock(ip, XFS_IOLOCK_EXCL);
 374
 375        trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
 376
 377        ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
 378        if (ret > 0)
 379                XFS_STATS_ADD(xs_write_bytes, ret);
 380
 381        xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 382        return ret;
 383}
 384
 385/*
 386 * This routine is called to handle zeroing any space in the last block of the
 387 * file that is beyond the EOF.  We do this since the size is being increased
 388 * without writing anything to that block and we don't want to read the
 389 * garbage on the disk.
 390 */
 391STATIC int                              /* error (positive) */
 392xfs_zero_last_block(
 393        struct xfs_inode        *ip,
 394        xfs_fsize_t             offset,
 395        xfs_fsize_t             isize)
 396{
 397        struct xfs_mount        *mp = ip->i_mount;
 398        xfs_fileoff_t           last_fsb = XFS_B_TO_FSBT(mp, isize);
 399        int                     zero_offset = XFS_B_FSB_OFFSET(mp, isize);
 400        int                     zero_len;
 401        int                     nimaps = 1;
 402        int                     error = 0;
 403        struct xfs_bmbt_irec    imap;
 404
 405        xfs_ilock(ip, XFS_ILOCK_EXCL);
 406        error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
 407        xfs_iunlock(ip, XFS_ILOCK_EXCL);
 408        if (error)
 409                return error;
 410
 411        ASSERT(nimaps > 0);
 412
 413        /*
 414         * If the block underlying isize is just a hole, then there
 415         * is nothing to zero.
 416         */
 417        if (imap.br_startblock == HOLESTARTBLOCK)
 418                return 0;
 419
 420        zero_len = mp->m_sb.sb_blocksize - zero_offset;
 421        if (isize + zero_len > offset)
 422                zero_len = offset - isize;
 423        return xfs_iozero(ip, isize, zero_len);
 424}
 425
 426/*
 427 * Zero any on disk space between the current EOF and the new, larger EOF.
 428 *
 429 * This handles the normal case of zeroing the remainder of the last block in
 430 * the file and the unusual case of zeroing blocks out beyond the size of the
 431 * file.  This second case only happens with fixed size extents and when the
 432 * system crashes before the inode size was updated but after blocks were
 433 * allocated.
 434 *
 435 * Expects the iolock to be held exclusive, and will take the ilock internally.
 436 */
 437int                                     /* error (positive) */
 438xfs_zero_eof(
 439        struct xfs_inode        *ip,
 440        xfs_off_t               offset,         /* starting I/O offset */
 441        xfs_fsize_t             isize)          /* current inode size */
 442{
 443        struct xfs_mount        *mp = ip->i_mount;
 444        xfs_fileoff_t           start_zero_fsb;
 445        xfs_fileoff_t           end_zero_fsb;
 446        xfs_fileoff_t           zero_count_fsb;
 447        xfs_fileoff_t           last_fsb;
 448        xfs_fileoff_t           zero_off;
 449        xfs_fsize_t             zero_len;
 450        int                     nimaps;
 451        int                     error = 0;
 452        struct xfs_bmbt_irec    imap;
 453
 454        ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
 455        ASSERT(offset > isize);
 456
 457        /*
 458         * First handle zeroing the block on which isize resides.
 459         *
 460         * We only zero a part of that block so it is handled specially.
 461         */
 462        if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
 463                error = xfs_zero_last_block(ip, offset, isize);
 464                if (error)
 465                        return error;
 466        }
 467
 468        /*
 469         * Calculate the range between the new size and the old where blocks
 470         * needing to be zeroed may exist.
 471         *
 472         * To get the block where the last byte in the file currently resides,
 473         * we need to subtract one from the size and truncate back to a block
 474         * boundary.  We subtract 1 in case the size is exactly on a block
 475         * boundary.
 476         */
 477        last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
 478        start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
 479        end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
 480        ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
 481        if (last_fsb == end_zero_fsb) {
 482                /*
 483                 * The size was only incremented on its last block.
 484                 * We took care of that above, so just return.
 485                 */
 486                return 0;
 487        }
 488
 489        ASSERT(start_zero_fsb <= end_zero_fsb);
 490        while (start_zero_fsb <= end_zero_fsb) {
 491                nimaps = 1;
 492                zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
 493
 494                xfs_ilock(ip, XFS_ILOCK_EXCL);
 495                error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
 496                                          &imap, &nimaps, 0);
 497                xfs_iunlock(ip, XFS_ILOCK_EXCL);
 498                if (error)
 499                        return error;
 500
 501                ASSERT(nimaps > 0);
 502
 503                if (imap.br_state == XFS_EXT_UNWRITTEN ||
 504                    imap.br_startblock == HOLESTARTBLOCK) {
 505                        start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 506                        ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 507                        continue;
 508                }
 509
 510                /*
 511                 * There are blocks we need to zero.
 512                 */
 513                zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
 514                zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
 515
 516                if ((zero_off + zero_len) > offset)
 517                        zero_len = offset - zero_off;
 518
 519                error = xfs_iozero(ip, zero_off, zero_len);
 520                if (error)
 521                        return error;
 522
 523                start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 524                ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 525        }
 526
 527        return 0;
 528}
 529
 530/*
 531 * Common pre-write limit and setup checks.
 532 *
 533 * Called with the iolocked held either shared and exclusive according to
 534 * @iolock, and returns with it held.  Might upgrade the iolock to exclusive
 535 * if called for a direct write beyond i_size.
 536 */
 537STATIC ssize_t
 538xfs_file_aio_write_checks(
 539        struct file             *file,
 540        loff_t                  *pos,
 541        size_t                  *count,
 542        int                     *iolock)
 543{
 544        struct inode            *inode = file->f_mapping->host;
 545        struct xfs_inode        *ip = XFS_I(inode);
 546        int                     error = 0;
 547
 548restart:
 549        error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
 550        if (error)
 551                return error;
 552
 553        /*
 554         * If the offset is beyond the size of the file, we need to zero any
 555         * blocks that fall between the existing EOF and the start of this
 556         * write.  If zeroing is needed and we are currently holding the
 557         * iolock shared, we need to update it to exclusive which implies
 558         * having to redo all checks before.
 559         */
 560        if (*pos > i_size_read(inode)) {
 561                if (*iolock == XFS_IOLOCK_SHARED) {
 562                        xfs_rw_iunlock(ip, *iolock);
 563                        *iolock = XFS_IOLOCK_EXCL;
 564                        xfs_rw_ilock(ip, *iolock);
 565                        goto restart;
 566                }
 567                error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
 568                if (error)
 569                        return error;
 570        }
 571
 572        /*
 573         * Updating the timestamps will grab the ilock again from
 574         * xfs_fs_dirty_inode, so we have to call it after dropping the
 575         * lock above.  Eventually we should look into a way to avoid
 576         * the pointless lock roundtrip.
 577         */
 578        if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
 579                error = file_update_time(file);
 580                if (error)
 581                        return error;
 582        }
 583
 584        /*
 585         * If we're writing the file then make sure to clear the setuid and
 586         * setgid bits if the process is not being run by root.  This keeps
 587         * people from modifying setuid and setgid binaries.
 588         */
 589        return file_remove_suid(file);
 590}
 591
 592/*
 593 * xfs_file_dio_aio_write - handle direct IO writes
 594 *
 595 * Lock the inode appropriately to prepare for and issue a direct IO write.
 596 * By separating it from the buffered write path we remove all the tricky to
 597 * follow locking changes and looping.
 598 *
 599 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
 600 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
 601 * pages are flushed out.
 602 *
 603 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
 604 * allowing them to be done in parallel with reads and other direct IO writes.
 605 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
 606 * needs to do sub-block zeroing and that requires serialisation against other
 607 * direct IOs to the same block. In this case we need to serialise the
 608 * submission of the unaligned IOs so that we don't get racing block zeroing in
 609 * the dio layer.  To avoid the problem with aio, we also need to wait for
 610 * outstanding IOs to complete so that unwritten extent conversion is completed
 611 * before we try to map the overlapping block. This is currently implemented by
 612 * hitting it with a big hammer (i.e. inode_dio_wait()).
 613 *
 614 * Returns with locks held indicated by @iolock and errors indicated by
 615 * negative return values.
 616 */
 617STATIC ssize_t
 618xfs_file_dio_aio_write(
 619        struct kiocb            *iocb,
 620        const struct iovec      *iovp,
 621        unsigned long           nr_segs,
 622        loff_t                  pos,
 623        size_t                  ocount)
 624{
 625        struct file             *file = iocb->ki_filp;
 626        struct address_space    *mapping = file->f_mapping;
 627        struct inode            *inode = mapping->host;
 628        struct xfs_inode        *ip = XFS_I(inode);
 629        struct xfs_mount        *mp = ip->i_mount;
 630        ssize_t                 ret = 0;
 631        size_t                  count = ocount;
 632        int                     unaligned_io = 0;
 633        int                     iolock;
 634        struct xfs_buftarg      *target = XFS_IS_REALTIME_INODE(ip) ?
 635                                        mp->m_rtdev_targp : mp->m_ddev_targp;
 636
 637        if ((pos & target->bt_smask) || (count & target->bt_smask))
 638                return -XFS_ERROR(EINVAL);
 639
 640        if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
 641                unaligned_io = 1;
 642
 643        /*
 644         * We don't need to take an exclusive lock unless there page cache needs
 645         * to be invalidated or unaligned IO is being executed. We don't need to
 646         * consider the EOF extension case here because
 647         * xfs_file_aio_write_checks() will relock the inode as necessary for
 648         * EOF zeroing cases and fill out the new inode size as appropriate.
 649         */
 650        if (unaligned_io || mapping->nrpages)
 651                iolock = XFS_IOLOCK_EXCL;
 652        else
 653                iolock = XFS_IOLOCK_SHARED;
 654        xfs_rw_ilock(ip, iolock);
 655
 656        /*
 657         * Recheck if there are cached pages that need invalidate after we got
 658         * the iolock to protect against other threads adding new pages while
 659         * we were waiting for the iolock.
 660         */
 661        if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
 662                xfs_rw_iunlock(ip, iolock);
 663                iolock = XFS_IOLOCK_EXCL;
 664                xfs_rw_ilock(ip, iolock);
 665        }
 666
 667        ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
 668        if (ret)
 669                goto out;
 670
 671        if (mapping->nrpages) {
 672                ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
 673                                                        FI_REMAPF_LOCKED);
 674                if (ret)
 675                        goto out;
 676        }
 677
 678        /*
 679         * If we are doing unaligned IO, wait for all other IO to drain,
 680         * otherwise demote the lock if we had to flush cached pages
 681         */
 682        if (unaligned_io)
 683                inode_dio_wait(inode);
 684        else if (iolock == XFS_IOLOCK_EXCL) {
 685                xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
 686                iolock = XFS_IOLOCK_SHARED;
 687        }
 688
 689        trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
 690        ret = generic_file_direct_write(iocb, iovp,
 691                        &nr_segs, pos, &iocb->ki_pos, count, ocount);
 692
 693out:
 694        xfs_rw_iunlock(ip, iolock);
 695
 696        /* No fallback to buffered IO on errors for XFS. */
 697        ASSERT(ret < 0 || ret == count);
 698        return ret;
 699}
 700
 701STATIC ssize_t
 702xfs_file_buffered_aio_write(
 703        struct kiocb            *iocb,
 704        const struct iovec      *iovp,
 705        unsigned long           nr_segs,
 706        loff_t                  pos,
 707        size_t                  ocount)
 708{
 709        struct file             *file = iocb->ki_filp;
 710        struct address_space    *mapping = file->f_mapping;
 711        struct inode            *inode = mapping->host;
 712        struct xfs_inode        *ip = XFS_I(inode);
 713        ssize_t                 ret;
 714        int                     enospc = 0;
 715        int                     iolock = XFS_IOLOCK_EXCL;
 716        size_t                  count = ocount;
 717
 718        xfs_rw_ilock(ip, iolock);
 719
 720        ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
 721        if (ret)
 722                goto out;
 723
 724        /* We can write back this queue in page reclaim */
 725        current->backing_dev_info = mapping->backing_dev_info;
 726
 727write_retry:
 728        trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
 729        ret = generic_file_buffered_write(iocb, iovp, nr_segs,
 730                        pos, &iocb->ki_pos, count, ret);
 731        /*
 732         * if we just got an ENOSPC, flush the inode now we aren't holding any
 733         * page locks and retry *once*
 734         */
 735        if (ret == -ENOSPC && !enospc) {
 736                enospc = 1;
 737                ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
 738                if (!ret)
 739                        goto write_retry;
 740        }
 741
 742        current->backing_dev_info = NULL;
 743out:
 744        xfs_rw_iunlock(ip, iolock);
 745        return ret;
 746}
 747
 748STATIC ssize_t
 749xfs_file_aio_write(
 750        struct kiocb            *iocb,
 751        const struct iovec      *iovp,
 752        unsigned long           nr_segs,
 753        loff_t                  pos)
 754{
 755        struct file             *file = iocb->ki_filp;
 756        struct address_space    *mapping = file->f_mapping;
 757        struct inode            *inode = mapping->host;
 758        struct xfs_inode        *ip = XFS_I(inode);
 759        ssize_t                 ret;
 760        size_t                  ocount = 0;
 761
 762        XFS_STATS_INC(xs_write_calls);
 763
 764        BUG_ON(iocb->ki_pos != pos);
 765
 766        ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
 767        if (ret)
 768                return ret;
 769
 770        if (ocount == 0)
 771                return 0;
 772
 773        sb_start_write(inode->i_sb);
 774
 775        if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 776                ret = -EIO;
 777                goto out;
 778        }
 779
 780        if (unlikely(file->f_flags & O_DIRECT))
 781                ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
 782        else
 783                ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
 784                                                  ocount);
 785
 786        if (ret > 0) {
 787                ssize_t err;
 788
 789                XFS_STATS_ADD(xs_write_bytes, ret);
 790
 791                /* Handle various SYNC-type writes */
 792                err = generic_write_sync(file, pos, ret);
 793                if (err < 0)
 794                        ret = err;
 795        }
 796
 797out:
 798        sb_end_write(inode->i_sb);
 799        return ret;
 800}
 801
 802STATIC long
 803xfs_file_fallocate(
 804        struct file     *file,
 805        int             mode,
 806        loff_t          offset,
 807        loff_t          len)
 808{
 809        struct inode    *inode = file->f_path.dentry->d_inode;
 810        long            error;
 811        loff_t          new_size = 0;
 812        xfs_flock64_t   bf;
 813        xfs_inode_t     *ip = XFS_I(inode);
 814        int             cmd = XFS_IOC_RESVSP;
 815        int             attr_flags = XFS_ATTR_NOLOCK;
 816
 817        if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 818                return -EOPNOTSUPP;
 819
 820        bf.l_whence = 0;
 821        bf.l_start = offset;
 822        bf.l_len = len;
 823
 824        xfs_ilock(ip, XFS_IOLOCK_EXCL);
 825
 826        if (mode & FALLOC_FL_PUNCH_HOLE)
 827                cmd = XFS_IOC_UNRESVSP;
 828
 829        /* check the new inode size is valid before allocating */
 830        if (!(mode & FALLOC_FL_KEEP_SIZE) &&
 831            offset + len > i_size_read(inode)) {
 832                new_size = offset + len;
 833                error = inode_newsize_ok(inode, new_size);
 834                if (error)
 835                        goto out_unlock;
 836        }
 837
 838        if (file->f_flags & O_DSYNC)
 839                attr_flags |= XFS_ATTR_SYNC;
 840
 841        error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
 842        if (error)
 843                goto out_unlock;
 844
 845        /* Change file size if needed */
 846        if (new_size) {
 847                struct iattr iattr;
 848
 849                iattr.ia_valid = ATTR_SIZE;
 850                iattr.ia_size = new_size;
 851                error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
 852        }
 853
 854out_unlock:
 855        xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 856        return error;
 857}
 858
 859
 860STATIC int
 861xfs_file_open(
 862        struct inode    *inode,
 863        struct file     *file)
 864{
 865        if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
 866                return -EFBIG;
 867        if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
 868                return -EIO;
 869        return 0;
 870}
 871
 872STATIC int
 873xfs_dir_open(
 874        struct inode    *inode,
 875        struct file     *file)
 876{
 877        struct xfs_inode *ip = XFS_I(inode);
 878        int             mode;
 879        int             error;
 880
 881        error = xfs_file_open(inode, file);
 882        if (error)
 883                return error;
 884
 885        /*
 886         * If there are any blocks, read-ahead block 0 as we're almost
 887         * certain to have the next operation be a read there.
 888         */
 889        mode = xfs_ilock_map_shared(ip);
 890        if (ip->i_d.di_nextents > 0)
 891                xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
 892        xfs_iunlock(ip, mode);
 893        return 0;
 894}
 895
 896STATIC int
 897xfs_file_release(
 898        struct inode    *inode,
 899        struct file     *filp)
 900{
 901        return -xfs_release(XFS_I(inode));
 902}
 903
 904STATIC int
 905xfs_file_readdir(
 906        struct file     *filp,
 907        void            *dirent,
 908        filldir_t       filldir)
 909{
 910        struct inode    *inode = filp->f_path.dentry->d_inode;
 911        xfs_inode_t     *ip = XFS_I(inode);
 912        int             error;
 913        size_t          bufsize;
 914
 915        /*
 916         * The Linux API doesn't pass down the total size of the buffer
 917         * we read into down to the filesystem.  With the filldir concept
 918         * it's not needed for correct information, but the XFS dir2 leaf
 919         * code wants an estimate of the buffer size to calculate it's
 920         * readahead window and size the buffers used for mapping to
 921         * physical blocks.
 922         *
 923         * Try to give it an estimate that's good enough, maybe at some
 924         * point we can change the ->readdir prototype to include the
 925         * buffer size.  For now we use the current glibc buffer size.
 926         */
 927        bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
 928
 929        error = xfs_readdir(ip, dirent, bufsize,
 930                                (xfs_off_t *)&filp->f_pos, filldir);
 931        if (error)
 932                return -error;
 933        return 0;
 934}
 935
 936STATIC int
 937xfs_file_mmap(
 938        struct file     *filp,
 939        struct vm_area_struct *vma)
 940{
 941        vma->vm_ops = &xfs_file_vm_ops;
 942        vma->vm_flags |= VM_CAN_NONLINEAR;
 943
 944        file_accessed(filp);
 945        return 0;
 946}
 947
 948/*
 949 * mmap()d file has taken write protection fault and is being made
 950 * writable. We can set the page state up correctly for a writable
 951 * page, which means we can do correct delalloc accounting (ENOSPC
 952 * checking!) and unwritten extent mapping.
 953 */
 954STATIC int
 955xfs_vm_page_mkwrite(
 956        struct vm_area_struct   *vma,
 957        struct vm_fault         *vmf)
 958{
 959        return block_page_mkwrite(vma, vmf, xfs_get_blocks);
 960}
 961
 962STATIC loff_t
 963xfs_seek_data(
 964        struct file             *file,
 965        loff_t                  start,
 966        u32                     type)
 967{
 968        struct inode            *inode = file->f_mapping->host;
 969        struct xfs_inode        *ip = XFS_I(inode);
 970        struct xfs_mount        *mp = ip->i_mount;
 971        struct xfs_bmbt_irec    map[2];
 972        int                     nmap = 2;
 973        loff_t                  uninitialized_var(offset);
 974        xfs_fsize_t             isize;
 975        xfs_fileoff_t           fsbno;
 976        xfs_filblks_t           end;
 977        uint                    lock;
 978        int                     error;
 979
 980        lock = xfs_ilock_map_shared(ip);
 981
 982        isize = i_size_read(inode);
 983        if (start >= isize) {
 984                error = ENXIO;
 985                goto out_unlock;
 986        }
 987
 988        fsbno = XFS_B_TO_FSBT(mp, start);
 989
 990        /*
 991         * Try to read extents from the first block indicated
 992         * by fsbno to the end block of the file.
 993         */
 994        end = XFS_B_TO_FSB(mp, isize);
 995
 996        error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
 997                               XFS_BMAPI_ENTIRE);
 998        if (error)
 999                goto out_unlock;
1000
1001        /*
1002         * Treat unwritten extent as data extent since it might
1003         * contains dirty data in page cache.
1004         */
1005        if (map[0].br_startblock != HOLESTARTBLOCK) {
1006                offset = max_t(loff_t, start,
1007                               XFS_FSB_TO_B(mp, map[0].br_startoff));
1008        } else {
1009                if (nmap == 1) {
1010                        error = ENXIO;
1011                        goto out_unlock;
1012                }
1013
1014                offset = max_t(loff_t, start,
1015                               XFS_FSB_TO_B(mp, map[1].br_startoff));
1016        }
1017
1018        if (offset != file->f_pos)
1019                file->f_pos = offset;
1020
1021out_unlock:
1022        xfs_iunlock_map_shared(ip, lock);
1023
1024        if (error)
1025                return -error;
1026        return offset;
1027}
1028
1029STATIC loff_t
1030xfs_seek_hole(
1031        struct file             *file,
1032        loff_t                  start,
1033        u32                     type)
1034{
1035        struct inode            *inode = file->f_mapping->host;
1036        struct xfs_inode        *ip = XFS_I(inode);
1037        struct xfs_mount        *mp = ip->i_mount;
1038        loff_t                  uninitialized_var(offset);
1039        loff_t                  holeoff;
1040        xfs_fsize_t             isize;
1041        xfs_fileoff_t           fsbno;
1042        uint                    lock;
1043        int                     error;
1044
1045        if (XFS_FORCED_SHUTDOWN(mp))
1046                return -XFS_ERROR(EIO);
1047
1048        lock = xfs_ilock_map_shared(ip);
1049
1050        isize = i_size_read(inode);
1051        if (start >= isize) {
1052                error = ENXIO;
1053                goto out_unlock;
1054        }
1055
1056        fsbno = XFS_B_TO_FSBT(mp, start);
1057        error = xfs_bmap_first_unused(NULL, ip, 1, &fsbno, XFS_DATA_FORK);
1058        if (error)
1059                goto out_unlock;
1060
1061        holeoff = XFS_FSB_TO_B(mp, fsbno);
1062        if (holeoff <= start)
1063                offset = start;
1064        else {
1065                /*
1066                 * xfs_bmap_first_unused() could return a value bigger than
1067                 * isize if there are no more holes past the supplied offset.
1068                 */
1069                offset = min_t(loff_t, holeoff, isize);
1070        }
1071
1072        if (offset != file->f_pos)
1073                file->f_pos = offset;
1074
1075out_unlock:
1076        xfs_iunlock_map_shared(ip, lock);
1077
1078        if (error)
1079                return -error;
1080        return offset;
1081}
1082
1083STATIC loff_t
1084xfs_file_llseek(
1085        struct file     *file,
1086        loff_t          offset,
1087        int             origin)
1088{
1089        switch (origin) {
1090        case SEEK_END:
1091        case SEEK_CUR:
1092        case SEEK_SET:
1093                return generic_file_llseek(file, offset, origin);
1094        case SEEK_DATA:
1095                return xfs_seek_data(file, offset, origin);
1096        case SEEK_HOLE:
1097                return xfs_seek_hole(file, offset, origin);
1098        default:
1099                return -EINVAL;
1100        }
1101}
1102
1103const struct file_operations xfs_file_operations = {
1104        .llseek         = xfs_file_llseek,
1105        .read           = do_sync_read,
1106        .write          = do_sync_write,
1107        .aio_read       = xfs_file_aio_read,
1108        .aio_write      = xfs_file_aio_write,
1109        .splice_read    = xfs_file_splice_read,
1110        .splice_write   = xfs_file_splice_write,
1111        .unlocked_ioctl = xfs_file_ioctl,
1112#ifdef CONFIG_COMPAT
1113        .compat_ioctl   = xfs_file_compat_ioctl,
1114#endif
1115        .mmap           = xfs_file_mmap,
1116        .open           = xfs_file_open,
1117        .release        = xfs_file_release,
1118        .fsync          = xfs_file_fsync,
1119        .fallocate      = xfs_file_fallocate,
1120};
1121
1122const struct file_operations xfs_dir_file_operations = {
1123        .open           = xfs_dir_open,
1124        .read           = generic_read_dir,
1125        .readdir        = xfs_file_readdir,
1126        .llseek         = generic_file_llseek,
1127        .unlocked_ioctl = xfs_file_ioctl,
1128#ifdef CONFIG_COMPAT
1129        .compat_ioctl   = xfs_file_compat_ioctl,
1130#endif
1131        .fsync          = xfs_dir_fsync,
1132};
1133
1134static const struct vm_operations_struct xfs_file_vm_ops = {
1135        .fault          = filemap_fault,
1136        .page_mkwrite   = xfs_vm_page_mkwrite,
1137};
1138
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.