fs/xfs/linux-2.6/xfs_lrw.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 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_bit.h"
  21 #include "xfs_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_alloc.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_quota.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_bmap.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_ialloc.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_itable.h"
  44 #include "xfs_rw.h"
  45 #include "xfs_attr.h"
  46 #include "xfs_inode_item.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_utils.h"
  49 #include "xfs_iomap.h"
  50 #include "xfs_vnodeops.h"
  51 #include "xfs_trace.h"
  52
  53 #include <linux/capability.h>
  54 #include <linux/writeback.h>
  55
  56
  57 /*
  58  *      xfs_iozero
  59  *
  60  *      xfs_iozero clears the specified range of buffer supplied,
  61  *      and marks all the affected blocks as valid and modified.  If
  62  *      an affected block is not allocated, it will be allocated.  If
  63  *      an affected block is not completely overwritten, and is not
  64  *      valid before the operation, it will be read from disk before
  65  *      being partially zeroed.
  66  */
  67 STATIC int
  68 xfs_iozero(
  69         struct xfs_inode        *ip,    /* inode                        */
  70         loff_t                  pos,    /* offset in file               */
  71         size_t                  count)  /* size of data to zero         */
  72 {
  73         struct page             *page;
  74         struct address_space    *mapping;
  75         int                     status;
  76
  77         mapping = VFS_I(ip)->i_mapping;
  78         do {
  79                 unsigned offset, bytes;
  80                 void *fsdata;
  81
  82                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
  83                 bytes = PAGE_CACHE_SIZE - offset;
  84                 if (bytes > count)
  85                         bytes = count;
  86
  87                 status = pagecache_write_begin(NULL, mapping, pos, bytes,
  88                                         AOP_FLAG_UNINTERRUPTIBLE,
  89                                         &page, &fsdata);
  90                 if (status)
  91                         break;
  92
  93                 zero_user(page, offset, bytes);
  94
  95                 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
  96                                         page, fsdata);
  97                 WARN_ON(status <= 0); /* can't return less than zero! */
  98                 pos += bytes;
  99                 count -= bytes;
 100                 status = 0;
 101         } while (count);
 102
 103         return (-status);
 104 }
 105
 106 ssize_t                 /* bytes read, or (-)  error */
 107 xfs_read(
 108         xfs_inode_t             *ip,
 109         struct kiocb            *iocb,
 110         const struct iovec      *iovp,
 111         unsigned int            segs,
 112         loff_t                  *offset,
 113         int                     ioflags)
 114 {
 115         struct file             *file = iocb->ki_filp;
 116         struct inode            *inode = file->f_mapping->host;
 117         xfs_mount_t             *mp = ip->i_mount;
 118         size_t                  size = 0;
 119         ssize_t                 ret = 0;
 120         xfs_fsize_t             n;
 121         unsigned long           seg;
 122
 123
 124         XFS_STATS_INC(xs_read_calls);
 125
 126         /* START copy & waste from filemap.c */
 127         for (seg = 0; seg < segs; seg++) {
 128                 const struct iovec *iv = &iovp[seg];
 129
 130                 /*
 131                  * If any segment has a negative length, or the cumulative
 132                  * length ever wraps negative then return -EINVAL.
 133                  */
 134                 size += iv->iov_len;
 135                 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
 136                         return XFS_ERROR(-EINVAL);
 137         }
 138         /* END copy & waste from filemap.c */
 139
 140         if (unlikely(ioflags & IO_ISDIRECT)) {
 141                 xfs_buftarg_t   *target =
 142                         XFS_IS_REALTIME_INODE(ip) ?
 143                                 mp->m_rtdev_targp : mp->m_ddev_targp;
 144                 if ((*offset & target->bt_smask) ||
 145                     (size & target->bt_smask)) {
 146                         if (*offset == ip->i_size) {
 147                                 return (0);
 148                         }
 149                         return -XFS_ERROR(EINVAL);
 150                 }
 151         }
 152
 153         n = XFS_MAXIOFFSET(mp) - *offset;
 154         if ((n <= 0) || (size == 0))
 155                 return 0;
 156
 157         if (n < size)
 158                 size = n;
 159
 160         if (XFS_FORCED_SHUTDOWN(mp))
 161                 return -EIO;
 162
 163         if (unlikely(ioflags & IO_ISDIRECT))
 164                 mutex_lock(&inode->i_mutex);
 165         xfs_ilock(ip, XFS_IOLOCK_SHARED);
 166
 167         if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) {
 168                 int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
 169                 int iolock = XFS_IOLOCK_SHARED;
 170
 171                 ret = -XFS_SEND_DATA(mp, DM_EVENT_READ, ip, *offset, size,
 172                                         dmflags, &iolock);
 173                 if (ret) {
 174                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 175                         if (unlikely(ioflags & IO_ISDIRECT))
 176                                 mutex_unlock(&inode->i_mutex);
 177                         return ret;
 178                 }
 179         }
 180
 181         if (unlikely(ioflags & IO_ISDIRECT)) {
 182                 if (inode->i_mapping->nrpages)
 183                         ret = -xfs_flushinval_pages(ip, (*offset & PAGE_CACHE_MASK),
 184                                                     -1, FI_REMAPF_LOCKED);
 185                 mutex_unlock(&inode->i_mutex);
 186                 if (ret) {
 187                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 188                         return ret;
 189                 }
 190         }
 191
 192         trace_xfs_file_read(ip, size, *offset, ioflags);
 193
 194         iocb->ki_pos = *offset;
 195         ret = generic_file_aio_read(iocb, iovp, segs, *offset);
 196         if (ret > 0)
 197                 XFS_STATS_ADD(xs_read_bytes, ret);
 198
 199         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 200         return ret;
 201 }
 202
 203 ssize_t
 204 xfs_splice_read(
 205         xfs_inode_t             *ip,
 206         struct file             *infilp,
 207         loff_t                  *ppos,
 208         struct pipe_inode_info  *pipe,
 209         size_t                  count,
 210         int                     flags,
 211         int                     ioflags)
 212 {
 213         xfs_mount_t             *mp = ip->i_mount;
 214         ssize_t                 ret;
 215
 216         XFS_STATS_INC(xs_read_calls);
 217         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 218                 return -EIO;
 219
 220         xfs_ilock(ip, XFS_IOLOCK_SHARED);
 221
 222         if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) {
 223                 int iolock = XFS_IOLOCK_SHARED;
 224                 int error;
 225
 226                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, ip, *ppos, count,
 227                                         FILP_DELAY_FLAG(infilp), &iolock);
 228                 if (error) {
 229                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 230                         return -error;
 231                 }
 232         }
 233
 234         trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
 235
 236         ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
 237         if (ret > 0)
 238                 XFS_STATS_ADD(xs_read_bytes, ret);
 239
 240         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 241         return ret;
 242 }
 243
 244 ssize_t
 245 xfs_splice_write(
 246         xfs_inode_t             *ip,
 247         struct pipe_inode_info  *pipe,
 248         struct file             *outfilp,
 249         loff_t                  *ppos,
 250         size_t                  count,
 251         int                     flags,
 252         int                     ioflags)
 253 {
 254         xfs_mount_t             *mp = ip->i_mount;
 255         ssize_t                 ret;
 256         struct inode            *inode = outfilp->f_mapping->host;
 257         xfs_fsize_t             isize, new_size;
 258
 259         XFS_STATS_INC(xs_write_calls);
 260         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 261                 return -EIO;
 262
 263         xfs_ilock(ip, XFS_IOLOCK_EXCL);
 264
 265         if (DM_EVENT_ENABLED(ip, DM_EVENT_WRITE) && !(ioflags & IO_INVIS)) {
 266                 int iolock = XFS_IOLOCK_EXCL;
 267                 int error;
 268
 269                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, *ppos, count,
 270                                         FILP_DELAY_FLAG(outfilp), &iolock);
 271                 if (error) {
 272                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 273                         return -error;
 274                 }
 275         }
 276
 277         new_size = *ppos + count;
 278
 279         xfs_ilock(ip, XFS_ILOCK_EXCL);
 280         if (new_size > ip->i_size)
 281                 ip->i_new_size = new_size;
 282         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 283
 284         trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
 285
 286         ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
 287         if (ret > 0)
 288                 XFS_STATS_ADD(xs_write_bytes, ret);
 289
 290         isize = i_size_read(inode);
 291         if (unlikely(ret < 0 && ret != -EFAULT && *ppos > isize))
 292                 *ppos = isize;
 293
 294         if (*ppos > ip->i_size) {
 295                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 296                 if (*ppos > ip->i_size)
 297                         ip->i_size = *ppos;
 298                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 299         }
 300
 301         if (ip->i_new_size) {
 302                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 303                 ip->i_new_size = 0;
 304                 if (ip->i_d.di_size > ip->i_size)
 305                         ip->i_d.di_size = ip->i_size;
 306                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 307         }
 308         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 309         return ret;
 310 }
 311
 312 /*
 313  * This routine is called to handle zeroing any space in the last
 314  * block of the file that is beyond the EOF.  We do this since the
 315  * size is being increased without writing anything to that block
 316  * and we don't want anyone to read the garbage on the disk.
 317  */
 318 STATIC int                              /* error (positive) */
 319 xfs_zero_last_block(
 320         xfs_inode_t     *ip,
 321         xfs_fsize_t     offset,
 322         xfs_fsize_t     isize)
 323 {
 324         xfs_fileoff_t   last_fsb;
 325         xfs_mount_t     *mp = ip->i_mount;
 326         int             nimaps;
 327         int             zero_offset;
 328         int             zero_len;
 329         int             error = 0;
 330         xfs_bmbt_irec_t imap;
 331
 332         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 333
 334         zero_offset = XFS_B_FSB_OFFSET(mp, isize);
 335         if (zero_offset == 0) {
 336                 /*
 337                  * There are no extra bytes in the last block on disk to
 338                  * zero, so return.
 339                  */
 340                 return 0;
 341         }
 342
 343         last_fsb = XFS_B_TO_FSBT(mp, isize);
 344         nimaps = 1;
 345         error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, NULL, 0, &imap,
 346                           &nimaps, NULL, NULL);
 347         if (error) {
 348                 return error;
 349         }
 350         ASSERT(nimaps > 0);
 351         /*
 352          * If the block underlying isize is just a hole, then there
 353          * is nothing to zero.
 354          */
 355         if (imap.br_startblock == HOLESTARTBLOCK) {
 356                 return 0;
 357         }
 358         /*
 359          * Zero the part of the last block beyond the EOF, and write it
 360          * out sync.  We need to drop the ilock while we do this so we
 361          * don't deadlock when the buffer cache calls back to us.
 362          */
 363         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 364
 365         zero_len = mp->m_sb.sb_blocksize - zero_offset;
 366         if (isize + zero_len > offset)
 367                 zero_len = offset - isize;
 368         error = xfs_iozero(ip, isize, zero_len);
 369
 370         xfs_ilock(ip, XFS_ILOCK_EXCL);
 371         ASSERT(error >= 0);
 372         return error;
 373 }
 374
 375 /*
 376  * Zero any on disk space between the current EOF and the new,
 377  * larger EOF.  This handles the normal case of zeroing the remainder
 378  * of the last block in the file and the unusual case of zeroing blocks
 379  * out beyond the size of the file.  This second case only happens
 380  * with fixed size extents and when the system crashes before the inode
 381  * size was updated but after blocks were allocated.  If fill is set,
 382  * then any holes in the range are filled and zeroed.  If not, the holes
 383  * are left alone as holes.
 384  */
 385
 386 int                                     /* error (positive) */
 387 xfs_zero_eof(
 388         xfs_inode_t     *ip,
 389         xfs_off_t       offset,         /* starting I/O offset */
 390         xfs_fsize_t     isize)          /* current inode size */
 391 {
 392         xfs_mount_t     *mp = ip->i_mount;
 393         xfs_fileoff_t   start_zero_fsb;
 394         xfs_fileoff_t   end_zero_fsb;
 395         xfs_fileoff_t   zero_count_fsb;
 396         xfs_fileoff_t   last_fsb;
 397         xfs_fileoff_t   zero_off;
 398         xfs_fsize_t     zero_len;
 399         int             nimaps;
 400         int             error = 0;
 401         xfs_bmbt_irec_t imap;
 402
 403         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
 404         ASSERT(offset > isize);
 405
 406         /*
 407          * First handle zeroing the block on which isize resides.
 408          * We only zero a part of that block so it is handled specially.
 409          */
 410         error = xfs_zero_last_block(ip, offset, isize);
 411         if (error) {
 412                 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
 413                 return error;
 414         }
 415
 416         /*
 417          * Calculate the range between the new size and the old
 418          * where blocks needing to be zeroed may exist.  To get the
 419          * block where the last byte in the file currently resides,
 420          * we need to subtract one from the size and truncate back
 421          * to a block boundary.  We subtract 1 in case the size is
 422          * exactly on a block boundary.
 423          */
 424         last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
 425         start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
 426         end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
 427         ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
 428         if (last_fsb == end_zero_fsb) {
 429                 /*
 430                  * The size was only incremented on its last block.
 431                  * We took care of that above, so just return.
 432                  */
 433                 return 0;
 434         }
 435
 436         ASSERT(start_zero_fsb <= end_zero_fsb);
 437         while (start_zero_fsb <= end_zero_fsb) {
 438                 nimaps = 1;
 439                 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
 440                 error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb,
 441                                   0, NULL, 0, &imap, &nimaps, NULL, NULL);
 442                 if (error) {
 443                         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
 444                         return error;
 445                 }
 446                 ASSERT(nimaps > 0);
 447
 448                 if (imap.br_state == XFS_EXT_UNWRITTEN ||
 449                     imap.br_startblock == HOLESTARTBLOCK) {
 450                         /*
 451                          * This loop handles initializing pages that were
 452                          * partially initialized by the code below this
 453                          * loop. It basically zeroes the part of the page
 454                          * that sits on a hole and sets the page as P_HOLE
 455                          * and calls remapf if it is a mapped file.
 456                          */
 457                         start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 458                         ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 459                         continue;
 460                 }
 461
 462                 /*
 463                  * There are blocks we need to zero.
 464                  * Drop the inode lock while we're doing the I/O.
 465                  * We'll still have the iolock to protect us.
 466                  */
 467                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 468
 469                 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
 470                 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
 471
 472                 if ((zero_off + zero_len) > offset)
 473                         zero_len = offset - zero_off;
 474
 475                 error = xfs_iozero(ip, zero_off, zero_len);
 476                 if (error) {
 477                         goto out_lock;
 478                 }
 479
 480                 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 481                 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 482
 483                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 484         }
 485
 486         return 0;
 487
 488 out_lock:
 489         xfs_ilock(ip, XFS_ILOCK_EXCL);
 490         ASSERT(error >= 0);
 491         return error;
 492 }
 493
 494 ssize_t                         /* bytes written, or (-) error */
 495 xfs_write(
 496         struct xfs_inode        *xip,
 497         struct kiocb            *iocb,
 498         const struct iovec      *iovp,
 499         unsigned int            nsegs,
 500         loff_t                  *offset,
 501         int                     ioflags)
 502 {
 503         struct file             *file = iocb->ki_filp;
 504         struct address_space    *mapping = file->f_mapping;
 505         struct inode            *inode = mapping->host;
 506         unsigned long           segs = nsegs;
 507         xfs_mount_t             *mp;
 508         ssize_t                 ret = 0, error = 0;
 509         xfs_fsize_t             isize, new_size;
 510         int                     iolock;
 511         int                     eventsent = 0;
 512         size_t                  ocount = 0, count;
 513         loff_t                  pos;
 514         int                     need_i_mutex;
 515
 516         XFS_STATS_INC(xs_write_calls);
 517
 518         error = generic_segment_checks(iovp, &segs, &ocount, VERIFY_READ);
 519         if (error)
 520                 return error;
 521
 522         count = ocount;
 523         pos = *offset;
 524
 525         if (count == 0)
 526                 return 0;
 527
 528         mp = xip->i_mount;
 529
 530         xfs_wait_for_freeze(mp, SB_FREEZE_WRITE);
 531
 532         if (XFS_FORCED_SHUTDOWN(mp))
 533                 return -EIO;
 534
 535 relock:
 536         if (ioflags & IO_ISDIRECT) {
 537                 iolock = XFS_IOLOCK_SHARED;
 538                 need_i_mutex = 0;
 539         } else {
 540                 iolock = XFS_IOLOCK_EXCL;
 541                 need_i_mutex = 1;
 542                 mutex_lock(&inode->i_mutex);
 543         }
 544
 545         xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
 546
 547 start:
 548         error = -generic_write_checks(file, &pos, &count,
 549                                         S_ISBLK(inode->i_mode));
 550         if (error) {
 551                 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 552                 goto out_unlock_mutex;
 553         }
 554
 555         if ((DM_EVENT_ENABLED(xip, DM_EVENT_WRITE) &&
 556             !(ioflags & IO_INVIS) && !eventsent)) {
 557                 int             dmflags = FILP_DELAY_FLAG(file);
 558
 559                 if (need_i_mutex)
 560                         dmflags |= DM_FLAGS_IMUX;
 561
 562                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 563                 error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, xip,
 564                                       pos, count, dmflags, &iolock);
 565                 if (error) {
 566                         goto out_unlock_internal;
 567                 }
 568                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 569                 eventsent = 1;
 570
 571                 /*
 572                  * The iolock was dropped and reacquired in XFS_SEND_DATA
 573                  * so we have to recheck the size when appending.
 574                  * We will only "goto start;" once, since having sent the
 575                  * event prevents another call to XFS_SEND_DATA, which is
 576                  * what allows the size to change in the first place.
 577                  */
 578                 if ((file->f_flags & O_APPEND) && pos != xip->i_size)
 579                         goto start;
 580         }
 581
 582         if (ioflags & IO_ISDIRECT) {
 583                 xfs_buftarg_t   *target =
 584                         XFS_IS_REALTIME_INODE(xip) ?
 585                                 mp->m_rtdev_targp : mp->m_ddev_targp;
 586
 587                 if ((pos & target->bt_smask) || (count & target->bt_smask)) {
 588                         xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 589                         return XFS_ERROR(-EINVAL);
 590                 }
 591
 592                 if (!need_i_mutex && (mapping->nrpages || pos > xip->i_size)) {
 593                         xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 594                         iolock = XFS_IOLOCK_EXCL;
 595                         need_i_mutex = 1;
 596                         mutex_lock(&inode->i_mutex);
 597                         xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
 598                         goto start;
 599                 }
 600         }
 601
 602         new_size = pos + count;
 603         if (new_size > xip->i_size)
 604                 xip->i_new_size = new_size;
 605
 606         if (likely(!(ioflags & IO_INVIS)))
 607                 file_update_time(file);
 608
 609         /*
 610          * If the offset is beyond the size of the file, we have a couple
 611          * of things to do. First, if there is already space allocated
 612          * we need to either create holes or zero the disk or ...
 613          *
 614          * If there is a page where the previous size lands, we need
 615          * to zero it out up to the new size.
 616          */
 617
 618         if (pos > xip->i_size) {
 619                 error = xfs_zero_eof(xip, pos, xip->i_size);
 620                 if (error) {
 621                         xfs_iunlock(xip, XFS_ILOCK_EXCL);
 622                         goto out_unlock_internal;
 623                 }
 624         }
 625         xfs_iunlock(xip, XFS_ILOCK_EXCL);
 626
 627         /*
 628          * If we're writing the file then make sure to clear the
 629          * setuid and setgid bits if the process is not being run
 630          * by root.  This keeps people from modifying setuid and
 631          * setgid binaries.
 632          */
 633
 634         if (((xip->i_d.di_mode & S_ISUID) ||
 635             ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
 636                 (S_ISGID | S_IXGRP))) &&
 637              !capable(CAP_FSETID)) {
 638                 error = xfs_write_clear_setuid(xip);
 639                 if (likely(!error))
 640                         error = -file_remove_suid(file);
 641                 if (unlikely(error)) {
 642                         goto out_unlock_internal;
 643                 }
 644         }
 645
 646         /* We can write back this queue in page reclaim */
 647         current->backing_dev_info = mapping->backing_dev_info;
 648
 649         if ((ioflags & IO_ISDIRECT)) {
 650                 if (mapping->nrpages) {
 651                         WARN_ON(need_i_mutex == 0);
 652                         error = xfs_flushinval_pages(xip,
 653                                         (pos & PAGE_CACHE_MASK),
 654                                         -1, FI_REMAPF_LOCKED);
 655                         if (error)
 656                                 goto out_unlock_internal;
 657                 }
 658
 659                 if (need_i_mutex) {
 660                         /* demote the lock now the cached pages are gone */
 661                         xfs_ilock_demote(xip, XFS_IOLOCK_EXCL);
 662                         mutex_unlock(&inode->i_mutex);
 663
 664                         iolock = XFS_IOLOCK_SHARED;
 665                         need_i_mutex = 0;
 666                 }
 667
 668                 trace_xfs_file_direct_write(xip, count, *offset, ioflags);
 669                 ret = generic_file_direct_write(iocb, iovp,
 670                                 &segs, pos, offset, count, ocount);
 671
 672                 /*
 673                  * direct-io write to a hole: fall through to buffered I/O
 674                  * for completing the rest of the request.
 675                  */
 676                 if (ret >= 0 && ret != count) {
 677                         XFS_STATS_ADD(xs_write_bytes, ret);
 678
 679                         pos += ret;
 680                         count -= ret;
 681
 682                         ioflags &= ~IO_ISDIRECT;
 683                         xfs_iunlock(xip, iolock);
 684                         goto relock;
 685                 }
 686         } else {
 687                 int enospc = 0;
 688                 ssize_t ret2 = 0;
 689
 690 write_retry:
 691                 trace_xfs_file_buffered_write(xip, count, *offset, ioflags);
 692                 ret2 = generic_file_buffered_write(iocb, iovp, segs,
 693                                 pos, offset, count, ret);
 694                 /*
 695                  * if we just got an ENOSPC, flush the inode now we
 696                  * aren't holding any page locks and retry *once*
 697                  */
 698                 if (ret2 == -ENOSPC && !enospc) {
 699                         error = xfs_flush_pages(xip, 0, -1, 0, FI_NONE);
 700                         if (error)
 701                                 goto out_unlock_internal;
 702                         enospc = 1;
 703                         goto write_retry;
 704                 }
 705                 ret = ret2;
 706         }
 707
 708         current->backing_dev_info = NULL;
 709
 710         isize = i_size_read(inode);
 711         if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize))
 712                 *offset = isize;
 713
 714         if (*offset > xip->i_size) {
 715                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 716                 if (*offset > xip->i_size)
 717                         xip->i_size = *offset;
 718                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 719         }
 720
 721         if (ret == -ENOSPC &&
 722             DM_EVENT_ENABLED(xip, DM_EVENT_NOSPACE) && !(ioflags & IO_INVIS)) {
 723                 xfs_iunlock(xip, iolock);
 724                 if (need_i_mutex)
 725                         mutex_unlock(&inode->i_mutex);
 726                 error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, xip,
 727                                 DM_RIGHT_NULL, xip, DM_RIGHT_NULL, NULL, NULL,
 728                                 0, 0, 0); /* Delay flag intentionally  unused */
 729                 if (need_i_mutex)
 730                         mutex_lock(&inode->i_mutex);
 731                 xfs_ilock(xip, iolock);
 732                 if (error)
 733                         goto out_unlock_internal;
 734                 goto start;
 735         }
 736
 737         error = -ret;
 738         if (ret <= 0)
 739                 goto out_unlock_internal;
 740
 741         XFS_STATS_ADD(xs_write_bytes, ret);
 742
 743         /* Handle various SYNC-type writes */
 744         if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
 745                 loff_t end = pos + ret - 1;
 746                 int error2;
 747
 748                 xfs_iunlock(xip, iolock);
 749                 if (need_i_mutex)
 750                         mutex_unlock(&inode->i_mutex);
 751
 752                 error2 = filemap_write_and_wait_range(mapping, pos, end);
 753                 if (!error)
 754                         error = error2;
 755                 if (need_i_mutex)
 756                         mutex_lock(&inode->i_mutex);
 757                 xfs_ilock(xip, iolock);
 758
 759                 error2 = xfs_fsync(xip);
 760                 if (!error)
 761                         error = error2;
 762         }
 763
 764  out_unlock_internal:
 765         if (xip->i_new_size) {
 766                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 767                 xip->i_new_size = 0;
 768                 /*
 769                  * If this was a direct or synchronous I/O that failed (such
 770                  * as ENOSPC) then part of the I/O may have been written to
 771                  * disk before the error occured.  In this case the on-disk
 772                  * file size may have been adjusted beyond the in-memory file
 773                  * size and now needs to be truncated back.
 774                  */
 775                 if (xip->i_d.di_size > xip->i_size)
 776                         xip->i_d.di_size = xip->i_size;
 777                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 778         }
 779         xfs_iunlock(xip, iolock);
 780  out_unlock_mutex:
 781         if (need_i_mutex)
 782                 mutex_unlock(&inode->i_mutex);
 783         return -error;
 784 }
 785
 786 /*
 787  * All xfs metadata buffers except log state machine buffers
 788  * get this attached as their b_bdstrat callback function.
 789  * This is so that we can catch a buffer
 790  * after prematurely unpinning it to forcibly shutdown the filesystem.
 791  */
 792 int
 793 xfs_bdstrat_cb(struct xfs_buf *bp)
 794 {
 795         if (XFS_FORCED_SHUTDOWN(bp->b_mount)) {
 796                 trace_xfs_bdstrat_shut(bp, _RET_IP_);
 797                 /*
 798                  * Metadata write that didn't get logged but
 799                  * written delayed anyway. These aren't associated
 800                  * with a transaction, and can be ignored.
 801                  */
 802                 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
 803                     (XFS_BUF_ISREAD(bp)) == 0)
 804                         return (xfs_bioerror_relse(bp));
 805                 else
 806                         return (xfs_bioerror(bp));
 807         }
 808
 809         xfs_buf_iorequest(bp);
 810         return 0;
 811 }
 812
 813 /*
 814  * Wrapper around bdstrat so that we can stop data from going to disk in case
 815  * we are shutting down the filesystem.  Typically user data goes thru this
 816  * path; one of the exceptions is the superblock.
 817  */
 818 void
 819 xfsbdstrat(
 820         struct xfs_mount        *mp,
 821         struct xfs_buf          *bp)
 822 {
 823         ASSERT(mp);
 824         if (!XFS_FORCED_SHUTDOWN(mp)) {
 825                 xfs_buf_iorequest(bp);
 826                 return;
 827         }
 828
 829         trace_xfs_bdstrat_shut(bp, _RET_IP_);
 830         xfs_bioerror_relse(bp);
 831 }
 832
 833 /*
 834  * If the underlying (data/log/rt) device is readonly, there are some
 835  * operations that cannot proceed.
 836  */
 837 int
 838 xfs_dev_is_read_only(
 839         xfs_mount_t             *mp,
 840         char                    *message)
 841 {
 842         if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
 843             xfs_readonly_buftarg(mp->m_logdev_targp) ||
 844             (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
 845                 cmn_err(CE_NOTE,
 846                         "XFS: %s required on read-only device.", message);
 847                 cmn_err(CE_NOTE,
 848                         "XFS: write access unavailable, cannot proceed.");
 849                 return EROFS;
 850         }
 851         return 0;
 852 }