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_shared.h" 21 #include "xfs_format.h" 22 #include "xfs_log_format.h" 23 #include "xfs_trans_resv.h" 24 #include "xfs_mount.h" 25 #include "xfs_da_format.h" 26 #include "xfs_da_btree.h" 27 #include "xfs_inode.h" 28 #include "xfs_trans.h" 29 #include "xfs_inode_item.h" 30 #include "xfs_bmap.h" 31 #include "xfs_bmap_util.h" 32 #include "xfs_error.h" 33 #include "xfs_dir2.h" 34 #include "xfs_dir2_priv.h" 35 #include "xfs_ioctl.h" 36 #include "xfs_trace.h" 37 #include "xfs_log.h" 38 #include "xfs_icache.h" 39 #include "xfs_pnfs.h" 40 41 #include <linux/aio.h> 42 #include <linux/dcache.h> 43 #include <linux/falloc.h> 44 #include <linux/pagevec.h> 45 46 static const struct vm_operations_struct xfs_file_vm_ops; 47 48 /* 49 * Locking primitives for read and write IO paths to ensure we consistently use 50 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. 51 */ 52 static inline void 53 xfs_rw_ilock( 54 struct xfs_inode *ip, 55 int type) 56 { 57 if (type & XFS_IOLOCK_EXCL) 58 mutex_lock(&VFS_I(ip)->i_mutex); 59 xfs_ilock(ip, type); 60 } 61 62 static inline void 63 xfs_rw_iunlock( 64 struct xfs_inode *ip, 65 int type) 66 { 67 xfs_iunlock(ip, type); 68 if (type & XFS_IOLOCK_EXCL) 69 mutex_unlock(&VFS_I(ip)->i_mutex); 70 } 71 72 static inline void 73 xfs_rw_ilock_demote( 74 struct xfs_inode *ip, 75 int type) 76 { 77 xfs_ilock_demote(ip, type); 78 if (type & XFS_IOLOCK_EXCL) 79 mutex_unlock(&VFS_I(ip)->i_mutex); 80 } 81 82 /* 83 * xfs_iozero 84 * 85 * xfs_iozero clears the specified range of buffer supplied, 86 * and marks all the affected blocks as valid and modified. If 87 * an affected block is not allocated, it will be allocated. If 88 * an affected block is not completely overwritten, and is not 89 * valid before the operation, it will be read from disk before 90 * being partially zeroed. 91 */ 92 int 93 xfs_iozero( 94 struct xfs_inode *ip, /* inode */ 95 loff_t pos, /* offset in file */ 96 size_t count) /* size of data to zero */ 97 { 98 struct page *page; 99 struct address_space *mapping; 100 int status; 101 102 mapping = VFS_I(ip)->i_mapping; 103 do { 104 unsigned offset, bytes; 105 void *fsdata; 106 107 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ 108 bytes = PAGE_CACHE_SIZE - offset; 109 if (bytes > count) 110 bytes = count; 111 112 status = pagecache_write_begin(NULL, mapping, pos, bytes, 113 AOP_FLAG_UNINTERRUPTIBLE, 114 &page, &fsdata); 115 if (status) 116 break; 117 118 zero_user(page, offset, bytes); 119 120 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, 121 page, fsdata); 122 WARN_ON(status <= 0); /* can't return less than zero! */ 123 pos += bytes; 124 count -= bytes; 125 status = 0; 126 } while (count); 127 128 return (-status); 129 } 130 131 int 132 xfs_update_prealloc_flags( 133 struct xfs_inode *ip, 134 enum xfs_prealloc_flags flags) 135 { 136 struct xfs_trans *tp; 137 int error; 138 139 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); 140 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); 141 if (error) { 142 xfs_trans_cancel(tp, 0); 143 return error; 144 } 145 146 xfs_ilock(ip, XFS_ILOCK_EXCL); 147 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 148 149 if (!(flags & XFS_PREALLOC_INVISIBLE)) { 150 ip->i_d.di_mode &= ~S_ISUID; 151 if (ip->i_d.di_mode & S_IXGRP) 152 ip->i_d.di_mode &= ~S_ISGID; 153 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 154 } 155 156 if (flags & XFS_PREALLOC_SET) 157 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; 158 if (flags & XFS_PREALLOC_CLEAR) 159 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; 160 161 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 162 if (flags & XFS_PREALLOC_SYNC) 163 xfs_trans_set_sync(tp); 164 return xfs_trans_commit(tp, 0); 165 } 166 167 /* 168 * Fsync operations on directories are much simpler than on regular files, 169 * as there is no file data to flush, and thus also no need for explicit 170 * cache flush operations, and there are no non-transaction metadata updates 171 * on directories either. 172 */ 173 STATIC int 174 xfs_dir_fsync( 175 struct file *file, 176 loff_t start, 177 loff_t end, 178 int datasync) 179 { 180 struct xfs_inode *ip = XFS_I(file->f_mapping->host); 181 struct xfs_mount *mp = ip->i_mount; 182 xfs_lsn_t lsn = 0; 183 184 trace_xfs_dir_fsync(ip); 185 186 xfs_ilock(ip, XFS_ILOCK_SHARED); 187 if (xfs_ipincount(ip)) 188 lsn = ip->i_itemp->ili_last_lsn; 189 xfs_iunlock(ip, XFS_ILOCK_SHARED); 190 191 if (!lsn) 192 return 0; 193 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); 194 } 195 196 STATIC int 197 xfs_file_fsync( 198 struct file *file, 199 loff_t start, 200 loff_t end, 201 int datasync) 202 { 203 struct inode *inode = file->f_mapping->host; 204 struct xfs_inode *ip = XFS_I(inode); 205 struct xfs_mount *mp = ip->i_mount; 206 int error = 0; 207 int log_flushed = 0; 208 xfs_lsn_t lsn = 0; 209 210 trace_xfs_file_fsync(ip); 211 212 error = filemap_write_and_wait_range(inode->i_mapping, start, end); 213 if (error) 214 return error; 215 216 if (XFS_FORCED_SHUTDOWN(mp)) 217 return -EIO; 218 219 xfs_iflags_clear(ip, XFS_ITRUNCATED); 220 221 if (mp->m_flags & XFS_MOUNT_BARRIER) { 222 /* 223 * If we have an RT and/or log subvolume we need to make sure 224 * to flush the write cache the device used for file data 225 * first. This is to ensure newly written file data make 226 * it to disk before logging the new inode size in case of 227 * an extending write. 228 */ 229 if (XFS_IS_REALTIME_INODE(ip)) 230 xfs_blkdev_issue_flush(mp->m_rtdev_targp); 231 else if (mp->m_logdev_targp != mp->m_ddev_targp) 232 xfs_blkdev_issue_flush(mp->m_ddev_targp); 233 } 234 235 /* 236 * All metadata updates are logged, which means that we just have 237 * to flush the log up to the latest LSN that touched the inode. 238 */ 239 xfs_ilock(ip, XFS_ILOCK_SHARED); 240 if (xfs_ipincount(ip)) { 241 if (!datasync || 242 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) 243 lsn = ip->i_itemp->ili_last_lsn; 244 } 245 xfs_iunlock(ip, XFS_ILOCK_SHARED); 246 247 if (lsn) 248 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); 249 250 /* 251 * If we only have a single device, and the log force about was 252 * a no-op we might have to flush the data device cache here. 253 * This can only happen for fdatasync/O_DSYNC if we were overwriting 254 * an already allocated file and thus do not have any metadata to 255 * commit. 256 */ 257 if ((mp->m_flags & XFS_MOUNT_BARRIER) && 258 mp->m_logdev_targp == mp->m_ddev_targp && 259 !XFS_IS_REALTIME_INODE(ip) && 260 !log_flushed) 261 xfs_blkdev_issue_flush(mp->m_ddev_targp); 262 263 return error; 264 } 265 266 STATIC ssize_t 267 xfs_file_read_iter( 268 struct kiocb *iocb, 269 struct iov_iter *to) 270 { 271 struct file *file = iocb->ki_filp; 272 struct inode *inode = file->f_mapping->host; 273 struct xfs_inode *ip = XFS_I(inode); 274 struct xfs_mount *mp = ip->i_mount; 275 size_t size = iov_iter_count(to); 276 ssize_t ret = 0; 277 int ioflags = 0; 278 xfs_fsize_t n; 279 loff_t pos = iocb->ki_pos; 280 281 XFS_STATS_INC(xs_read_calls); 282 283 if (unlikely(file->f_flags & O_DIRECT)) 284 ioflags |= XFS_IO_ISDIRECT; 285 if (file->f_mode & FMODE_NOCMTIME) 286 ioflags |= XFS_IO_INVIS; 287 288 if (unlikely(ioflags & XFS_IO_ISDIRECT)) { 289 xfs_buftarg_t *target = 290 XFS_IS_REALTIME_INODE(ip) ? 291 mp->m_rtdev_targp : mp->m_ddev_targp; 292 /* DIO must be aligned to device logical sector size */ 293 if ((pos | size) & target->bt_logical_sectormask) { 294 if (pos == i_size_read(inode)) 295 return 0; 296 return -EINVAL; 297 } 298 } 299 300 n = mp->m_super->s_maxbytes - pos; 301 if (n <= 0 || size == 0) 302 return 0; 303 304 if (n < size) 305 size = n; 306 307 if (XFS_FORCED_SHUTDOWN(mp)) 308 return -EIO; 309 310 /* 311 * Locking is a bit tricky here. If we take an exclusive lock 312 * for direct IO, we effectively serialise all new concurrent 313 * read IO to this file and block it behind IO that is currently in 314 * progress because IO in progress holds the IO lock shared. We only 315 * need to hold the lock exclusive to blow away the page cache, so 316 * only take lock exclusively if the page cache needs invalidation. 317 * This allows the normal direct IO case of no page cache pages to 318 * proceeed concurrently without serialisation. 319 */ 320 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); 321 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { 322 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); 323 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); 324 325 if (inode->i_mapping->nrpages) { 326 ret = filemap_write_and_wait_range( 327 VFS_I(ip)->i_mapping, 328 pos, pos + size - 1); 329 if (ret) { 330 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); 331 return ret; 332 } 333 334 /* 335 * Invalidate whole pages. This can return an error if 336 * we fail to invalidate a page, but this should never 337 * happen on XFS. Warn if it does fail. 338 */ 339 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, 340 pos >> PAGE_CACHE_SHIFT, 341 (pos + size - 1) >> PAGE_CACHE_SHIFT); 342 WARN_ON_ONCE(ret); 343 ret = 0; 344 } 345 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); 346 } 347 348 trace_xfs_file_read(ip, size, pos, ioflags); 349 350 ret = generic_file_read_iter(iocb, to); 351 if (ret > 0) 352 XFS_STATS_ADD(xs_read_bytes, ret); 353 354 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); 355 return ret; 356 } 357 358 STATIC ssize_t 359 xfs_file_splice_read( 360 struct file *infilp, 361 loff_t *ppos, 362 struct pipe_inode_info *pipe, 363 size_t count, 364 unsigned int flags) 365 { 366 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); 367 int ioflags = 0; 368 ssize_t ret; 369 370 XFS_STATS_INC(xs_read_calls); 371 372 if (infilp->f_mode & FMODE_NOCMTIME) 373 ioflags |= XFS_IO_INVIS; 374 375 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) 376 return -EIO; 377 378 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); 379 380 trace_xfs_file_splice_read(ip, count, *ppos, ioflags); 381 382 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); 383 if (ret > 0) 384 XFS_STATS_ADD(xs_read_bytes, ret); 385 386 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); 387 return ret; 388 } 389 390 /* 391 * This routine is called to handle zeroing any space in the last block of the 392 * file that is beyond the EOF. We do this since the size is being increased 393 * without writing anything to that block and we don't want to read the 394 * garbage on the disk. 395 */ 396 STATIC int /* error (positive) */ 397 xfs_zero_last_block( 398 struct xfs_inode *ip, 399 xfs_fsize_t offset, 400 xfs_fsize_t isize, 401 bool *did_zeroing) 402 { 403 struct xfs_mount *mp = ip->i_mount; 404 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); 405 int zero_offset = XFS_B_FSB_OFFSET(mp, isize); 406 int zero_len; 407 int nimaps = 1; 408 int error = 0; 409 struct xfs_bmbt_irec imap; 410 411 xfs_ilock(ip, XFS_ILOCK_EXCL); 412 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); 413 xfs_iunlock(ip, XFS_ILOCK_EXCL); 414 if (error) 415 return error; 416 417 ASSERT(nimaps > 0); 418 419 /* 420 * If the block underlying isize is just a hole, then there 421 * is nothing to zero. 422 */ 423 if (imap.br_startblock == HOLESTARTBLOCK) 424 return 0; 425 426 zero_len = mp->m_sb.sb_blocksize - zero_offset; 427 if (isize + zero_len > offset) 428 zero_len = offset - isize; 429 *did_zeroing = true; 430 return xfs_iozero(ip, isize, zero_len); 431 } 432 433 /* 434 * Zero any on disk space between the current EOF and the new, larger EOF. 435 * 436 * This handles the normal case of zeroing the remainder of the last block in 437 * the file and the unusual case of zeroing blocks out beyond the size of the 438 * file. This second case only happens with fixed size extents and when the 439 * system crashes before the inode size was updated but after blocks were 440 * allocated. 441 * 442 * Expects the iolock to be held exclusive, and will take the ilock internally. 443 */ 444 int /* error (positive) */ 445 xfs_zero_eof( 446 struct xfs_inode *ip, 447 xfs_off_t offset, /* starting I/O offset */ 448 xfs_fsize_t isize, /* current inode size */ 449 bool *did_zeroing) 450 { 451 struct xfs_mount *mp = ip->i_mount; 452 xfs_fileoff_t start_zero_fsb; 453 xfs_fileoff_t end_zero_fsb; 454 xfs_fileoff_t zero_count_fsb; 455 xfs_fileoff_t last_fsb; 456 xfs_fileoff_t zero_off; 457 xfs_fsize_t zero_len; 458 int nimaps; 459 int error = 0; 460 struct xfs_bmbt_irec imap; 461 462 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 463 ASSERT(offset > isize); 464 465 /* 466 * First handle zeroing the block on which isize resides. 467 * 468 * We only zero a part of that block so it is handled specially. 469 */ 470 if (XFS_B_FSB_OFFSET(mp, isize) != 0) { 471 error = xfs_zero_last_block(ip, offset, isize, did_zeroing); 472 if (error) 473 return error; 474 } 475 476 /* 477 * Calculate the range between the new size and the old where blocks 478 * needing to be zeroed may exist. 479 * 480 * To get the block where the last byte in the file currently resides, 481 * we need to subtract one from the size and truncate back to a block 482 * boundary. We subtract 1 in case the size is exactly on a block 483 * boundary. 484 */ 485 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; 486 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); 487 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); 488 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); 489 if (last_fsb == end_zero_fsb) { 490 /* 491 * The size was only incremented on its last block. 492 * We took care of that above, so just return. 493 */ 494 return 0; 495 } 496 497 ASSERT(start_zero_fsb <= end_zero_fsb); 498 while (start_zero_fsb <= end_zero_fsb) { 499 nimaps = 1; 500 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; 501 502 xfs_ilock(ip, XFS_ILOCK_EXCL); 503 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, 504 &imap, &nimaps, 0); 505 xfs_iunlock(ip, XFS_ILOCK_EXCL); 506 if (error) 507 return error; 508 509 ASSERT(nimaps > 0); 510 511 if (imap.br_state == XFS_EXT_UNWRITTEN || 512 imap.br_startblock == HOLESTARTBLOCK) { 513 start_zero_fsb = imap.br_startoff + imap.br_blockcount; 514 ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); 515 continue; 516 } 517 518 /* 519 * There are blocks we need to zero. 520 */ 521 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); 522 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); 523 524 if ((zero_off + zero_len) > offset) 525 zero_len = offset - zero_off; 526 527 error = xfs_iozero(ip, zero_off, zero_len); 528 if (error) 529 return error; 530 531 *did_zeroing = true; 532 start_zero_fsb = imap.br_startoff + imap.br_blockcount; 533 ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); 534 } 535 536 return 0; 537 } 538 539 /* 540 * Common pre-write limit and setup checks. 541 * 542 * Called with the iolocked held either shared and exclusive according to 543 * @iolock, and returns with it held. Might upgrade the iolock to exclusive 544 * if called for a direct write beyond i_size. 545 */ 546 STATIC ssize_t 547 xfs_file_aio_write_checks( 548 struct file *file, 549 loff_t *pos, 550 size_t *count, 551 int *iolock) 552 { 553 struct inode *inode = file->f_mapping->host; 554 struct xfs_inode *ip = XFS_I(inode); 555 int error = 0; 556 557 restart: 558 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode)); 559 if (error) 560 return error; 561 562 error = xfs_break_layouts(inode, iolock); 563 if (error) 564 return error; 565 566 /* 567 * If the offset is beyond the size of the file, we need to zero any 568 * blocks that fall between the existing EOF and the start of this 569 * write. If zeroing is needed and we are currently holding the 570 * iolock shared, we need to update it to exclusive which implies 571 * having to redo all checks before. 572 */ 573 if (*pos > i_size_read(inode)) { 574 bool zero = false; 575 576 if (*iolock == XFS_IOLOCK_SHARED) { 577 xfs_rw_iunlock(ip, *iolock); 578 *iolock = XFS_IOLOCK_EXCL; 579 xfs_rw_ilock(ip, *iolock); 580 goto restart; 581 } 582 error = xfs_zero_eof(ip, *pos, i_size_read(inode), &zero); 583 if (error) 584 return error; 585 } 586 587 /* 588 * Updating the timestamps will grab the ilock again from 589 * xfs_fs_dirty_inode, so we have to call it after dropping the 590 * lock above. Eventually we should look into a way to avoid 591 * the pointless lock roundtrip. 592 */ 593 if (likely(!(file->f_mode & FMODE_NOCMTIME))) { 594 error = file_update_time(file); 595 if (error) 596 return error; 597 } 598 599 /* 600 * If we're writing the file then make sure to clear the setuid and 601 * setgid bits if the process is not being run by root. This keeps 602 * people from modifying setuid and setgid binaries. 603 */ 604 return file_remove_suid(file); 605 } 606 607 /* 608 * xfs_file_dio_aio_write - handle direct IO writes 609 * 610 * Lock the inode appropriately to prepare for and issue a direct IO write. 611 * By separating it from the buffered write path we remove all the tricky to 612 * follow locking changes and looping. 613 * 614 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL 615 * until we're sure the bytes at the new EOF have been zeroed and/or the cached 616 * pages are flushed out. 617 * 618 * In most cases the direct IO writes will be done holding IOLOCK_SHARED 619 * allowing them to be done in parallel with reads and other direct IO writes. 620 * However, if the IO is not aligned to filesystem blocks, the direct IO layer 621 * needs to do sub-block zeroing and that requires serialisation against other 622 * direct IOs to the same block. In this case we need to serialise the 623 * submission of the unaligned IOs so that we don't get racing block zeroing in 624 * the dio layer. To avoid the problem with aio, we also need to wait for 625 * outstanding IOs to complete so that unwritten extent conversion is completed 626 * before we try to map the overlapping block. This is currently implemented by 627 * hitting it with a big hammer (i.e. inode_dio_wait()). 628 * 629 * Returns with locks held indicated by @iolock and errors indicated by 630 * negative return values. 631 */ 632 STATIC ssize_t 633 xfs_file_dio_aio_write( 634 struct kiocb *iocb, 635 struct iov_iter *from) 636 { 637 struct file *file = iocb->ki_filp; 638 struct address_space *mapping = file->f_mapping; 639 struct inode *inode = mapping->host; 640 struct xfs_inode *ip = XFS_I(inode); 641 struct xfs_mount *mp = ip->i_mount; 642 ssize_t ret = 0; 643 int unaligned_io = 0; 644 int iolock; 645 size_t count = iov_iter_count(from); 646 loff_t pos = iocb->ki_pos; 647 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? 648 mp->m_rtdev_targp : mp->m_ddev_targp; 649 650 /* DIO must be aligned to device logical sector size */ 651 if ((pos | count) & target->bt_logical_sectormask) 652 return -EINVAL; 653 654 /* "unaligned" here means not aligned to a filesystem block */ 655 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) 656 unaligned_io = 1; 657 658 /* 659 * We don't need to take an exclusive lock unless there page cache needs 660 * to be invalidated or unaligned IO is being executed. We don't need to 661 * consider the EOF extension case here because 662 * xfs_file_aio_write_checks() will relock the inode as necessary for 663 * EOF zeroing cases and fill out the new inode size as appropriate. 664 */ 665 if (unaligned_io || mapping->nrpages) 666 iolock = XFS_IOLOCK_EXCL; 667 else 668 iolock = XFS_IOLOCK_SHARED; 669 xfs_rw_ilock(ip, iolock); 670 671 /* 672 * Recheck if there are cached pages that need invalidate after we got 673 * the iolock to protect against other threads adding new pages while 674 * we were waiting for the iolock. 675 */ 676 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { 677 xfs_rw_iunlock(ip, iolock); 678 iolock = XFS_IOLOCK_EXCL; 679 xfs_rw_ilock(ip, iolock); 680 } 681 682 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); 683 if (ret) 684 goto out; 685 iov_iter_truncate(from, count); 686 687 if (mapping->nrpages) { 688 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 689 pos, pos + count - 1); 690 if (ret) 691 goto out; 692 /* 693 * Invalidate whole pages. This can return an error if 694 * we fail to invalidate a page, but this should never 695 * happen on XFS. Warn if it does fail. 696 */ 697 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, 698 pos >> PAGE_CACHE_SHIFT, 699 (pos + count - 1) >> PAGE_CACHE_SHIFT); 700 WARN_ON_ONCE(ret); 701 ret = 0; 702 } 703 704 /* 705 * If we are doing unaligned IO, wait for all other IO to drain, 706 * otherwise demote the lock if we had to flush cached pages 707 */ 708 if (unaligned_io) 709 inode_dio_wait(inode); 710 else if (iolock == XFS_IOLOCK_EXCL) { 711 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); 712 iolock = XFS_IOLOCK_SHARED; 713 } 714 715 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); 716 ret = generic_file_direct_write(iocb, from, pos); 717 718 out: 719 xfs_rw_iunlock(ip, iolock); 720 721 /* No fallback to buffered IO on errors for XFS. */ 722 ASSERT(ret < 0 || ret == count); 723 return ret; 724 } 725 726 STATIC ssize_t 727 xfs_file_buffered_aio_write( 728 struct kiocb *iocb, 729 struct iov_iter *from) 730 { 731 struct file *file = iocb->ki_filp; 732 struct address_space *mapping = file->f_mapping; 733 struct inode *inode = mapping->host; 734 struct xfs_inode *ip = XFS_I(inode); 735 ssize_t ret; 736 int enospc = 0; 737 int iolock = XFS_IOLOCK_EXCL; 738 loff_t pos = iocb->ki_pos; 739 size_t count = iov_iter_count(from); 740 741 xfs_rw_ilock(ip, iolock); 742 743 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); 744 if (ret) 745 goto out; 746 747 iov_iter_truncate(from, count); 748 /* We can write back this queue in page reclaim */ 749 current->backing_dev_info = inode_to_bdi(inode); 750 751 write_retry: 752 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0); 753 ret = generic_perform_write(file, from, pos); 754 if (likely(ret >= 0)) 755 iocb->ki_pos = pos + ret; 756 757 /* 758 * If we hit a space limit, try to free up some lingering preallocated 759 * space before returning an error. In the case of ENOSPC, first try to 760 * write back all dirty inodes to free up some of the excess reserved 761 * metadata space. This reduces the chances that the eofblocks scan 762 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this 763 * also behaves as a filter to prevent too many eofblocks scans from 764 * running at the same time. 765 */ 766 if (ret == -EDQUOT && !enospc) { 767 enospc = xfs_inode_free_quota_eofblocks(ip); 768 if (enospc) 769 goto write_retry; 770 } else if (ret == -ENOSPC && !enospc) { 771 struct xfs_eofblocks eofb = {0}; 772 773 enospc = 1; 774 xfs_flush_inodes(ip->i_mount); 775 eofb.eof_scan_owner = ip->i_ino; /* for locking */ 776 eofb.eof_flags = XFS_EOF_FLAGS_SYNC; 777 xfs_icache_free_eofblocks(ip->i_mount, &eofb); 778 goto write_retry; 779 } 780 781 current->backing_dev_info = NULL; 782 out: 783 xfs_rw_iunlock(ip, iolock); 784 return ret; 785 } 786 787 STATIC ssize_t 788 xfs_file_write_iter( 789 struct kiocb *iocb, 790 struct iov_iter *from) 791 { 792 struct file *file = iocb->ki_filp; 793 struct address_space *mapping = file->f_mapping; 794 struct inode *inode = mapping->host; 795 struct xfs_inode *ip = XFS_I(inode); 796 ssize_t ret; 797 size_t ocount = iov_iter_count(from); 798 799 XFS_STATS_INC(xs_write_calls); 800 801 if (ocount == 0) 802 return 0; 803 804 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) 805 return -EIO; 806 807 if (unlikely(file->f_flags & O_DIRECT)) 808 ret = xfs_file_dio_aio_write(iocb, from); 809 else 810 ret = xfs_file_buffered_aio_write(iocb, from); 811 812 if (ret > 0) { 813 ssize_t err; 814 815 XFS_STATS_ADD(xs_write_bytes, ret); 816 817 /* Handle various SYNC-type writes */ 818 err = generic_write_sync(file, iocb->ki_pos - ret, ret); 819 if (err < 0) 820 ret = err; 821 } 822 return ret; 823 } 824 825 STATIC long 826 xfs_file_fallocate( 827 struct file *file, 828 int mode, 829 loff_t offset, 830 loff_t len) 831 { 832 struct inode *inode = file_inode(file); 833 struct xfs_inode *ip = XFS_I(inode); 834 long error; 835 enum xfs_prealloc_flags flags = 0; 836 uint iolock = XFS_IOLOCK_EXCL; 837 loff_t new_size = 0; 838 839 if (!S_ISREG(inode->i_mode)) 840 return -EINVAL; 841 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | 842 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) 843 return -EOPNOTSUPP; 844 845 xfs_ilock(ip, iolock); 846 error = xfs_break_layouts(inode, &iolock); 847 if (error) 848 goto out_unlock; 849 850 if (mode & FALLOC_FL_PUNCH_HOLE) { 851 error = xfs_free_file_space(ip, offset, len); 852 if (error) 853 goto out_unlock; 854 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { 855 unsigned blksize_mask = (1 << inode->i_blkbits) - 1; 856 857 if (offset & blksize_mask || len & blksize_mask) { 858 error = -EINVAL; 859 goto out_unlock; 860 } 861 862 /* 863 * There is no need to overlap collapse range with EOF, 864 * in which case it is effectively a truncate operation 865 */ 866 if (offset + len >= i_size_read(inode)) { 867 error = -EINVAL; 868 goto out_unlock; 869 } 870 871 new_size = i_size_read(inode) - len; 872 873 error = xfs_collapse_file_space(ip, offset, len); 874 if (error) 875 goto out_unlock; 876 } else { 877 flags |= XFS_PREALLOC_SET; 878 879 if (!(mode & FALLOC_FL_KEEP_SIZE) && 880 offset + len > i_size_read(inode)) { 881 new_size = offset + len; 882 error = inode_newsize_ok(inode, new_size); 883 if (error) 884 goto out_unlock; 885 } 886 887 if (mode & FALLOC_FL_ZERO_RANGE) 888 error = xfs_zero_file_space(ip, offset, len); 889 else 890 error = xfs_alloc_file_space(ip, offset, len, 891 XFS_BMAPI_PREALLOC); 892 if (error) 893 goto out_unlock; 894 } 895 896 if (file->f_flags & O_DSYNC) 897 flags |= XFS_PREALLOC_SYNC; 898 899 error = xfs_update_prealloc_flags(ip, flags); 900 if (error) 901 goto out_unlock; 902 903 /* Change file size if needed */ 904 if (new_size) { 905 struct iattr iattr; 906 907 iattr.ia_valid = ATTR_SIZE; 908 iattr.ia_size = new_size; 909 error = xfs_setattr_size(ip, &iattr); 910 } 911 912 out_unlock: 913 xfs_iunlock(ip, iolock); 914 return error; 915 } 916 917 918 STATIC int 919 xfs_file_open( 920 struct inode *inode, 921 struct file *file) 922 { 923 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) 924 return -EFBIG; 925 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) 926 return -EIO; 927 return 0; 928 } 929 930 STATIC int 931 xfs_dir_open( 932 struct inode *inode, 933 struct file *file) 934 { 935 struct xfs_inode *ip = XFS_I(inode); 936 int mode; 937 int error; 938 939 error = xfs_file_open(inode, file); 940 if (error) 941 return error; 942 943 /* 944 * If there are any blocks, read-ahead block 0 as we're almost 945 * certain to have the next operation be a read there. 946 */ 947 mode = xfs_ilock_data_map_shared(ip); 948 if (ip->i_d.di_nextents > 0) 949 xfs_dir3_data_readahead(ip, 0, -1); 950 xfs_iunlock(ip, mode); 951 return 0; 952 } 953 954 STATIC int 955 xfs_file_release( 956 struct inode *inode, 957 struct file *filp) 958 { 959 return xfs_release(XFS_I(inode)); 960 } 961 962 STATIC int 963 xfs_file_readdir( 964 struct file *file, 965 struct dir_context *ctx) 966 { 967 struct inode *inode = file_inode(file); 968 xfs_inode_t *ip = XFS_I(inode); 969 size_t bufsize; 970 971 /* 972 * The Linux API doesn't pass down the total size of the buffer 973 * we read into down to the filesystem. With the filldir concept 974 * it's not needed for correct information, but the XFS dir2 leaf 975 * code wants an estimate of the buffer size to calculate it's 976 * readahead window and size the buffers used for mapping to 977 * physical blocks. 978 * 979 * Try to give it an estimate that's good enough, maybe at some 980 * point we can change the ->readdir prototype to include the 981 * buffer size. For now we use the current glibc buffer size. 982 */ 983 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); 984 985 return xfs_readdir(ip, ctx, bufsize); 986 } 987 988 STATIC int 989 xfs_file_mmap( 990 struct file *filp, 991 struct vm_area_struct *vma) 992 { 993 vma->vm_ops = &xfs_file_vm_ops; 994 995 file_accessed(filp); 996 return 0; 997 } 998 999 /* 1000 * mmap()d file has taken write protection fault and is being made 1001 * writable. We can set the page state up correctly for a writable 1002 * page, which means we can do correct delalloc accounting (ENOSPC 1003 * checking!) and unwritten extent mapping. 1004 */ 1005 STATIC int 1006 xfs_vm_page_mkwrite( 1007 struct vm_area_struct *vma, 1008 struct vm_fault *vmf) 1009 { 1010 return block_page_mkwrite(vma, vmf, xfs_get_blocks); 1011 } 1012 1013 /* 1014 * This type is designed to indicate the type of offset we would like 1015 * to search from page cache for xfs_seek_hole_data(). 1016 */ 1017 enum { 1018 HOLE_OFF = 0, 1019 DATA_OFF, 1020 }; 1021 1022 /* 1023 * Lookup the desired type of offset from the given page. 1024 * 1025 * On success, return true and the offset argument will point to the 1026 * start of the region that was found. Otherwise this function will 1027 * return false and keep the offset argument unchanged. 1028 */ 1029 STATIC bool 1030 xfs_lookup_buffer_offset( 1031 struct page *page, 1032 loff_t *offset, 1033 unsigned int type) 1034 { 1035 loff_t lastoff = page_offset(page); 1036 bool found = false; 1037 struct buffer_head *bh, *head; 1038 1039 bh = head = page_buffers(page); 1040 do { 1041 /* 1042 * Unwritten extents that have data in the page 1043 * cache covering them can be identified by the 1044 * BH_Unwritten state flag. Pages with multiple 1045 * buffers might have a mix of holes, data and 1046 * unwritten extents - any buffer with valid 1047 * data in it should have BH_Uptodate flag set 1048 * on it. 1049 */ 1050 if (buffer_unwritten(bh) || 1051 buffer_uptodate(bh)) { 1052 if (type == DATA_OFF) 1053 found = true; 1054 } else { 1055 if (type == HOLE_OFF) 1056 found = true; 1057 } 1058 1059 if (found) { 1060 *offset = lastoff; 1061 break; 1062 } 1063 lastoff += bh->b_size; 1064 } while ((bh = bh->b_this_page) != head); 1065 1066 return found; 1067 } 1068 1069 /* 1070 * This routine is called to find out and return a data or hole offset 1071 * from the page cache for unwritten extents according to the desired 1072 * type for xfs_seek_hole_data(). 1073 * 1074 * The argument offset is used to tell where we start to search from the 1075 * page cache. Map is used to figure out the end points of the range to 1076 * lookup pages. 1077 * 1078 * Return true if the desired type of offset was found, and the argument 1079 * offset is filled with that address. Otherwise, return false and keep 1080 * offset unchanged. 1081 */ 1082 STATIC bool 1083 xfs_find_get_desired_pgoff( 1084 struct inode *inode, 1085 struct xfs_bmbt_irec *map, 1086 unsigned int type, 1087 loff_t *offset) 1088 { 1089 struct xfs_inode *ip = XFS_I(inode); 1090 struct xfs_mount *mp = ip->i_mount; 1091 struct pagevec pvec; 1092 pgoff_t index; 1093 pgoff_t end; 1094 loff_t endoff; 1095 loff_t startoff = *offset; 1096 loff_t lastoff = startoff; 1097 bool found = false; 1098 1099 pagevec_init(&pvec, 0); 1100 1101 index = startoff >> PAGE_CACHE_SHIFT; 1102 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); 1103 end = endoff >> PAGE_CACHE_SHIFT; 1104 do { 1105 int want; 1106 unsigned nr_pages; 1107 unsigned int i; 1108 1109 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); 1110 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, 1111 want); 1112 /* 1113 * No page mapped into given range. If we are searching holes 1114 * and if this is the first time we got into the loop, it means 1115 * that the given offset is landed in a hole, return it. 1116 * 1117 * If we have already stepped through some block buffers to find 1118 * holes but they all contains data. In this case, the last 1119 * offset is already updated and pointed to the end of the last 1120 * mapped page, if it does not reach the endpoint to search, 1121 * that means there should be a hole between them. 1122 */ 1123 if (nr_pages == 0) { 1124 /* Data search found nothing */ 1125 if (type == DATA_OFF) 1126 break; 1127 1128 ASSERT(type == HOLE_OFF); 1129 if (lastoff == startoff || lastoff < endoff) { 1130 found = true; 1131 *offset = lastoff; 1132 } 1133 break; 1134 } 1135 1136 /* 1137 * At lease we found one page. If this is the first time we 1138 * step into the loop, and if the first page index offset is 1139 * greater than the given search offset, a hole was found. 1140 */ 1141 if (type == HOLE_OFF && lastoff == startoff && 1142 lastoff < page_offset(pvec.pages[0])) { 1143 found = true; 1144 break; 1145 } 1146 1147 for (i = 0; i < nr_pages; i++) { 1148 struct page *page = pvec.pages[i]; 1149 loff_t b_offset; 1150 1151 /* 1152 * At this point, the page may be truncated or 1153 * invalidated (changing page->mapping to NULL), 1154 * or even swizzled back from swapper_space to tmpfs 1155 * file mapping. However, page->index will not change 1156 * because we have a reference on the page. 1157 * 1158 * Searching done if the page index is out of range. 1159 * If the current offset is not reaches the end of 1160 * the specified search range, there should be a hole 1161 * between them. 1162 */ 1163 if (page->index > end) { 1164 if (type == HOLE_OFF && lastoff < endoff) { 1165 *offset = lastoff; 1166 found = true; 1167 } 1168 goto out; 1169 } 1170 1171 lock_page(page); 1172 /* 1173 * Page truncated or invalidated(page->mapping == NULL). 1174 * We can freely skip it and proceed to check the next 1175 * page. 1176 */ 1177 if (unlikely(page->mapping != inode->i_mapping)) { 1178 unlock_page(page); 1179 continue; 1180 } 1181 1182 if (!page_has_buffers(page)) { 1183 unlock_page(page); 1184 continue; 1185 } 1186 1187 found = xfs_lookup_buffer_offset(page, &b_offset, type); 1188 if (found) { 1189 /* 1190 * The found offset may be less than the start 1191 * point to search if this is the first time to 1192 * come here. 1193 */ 1194 *offset = max_t(loff_t, startoff, b_offset); 1195 unlock_page(page); 1196 goto out; 1197 } 1198 1199 /* 1200 * We either searching data but nothing was found, or 1201 * searching hole but found a data buffer. In either 1202 * case, probably the next page contains the desired 1203 * things, update the last offset to it so. 1204 */ 1205 lastoff = page_offset(page) + PAGE_SIZE; 1206 unlock_page(page); 1207 } 1208 1209 /* 1210 * The number of returned pages less than our desired, search 1211 * done. In this case, nothing was found for searching data, 1212 * but we found a hole behind the last offset. 1213 */ 1214 if (nr_pages < want) { 1215 if (type == HOLE_OFF) { 1216 *offset = lastoff; 1217 found = true; 1218 } 1219 break; 1220 } 1221 1222 index = pvec.pages[i - 1]->index + 1; 1223 pagevec_release(&pvec); 1224 } while (index <= end); 1225 1226 out: 1227 pagevec_release(&pvec); 1228 return found; 1229 } 1230 1231 STATIC loff_t 1232 xfs_seek_hole_data( 1233 struct file *file, 1234 loff_t start, 1235 int whence) 1236 { 1237 struct inode *inode = file->f_mapping->host; 1238 struct xfs_inode *ip = XFS_I(inode); 1239 struct xfs_mount *mp = ip->i_mount; 1240 loff_t uninitialized_var(offset); 1241 xfs_fsize_t isize; 1242 xfs_fileoff_t fsbno; 1243 xfs_filblks_t end; 1244 uint lock; 1245 int error; 1246 1247 if (XFS_FORCED_SHUTDOWN(mp)) 1248 return -EIO; 1249 1250 lock = xfs_ilock_data_map_shared(ip); 1251 1252 isize = i_size_read(inode); 1253 if (start >= isize) { 1254 error = -ENXIO; 1255 goto out_unlock; 1256 } 1257 1258 /* 1259 * Try to read extents from the first block indicated 1260 * by fsbno to the end block of the file. 1261 */ 1262 fsbno = XFS_B_TO_FSBT(mp, start); 1263 end = XFS_B_TO_FSB(mp, isize); 1264 1265 for (;;) { 1266 struct xfs_bmbt_irec map[2]; 1267 int nmap = 2; 1268 unsigned int i; 1269 1270 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, 1271 XFS_BMAPI_ENTIRE); 1272 if (error) 1273 goto out_unlock; 1274 1275 /* No extents at given offset, must be beyond EOF */ 1276 if (nmap == 0) { 1277 error = -ENXIO; 1278 goto out_unlock; 1279 } 1280 1281 for (i = 0; i < nmap; i++) { 1282 offset = max_t(loff_t, start, 1283 XFS_FSB_TO_B(mp, map[i].br_startoff)); 1284 1285 /* Landed in the hole we wanted? */ 1286 if (whence == SEEK_HOLE && 1287 map[i].br_startblock == HOLESTARTBLOCK) 1288 goto out; 1289 1290 /* Landed in the data extent we wanted? */ 1291 if (whence == SEEK_DATA && 1292 (map[i].br_startblock == DELAYSTARTBLOCK || 1293 (map[i].br_state == XFS_EXT_NORM && 1294 !isnullstartblock(map[i].br_startblock)))) 1295 goto out; 1296 1297 /* 1298 * Landed in an unwritten extent, try to search 1299 * for hole or data from page cache. 1300 */ 1301 if (map[i].br_state == XFS_EXT_UNWRITTEN) { 1302 if (xfs_find_get_desired_pgoff(inode, &map[i], 1303 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, 1304 &offset)) 1305 goto out; 1306 } 1307 } 1308 1309 /* 1310 * We only received one extent out of the two requested. This 1311 * means we've hit EOF and didn't find what we are looking for. 1312 */ 1313 if (nmap == 1) { 1314 /* 1315 * If we were looking for a hole, set offset to 1316 * the end of the file (i.e., there is an implicit 1317 * hole at the end of any file). 1318 */ 1319 if (whence == SEEK_HOLE) { 1320 offset = isize; 1321 break; 1322 } 1323 /* 1324 * If we were looking for data, it's nowhere to be found 1325 */ 1326 ASSERT(whence == SEEK_DATA); 1327 error = -ENXIO; 1328 goto out_unlock; 1329 } 1330 1331 ASSERT(i > 1); 1332 1333 /* 1334 * Nothing was found, proceed to the next round of search 1335 * if the next reading offset is not at or beyond EOF. 1336 */ 1337 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; 1338 start = XFS_FSB_TO_B(mp, fsbno); 1339 if (start >= isize) { 1340 if (whence == SEEK_HOLE) { 1341 offset = isize; 1342 break; 1343 } 1344 ASSERT(whence == SEEK_DATA); 1345 error = -ENXIO; 1346 goto out_unlock; 1347 } 1348 } 1349 1350 out: 1351 /* 1352 * If at this point we have found the hole we wanted, the returned 1353 * offset may be bigger than the file size as it may be aligned to 1354 * page boundary for unwritten extents. We need to deal with this 1355 * situation in particular. 1356 */ 1357 if (whence == SEEK_HOLE) 1358 offset = min_t(loff_t, offset, isize); 1359 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); 1360 1361 out_unlock: 1362 xfs_iunlock(ip, lock); 1363 1364 if (error) 1365 return error; 1366 return offset; 1367 } 1368 1369 STATIC loff_t 1370 xfs_file_llseek( 1371 struct file *file, 1372 loff_t offset, 1373 int whence) 1374 { 1375 switch (whence) { 1376 case SEEK_END: 1377 case SEEK_CUR: 1378 case SEEK_SET: 1379 return generic_file_llseek(file, offset, whence); 1380 case SEEK_HOLE: 1381 case SEEK_DATA: 1382 return xfs_seek_hole_data(file, offset, whence); 1383 default: 1384 return -EINVAL; 1385 } 1386 } 1387 1388 const struct file_operations xfs_file_operations = { 1389 .llseek = xfs_file_llseek, 1390 .read = new_sync_read, 1391 .write = new_sync_write, 1392 .read_iter = xfs_file_read_iter, 1393 .write_iter = xfs_file_write_iter, 1394 .splice_read = xfs_file_splice_read, 1395 .splice_write = iter_file_splice_write, 1396 .unlocked_ioctl = xfs_file_ioctl, 1397 #ifdef CONFIG_COMPAT 1398 .compat_ioctl = xfs_file_compat_ioctl, 1399 #endif 1400 .mmap = xfs_file_mmap, 1401 .open = xfs_file_open, 1402 .release = xfs_file_release, 1403 .fsync = xfs_file_fsync, 1404 .fallocate = xfs_file_fallocate, 1405 }; 1406 1407 const struct file_operations xfs_dir_file_operations = { 1408 .open = xfs_dir_open, 1409 .read = generic_read_dir, 1410 .iterate = xfs_file_readdir, 1411 .llseek = generic_file_llseek, 1412 .unlocked_ioctl = xfs_file_ioctl, 1413 #ifdef CONFIG_COMPAT 1414 .compat_ioctl = xfs_file_compat_ioctl, 1415 #endif 1416 .fsync = xfs_dir_fsync, 1417 }; 1418 1419 static const struct vm_operations_struct xfs_file_vm_ops = { 1420 .fault = filemap_fault, 1421 .map_pages = filemap_map_pages, 1422 .page_mkwrite = xfs_vm_page_mkwrite, 1423 }; 1424