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/dcache.h> 42 #include <linux/falloc.h> 43 #include <linux/pagevec.h> 44 45 static const struct vm_operations_struct xfs_file_vm_ops; 46 47 /* 48 * Locking primitives for read and write IO paths to ensure we consistently use 49 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. 50 */ 51 static inline void 52 xfs_rw_ilock( 53 struct xfs_inode *ip, 54 int type) 55 { 56 if (type & XFS_IOLOCK_EXCL) 57 mutex_lock(&VFS_I(ip)->i_mutex); 58 xfs_ilock(ip, type); 59 } 60 61 static inline void 62 xfs_rw_iunlock( 63 struct xfs_inode *ip, 64 int type) 65 { 66 xfs_iunlock(ip, type); 67 if (type & XFS_IOLOCK_EXCL) 68 mutex_unlock(&VFS_I(ip)->i_mutex); 69 } 70 71 static inline void 72 xfs_rw_ilock_demote( 73 struct xfs_inode *ip, 74 int type) 75 { 76 xfs_ilock_demote(ip, type); 77 if (type & XFS_IOLOCK_EXCL) 78 mutex_unlock(&VFS_I(ip)->i_mutex); 79 } 80 81 /* 82 * xfs_iozero 83 * 84 * xfs_iozero clears the specified range of buffer supplied, 85 * and marks all the affected blocks as valid and modified. If 86 * an affected block is not allocated, it will be allocated. If 87 * an affected block is not completely overwritten, and is not 88 * valid before the operation, it will be read from disk before 89 * being partially zeroed. 90 */ 91 int 92 xfs_iozero( 93 struct xfs_inode *ip, /* inode */ 94 loff_t pos, /* offset in file */ 95 size_t count) /* size of data to zero */ 96 { 97 struct page *page; 98 struct address_space *mapping; 99 int status; 100 101 mapping = VFS_I(ip)->i_mapping; 102 do { 103 unsigned offset, bytes; 104 void *fsdata; 105 106 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ 107 bytes = PAGE_CACHE_SIZE - offset; 108 if (bytes > count) 109 bytes = count; 110 111 status = pagecache_write_begin(NULL, mapping, pos, bytes, 112 AOP_FLAG_UNINTERRUPTIBLE, 113 &page, &fsdata); 114 if (status) 115 break; 116 117 zero_user(page, offset, bytes); 118 119 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, 120 page, fsdata); 121 WARN_ON(status <= 0); /* can't return less than zero! */ 122 pos += bytes; 123 count -= bytes; 124 status = 0; 125 } while (count); 126 127 return status; 128 } 129 130 int 131 xfs_update_prealloc_flags( 132 struct xfs_inode *ip, 133 enum xfs_prealloc_flags flags) 134 { 135 struct xfs_trans *tp; 136 int error; 137 138 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); 139 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); 140 if (error) { 141 xfs_trans_cancel(tp, 0); 142 return error; 143 } 144 145 xfs_ilock(ip, XFS_ILOCK_EXCL); 146 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 147 148 if (!(flags & XFS_PREALLOC_INVISIBLE)) { 149 ip->i_d.di_mode &= ~S_ISUID; 150 if (ip->i_d.di_mode & S_IXGRP) 151 ip->i_d.di_mode &= ~S_ISGID; 152 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 153 } 154 155 if (flags & XFS_PREALLOC_SET) 156 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; 157 if (flags & XFS_PREALLOC_CLEAR) 158 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; 159 160 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 161 if (flags & XFS_PREALLOC_SYNC) 162 xfs_trans_set_sync(tp); 163 return xfs_trans_commit(tp, 0); 164 } 165 166 /* 167 * Fsync operations on directories are much simpler than on regular files, 168 * as there is no file data to flush, and thus also no need for explicit 169 * cache flush operations, and there are no non-transaction metadata updates 170 * on directories either. 171 */ 172 STATIC int 173 xfs_dir_fsync( 174 struct file *file, 175 loff_t start, 176 loff_t end, 177 int datasync) 178 { 179 struct xfs_inode *ip = XFS_I(file->f_mapping->host); 180 struct xfs_mount *mp = ip->i_mount; 181 xfs_lsn_t lsn = 0; 182 183 trace_xfs_dir_fsync(ip); 184 185 xfs_ilock(ip, XFS_ILOCK_SHARED); 186 if (xfs_ipincount(ip)) 187 lsn = ip->i_itemp->ili_last_lsn; 188 xfs_iunlock(ip, XFS_ILOCK_SHARED); 189 190 if (!lsn) 191 return 0; 192 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); 193 } 194 195 STATIC int 196 xfs_file_fsync( 197 struct file *file, 198 loff_t start, 199 loff_t end, 200 int datasync) 201 { 202 struct inode *inode = file->f_mapping->host; 203 struct xfs_inode *ip = XFS_I(inode); 204 struct xfs_mount *mp = ip->i_mount; 205 int error = 0; 206 int log_flushed = 0; 207 xfs_lsn_t lsn = 0; 208 209 trace_xfs_file_fsync(ip); 210 211 error = filemap_write_and_wait_range(inode->i_mapping, start, end); 212 if (error) 213 return error; 214 215 if (XFS_FORCED_SHUTDOWN(mp)) 216 return -EIO; 217 218 xfs_iflags_clear(ip, XFS_ITRUNCATED); 219 220 if (mp->m_flags & XFS_MOUNT_BARRIER) { 221 /* 222 * If we have an RT and/or log subvolume we need to make sure 223 * to flush the write cache the device used for file data 224 * first. This is to ensure newly written file data make 225 * it to disk before logging the new inode size in case of 226 * an extending write. 227 */ 228 if (XFS_IS_REALTIME_INODE(ip)) 229 xfs_blkdev_issue_flush(mp->m_rtdev_targp); 230 else if (mp->m_logdev_targp != mp->m_ddev_targp) 231 xfs_blkdev_issue_flush(mp->m_ddev_targp); 232 } 233 234 /* 235 * All metadata updates are logged, which means that we just have 236 * to flush the log up to the latest LSN that touched the inode. 237 */ 238 xfs_ilock(ip, XFS_ILOCK_SHARED); 239 if (xfs_ipincount(ip)) { 240 if (!datasync || 241 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) 242 lsn = ip->i_itemp->ili_last_lsn; 243 } 244 xfs_iunlock(ip, XFS_ILOCK_SHARED); 245 246 if (lsn) 247 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); 248 249 /* 250 * If we only have a single device, and the log force about was 251 * a no-op we might have to flush the data device cache here. 252 * This can only happen for fdatasync/O_DSYNC if we were overwriting 253 * an already allocated file and thus do not have any metadata to 254 * commit. 255 */ 256 if ((mp->m_flags & XFS_MOUNT_BARRIER) && 257 mp->m_logdev_targp == mp->m_ddev_targp && 258 !XFS_IS_REALTIME_INODE(ip) && 259 !log_flushed) 260 xfs_blkdev_issue_flush(mp->m_ddev_targp); 261 262 return error; 263 } 264 265 STATIC ssize_t 266 xfs_file_read_iter( 267 struct kiocb *iocb, 268 struct iov_iter *to) 269 { 270 struct file *file = iocb->ki_filp; 271 struct inode *inode = file->f_mapping->host; 272 struct xfs_inode *ip = XFS_I(inode); 273 struct xfs_mount *mp = ip->i_mount; 274 size_t size = iov_iter_count(to); 275 ssize_t ret = 0; 276 int ioflags = 0; 277 xfs_fsize_t n; 278 loff_t pos = iocb->ki_pos; 279 280 XFS_STATS_INC(xs_read_calls); 281 282 if (unlikely(iocb->ki_flags & IOCB_DIRECT)) 283 ioflags |= XFS_IO_ISDIRECT; 284 if (file->f_mode & FMODE_NOCMTIME) 285 ioflags |= XFS_IO_INVIS; 286 287 if (unlikely(ioflags & XFS_IO_ISDIRECT)) { 288 xfs_buftarg_t *target = 289 XFS_IS_REALTIME_INODE(ip) ? 290 mp->m_rtdev_targp : mp->m_ddev_targp; 291 /* DIO must be aligned to device logical sector size */ 292 if ((pos | size) & target->bt_logical_sectormask) { 293 if (pos == i_size_read(inode)) 294 return 0; 295 return -EINVAL; 296 } 297 } 298 299 n = mp->m_super->s_maxbytes - pos; 300 if (n <= 0 || size == 0) 301 return 0; 302 303 if (n < size) 304 size = n; 305 306 if (XFS_FORCED_SHUTDOWN(mp)) 307 return -EIO; 308 309 /* 310 * Locking is a bit tricky here. If we take an exclusive lock 311 * for direct IO, we effectively serialise all new concurrent 312 * read IO to this file and block it behind IO that is currently in 313 * progress because IO in progress holds the IO lock shared. We only 314 * need to hold the lock exclusive to blow away the page cache, so 315 * only take lock exclusively if the page cache needs invalidation. 316 * This allows the normal direct IO case of no page cache pages to 317 * proceeed concurrently without serialisation. 318 */ 319 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); 320 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { 321 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); 322 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); 323 324 if (inode->i_mapping->nrpages) { 325 ret = filemap_write_and_wait_range( 326 VFS_I(ip)->i_mapping, 327 pos, pos + size - 1); 328 if (ret) { 329 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); 330 return ret; 331 } 332 333 /* 334 * Invalidate whole pages. This can return an error if 335 * we fail to invalidate a page, but this should never 336 * happen on XFS. Warn if it does fail. 337 */ 338 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, 339 pos >> PAGE_CACHE_SHIFT, 340 (pos + size - 1) >> PAGE_CACHE_SHIFT); 341 WARN_ON_ONCE(ret); 342 ret = 0; 343 } 344 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); 345 } 346 347 trace_xfs_file_read(ip, size, pos, ioflags); 348 349 ret = generic_file_read_iter(iocb, to); 350 if (ret > 0) 351 XFS_STATS_ADD(xs_read_bytes, ret); 352 353 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); 354 return ret; 355 } 356 357 STATIC ssize_t 358 xfs_file_splice_read( 359 struct file *infilp, 360 loff_t *ppos, 361 struct pipe_inode_info *pipe, 362 size_t count, 363 unsigned int flags) 364 { 365 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); 366 int ioflags = 0; 367 ssize_t ret; 368 369 XFS_STATS_INC(xs_read_calls); 370 371 if (infilp->f_mode & FMODE_NOCMTIME) 372 ioflags |= XFS_IO_INVIS; 373 374 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) 375 return -EIO; 376 377 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); 378 379 trace_xfs_file_splice_read(ip, count, *ppos, ioflags); 380 381 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); 382 if (ret > 0) 383 XFS_STATS_ADD(xs_read_bytes, ret); 384 385 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); 386 return ret; 387 } 388 389 /* 390 * This routine is called to handle zeroing any space in the last block of the 391 * file that is beyond the EOF. We do this since the size is being increased 392 * without writing anything to that block and we don't want to read the 393 * garbage on the disk. 394 */ 395 STATIC int /* error (positive) */ 396 xfs_zero_last_block( 397 struct xfs_inode *ip, 398 xfs_fsize_t offset, 399 xfs_fsize_t isize, 400 bool *did_zeroing) 401 { 402 struct xfs_mount *mp = ip->i_mount; 403 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); 404 int zero_offset = XFS_B_FSB_OFFSET(mp, isize); 405 int zero_len; 406 int nimaps = 1; 407 int error = 0; 408 struct xfs_bmbt_irec imap; 409 410 xfs_ilock(ip, XFS_ILOCK_EXCL); 411 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); 412 xfs_iunlock(ip, XFS_ILOCK_EXCL); 413 if (error) 414 return error; 415 416 ASSERT(nimaps > 0); 417 418 /* 419 * If the block underlying isize is just a hole, then there 420 * is nothing to zero. 421 */ 422 if (imap.br_startblock == HOLESTARTBLOCK) 423 return 0; 424 425 zero_len = mp->m_sb.sb_blocksize - zero_offset; 426 if (isize + zero_len > offset) 427 zero_len = offset - isize; 428 *did_zeroing = true; 429 return xfs_iozero(ip, isize, zero_len); 430 } 431 432 /* 433 * Zero any on disk space between the current EOF and the new, larger EOF. 434 * 435 * This handles the normal case of zeroing the remainder of the last block in 436 * the file and the unusual case of zeroing blocks out beyond the size of the 437 * file. This second case only happens with fixed size extents and when the 438 * system crashes before the inode size was updated but after blocks were 439 * allocated. 440 * 441 * Expects the iolock to be held exclusive, and will take the ilock internally. 442 */ 443 int /* error (positive) */ 444 xfs_zero_eof( 445 struct xfs_inode *ip, 446 xfs_off_t offset, /* starting I/O offset */ 447 xfs_fsize_t isize, /* current inode size */ 448 bool *did_zeroing) 449 { 450 struct xfs_mount *mp = ip->i_mount; 451 xfs_fileoff_t start_zero_fsb; 452 xfs_fileoff_t end_zero_fsb; 453 xfs_fileoff_t zero_count_fsb; 454 xfs_fileoff_t last_fsb; 455 xfs_fileoff_t zero_off; 456 xfs_fsize_t zero_len; 457 int nimaps; 458 int error = 0; 459 struct xfs_bmbt_irec imap; 460 461 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 462 ASSERT(offset > isize); 463 464 /* 465 * First handle zeroing the block on which isize resides. 466 * 467 * We only zero a part of that block so it is handled specially. 468 */ 469 if (XFS_B_FSB_OFFSET(mp, isize) != 0) { 470 error = xfs_zero_last_block(ip, offset, isize, did_zeroing); 471 if (error) 472 return error; 473 } 474 475 /* 476 * Calculate the range between the new size and the old where blocks 477 * needing to be zeroed may exist. 478 * 479 * To get the block where the last byte in the file currently resides, 480 * we need to subtract one from the size and truncate back to a block 481 * boundary. We subtract 1 in case the size is exactly on a block 482 * boundary. 483 */ 484 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; 485 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); 486 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); 487 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); 488 if (last_fsb == end_zero_fsb) { 489 /* 490 * The size was only incremented on its last block. 491 * We took care of that above, so just return. 492 */ 493 return 0; 494 } 495 496 ASSERT(start_zero_fsb <= end_zero_fsb); 497 while (start_zero_fsb <= end_zero_fsb) { 498 nimaps = 1; 499 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; 500 501 xfs_ilock(ip, XFS_ILOCK_EXCL); 502 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, 503 &imap, &nimaps, 0); 504 xfs_iunlock(ip, XFS_ILOCK_EXCL); 505 if (error) 506 return error; 507 508 ASSERT(nimaps > 0); 509 510 if (imap.br_state == XFS_EXT_UNWRITTEN || 511 imap.br_startblock == HOLESTARTBLOCK) { 512 start_zero_fsb = imap.br_startoff + imap.br_blockcount; 513 ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); 514 continue; 515 } 516 517 /* 518 * There are blocks we need to zero. 519 */ 520 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); 521 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); 522 523 if ((zero_off + zero_len) > offset) 524 zero_len = offset - zero_off; 525 526 error = xfs_iozero(ip, zero_off, zero_len); 527 if (error) 528 return error; 529 530 *did_zeroing = true; 531 start_zero_fsb = imap.br_startoff + imap.br_blockcount; 532 ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); 533 } 534 535 return 0; 536 } 537 538 /* 539 * Common pre-write limit and setup checks. 540 * 541 * Called with the iolocked held either shared and exclusive according to 542 * @iolock, and returns with it held. Might upgrade the iolock to exclusive 543 * if called for a direct write beyond i_size. 544 */ 545 STATIC ssize_t 546 xfs_file_aio_write_checks( 547 struct kiocb *iocb, 548 struct iov_iter *from, 549 int *iolock) 550 { 551 struct file *file = iocb->ki_filp; 552 struct inode *inode = file->f_mapping->host; 553 struct xfs_inode *ip = XFS_I(inode); 554 ssize_t error = 0; 555 size_t count = iov_iter_count(from); 556 557 restart: 558 error = generic_write_checks(iocb, from); 559 if (error <= 0) 560 return error; 561 562 error = xfs_break_layouts(inode, iolock, true); 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 * We need to serialise against EOF updates that occur in IO 574 * completions here. We want to make sure that nobody is changing the 575 * size while we do this check until we have placed an IO barrier (i.e. 576 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. 577 * The spinlock effectively forms a memory barrier once we have the 578 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value 579 * and hence be able to correctly determine if we need to run zeroing. 580 */ 581 spin_lock(&ip->i_flags_lock); 582 if (iocb->ki_pos > i_size_read(inode)) { 583 bool zero = false; 584 585 spin_unlock(&ip->i_flags_lock); 586 if (*iolock == XFS_IOLOCK_SHARED) { 587 xfs_rw_iunlock(ip, *iolock); 588 *iolock = XFS_IOLOCK_EXCL; 589 xfs_rw_ilock(ip, *iolock); 590 iov_iter_reexpand(from, count); 591 592 /* 593 * We now have an IO submission barrier in place, but 594 * AIO can do EOF updates during IO completion and hence 595 * we now need to wait for all of them to drain. Non-AIO 596 * DIO will have drained before we are given the 597 * XFS_IOLOCK_EXCL, and so for most cases this wait is a 598 * no-op. 599 */ 600 inode_dio_wait(inode); 601 goto restart; 602 } 603 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); 604 if (error) 605 return error; 606 } else 607 spin_unlock(&ip->i_flags_lock); 608 609 /* 610 * Updating the timestamps will grab the ilock again from 611 * xfs_fs_dirty_inode, so we have to call it after dropping the 612 * lock above. Eventually we should look into a way to avoid 613 * the pointless lock roundtrip. 614 */ 615 if (likely(!(file->f_mode & FMODE_NOCMTIME))) { 616 error = file_update_time(file); 617 if (error) 618 return error; 619 } 620 621 /* 622 * If we're writing the file then make sure to clear the setuid and 623 * setgid bits if the process is not being run by root. This keeps 624 * people from modifying setuid and setgid binaries. 625 */ 626 return file_remove_suid(file); 627 } 628 629 /* 630 * xfs_file_dio_aio_write - handle direct IO writes 631 * 632 * Lock the inode appropriately to prepare for and issue a direct IO write. 633 * By separating it from the buffered write path we remove all the tricky to 634 * follow locking changes and looping. 635 * 636 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL 637 * until we're sure the bytes at the new EOF have been zeroed and/or the cached 638 * pages are flushed out. 639 * 640 * In most cases the direct IO writes will be done holding IOLOCK_SHARED 641 * allowing them to be done in parallel with reads and other direct IO writes. 642 * However, if the IO is not aligned to filesystem blocks, the direct IO layer 643 * needs to do sub-block zeroing and that requires serialisation against other 644 * direct IOs to the same block. In this case we need to serialise the 645 * submission of the unaligned IOs so that we don't get racing block zeroing in 646 * the dio layer. To avoid the problem with aio, we also need to wait for 647 * outstanding IOs to complete so that unwritten extent conversion is completed 648 * before we try to map the overlapping block. This is currently implemented by 649 * hitting it with a big hammer (i.e. inode_dio_wait()). 650 * 651 * Returns with locks held indicated by @iolock and errors indicated by 652 * negative return values. 653 */ 654 STATIC ssize_t 655 xfs_file_dio_aio_write( 656 struct kiocb *iocb, 657 struct iov_iter *from) 658 { 659 struct file *file = iocb->ki_filp; 660 struct address_space *mapping = file->f_mapping; 661 struct inode *inode = mapping->host; 662 struct xfs_inode *ip = XFS_I(inode); 663 struct xfs_mount *mp = ip->i_mount; 664 ssize_t ret = 0; 665 int unaligned_io = 0; 666 int iolock; 667 size_t count = iov_iter_count(from); 668 loff_t pos = iocb->ki_pos; 669 loff_t end; 670 struct iov_iter data; 671 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? 672 mp->m_rtdev_targp : mp->m_ddev_targp; 673 674 /* DIO must be aligned to device logical sector size */ 675 if ((pos | count) & target->bt_logical_sectormask) 676 return -EINVAL; 677 678 /* "unaligned" here means not aligned to a filesystem block */ 679 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) 680 unaligned_io = 1; 681 682 /* 683 * We don't need to take an exclusive lock unless there page cache needs 684 * to be invalidated or unaligned IO is being executed. We don't need to 685 * consider the EOF extension case here because 686 * xfs_file_aio_write_checks() will relock the inode as necessary for 687 * EOF zeroing cases and fill out the new inode size as appropriate. 688 */ 689 if (unaligned_io || mapping->nrpages) 690 iolock = XFS_IOLOCK_EXCL; 691 else 692 iolock = XFS_IOLOCK_SHARED; 693 xfs_rw_ilock(ip, iolock); 694 695 /* 696 * Recheck if there are cached pages that need invalidate after we got 697 * the iolock to protect against other threads adding new pages while 698 * we were waiting for the iolock. 699 */ 700 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { 701 xfs_rw_iunlock(ip, iolock); 702 iolock = XFS_IOLOCK_EXCL; 703 xfs_rw_ilock(ip, iolock); 704 } 705 706 ret = xfs_file_aio_write_checks(iocb, from, &iolock); 707 if (ret) 708 goto out; 709 count = iov_iter_count(from); 710 pos = iocb->ki_pos; 711 end = pos + count - 1; 712 713 if (mapping->nrpages) { 714 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 715 pos, end); 716 if (ret) 717 goto out; 718 /* 719 * Invalidate whole pages. This can return an error if 720 * we fail to invalidate a page, but this should never 721 * happen on XFS. Warn if it does fail. 722 */ 723 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, 724 pos >> PAGE_CACHE_SHIFT, 725 end >> PAGE_CACHE_SHIFT); 726 WARN_ON_ONCE(ret); 727 ret = 0; 728 } 729 730 /* 731 * If we are doing unaligned IO, wait for all other IO to drain, 732 * otherwise demote the lock if we had to flush cached pages 733 */ 734 if (unaligned_io) 735 inode_dio_wait(inode); 736 else if (iolock == XFS_IOLOCK_EXCL) { 737 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); 738 iolock = XFS_IOLOCK_SHARED; 739 } 740 741 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); 742 743 data = *from; 744 ret = mapping->a_ops->direct_IO(iocb, &data, pos); 745 746 /* see generic_file_direct_write() for why this is necessary */ 747 if (mapping->nrpages) { 748 invalidate_inode_pages2_range(mapping, 749 pos >> PAGE_CACHE_SHIFT, 750 end >> PAGE_CACHE_SHIFT); 751 } 752 753 if (ret > 0) { 754 pos += ret; 755 iov_iter_advance(from, ret); 756 iocb->ki_pos = pos; 757 } 758 out: 759 xfs_rw_iunlock(ip, iolock); 760 761 /* No fallback to buffered IO on errors for XFS. */ 762 ASSERT(ret < 0 || ret == count); 763 return ret; 764 } 765 766 STATIC ssize_t 767 xfs_file_buffered_aio_write( 768 struct kiocb *iocb, 769 struct iov_iter *from) 770 { 771 struct file *file = iocb->ki_filp; 772 struct address_space *mapping = file->f_mapping; 773 struct inode *inode = mapping->host; 774 struct xfs_inode *ip = XFS_I(inode); 775 ssize_t ret; 776 int enospc = 0; 777 int iolock = XFS_IOLOCK_EXCL; 778 779 xfs_rw_ilock(ip, iolock); 780 781 ret = xfs_file_aio_write_checks(iocb, from, &iolock); 782 if (ret) 783 goto out; 784 785 /* We can write back this queue in page reclaim */ 786 current->backing_dev_info = inode_to_bdi(inode); 787 788 write_retry: 789 trace_xfs_file_buffered_write(ip, iov_iter_count(from), 790 iocb->ki_pos, 0); 791 ret = generic_perform_write(file, from, iocb->ki_pos); 792 if (likely(ret >= 0)) 793 iocb->ki_pos += ret; 794 795 /* 796 * If we hit a space limit, try to free up some lingering preallocated 797 * space before returning an error. In the case of ENOSPC, first try to 798 * write back all dirty inodes to free up some of the excess reserved 799 * metadata space. This reduces the chances that the eofblocks scan 800 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this 801 * also behaves as a filter to prevent too many eofblocks scans from 802 * running at the same time. 803 */ 804 if (ret == -EDQUOT && !enospc) { 805 enospc = xfs_inode_free_quota_eofblocks(ip); 806 if (enospc) 807 goto write_retry; 808 } else if (ret == -ENOSPC && !enospc) { 809 struct xfs_eofblocks eofb = {0}; 810 811 enospc = 1; 812 xfs_flush_inodes(ip->i_mount); 813 eofb.eof_scan_owner = ip->i_ino; /* for locking */ 814 eofb.eof_flags = XFS_EOF_FLAGS_SYNC; 815 xfs_icache_free_eofblocks(ip->i_mount, &eofb); 816 goto write_retry; 817 } 818 819 current->backing_dev_info = NULL; 820 out: 821 xfs_rw_iunlock(ip, iolock); 822 return ret; 823 } 824 825 STATIC ssize_t 826 xfs_file_write_iter( 827 struct kiocb *iocb, 828 struct iov_iter *from) 829 { 830 struct file *file = iocb->ki_filp; 831 struct address_space *mapping = file->f_mapping; 832 struct inode *inode = mapping->host; 833 struct xfs_inode *ip = XFS_I(inode); 834 ssize_t ret; 835 size_t ocount = iov_iter_count(from); 836 837 XFS_STATS_INC(xs_write_calls); 838 839 if (ocount == 0) 840 return 0; 841 842 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) 843 return -EIO; 844 845 if (unlikely(iocb->ki_flags & IOCB_DIRECT)) 846 ret = xfs_file_dio_aio_write(iocb, from); 847 else 848 ret = xfs_file_buffered_aio_write(iocb, from); 849 850 if (ret > 0) { 851 ssize_t err; 852 853 XFS_STATS_ADD(xs_write_bytes, ret); 854 855 /* Handle various SYNC-type writes */ 856 err = generic_write_sync(file, iocb->ki_pos - ret, ret); 857 if (err < 0) 858 ret = err; 859 } 860 return ret; 861 } 862 863 #define XFS_FALLOC_FL_SUPPORTED \ 864 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ 865 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ 866 FALLOC_FL_INSERT_RANGE) 867 868 STATIC long 869 xfs_file_fallocate( 870 struct file *file, 871 int mode, 872 loff_t offset, 873 loff_t len) 874 { 875 struct inode *inode = file_inode(file); 876 struct xfs_inode *ip = XFS_I(inode); 877 long error; 878 enum xfs_prealloc_flags flags = 0; 879 uint iolock = XFS_IOLOCK_EXCL; 880 loff_t new_size = 0; 881 bool do_file_insert = 0; 882 883 if (!S_ISREG(inode->i_mode)) 884 return -EINVAL; 885 if (mode & ~XFS_FALLOC_FL_SUPPORTED) 886 return -EOPNOTSUPP; 887 888 xfs_ilock(ip, iolock); 889 error = xfs_break_layouts(inode, &iolock, false); 890 if (error) 891 goto out_unlock; 892 893 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 894 iolock |= XFS_MMAPLOCK_EXCL; 895 896 if (mode & FALLOC_FL_PUNCH_HOLE) { 897 error = xfs_free_file_space(ip, offset, len); 898 if (error) 899 goto out_unlock; 900 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { 901 unsigned blksize_mask = (1 << inode->i_blkbits) - 1; 902 903 if (offset & blksize_mask || len & blksize_mask) { 904 error = -EINVAL; 905 goto out_unlock; 906 } 907 908 /* 909 * There is no need to overlap collapse range with EOF, 910 * in which case it is effectively a truncate operation 911 */ 912 if (offset + len >= i_size_read(inode)) { 913 error = -EINVAL; 914 goto out_unlock; 915 } 916 917 new_size = i_size_read(inode) - len; 918 919 error = xfs_collapse_file_space(ip, offset, len); 920 if (error) 921 goto out_unlock; 922 } else if (mode & FALLOC_FL_INSERT_RANGE) { 923 unsigned blksize_mask = (1 << inode->i_blkbits) - 1; 924 925 new_size = i_size_read(inode) + len; 926 if (offset & blksize_mask || len & blksize_mask) { 927 error = -EINVAL; 928 goto out_unlock; 929 } 930 931 /* check the new inode size does not wrap through zero */ 932 if (new_size > inode->i_sb->s_maxbytes) { 933 error = -EFBIG; 934 goto out_unlock; 935 } 936 937 /* Offset should be less than i_size */ 938 if (offset >= i_size_read(inode)) { 939 error = -EINVAL; 940 goto out_unlock; 941 } 942 do_file_insert = 1; 943 } else { 944 flags |= XFS_PREALLOC_SET; 945 946 if (!(mode & FALLOC_FL_KEEP_SIZE) && 947 offset + len > i_size_read(inode)) { 948 new_size = offset + len; 949 error = inode_newsize_ok(inode, new_size); 950 if (error) 951 goto out_unlock; 952 } 953 954 if (mode & FALLOC_FL_ZERO_RANGE) 955 error = xfs_zero_file_space(ip, offset, len); 956 else 957 error = xfs_alloc_file_space(ip, offset, len, 958 XFS_BMAPI_PREALLOC); 959 if (error) 960 goto out_unlock; 961 } 962 963 if (file->f_flags & O_DSYNC) 964 flags |= XFS_PREALLOC_SYNC; 965 966 error = xfs_update_prealloc_flags(ip, flags); 967 if (error) 968 goto out_unlock; 969 970 /* Change file size if needed */ 971 if (new_size) { 972 struct iattr iattr; 973 974 iattr.ia_valid = ATTR_SIZE; 975 iattr.ia_size = new_size; 976 error = xfs_setattr_size(ip, &iattr); 977 if (error) 978 goto out_unlock; 979 } 980 981 /* 982 * Perform hole insertion now that the file size has been 983 * updated so that if we crash during the operation we don't 984 * leave shifted extents past EOF and hence losing access to 985 * the data that is contained within them. 986 */ 987 if (do_file_insert) 988 error = xfs_insert_file_space(ip, offset, len); 989 990 out_unlock: 991 xfs_iunlock(ip, iolock); 992 return error; 993 } 994 995 996 STATIC int 997 xfs_file_open( 998 struct inode *inode, 999 struct file *file) 1000 { 1001 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) 1002 return -EFBIG; 1003 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) 1004 return -EIO; 1005 return 0; 1006 } 1007 1008 STATIC int 1009 xfs_dir_open( 1010 struct inode *inode, 1011 struct file *file) 1012 { 1013 struct xfs_inode *ip = XFS_I(inode); 1014 int mode; 1015 int error; 1016 1017 error = xfs_file_open(inode, file); 1018 if (error) 1019 return error; 1020 1021 /* 1022 * If there are any blocks, read-ahead block 0 as we're almost 1023 * certain to have the next operation be a read there. 1024 */ 1025 mode = xfs_ilock_data_map_shared(ip); 1026 if (ip->i_d.di_nextents > 0) 1027 xfs_dir3_data_readahead(ip, 0, -1); 1028 xfs_iunlock(ip, mode); 1029 return 0; 1030 } 1031 1032 STATIC int 1033 xfs_file_release( 1034 struct inode *inode, 1035 struct file *filp) 1036 { 1037 return xfs_release(XFS_I(inode)); 1038 } 1039 1040 STATIC int 1041 xfs_file_readdir( 1042 struct file *file, 1043 struct dir_context *ctx) 1044 { 1045 struct inode *inode = file_inode(file); 1046 xfs_inode_t *ip = XFS_I(inode); 1047 size_t bufsize; 1048 1049 /* 1050 * The Linux API doesn't pass down the total size of the buffer 1051 * we read into down to the filesystem. With the filldir concept 1052 * it's not needed for correct information, but the XFS dir2 leaf 1053 * code wants an estimate of the buffer size to calculate it's 1054 * readahead window and size the buffers used for mapping to 1055 * physical blocks. 1056 * 1057 * Try to give it an estimate that's good enough, maybe at some 1058 * point we can change the ->readdir prototype to include the 1059 * buffer size. For now we use the current glibc buffer size. 1060 */ 1061 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); 1062 1063 return xfs_readdir(ip, ctx, bufsize); 1064 } 1065 1066 STATIC int 1067 xfs_file_mmap( 1068 struct file *filp, 1069 struct vm_area_struct *vma) 1070 { 1071 vma->vm_ops = &xfs_file_vm_ops; 1072 1073 file_accessed(filp); 1074 return 0; 1075 } 1076 1077 /* 1078 * This type is designed to indicate the type of offset we would like 1079 * to search from page cache for xfs_seek_hole_data(). 1080 */ 1081 enum { 1082 HOLE_OFF = 0, 1083 DATA_OFF, 1084 }; 1085 1086 /* 1087 * Lookup the desired type of offset from the given page. 1088 * 1089 * On success, return true and the offset argument will point to the 1090 * start of the region that was found. Otherwise this function will 1091 * return false and keep the offset argument unchanged. 1092 */ 1093 STATIC bool 1094 xfs_lookup_buffer_offset( 1095 struct page *page, 1096 loff_t *offset, 1097 unsigned int type) 1098 { 1099 loff_t lastoff = page_offset(page); 1100 bool found = false; 1101 struct buffer_head *bh, *head; 1102 1103 bh = head = page_buffers(page); 1104 do { 1105 /* 1106 * Unwritten extents that have data in the page 1107 * cache covering them can be identified by the 1108 * BH_Unwritten state flag. Pages with multiple 1109 * buffers might have a mix of holes, data and 1110 * unwritten extents - any buffer with valid 1111 * data in it should have BH_Uptodate flag set 1112 * on it. 1113 */ 1114 if (buffer_unwritten(bh) || 1115 buffer_uptodate(bh)) { 1116 if (type == DATA_OFF) 1117 found = true; 1118 } else { 1119 if (type == HOLE_OFF) 1120 found = true; 1121 } 1122 1123 if (found) { 1124 *offset = lastoff; 1125 break; 1126 } 1127 lastoff += bh->b_size; 1128 } while ((bh = bh->b_this_page) != head); 1129 1130 return found; 1131 } 1132 1133 /* 1134 * This routine is called to find out and return a data or hole offset 1135 * from the page cache for unwritten extents according to the desired 1136 * type for xfs_seek_hole_data(). 1137 * 1138 * The argument offset is used to tell where we start to search from the 1139 * page cache. Map is used to figure out the end points of the range to 1140 * lookup pages. 1141 * 1142 * Return true if the desired type of offset was found, and the argument 1143 * offset is filled with that address. Otherwise, return false and keep 1144 * offset unchanged. 1145 */ 1146 STATIC bool 1147 xfs_find_get_desired_pgoff( 1148 struct inode *inode, 1149 struct xfs_bmbt_irec *map, 1150 unsigned int type, 1151 loff_t *offset) 1152 { 1153 struct xfs_inode *ip = XFS_I(inode); 1154 struct xfs_mount *mp = ip->i_mount; 1155 struct pagevec pvec; 1156 pgoff_t index; 1157 pgoff_t end; 1158 loff_t endoff; 1159 loff_t startoff = *offset; 1160 loff_t lastoff = startoff; 1161 bool found = false; 1162 1163 pagevec_init(&pvec, 0); 1164 1165 index = startoff >> PAGE_CACHE_SHIFT; 1166 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); 1167 end = endoff >> PAGE_CACHE_SHIFT; 1168 do { 1169 int want; 1170 unsigned nr_pages; 1171 unsigned int i; 1172 1173 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); 1174 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, 1175 want); 1176 /* 1177 * No page mapped into given range. If we are searching holes 1178 * and if this is the first time we got into the loop, it means 1179 * that the given offset is landed in a hole, return it. 1180 * 1181 * If we have already stepped through some block buffers to find 1182 * holes but they all contains data. In this case, the last 1183 * offset is already updated and pointed to the end of the last 1184 * mapped page, if it does not reach the endpoint to search, 1185 * that means there should be a hole between them. 1186 */ 1187 if (nr_pages == 0) { 1188 /* Data search found nothing */ 1189 if (type == DATA_OFF) 1190 break; 1191 1192 ASSERT(type == HOLE_OFF); 1193 if (lastoff == startoff || lastoff < endoff) { 1194 found = true; 1195 *offset = lastoff; 1196 } 1197 break; 1198 } 1199 1200 /* 1201 * At lease we found one page. If this is the first time we 1202 * step into the loop, and if the first page index offset is 1203 * greater than the given search offset, a hole was found. 1204 */ 1205 if (type == HOLE_OFF && lastoff == startoff && 1206 lastoff < page_offset(pvec.pages[0])) { 1207 found = true; 1208 break; 1209 } 1210 1211 for (i = 0; i < nr_pages; i++) { 1212 struct page *page = pvec.pages[i]; 1213 loff_t b_offset; 1214 1215 /* 1216 * At this point, the page may be truncated or 1217 * invalidated (changing page->mapping to NULL), 1218 * or even swizzled back from swapper_space to tmpfs 1219 * file mapping. However, page->index will not change 1220 * because we have a reference on the page. 1221 * 1222 * Searching done if the page index is out of range. 1223 * If the current offset is not reaches the end of 1224 * the specified search range, there should be a hole 1225 * between them. 1226 */ 1227 if (page->index > end) { 1228 if (type == HOLE_OFF && lastoff < endoff) { 1229 *offset = lastoff; 1230 found = true; 1231 } 1232 goto out; 1233 } 1234 1235 lock_page(page); 1236 /* 1237 * Page truncated or invalidated(page->mapping == NULL). 1238 * We can freely skip it and proceed to check the next 1239 * page. 1240 */ 1241 if (unlikely(page->mapping != inode->i_mapping)) { 1242 unlock_page(page); 1243 continue; 1244 } 1245 1246 if (!page_has_buffers(page)) { 1247 unlock_page(page); 1248 continue; 1249 } 1250 1251 found = xfs_lookup_buffer_offset(page, &b_offset, type); 1252 if (found) { 1253 /* 1254 * The found offset may be less than the start 1255 * point to search if this is the first time to 1256 * come here. 1257 */ 1258 *offset = max_t(loff_t, startoff, b_offset); 1259 unlock_page(page); 1260 goto out; 1261 } 1262 1263 /* 1264 * We either searching data but nothing was found, or 1265 * searching hole but found a data buffer. In either 1266 * case, probably the next page contains the desired 1267 * things, update the last offset to it so. 1268 */ 1269 lastoff = page_offset(page) + PAGE_SIZE; 1270 unlock_page(page); 1271 } 1272 1273 /* 1274 * The number of returned pages less than our desired, search 1275 * done. In this case, nothing was found for searching data, 1276 * but we found a hole behind the last offset. 1277 */ 1278 if (nr_pages < want) { 1279 if (type == HOLE_OFF) { 1280 *offset = lastoff; 1281 found = true; 1282 } 1283 break; 1284 } 1285 1286 index = pvec.pages[i - 1]->index + 1; 1287 pagevec_release(&pvec); 1288 } while (index <= end); 1289 1290 out: 1291 pagevec_release(&pvec); 1292 return found; 1293 } 1294 1295 STATIC loff_t 1296 xfs_seek_hole_data( 1297 struct file *file, 1298 loff_t start, 1299 int whence) 1300 { 1301 struct inode *inode = file->f_mapping->host; 1302 struct xfs_inode *ip = XFS_I(inode); 1303 struct xfs_mount *mp = ip->i_mount; 1304 loff_t uninitialized_var(offset); 1305 xfs_fsize_t isize; 1306 xfs_fileoff_t fsbno; 1307 xfs_filblks_t end; 1308 uint lock; 1309 int error; 1310 1311 if (XFS_FORCED_SHUTDOWN(mp)) 1312 return -EIO; 1313 1314 lock = xfs_ilock_data_map_shared(ip); 1315 1316 isize = i_size_read(inode); 1317 if (start >= isize) { 1318 error = -ENXIO; 1319 goto out_unlock; 1320 } 1321 1322 /* 1323 * Try to read extents from the first block indicated 1324 * by fsbno to the end block of the file. 1325 */ 1326 fsbno = XFS_B_TO_FSBT(mp, start); 1327 end = XFS_B_TO_FSB(mp, isize); 1328 1329 for (;;) { 1330 struct xfs_bmbt_irec map[2]; 1331 int nmap = 2; 1332 unsigned int i; 1333 1334 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, 1335 XFS_BMAPI_ENTIRE); 1336 if (error) 1337 goto out_unlock; 1338 1339 /* No extents at given offset, must be beyond EOF */ 1340 if (nmap == 0) { 1341 error = -ENXIO; 1342 goto out_unlock; 1343 } 1344 1345 for (i = 0; i < nmap; i++) { 1346 offset = max_t(loff_t, start, 1347 XFS_FSB_TO_B(mp, map[i].br_startoff)); 1348 1349 /* Landed in the hole we wanted? */ 1350 if (whence == SEEK_HOLE && 1351 map[i].br_startblock == HOLESTARTBLOCK) 1352 goto out; 1353 1354 /* Landed in the data extent we wanted? */ 1355 if (whence == SEEK_DATA && 1356 (map[i].br_startblock == DELAYSTARTBLOCK || 1357 (map[i].br_state == XFS_EXT_NORM && 1358 !isnullstartblock(map[i].br_startblock)))) 1359 goto out; 1360 1361 /* 1362 * Landed in an unwritten extent, try to search 1363 * for hole or data from page cache. 1364 */ 1365 if (map[i].br_state == XFS_EXT_UNWRITTEN) { 1366 if (xfs_find_get_desired_pgoff(inode, &map[i], 1367 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, 1368 &offset)) 1369 goto out; 1370 } 1371 } 1372 1373 /* 1374 * We only received one extent out of the two requested. This 1375 * means we've hit EOF and didn't find what we are looking for. 1376 */ 1377 if (nmap == 1) { 1378 /* 1379 * If we were looking for a hole, set offset to 1380 * the end of the file (i.e., there is an implicit 1381 * hole at the end of any file). 1382 */ 1383 if (whence == SEEK_HOLE) { 1384 offset = isize; 1385 break; 1386 } 1387 /* 1388 * If we were looking for data, it's nowhere to be found 1389 */ 1390 ASSERT(whence == SEEK_DATA); 1391 error = -ENXIO; 1392 goto out_unlock; 1393 } 1394 1395 ASSERT(i > 1); 1396 1397 /* 1398 * Nothing was found, proceed to the next round of search 1399 * if the next reading offset is not at or beyond EOF. 1400 */ 1401 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; 1402 start = XFS_FSB_TO_B(mp, fsbno); 1403 if (start >= isize) { 1404 if (whence == SEEK_HOLE) { 1405 offset = isize; 1406 break; 1407 } 1408 ASSERT(whence == SEEK_DATA); 1409 error = -ENXIO; 1410 goto out_unlock; 1411 } 1412 } 1413 1414 out: 1415 /* 1416 * If at this point we have found the hole we wanted, the returned 1417 * offset may be bigger than the file size as it may be aligned to 1418 * page boundary for unwritten extents. We need to deal with this 1419 * situation in particular. 1420 */ 1421 if (whence == SEEK_HOLE) 1422 offset = min_t(loff_t, offset, isize); 1423 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); 1424 1425 out_unlock: 1426 xfs_iunlock(ip, lock); 1427 1428 if (error) 1429 return error; 1430 return offset; 1431 } 1432 1433 STATIC loff_t 1434 xfs_file_llseek( 1435 struct file *file, 1436 loff_t offset, 1437 int whence) 1438 { 1439 switch (whence) { 1440 case SEEK_END: 1441 case SEEK_CUR: 1442 case SEEK_SET: 1443 return generic_file_llseek(file, offset, whence); 1444 case SEEK_HOLE: 1445 case SEEK_DATA: 1446 return xfs_seek_hole_data(file, offset, whence); 1447 default: 1448 return -EINVAL; 1449 } 1450 } 1451 1452 /* 1453 * Locking for serialisation of IO during page faults. This results in a lock 1454 * ordering of: 1455 * 1456 * mmap_sem (MM) 1457 * i_mmap_lock (XFS - truncate serialisation) 1458 * page_lock (MM) 1459 * i_lock (XFS - extent map serialisation) 1460 */ 1461 STATIC int 1462 xfs_filemap_fault( 1463 struct vm_area_struct *vma, 1464 struct vm_fault *vmf) 1465 { 1466 struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host); 1467 int error; 1468 1469 trace_xfs_filemap_fault(ip); 1470 1471 xfs_ilock(ip, XFS_MMAPLOCK_SHARED); 1472 error = filemap_fault(vma, vmf); 1473 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); 1474 1475 return error; 1476 } 1477 1478 /* 1479 * mmap()d file has taken write protection fault and is being made writable. We 1480 * can set the page state up correctly for a writable page, which means we can 1481 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent 1482 * mapping. 1483 */ 1484 STATIC int 1485 xfs_filemap_page_mkwrite( 1486 struct vm_area_struct *vma, 1487 struct vm_fault *vmf) 1488 { 1489 struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host); 1490 int error; 1491 1492 trace_xfs_filemap_page_mkwrite(ip); 1493 1494 xfs_ilock(ip, XFS_MMAPLOCK_SHARED); 1495 error = block_page_mkwrite(vma, vmf, xfs_get_blocks); 1496 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); 1497 1498 return error; 1499 } 1500 1501 const struct file_operations xfs_file_operations = { 1502 .llseek = xfs_file_llseek, 1503 .read_iter = xfs_file_read_iter, 1504 .write_iter = xfs_file_write_iter, 1505 .splice_read = xfs_file_splice_read, 1506 .splice_write = iter_file_splice_write, 1507 .unlocked_ioctl = xfs_file_ioctl, 1508 #ifdef CONFIG_COMPAT 1509 .compat_ioctl = xfs_file_compat_ioctl, 1510 #endif 1511 .mmap = xfs_file_mmap, 1512 .open = xfs_file_open, 1513 .release = xfs_file_release, 1514 .fsync = xfs_file_fsync, 1515 .fallocate = xfs_file_fallocate, 1516 }; 1517 1518 const struct file_operations xfs_dir_file_operations = { 1519 .open = xfs_dir_open, 1520 .read = generic_read_dir, 1521 .iterate = xfs_file_readdir, 1522 .llseek = generic_file_llseek, 1523 .unlocked_ioctl = xfs_file_ioctl, 1524 #ifdef CONFIG_COMPAT 1525 .compat_ioctl = xfs_file_compat_ioctl, 1526 #endif 1527 .fsync = xfs_dir_fsync, 1528 }; 1529 1530 static const struct vm_operations_struct xfs_file_vm_ops = { 1531 .fault = xfs_filemap_fault, 1532 .map_pages = filemap_map_pages, 1533 .page_mkwrite = xfs_filemap_page_mkwrite, 1534 }; 1535