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 #include "xfs_iomap.h" 41 #include "xfs_reflink.h" 42 43 #include <linux/dcache.h> 44 #include <linux/falloc.h> 45 #include <linux/pagevec.h> 46 #include <linux/backing-dev.h> 47 48 static const struct vm_operations_struct xfs_file_vm_ops; 49 50 /* 51 * Clear the specified ranges to zero through either the pagecache or DAX. 52 * Holes and unwritten extents will be left as-is as they already are zeroed. 53 */ 54 int 55 xfs_zero_range( 56 struct xfs_inode *ip, 57 xfs_off_t pos, 58 xfs_off_t count, 59 bool *did_zero) 60 { 61 return iomap_zero_range(VFS_I(ip), pos, count, NULL, &xfs_iomap_ops); 62 } 63 64 int 65 xfs_update_prealloc_flags( 66 struct xfs_inode *ip, 67 enum xfs_prealloc_flags flags) 68 { 69 struct xfs_trans *tp; 70 int error; 71 72 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid, 73 0, 0, 0, &tp); 74 if (error) 75 return error; 76 77 xfs_ilock(ip, XFS_ILOCK_EXCL); 78 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 79 80 if (!(flags & XFS_PREALLOC_INVISIBLE)) { 81 VFS_I(ip)->i_mode &= ~S_ISUID; 82 if (VFS_I(ip)->i_mode & S_IXGRP) 83 VFS_I(ip)->i_mode &= ~S_ISGID; 84 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 85 } 86 87 if (flags & XFS_PREALLOC_SET) 88 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; 89 if (flags & XFS_PREALLOC_CLEAR) 90 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; 91 92 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 93 if (flags & XFS_PREALLOC_SYNC) 94 xfs_trans_set_sync(tp); 95 return xfs_trans_commit(tp); 96 } 97 98 /* 99 * Fsync operations on directories are much simpler than on regular files, 100 * as there is no file data to flush, and thus also no need for explicit 101 * cache flush operations, and there are no non-transaction metadata updates 102 * on directories either. 103 */ 104 STATIC int 105 xfs_dir_fsync( 106 struct file *file, 107 loff_t start, 108 loff_t end, 109 int datasync) 110 { 111 struct xfs_inode *ip = XFS_I(file->f_mapping->host); 112 struct xfs_mount *mp = ip->i_mount; 113 xfs_lsn_t lsn = 0; 114 115 trace_xfs_dir_fsync(ip); 116 117 xfs_ilock(ip, XFS_ILOCK_SHARED); 118 if (xfs_ipincount(ip)) 119 lsn = ip->i_itemp->ili_last_lsn; 120 xfs_iunlock(ip, XFS_ILOCK_SHARED); 121 122 if (!lsn) 123 return 0; 124 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); 125 } 126 127 STATIC int 128 xfs_file_fsync( 129 struct file *file, 130 loff_t start, 131 loff_t end, 132 int datasync) 133 { 134 struct inode *inode = file->f_mapping->host; 135 struct xfs_inode *ip = XFS_I(inode); 136 struct xfs_mount *mp = ip->i_mount; 137 int error = 0; 138 int log_flushed = 0; 139 xfs_lsn_t lsn = 0; 140 141 trace_xfs_file_fsync(ip); 142 143 error = file_write_and_wait_range(file, start, end); 144 if (error) 145 return error; 146 147 if (XFS_FORCED_SHUTDOWN(mp)) 148 return -EIO; 149 150 xfs_iflags_clear(ip, XFS_ITRUNCATED); 151 152 /* 153 * If we have an RT and/or log subvolume we need to make sure to flush 154 * the write cache the device used for file data first. This is to 155 * ensure newly written file data make it to disk before logging the new 156 * inode size in case of an extending write. 157 */ 158 if (XFS_IS_REALTIME_INODE(ip)) 159 xfs_blkdev_issue_flush(mp->m_rtdev_targp); 160 else if (mp->m_logdev_targp != mp->m_ddev_targp) 161 xfs_blkdev_issue_flush(mp->m_ddev_targp); 162 163 /* 164 * All metadata updates are logged, which means that we just have to 165 * flush the log up to the latest LSN that touched the inode. If we have 166 * concurrent fsync/fdatasync() calls, we need them to all block on the 167 * log force before we clear the ili_fsync_fields field. This ensures 168 * that we don't get a racing sync operation that does not wait for the 169 * metadata to hit the journal before returning. If we race with 170 * clearing the ili_fsync_fields, then all that will happen is the log 171 * force will do nothing as the lsn will already be on disk. We can't 172 * race with setting ili_fsync_fields because that is done under 173 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared 174 * until after the ili_fsync_fields is cleared. 175 */ 176 xfs_ilock(ip, XFS_ILOCK_SHARED); 177 if (xfs_ipincount(ip)) { 178 if (!datasync || 179 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) 180 lsn = ip->i_itemp->ili_last_lsn; 181 } 182 183 if (lsn) { 184 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); 185 ip->i_itemp->ili_fsync_fields = 0; 186 } 187 xfs_iunlock(ip, XFS_ILOCK_SHARED); 188 189 /* 190 * If we only have a single device, and the log force about was 191 * a no-op we might have to flush the data device cache here. 192 * This can only happen for fdatasync/O_DSYNC if we were overwriting 193 * an already allocated file and thus do not have any metadata to 194 * commit. 195 */ 196 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) && 197 mp->m_logdev_targp == mp->m_ddev_targp) 198 xfs_blkdev_issue_flush(mp->m_ddev_targp); 199 200 return error; 201 } 202 203 STATIC ssize_t 204 xfs_file_dio_aio_read( 205 struct kiocb *iocb, 206 struct iov_iter *to) 207 { 208 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); 209 size_t count = iov_iter_count(to); 210 ssize_t ret; 211 212 trace_xfs_file_direct_read(ip, count, iocb->ki_pos); 213 214 if (!count) 215 return 0; /* skip atime */ 216 217 file_accessed(iocb->ki_filp); 218 219 xfs_ilock(ip, XFS_IOLOCK_SHARED); 220 ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL); 221 xfs_iunlock(ip, XFS_IOLOCK_SHARED); 222 223 return ret; 224 } 225 226 static noinline ssize_t 227 xfs_file_dax_read( 228 struct kiocb *iocb, 229 struct iov_iter *to) 230 { 231 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host); 232 size_t count = iov_iter_count(to); 233 ssize_t ret = 0; 234 235 trace_xfs_file_dax_read(ip, count, iocb->ki_pos); 236 237 if (!count) 238 return 0; /* skip atime */ 239 240 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { 241 if (iocb->ki_flags & IOCB_NOWAIT) 242 return -EAGAIN; 243 xfs_ilock(ip, XFS_IOLOCK_SHARED); 244 } 245 ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops); 246 xfs_iunlock(ip, XFS_IOLOCK_SHARED); 247 248 file_accessed(iocb->ki_filp); 249 return ret; 250 } 251 252 STATIC ssize_t 253 xfs_file_buffered_aio_read( 254 struct kiocb *iocb, 255 struct iov_iter *to) 256 { 257 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); 258 ssize_t ret; 259 260 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos); 261 262 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { 263 if (iocb->ki_flags & IOCB_NOWAIT) 264 return -EAGAIN; 265 xfs_ilock(ip, XFS_IOLOCK_SHARED); 266 } 267 ret = generic_file_read_iter(iocb, to); 268 xfs_iunlock(ip, XFS_IOLOCK_SHARED); 269 270 return ret; 271 } 272 273 STATIC ssize_t 274 xfs_file_read_iter( 275 struct kiocb *iocb, 276 struct iov_iter *to) 277 { 278 struct inode *inode = file_inode(iocb->ki_filp); 279 struct xfs_mount *mp = XFS_I(inode)->i_mount; 280 ssize_t ret = 0; 281 282 XFS_STATS_INC(mp, xs_read_calls); 283 284 if (XFS_FORCED_SHUTDOWN(mp)) 285 return -EIO; 286 287 if (IS_DAX(inode)) 288 ret = xfs_file_dax_read(iocb, to); 289 else if (iocb->ki_flags & IOCB_DIRECT) 290 ret = xfs_file_dio_aio_read(iocb, to); 291 else 292 ret = xfs_file_buffered_aio_read(iocb, to); 293 294 if (ret > 0) 295 XFS_STATS_ADD(mp, xs_read_bytes, ret); 296 return ret; 297 } 298 299 /* 300 * Zero any on disk space between the current EOF and the new, larger EOF. 301 * 302 * This handles the normal case of zeroing the remainder of the last block in 303 * the file and the unusual case of zeroing blocks out beyond the size of the 304 * file. This second case only happens with fixed size extents and when the 305 * system crashes before the inode size was updated but after blocks were 306 * allocated. 307 * 308 * Expects the iolock to be held exclusive, and will take the ilock internally. 309 */ 310 int /* error (positive) */ 311 xfs_zero_eof( 312 struct xfs_inode *ip, 313 xfs_off_t offset, /* starting I/O offset */ 314 xfs_fsize_t isize, /* current inode size */ 315 bool *did_zeroing) 316 { 317 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 318 ASSERT(offset > isize); 319 320 trace_xfs_zero_eof(ip, isize, offset - isize); 321 return xfs_zero_range(ip, isize, offset - isize, did_zeroing); 322 } 323 324 /* 325 * Common pre-write limit and setup checks. 326 * 327 * Called with the iolocked held either shared and exclusive according to 328 * @iolock, and returns with it held. Might upgrade the iolock to exclusive 329 * if called for a direct write beyond i_size. 330 */ 331 STATIC ssize_t 332 xfs_file_aio_write_checks( 333 struct kiocb *iocb, 334 struct iov_iter *from, 335 int *iolock) 336 { 337 struct file *file = iocb->ki_filp; 338 struct inode *inode = file->f_mapping->host; 339 struct xfs_inode *ip = XFS_I(inode); 340 ssize_t error = 0; 341 size_t count = iov_iter_count(from); 342 bool drained_dio = false; 343 344 restart: 345 error = generic_write_checks(iocb, from); 346 if (error <= 0) 347 return error; 348 349 error = xfs_break_layouts(inode, iolock); 350 if (error) 351 return error; 352 353 /* 354 * For changing security info in file_remove_privs() we need i_rwsem 355 * exclusively. 356 */ 357 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { 358 xfs_iunlock(ip, *iolock); 359 *iolock = XFS_IOLOCK_EXCL; 360 xfs_ilock(ip, *iolock); 361 goto restart; 362 } 363 /* 364 * If the offset is beyond the size of the file, we need to zero any 365 * blocks that fall between the existing EOF and the start of this 366 * write. If zeroing is needed and we are currently holding the 367 * iolock shared, we need to update it to exclusive which implies 368 * having to redo all checks before. 369 * 370 * We need to serialise against EOF updates that occur in IO 371 * completions here. We want to make sure that nobody is changing the 372 * size while we do this check until we have placed an IO barrier (i.e. 373 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. 374 * The spinlock effectively forms a memory barrier once we have the 375 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value 376 * and hence be able to correctly determine if we need to run zeroing. 377 */ 378 spin_lock(&ip->i_flags_lock); 379 if (iocb->ki_pos > i_size_read(inode)) { 380 bool zero = false; 381 382 spin_unlock(&ip->i_flags_lock); 383 if (!drained_dio) { 384 if (*iolock == XFS_IOLOCK_SHARED) { 385 xfs_iunlock(ip, *iolock); 386 *iolock = XFS_IOLOCK_EXCL; 387 xfs_ilock(ip, *iolock); 388 iov_iter_reexpand(from, count); 389 } 390 /* 391 * We now have an IO submission barrier in place, but 392 * AIO can do EOF updates during IO completion and hence 393 * we now need to wait for all of them to drain. Non-AIO 394 * DIO will have drained before we are given the 395 * XFS_IOLOCK_EXCL, and so for most cases this wait is a 396 * no-op. 397 */ 398 inode_dio_wait(inode); 399 drained_dio = true; 400 goto restart; 401 } 402 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); 403 if (error) 404 return error; 405 } else 406 spin_unlock(&ip->i_flags_lock); 407 408 /* 409 * Updating the timestamps will grab the ilock again from 410 * xfs_fs_dirty_inode, so we have to call it after dropping the 411 * lock above. Eventually we should look into a way to avoid 412 * the pointless lock roundtrip. 413 */ 414 if (likely(!(file->f_mode & FMODE_NOCMTIME))) { 415 error = file_update_time(file); 416 if (error) 417 return error; 418 } 419 420 /* 421 * If we're writing the file then make sure to clear the setuid and 422 * setgid bits if the process is not being run by root. This keeps 423 * people from modifying setuid and setgid binaries. 424 */ 425 if (!IS_NOSEC(inode)) 426 return file_remove_privs(file); 427 return 0; 428 } 429 430 static int 431 xfs_dio_write_end_io( 432 struct kiocb *iocb, 433 ssize_t size, 434 unsigned flags) 435 { 436 struct inode *inode = file_inode(iocb->ki_filp); 437 struct xfs_inode *ip = XFS_I(inode); 438 loff_t offset = iocb->ki_pos; 439 bool update_size = false; 440 int error = 0; 441 442 trace_xfs_end_io_direct_write(ip, offset, size); 443 444 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) 445 return -EIO; 446 447 if (size <= 0) 448 return size; 449 450 /* 451 * We need to update the in-core inode size here so that we don't end up 452 * with the on-disk inode size being outside the in-core inode size. We 453 * have no other method of updating EOF for AIO, so always do it here 454 * if necessary. 455 * 456 * We need to lock the test/set EOF update as we can be racing with 457 * other IO completions here to update the EOF. Failing to serialise 458 * here can result in EOF moving backwards and Bad Things Happen when 459 * that occurs. 460 */ 461 spin_lock(&ip->i_flags_lock); 462 if (offset + size > i_size_read(inode)) { 463 i_size_write(inode, offset + size); 464 update_size = true; 465 } 466 spin_unlock(&ip->i_flags_lock); 467 468 if (flags & IOMAP_DIO_COW) { 469 error = xfs_reflink_end_cow(ip, offset, size); 470 if (error) 471 return error; 472 } 473 474 if (flags & IOMAP_DIO_UNWRITTEN) 475 error = xfs_iomap_write_unwritten(ip, offset, size); 476 else if (update_size) 477 error = xfs_setfilesize(ip, offset, size); 478 479 return error; 480 } 481 482 /* 483 * xfs_file_dio_aio_write - handle direct IO writes 484 * 485 * Lock the inode appropriately to prepare for and issue a direct IO write. 486 * By separating it from the buffered write path we remove all the tricky to 487 * follow locking changes and looping. 488 * 489 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL 490 * until we're sure the bytes at the new EOF have been zeroed and/or the cached 491 * pages are flushed out. 492 * 493 * In most cases the direct IO writes will be done holding IOLOCK_SHARED 494 * allowing them to be done in parallel with reads and other direct IO writes. 495 * However, if the IO is not aligned to filesystem blocks, the direct IO layer 496 * needs to do sub-block zeroing and that requires serialisation against other 497 * direct IOs to the same block. In this case we need to serialise the 498 * submission of the unaligned IOs so that we don't get racing block zeroing in 499 * the dio layer. To avoid the problem with aio, we also need to wait for 500 * outstanding IOs to complete so that unwritten extent conversion is completed 501 * before we try to map the overlapping block. This is currently implemented by 502 * hitting it with a big hammer (i.e. inode_dio_wait()). 503 * 504 * Returns with locks held indicated by @iolock and errors indicated by 505 * negative return values. 506 */ 507 STATIC ssize_t 508 xfs_file_dio_aio_write( 509 struct kiocb *iocb, 510 struct iov_iter *from) 511 { 512 struct file *file = iocb->ki_filp; 513 struct address_space *mapping = file->f_mapping; 514 struct inode *inode = mapping->host; 515 struct xfs_inode *ip = XFS_I(inode); 516 struct xfs_mount *mp = ip->i_mount; 517 ssize_t ret = 0; 518 int unaligned_io = 0; 519 int iolock; 520 size_t count = iov_iter_count(from); 521 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? 522 mp->m_rtdev_targp : mp->m_ddev_targp; 523 524 /* DIO must be aligned to device logical sector size */ 525 if ((iocb->ki_pos | count) & target->bt_logical_sectormask) 526 return -EINVAL; 527 528 /* 529 * Don't take the exclusive iolock here unless the I/O is unaligned to 530 * the file system block size. We don't need to consider the EOF 531 * extension case here because xfs_file_aio_write_checks() will relock 532 * the inode as necessary for EOF zeroing cases and fill out the new 533 * inode size as appropriate. 534 */ 535 if ((iocb->ki_pos & mp->m_blockmask) || 536 ((iocb->ki_pos + count) & mp->m_blockmask)) { 537 unaligned_io = 1; 538 539 /* 540 * We can't properly handle unaligned direct I/O to reflink 541 * files yet, as we can't unshare a partial block. 542 */ 543 if (xfs_is_reflink_inode(ip)) { 544 trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count); 545 return -EREMCHG; 546 } 547 iolock = XFS_IOLOCK_EXCL; 548 } else { 549 iolock = XFS_IOLOCK_SHARED; 550 } 551 552 if (!xfs_ilock_nowait(ip, iolock)) { 553 if (iocb->ki_flags & IOCB_NOWAIT) 554 return -EAGAIN; 555 xfs_ilock(ip, iolock); 556 } 557 558 ret = xfs_file_aio_write_checks(iocb, from, &iolock); 559 if (ret) 560 goto out; 561 count = iov_iter_count(from); 562 563 /* 564 * If we are doing unaligned IO, wait for all other IO to drain, 565 * otherwise demote the lock if we had to take the exclusive lock 566 * for other reasons in xfs_file_aio_write_checks. 567 */ 568 if (unaligned_io) { 569 /* If we are going to wait for other DIO to finish, bail */ 570 if (iocb->ki_flags & IOCB_NOWAIT) { 571 if (atomic_read(&inode->i_dio_count)) 572 return -EAGAIN; 573 } else { 574 inode_dio_wait(inode); 575 } 576 } else if (iolock == XFS_IOLOCK_EXCL) { 577 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL); 578 iolock = XFS_IOLOCK_SHARED; 579 } 580 581 trace_xfs_file_direct_write(ip, count, iocb->ki_pos); 582 ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io); 583 out: 584 xfs_iunlock(ip, iolock); 585 586 /* 587 * No fallback to buffered IO on errors for XFS, direct IO will either 588 * complete fully or fail. 589 */ 590 ASSERT(ret < 0 || ret == count); 591 return ret; 592 } 593 594 static noinline ssize_t 595 xfs_file_dax_write( 596 struct kiocb *iocb, 597 struct iov_iter *from) 598 { 599 struct inode *inode = iocb->ki_filp->f_mapping->host; 600 struct xfs_inode *ip = XFS_I(inode); 601 int iolock = XFS_IOLOCK_EXCL; 602 ssize_t ret, error = 0; 603 size_t count; 604 loff_t pos; 605 606 if (!xfs_ilock_nowait(ip, iolock)) { 607 if (iocb->ki_flags & IOCB_NOWAIT) 608 return -EAGAIN; 609 xfs_ilock(ip, iolock); 610 } 611 612 ret = xfs_file_aio_write_checks(iocb, from, &iolock); 613 if (ret) 614 goto out; 615 616 pos = iocb->ki_pos; 617 count = iov_iter_count(from); 618 619 trace_xfs_file_dax_write(ip, count, pos); 620 ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops); 621 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) { 622 i_size_write(inode, iocb->ki_pos); 623 error = xfs_setfilesize(ip, pos, ret); 624 } 625 out: 626 xfs_iunlock(ip, iolock); 627 return error ? error : ret; 628 } 629 630 STATIC ssize_t 631 xfs_file_buffered_aio_write( 632 struct kiocb *iocb, 633 struct iov_iter *from) 634 { 635 struct file *file = iocb->ki_filp; 636 struct address_space *mapping = file->f_mapping; 637 struct inode *inode = mapping->host; 638 struct xfs_inode *ip = XFS_I(inode); 639 ssize_t ret; 640 int enospc = 0; 641 int iolock; 642 643 if (iocb->ki_flags & IOCB_NOWAIT) 644 return -EOPNOTSUPP; 645 646 write_retry: 647 iolock = XFS_IOLOCK_EXCL; 648 xfs_ilock(ip, iolock); 649 650 ret = xfs_file_aio_write_checks(iocb, from, &iolock); 651 if (ret) 652 goto out; 653 654 /* We can write back this queue in page reclaim */ 655 current->backing_dev_info = inode_to_bdi(inode); 656 657 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos); 658 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops); 659 if (likely(ret >= 0)) 660 iocb->ki_pos += ret; 661 662 /* 663 * If we hit a space limit, try to free up some lingering preallocated 664 * space before returning an error. In the case of ENOSPC, first try to 665 * write back all dirty inodes to free up some of the excess reserved 666 * metadata space. This reduces the chances that the eofblocks scan 667 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this 668 * also behaves as a filter to prevent too many eofblocks scans from 669 * running at the same time. 670 */ 671 if (ret == -EDQUOT && !enospc) { 672 xfs_iunlock(ip, iolock); 673 enospc = xfs_inode_free_quota_eofblocks(ip); 674 if (enospc) 675 goto write_retry; 676 enospc = xfs_inode_free_quota_cowblocks(ip); 677 if (enospc) 678 goto write_retry; 679 iolock = 0; 680 } else if (ret == -ENOSPC && !enospc) { 681 struct xfs_eofblocks eofb = {0}; 682 683 enospc = 1; 684 xfs_flush_inodes(ip->i_mount); 685 686 xfs_iunlock(ip, iolock); 687 eofb.eof_flags = XFS_EOF_FLAGS_SYNC; 688 xfs_icache_free_eofblocks(ip->i_mount, &eofb); 689 xfs_icache_free_cowblocks(ip->i_mount, &eofb); 690 goto write_retry; 691 } 692 693 current->backing_dev_info = NULL; 694 out: 695 if (iolock) 696 xfs_iunlock(ip, iolock); 697 return ret; 698 } 699 700 STATIC ssize_t 701 xfs_file_write_iter( 702 struct kiocb *iocb, 703 struct iov_iter *from) 704 { 705 struct file *file = iocb->ki_filp; 706 struct address_space *mapping = file->f_mapping; 707 struct inode *inode = mapping->host; 708 struct xfs_inode *ip = XFS_I(inode); 709 ssize_t ret; 710 size_t ocount = iov_iter_count(from); 711 712 XFS_STATS_INC(ip->i_mount, xs_write_calls); 713 714 if (ocount == 0) 715 return 0; 716 717 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) 718 return -EIO; 719 720 if (IS_DAX(inode)) 721 ret = xfs_file_dax_write(iocb, from); 722 else if (iocb->ki_flags & IOCB_DIRECT) { 723 /* 724 * Allow a directio write to fall back to a buffered 725 * write *only* in the case that we're doing a reflink 726 * CoW. In all other directio scenarios we do not 727 * allow an operation to fall back to buffered mode. 728 */ 729 ret = xfs_file_dio_aio_write(iocb, from); 730 if (ret == -EREMCHG) 731 goto buffered; 732 } else { 733 buffered: 734 ret = xfs_file_buffered_aio_write(iocb, from); 735 } 736 737 if (ret > 0) { 738 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); 739 740 /* Handle various SYNC-type writes */ 741 ret = generic_write_sync(iocb, ret); 742 } 743 return ret; 744 } 745 746 #define XFS_FALLOC_FL_SUPPORTED \ 747 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ 748 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ 749 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE) 750 751 STATIC long 752 xfs_file_fallocate( 753 struct file *file, 754 int mode, 755 loff_t offset, 756 loff_t len) 757 { 758 struct inode *inode = file_inode(file); 759 struct xfs_inode *ip = XFS_I(inode); 760 long error; 761 enum xfs_prealloc_flags flags = 0; 762 uint iolock = XFS_IOLOCK_EXCL; 763 loff_t new_size = 0; 764 bool do_file_insert = 0; 765 766 if (!S_ISREG(inode->i_mode)) 767 return -EINVAL; 768 if (mode & ~XFS_FALLOC_FL_SUPPORTED) 769 return -EOPNOTSUPP; 770 771 xfs_ilock(ip, iolock); 772 error = xfs_break_layouts(inode, &iolock); 773 if (error) 774 goto out_unlock; 775 776 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 777 iolock |= XFS_MMAPLOCK_EXCL; 778 779 if (mode & FALLOC_FL_PUNCH_HOLE) { 780 error = xfs_free_file_space(ip, offset, len); 781 if (error) 782 goto out_unlock; 783 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { 784 unsigned int blksize_mask = i_blocksize(inode) - 1; 785 786 if (offset & blksize_mask || len & blksize_mask) { 787 error = -EINVAL; 788 goto out_unlock; 789 } 790 791 /* 792 * There is no need to overlap collapse range with EOF, 793 * in which case it is effectively a truncate operation 794 */ 795 if (offset + len >= i_size_read(inode)) { 796 error = -EINVAL; 797 goto out_unlock; 798 } 799 800 new_size = i_size_read(inode) - len; 801 802 error = xfs_collapse_file_space(ip, offset, len); 803 if (error) 804 goto out_unlock; 805 } else if (mode & FALLOC_FL_INSERT_RANGE) { 806 unsigned int blksize_mask = i_blocksize(inode) - 1; 807 808 new_size = i_size_read(inode) + len; 809 if (offset & blksize_mask || len & blksize_mask) { 810 error = -EINVAL; 811 goto out_unlock; 812 } 813 814 /* check the new inode size does not wrap through zero */ 815 if (new_size > inode->i_sb->s_maxbytes) { 816 error = -EFBIG; 817 goto out_unlock; 818 } 819 820 /* Offset should be less than i_size */ 821 if (offset >= i_size_read(inode)) { 822 error = -EINVAL; 823 goto out_unlock; 824 } 825 do_file_insert = 1; 826 } else { 827 flags |= XFS_PREALLOC_SET; 828 829 if (!(mode & FALLOC_FL_KEEP_SIZE) && 830 offset + len > i_size_read(inode)) { 831 new_size = offset + len; 832 error = inode_newsize_ok(inode, new_size); 833 if (error) 834 goto out_unlock; 835 } 836 837 if (mode & FALLOC_FL_ZERO_RANGE) 838 error = xfs_zero_file_space(ip, offset, len); 839 else { 840 if (mode & FALLOC_FL_UNSHARE_RANGE) { 841 error = xfs_reflink_unshare(ip, offset, len); 842 if (error) 843 goto out_unlock; 844 } 845 error = xfs_alloc_file_space(ip, offset, len, 846 XFS_BMAPI_PREALLOC); 847 } 848 if (error) 849 goto out_unlock; 850 } 851 852 if (file->f_flags & O_DSYNC) 853 flags |= XFS_PREALLOC_SYNC; 854 855 error = xfs_update_prealloc_flags(ip, flags); 856 if (error) 857 goto out_unlock; 858 859 /* Change file size if needed */ 860 if (new_size) { 861 struct iattr iattr; 862 863 iattr.ia_valid = ATTR_SIZE; 864 iattr.ia_size = new_size; 865 error = xfs_vn_setattr_size(file_dentry(file), &iattr); 866 if (error) 867 goto out_unlock; 868 } 869 870 /* 871 * Perform hole insertion now that the file size has been 872 * updated so that if we crash during the operation we don't 873 * leave shifted extents past EOF and hence losing access to 874 * the data that is contained within them. 875 */ 876 if (do_file_insert) 877 error = xfs_insert_file_space(ip, offset, len); 878 879 out_unlock: 880 xfs_iunlock(ip, iolock); 881 return error; 882 } 883 884 STATIC int 885 xfs_file_clone_range( 886 struct file *file_in, 887 loff_t pos_in, 888 struct file *file_out, 889 loff_t pos_out, 890 u64 len) 891 { 892 return xfs_reflink_remap_range(file_in, pos_in, file_out, pos_out, 893 len, false); 894 } 895 896 STATIC ssize_t 897 xfs_file_dedupe_range( 898 struct file *src_file, 899 u64 loff, 900 u64 len, 901 struct file *dst_file, 902 u64 dst_loff) 903 { 904 int error; 905 906 error = xfs_reflink_remap_range(src_file, loff, dst_file, dst_loff, 907 len, true); 908 if (error) 909 return error; 910 return len; 911 } 912 913 STATIC int 914 xfs_file_open( 915 struct inode *inode, 916 struct file *file) 917 { 918 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) 919 return -EFBIG; 920 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) 921 return -EIO; 922 file->f_mode |= FMODE_NOWAIT; 923 return 0; 924 } 925 926 STATIC int 927 xfs_dir_open( 928 struct inode *inode, 929 struct file *file) 930 { 931 struct xfs_inode *ip = XFS_I(inode); 932 int mode; 933 int error; 934 935 error = xfs_file_open(inode, file); 936 if (error) 937 return error; 938 939 /* 940 * If there are any blocks, read-ahead block 0 as we're almost 941 * certain to have the next operation be a read there. 942 */ 943 mode = xfs_ilock_data_map_shared(ip); 944 if (ip->i_d.di_nextents > 0) 945 error = xfs_dir3_data_readahead(ip, 0, -1); 946 xfs_iunlock(ip, mode); 947 return error; 948 } 949 950 STATIC int 951 xfs_file_release( 952 struct inode *inode, 953 struct file *filp) 954 { 955 return xfs_release(XFS_I(inode)); 956 } 957 958 STATIC int 959 xfs_file_readdir( 960 struct file *file, 961 struct dir_context *ctx) 962 { 963 struct inode *inode = file_inode(file); 964 xfs_inode_t *ip = XFS_I(inode); 965 size_t bufsize; 966 967 /* 968 * The Linux API doesn't pass down the total size of the buffer 969 * we read into down to the filesystem. With the filldir concept 970 * it's not needed for correct information, but the XFS dir2 leaf 971 * code wants an estimate of the buffer size to calculate it's 972 * readahead window and size the buffers used for mapping to 973 * physical blocks. 974 * 975 * Try to give it an estimate that's good enough, maybe at some 976 * point we can change the ->readdir prototype to include the 977 * buffer size. For now we use the current glibc buffer size. 978 */ 979 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); 980 981 return xfs_readdir(NULL, ip, ctx, bufsize); 982 } 983 984 STATIC loff_t 985 xfs_file_llseek( 986 struct file *file, 987 loff_t offset, 988 int whence) 989 { 990 struct inode *inode = file->f_mapping->host; 991 992 if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount)) 993 return -EIO; 994 995 switch (whence) { 996 default: 997 return generic_file_llseek(file, offset, whence); 998 case SEEK_HOLE: 999 offset = iomap_seek_hole(inode, offset, &xfs_iomap_ops); 1000 break; 1001 case SEEK_DATA: 1002 offset = iomap_seek_data(inode, offset, &xfs_iomap_ops); 1003 break; 1004 } 1005 1006 if (offset < 0) 1007 return offset; 1008 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); 1009 } 1010 1011 /* 1012 * Locking for serialisation of IO during page faults. This results in a lock 1013 * ordering of: 1014 * 1015 * mmap_sem (MM) 1016 * sb_start_pagefault(vfs, freeze) 1017 * i_mmaplock (XFS - truncate serialisation) 1018 * page_lock (MM) 1019 * i_lock (XFS - extent map serialisation) 1020 */ 1021 static int 1022 __xfs_filemap_fault( 1023 struct vm_fault *vmf, 1024 enum page_entry_size pe_size, 1025 bool write_fault) 1026 { 1027 struct inode *inode = file_inode(vmf->vma->vm_file); 1028 struct xfs_inode *ip = XFS_I(inode); 1029 int ret; 1030 1031 trace_xfs_filemap_fault(ip, pe_size, write_fault); 1032 1033 if (write_fault) { 1034 sb_start_pagefault(inode->i_sb); 1035 file_update_time(vmf->vma->vm_file); 1036 } 1037 1038 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); 1039 if (IS_DAX(inode)) { 1040 ret = dax_iomap_fault(vmf, pe_size, &xfs_iomap_ops); 1041 } else { 1042 if (write_fault) 1043 ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops); 1044 else 1045 ret = filemap_fault(vmf); 1046 } 1047 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); 1048 1049 if (write_fault) 1050 sb_end_pagefault(inode->i_sb); 1051 return ret; 1052 } 1053 1054 static int 1055 xfs_filemap_fault( 1056 struct vm_fault *vmf) 1057 { 1058 /* DAX can shortcut the normal fault path on write faults! */ 1059 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, 1060 IS_DAX(file_inode(vmf->vma->vm_file)) && 1061 (vmf->flags & FAULT_FLAG_WRITE)); 1062 } 1063 1064 static int 1065 xfs_filemap_huge_fault( 1066 struct vm_fault *vmf, 1067 enum page_entry_size pe_size) 1068 { 1069 if (!IS_DAX(file_inode(vmf->vma->vm_file))) 1070 return VM_FAULT_FALLBACK; 1071 1072 /* DAX can shortcut the normal fault path on write faults! */ 1073 return __xfs_filemap_fault(vmf, pe_size, 1074 (vmf->flags & FAULT_FLAG_WRITE)); 1075 } 1076 1077 static int 1078 xfs_filemap_page_mkwrite( 1079 struct vm_fault *vmf) 1080 { 1081 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true); 1082 } 1083 1084 /* 1085 * pfn_mkwrite was originally inteneded to ensure we capture time stamp 1086 * updates on write faults. In reality, it's need to serialise against 1087 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED 1088 * to ensure we serialise the fault barrier in place. 1089 */ 1090 static int 1091 xfs_filemap_pfn_mkwrite( 1092 struct vm_fault *vmf) 1093 { 1094 1095 struct inode *inode = file_inode(vmf->vma->vm_file); 1096 struct xfs_inode *ip = XFS_I(inode); 1097 int ret = VM_FAULT_NOPAGE; 1098 loff_t size; 1099 1100 trace_xfs_filemap_pfn_mkwrite(ip); 1101 1102 sb_start_pagefault(inode->i_sb); 1103 file_update_time(vmf->vma->vm_file); 1104 1105 /* check if the faulting page hasn't raced with truncate */ 1106 xfs_ilock(ip, XFS_MMAPLOCK_SHARED); 1107 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; 1108 if (vmf->pgoff >= size) 1109 ret = VM_FAULT_SIGBUS; 1110 else if (IS_DAX(inode)) 1111 ret = dax_iomap_fault(vmf, PE_SIZE_PTE, &xfs_iomap_ops); 1112 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); 1113 sb_end_pagefault(inode->i_sb); 1114 return ret; 1115 1116 } 1117 1118 static const struct vm_operations_struct xfs_file_vm_ops = { 1119 .fault = xfs_filemap_fault, 1120 .huge_fault = xfs_filemap_huge_fault, 1121 .map_pages = filemap_map_pages, 1122 .page_mkwrite = xfs_filemap_page_mkwrite, 1123 .pfn_mkwrite = xfs_filemap_pfn_mkwrite, 1124 }; 1125 1126 STATIC int 1127 xfs_file_mmap( 1128 struct file *filp, 1129 struct vm_area_struct *vma) 1130 { 1131 file_accessed(filp); 1132 vma->vm_ops = &xfs_file_vm_ops; 1133 if (IS_DAX(file_inode(filp))) 1134 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; 1135 return 0; 1136 } 1137 1138 const struct file_operations xfs_file_operations = { 1139 .llseek = xfs_file_llseek, 1140 .read_iter = xfs_file_read_iter, 1141 .write_iter = xfs_file_write_iter, 1142 .splice_read = generic_file_splice_read, 1143 .splice_write = iter_file_splice_write, 1144 .unlocked_ioctl = xfs_file_ioctl, 1145 #ifdef CONFIG_COMPAT 1146 .compat_ioctl = xfs_file_compat_ioctl, 1147 #endif 1148 .mmap = xfs_file_mmap, 1149 .open = xfs_file_open, 1150 .release = xfs_file_release, 1151 .fsync = xfs_file_fsync, 1152 .get_unmapped_area = thp_get_unmapped_area, 1153 .fallocate = xfs_file_fallocate, 1154 .clone_file_range = xfs_file_clone_range, 1155 .dedupe_file_range = xfs_file_dedupe_range, 1156 }; 1157 1158 const struct file_operations xfs_dir_file_operations = { 1159 .open = xfs_dir_open, 1160 .read = generic_read_dir, 1161 .iterate_shared = xfs_file_readdir, 1162 .llseek = generic_file_llseek, 1163 .unlocked_ioctl = xfs_file_ioctl, 1164 #ifdef CONFIG_COMPAT 1165 .compat_ioctl = xfs_file_compat_ioctl, 1166 #endif 1167 .fsync = xfs_dir_fsync, 1168 }; 1169