개념 설명 전체 · v6.6 / fs/ext4/file.c
1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/file.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card ([email protected]) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/file.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * ext4 fs regular file handling primitives 17 * 18 * 64-bit file support on 64-bit platforms by Jakub Jelinek 19 * ([email protected]) 20 */ 21 22 #include <linux/time.h> 23 #include <linux/fs.h> 24 #include <linux/iomap.h> 25 #include <linux/mount.h> 26 #include <linux/path.h> 27 #include <linux/dax.h> 28 #include <linux/quotaops.h> 29 #include <linux/pagevec.h> 30 #include <linux/uio.h> 31 #include <linux/mman.h> 32 #include <linux/backing-dev.h> 33 #include "ext4.h" 34 #include "ext4_jbd2.h" 35 #include "xattr.h" 36 #include "acl.h" 37 #include "truncate.h" 38 39 /* 40 * Returns %true if the given DIO request should be attempted with DIO, or 41 * %false if it should fall back to buffered I/O. 42 * 43 * DIO isn't well specified; when it's unsupported (either due to the request 44 * being misaligned, or due to the file not supporting DIO at all), filesystems 45 * either fall back to buffered I/O or return EINVAL. For files that don't use 46 * any special features like encryption or verity, ext4 has traditionally 47 * returned EINVAL for misaligned DIO. iomap_dio_rw() uses this convention too. 48 * In this case, we should attempt the DIO, *not* fall back to buffered I/O. 49 * 50 * In contrast, in cases where DIO is unsupported due to ext4 features, ext4 51 * traditionally falls back to buffered I/O. 52 * 53 * This function implements the traditional ext4 behavior in all these cases. 54 */ 55 static bool ext4_should_use_dio(struct kiocb *iocb, struct iov_iter *iter) 56 { 57 struct inode *inode = file_inode(iocb->ki_filp); 58 u32 dio_align = ext4_dio_alignment(inode); 59 60 if (dio_align == 0) 61 return false; 62 63 if (dio_align == 1) 64 return true; 65 66 return IS_ALIGNED(iocb->ki_pos | iov_iter_alignment(iter), dio_align); 67 } 68 69 static ssize_t ext4_dio_read_iter(struct kiocb *iocb, struct iov_iter *to) 70 { 71 ssize_t ret; 72 struct inode *inode = file_inode(iocb->ki_filp); 73 74 if (iocb->ki_flags & IOCB_NOWAIT) { 75 if (!inode_trylock_shared(inode)) 76 return -EAGAIN; 77 } else { 78 inode_lock_shared(inode); 79 } 80 81 if (!ext4_should_use_dio(iocb, to)) { 82 inode_unlock_shared(inode); 83 /* 84 * Fallback to buffered I/O if the operation being performed on 85 * the inode is not supported by direct I/O. The IOCB_DIRECT 86 * flag needs to be cleared here in order to ensure that the 87 * direct I/O path within generic_file_read_iter() is not 88 * taken. 89 */ 90 iocb->ki_flags &= ~IOCB_DIRECT; 91 return generic_file_read_iter(iocb, to); 92 } 93 94 ret = iomap_dio_rw(iocb, to, &ext4_iomap_ops, NULL, 0, NULL, 0); 95 inode_unlock_shared(inode); 96 97 file_accessed(iocb->ki_filp); 98 return ret; 99 } 100 101 #ifdef CONFIG_FS_DAX 102 static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to) 103 { 104 struct inode *inode = file_inode(iocb->ki_filp); 105 ssize_t ret; 106 107 if (iocb->ki_flags & IOCB_NOWAIT) { 108 if (!inode_trylock_shared(inode)) 109 return -EAGAIN; 110 } else { 111 inode_lock_shared(inode); 112 } 113 /* 114 * Recheck under inode lock - at this point we are sure it cannot 115 * change anymore 116 */ 117 if (!IS_DAX(inode)) { 118 inode_unlock_shared(inode); 119 /* Fallback to buffered IO in case we cannot support DAX */ 120 return generic_file_read_iter(iocb, to); 121 } 122 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops); 123 inode_unlock_shared(inode); 124 125 file_accessed(iocb->ki_filp); 126 return ret; 127 } 128 #endif 129 130 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 131 { 132 struct inode *inode = file_inode(iocb->ki_filp); 133 134 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 135 return -EIO; 136 137 if (!iov_iter_count(to)) 138 return 0; /* skip atime */ 139 140 #ifdef CONFIG_FS_DAX 141 if (IS_DAX(inode)) 142 return ext4_dax_read_iter(iocb, to); 143 #endif 144 if (iocb->ki_flags & IOCB_DIRECT) 145 return ext4_dio_read_iter(iocb, to); 146 147 return generic_file_read_iter(iocb, to); 148 } 149 150 static ssize_t ext4_file_splice_read(struct file *in, loff_t *ppos, 151 struct pipe_inode_info *pipe, 152 size_t len, unsigned int flags) 153 { 154 struct inode *inode = file_inode(in); 155 156 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 157 return -EIO; 158 return filemap_splice_read(in, ppos, pipe, len, flags); 159 } 160 161 /* 162 * Called when an inode is released. Note that this is different 163 * from ext4_file_open: open gets called at every open, but release 164 * gets called only when /all/ the files are closed. 165 */ 166 static int ext4_release_file(struct inode *inode, struct file *filp) 167 { 168 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) { 169 ext4_alloc_da_blocks(inode); 170 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); 171 } 172 /* if we are the last writer on the inode, drop the block reservation */ 173 if ((filp->f_mode & FMODE_WRITE) && 174 (atomic_read(&inode->i_writecount) == 1) && 175 !EXT4_I(inode)->i_reserved_data_blocks) { 176 down_write(&EXT4_I(inode)->i_data_sem); 177 ext4_discard_preallocations(inode, 0); 178 up_write(&EXT4_I(inode)->i_data_sem); 179 } 180 if (is_dx(inode) && filp->private_data) 181 ext4_htree_free_dir_info(filp->private_data); 182 183 return 0; 184 } 185 186 /* 187 * This tests whether the IO in question is block-aligned or not. 188 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they 189 * are converted to written only after the IO is complete. Until they are 190 * mapped, these blocks appear as holes, so dio_zero_block() will assume that 191 * it needs to zero out portions of the start and/or end block. If 2 AIO 192 * threads are at work on the same unwritten block, they must be synchronized 193 * or one thread will zero the other's data, causing corruption. 194 */ 195 static bool 196 ext4_unaligned_io(struct inode *inode, struct iov_iter *from, loff_t pos) 197 { 198 struct super_block *sb = inode->i_sb; 199 unsigned long blockmask = sb->s_blocksize - 1; 200 201 if ((pos | iov_iter_alignment(from)) & blockmask) 202 return true; 203 204 return false; 205 } 206 207 static bool 208 ext4_extending_io(struct inode *inode, loff_t offset, size_t len) 209 { 210 if (offset + len > i_size_read(inode) || 211 offset + len > EXT4_I(inode)->i_disksize) 212 return true; 213 return false; 214 } 215 216 /* Is IO overwriting allocated or initialized blocks? */ 217 static bool ext4_overwrite_io(struct inode *inode, 218 loff_t pos, loff_t len, bool *unwritten) 219 { 220 struct ext4_map_blocks map; 221 unsigned int blkbits = inode->i_blkbits; 222 int err, blklen; 223 224 if (pos + len > i_size_read(inode)) 225 return false; 226 227 map.m_lblk = pos >> blkbits; 228 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits); 229 blklen = map.m_len; 230 231 err = ext4_map_blocks(NULL, inode, &map, 0); 232 if (err != blklen) 233 return false; 234 /* 235 * 'err==len' means that all of the blocks have been preallocated, 236 * regardless of whether they have been initialized or not. We need to 237 * check m_flags to distinguish the unwritten extents. 238 */ 239 *unwritten = !(map.m_flags & EXT4_MAP_MAPPED); 240 return true; 241 } 242 243 static ssize_t ext4_generic_write_checks(struct kiocb *iocb, 244 struct iov_iter *from) 245 { 246 struct inode *inode = file_inode(iocb->ki_filp); 247 ssize_t ret; 248 249 if (unlikely(IS_IMMUTABLE(inode))) 250 return -EPERM; 251 252 ret = generic_write_checks(iocb, from); 253 if (ret <= 0) 254 return ret; 255 256 /* 257 * If we have encountered a bitmap-format file, the size limit 258 * is smaller than s_maxbytes, which is for extent-mapped files. 259 */ 260 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { 261 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 262 263 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) 264 return -EFBIG; 265 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos); 266 } 267 268 return iov_iter_count(from); 269 } 270 271 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from) 272 { 273 ssize_t ret, count; 274 275 count = ext4_generic_write_checks(iocb, from); 276 if (count <= 0) 277 return count; 278 279 ret = file_modified(iocb->ki_filp); 280 if (ret) 281 return ret; 282 return count; 283 } 284 285 static ssize_t ext4_buffered_write_iter(struct kiocb *iocb, 286 struct iov_iter *from) 287 { 288 ssize_t ret; 289 struct inode *inode = file_inode(iocb->ki_filp); 290 291 if (iocb->ki_flags & IOCB_NOWAIT) 292 return -EOPNOTSUPP; 293 294 inode_lock(inode); 295 ret = ext4_write_checks(iocb, from); 296 if (ret <= 0) 297 goto out; 298 299 ret = generic_perform_write(iocb, from); 300 301 out: 302 inode_unlock(inode); 303 if (unlikely(ret <= 0)) 304 return ret; 305 return generic_write_sync(iocb, ret); 306 } 307 308 static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset, 309 ssize_t written, size_t count) 310 { 311 handle_t *handle; 312 bool truncate = false; 313 u8 blkbits = inode->i_blkbits; 314 ext4_lblk_t written_blk, end_blk; 315 int ret; 316 317 /* 318 * Note that EXT4_I(inode)->i_disksize can get extended up to 319 * inode->i_size while the I/O was running due to writeback of delalloc 320 * blocks. But, the code in ext4_iomap_alloc() is careful to use 321 * zeroed/unwritten extents if this is possible; thus we won't leave 322 * uninitialized blocks in a file even if we didn't succeed in writing 323 * as much as we intended. 324 */ 325 WARN_ON_ONCE(i_size_read(inode) < EXT4_I(inode)->i_disksize); 326 if (offset + count <= EXT4_I(inode)->i_disksize) { 327 /* 328 * We need to ensure that the inode is removed from the orphan 329 * list if it has been added prematurely, due to writeback of 330 * delalloc blocks. 331 */ 332 if (!list_empty(&EXT4_I(inode)->i_orphan) && inode->i_nlink) { 333 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); 334 335 if (IS_ERR(handle)) { 336 ext4_orphan_del(NULL, inode); 337 return PTR_ERR(handle); 338 } 339 340 ext4_orphan_del(handle, inode); 341 ext4_journal_stop(handle); 342 } 343 344 return written; 345 } 346 347 if (written < 0) 348 goto truncate; 349 350 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); 351 if (IS_ERR(handle)) { 352 written = PTR_ERR(handle); 353 goto truncate; 354 } 355 356 if (ext4_update_inode_size(inode, offset + written)) { 357 ret = ext4_mark_inode_dirty(handle, inode); 358 if (unlikely(ret)) { 359 written = ret; 360 ext4_journal_stop(handle); 361 goto truncate; 362 } 363 } 364 365 /* 366 * We may need to truncate allocated but not written blocks beyond EOF. 367 */ 368 written_blk = ALIGN(offset + written, 1 << blkbits); 369 end_blk = ALIGN(offset + count, 1 << blkbits); 370 if (written_blk < end_blk && ext4_can_truncate(inode)) 371 truncate = true; 372 373 /* 374 * Remove the inode from the orphan list if it has been extended and 375 * everything went OK. 376 */ 377 if (!truncate && inode->i_nlink) 378 ext4_orphan_del(handle, inode); 379 ext4_journal_stop(handle); 380 381 if (truncate) { 382 truncate: 383 ext4_truncate_failed_write(inode); 384 /* 385 * If the truncate operation failed early, then the inode may 386 * still be on the orphan list. In that case, we need to try 387 * remove the inode from the in-memory linked list. 388 */ 389 if (inode->i_nlink) 390 ext4_orphan_del(NULL, inode); 391 } 392 393 return written; 394 } 395 396 static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size, 397 int error, unsigned int flags) 398 { 399 loff_t pos = iocb->ki_pos; 400 struct inode *inode = file_inode(iocb->ki_filp); 401 402 if (error) 403 return error; 404 405 if (size && flags & IOMAP_DIO_UNWRITTEN) { 406 error = ext4_convert_unwritten_extents(NULL, inode, pos, size); 407 if (error < 0) 408 return error; 409 } 410 /* 411 * If we are extending the file, we have to update i_size here before 412 * page cache gets invalidated in iomap_dio_rw(). Otherwise racing 413 * buffered reads could zero out too much from page cache pages. Update 414 * of on-disk size will happen later in ext4_dio_write_iter() where 415 * we have enough information to also perform orphan list handling etc. 416 * Note that we perform all extending writes synchronously under 417 * i_rwsem held exclusively so i_size update is safe here in that case. 418 * If the write was not extending, we cannot see pos > i_size here 419 * because operations reducing i_size like truncate wait for all 420 * outstanding DIO before updating i_size. 421 */ 422 pos += size; 423 if (pos > i_size_read(inode)) 424 i_size_write(inode, pos); 425 426 return 0; 427 } 428 429 static const struct iomap_dio_ops ext4_dio_write_ops = { 430 .end_io = ext4_dio_write_end_io, 431 }; 432 433 /* 434 * The intention here is to start with shared lock acquired then see if any 435 * condition requires an exclusive inode lock. If yes, then we restart the 436 * whole operation by releasing the shared lock and acquiring exclusive lock. 437 * 438 * - For unaligned_io we never take shared lock as it may cause data corruption 439 * when two unaligned IO tries to modify the same block e.g. while zeroing. 440 * 441 * - For extending writes case we don't take the shared lock, since it requires 442 * updating inode i_disksize and/or orphan handling with exclusive lock. 443 * 444 * - shared locking will only be true mostly with overwrites, including 445 * initialized blocks and unwritten blocks. For overwrite unwritten blocks 446 * we protect splitting extents by i_data_sem in ext4_inode_info, so we can 447 * also release exclusive i_rwsem lock. 448 * 449 * - Otherwise we will switch to exclusive i_rwsem lock. 450 */ 451 static ssize_t ext4_dio_write_checks(struct kiocb *iocb, struct iov_iter *from, 452 bool *ilock_shared, bool *extend, 453 bool *unwritten, int *dio_flags) 454 { 455 struct file *file = iocb->ki_filp; 456 struct inode *inode = file_inode(file); 457 loff_t offset; 458 size_t count; 459 ssize_t ret; 460 bool overwrite, unaligned_io; 461 462 restart: 463 ret = ext4_generic_write_checks(iocb, from); 464 if (ret <= 0) 465 goto out; 466 467 offset = iocb->ki_pos; 468 count = ret; 469 470 unaligned_io = ext4_unaligned_io(inode, from, offset); 471 *extend = ext4_extending_io(inode, offset, count); 472 overwrite = ext4_overwrite_io(inode, offset, count, unwritten); 473 474 /* 475 * Determine whether we need to upgrade to an exclusive lock. This is 476 * required to change security info in file_modified(), for extending 477 * I/O, any form of non-overwrite I/O, and unaligned I/O to unwritten 478 * extents (as partial block zeroing may be required). 479 * 480 * Note that unaligned writes are allowed under shared lock so long as 481 * they are pure overwrites. Otherwise, concurrent unaligned writes risk 482 * data corruption due to partial block zeroing in the dio layer, and so 483 * the I/O must occur exclusively. 484 */ 485 if (*ilock_shared && 486 ((!IS_NOSEC(inode) || *extend || !overwrite || 487 (unaligned_io && *unwritten)))) { 488 if (iocb->ki_flags & IOCB_NOWAIT) { 489 ret = -EAGAIN; 490 goto out; 491 } 492 inode_unlock_shared(inode); 493 *ilock_shared = false; 494 inode_lock(inode); 495 goto restart; 496 } 497 498 /* 499 * Now that locking is settled, determine dio flags and exclusivity 500 * requirements. We don't use DIO_OVERWRITE_ONLY because we enforce 501 * behavior already. The inode lock is already held exclusive if the 502 * write is non-overwrite or extending, so drain all outstanding dio and 503 * set the force wait dio flag. 504 */ 505 if (!*ilock_shared && (unaligned_io || *extend)) { 506 if (iocb->ki_flags & IOCB_NOWAIT) { 507 ret = -EAGAIN; 508 goto out; 509 } 510 if (unaligned_io && (!overwrite || *unwritten)) 511 inode_dio_wait(inode); 512 *dio_flags = IOMAP_DIO_FORCE_WAIT; 513 } 514 515 ret = file_modified(file); 516 if (ret < 0) 517 goto out; 518 519 return count; 520 out: 521 if (*ilock_shared) 522 inode_unlock_shared(inode); 523 else 524 inode_unlock(inode); 525 return ret; 526 } 527 528 static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from) 529 { 530 ssize_t ret; 531 handle_t *handle; 532 struct inode *inode = file_inode(iocb->ki_filp); 533 loff_t offset = iocb->ki_pos; 534 size_t count = iov_iter_count(from); 535 const struct iomap_ops *iomap_ops = &ext4_iomap_ops; 536 bool extend = false, unwritten = false; 537 bool ilock_shared = true; 538 int dio_flags = 0; 539 540 /* 541 * Quick check here without any i_rwsem lock to see if it is extending 542 * IO. A more reliable check is done in ext4_dio_write_checks() with 543 * proper locking in place. 544 */ 545 if (offset + count > i_size_read(inode)) 546 ilock_shared = false; 547 548 if (iocb->ki_flags & IOCB_NOWAIT) { 549 if (ilock_shared) { 550 if (!inode_trylock_shared(inode)) 551 return -EAGAIN; 552 } else { 553 if (!inode_trylock(inode)) 554 return -EAGAIN; 555 } 556 } else { 557 if (ilock_shared) 558 inode_lock_shared(inode); 559 else 560 inode_lock(inode); 561 } 562 563 /* Fallback to buffered I/O if the inode does not support direct I/O. */ 564 if (!ext4_should_use_dio(iocb, from)) { 565 if (ilock_shared) 566 inode_unlock_shared(inode); 567 else 568 inode_unlock(inode); 569 return ext4_buffered_write_iter(iocb, from); 570 } 571 572 ret = ext4_dio_write_checks(iocb, from, &ilock_shared, &extend, 573 &unwritten, &dio_flags); 574 if (ret <= 0) 575 return ret; 576 577 /* 578 * Make sure inline data cannot be created anymore since we are going 579 * to allocate blocks for DIO. We know the inode does not have any 580 * inline data now because ext4_dio_supported() checked for that. 581 */ 582 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 583 584 offset = iocb->ki_pos; 585 count = ret; 586 587 if (extend) { 588 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); 589 if (IS_ERR(handle)) { 590 ret = PTR_ERR(handle); 591 goto out; 592 } 593 594 ret = ext4_orphan_add(handle, inode); 595 if (ret) { 596 ext4_journal_stop(handle); 597 goto out; 598 } 599 600 ext4_journal_stop(handle); 601 } 602 603 if (ilock_shared && !unwritten) 604 iomap_ops = &ext4_iomap_overwrite_ops; 605 ret = iomap_dio_rw(iocb, from, iomap_ops, &ext4_dio_write_ops, 606 dio_flags, NULL, 0); 607 if (ret == -ENOTBLK) 608 ret = 0; 609 610 if (extend) 611 ret = ext4_handle_inode_extension(inode, offset, ret, count); 612 613 out: 614 if (ilock_shared) 615 inode_unlock_shared(inode); 616 else 617 inode_unlock(inode); 618 619 if (ret >= 0 && iov_iter_count(from)) { 620 ssize_t err; 621 loff_t endbyte; 622 623 offset = iocb->ki_pos; 624 err = ext4_buffered_write_iter(iocb, from); 625 if (err < 0) 626 return err; 627 628 /* 629 * We need to ensure that the pages within the page cache for 630 * the range covered by this I/O are written to disk and 631 * invalidated. This is in attempt to preserve the expected 632 * direct I/O semantics in the case we fallback to buffered I/O 633 * to complete off the I/O request. 634 */ 635 ret += err; 636 endbyte = offset + err - 1; 637 err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping, 638 offset, endbyte); 639 if (!err) 640 invalidate_mapping_pages(iocb->ki_filp->f_mapping, 641 offset >> PAGE_SHIFT, 642 endbyte >> PAGE_SHIFT); 643 } 644 645 return ret; 646 } 647 648 #ifdef CONFIG_FS_DAX 649 static ssize_t 650 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from) 651 { 652 ssize_t ret; 653 size_t count; 654 loff_t offset; 655 handle_t *handle; 656 bool extend = false; 657 struct inode *inode = file_inode(iocb->ki_filp); 658 659 if (iocb->ki_flags & IOCB_NOWAIT) { 660 if (!inode_trylock(inode)) 661 return -EAGAIN; 662 } else { 663 inode_lock(inode); 664 } 665 666 ret = ext4_write_checks(iocb, from); 667 if (ret <= 0) 668 goto out; 669 670 offset = iocb->ki_pos; 671 count = iov_iter_count(from); 672 673 if (offset + count > EXT4_I(inode)->i_disksize) { 674 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); 675 if (IS_ERR(handle)) { 676 ret = PTR_ERR(handle); 677 goto out; 678 } 679 680 ret = ext4_orphan_add(handle, inode); 681 if (ret) { 682 ext4_journal_stop(handle); 683 goto out; 684 } 685 686 extend = true; 687 ext4_journal_stop(handle); 688 } 689 690 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops); 691 692 if (extend) 693 ret = ext4_handle_inode_extension(inode, offset, ret, count); 694 out: 695 inode_unlock(inode); 696 if (ret > 0) 697 ret = generic_write_sync(iocb, ret); 698 return ret; 699 } 700 #endif 701 702 static ssize_t 703 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 704 { 705 struct inode *inode = file_inode(iocb->ki_filp); 706 707 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 708 return -EIO; 709 710 #ifdef CONFIG_FS_DAX 711 if (IS_DAX(inode)) 712 return ext4_dax_write_iter(iocb, from); 713 #endif 714 if (iocb->ki_flags & IOCB_DIRECT) 715 return ext4_dio_write_iter(iocb, from); 716 else 717 return ext4_buffered_write_iter(iocb, from); 718 } 719 720 #ifdef CONFIG_FS_DAX 721 static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf, unsigned int order) 722 { 723 int error = 0; 724 vm_fault_t result; 725 int retries = 0; 726 handle_t *handle = NULL; 727 struct inode *inode = file_inode(vmf->vma->vm_file); 728 struct super_block *sb = inode->i_sb; 729 730 /* 731 * We have to distinguish real writes from writes which will result in a 732 * COW page; COW writes should *not* poke the journal (the file will not 733 * be changed). Doing so would cause unintended failures when mounted 734 * read-only. 735 * 736 * We check for VM_SHARED rather than vmf->cow_page since the latter is 737 * unset for order != 0 (i.e. only in do_cow_fault); for 738 * other sizes, dax_iomap_fault will handle splitting / fallback so that 739 * we eventually come back with a COW page. 740 */ 741 bool write = (vmf->flags & FAULT_FLAG_WRITE) && 742 (vmf->vma->vm_flags & VM_SHARED); 743 struct address_space *mapping = vmf->vma->vm_file->f_mapping; 744 pfn_t pfn; 745 746 if (write) { 747 sb_start_pagefault(sb); 748 file_update_time(vmf->vma->vm_file); 749 filemap_invalidate_lock_shared(mapping); 750 retry: 751 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE, 752 EXT4_DATA_TRANS_BLOCKS(sb)); 753 if (IS_ERR(handle)) { 754 filemap_invalidate_unlock_shared(mapping); 755 sb_end_pagefault(sb); 756 return VM_FAULT_SIGBUS; 757 } 758 } else { 759 filemap_invalidate_lock_shared(mapping); 760 } 761 result = dax_iomap_fault(vmf, order, &pfn, &error, &ext4_iomap_ops); 762 if (write) { 763 ext4_journal_stop(handle); 764 765 if ((result & VM_FAULT_ERROR) && error == -ENOSPC && 766 ext4_should_retry_alloc(sb, &retries)) 767 goto retry; 768 /* Handling synchronous page fault? */ 769 if (result & VM_FAULT_NEEDDSYNC) 770 result = dax_finish_sync_fault(vmf, order, pfn); 771 filemap_invalidate_unlock_shared(mapping); 772 sb_end_pagefault(sb); 773 } else { 774 filemap_invalidate_unlock_shared(mapping); 775 } 776 777 return result; 778 } 779 780 static vm_fault_t ext4_dax_fault(struct vm_fault *vmf) 781 { 782 return ext4_dax_huge_fault(vmf, 0); 783 } 784 785 static const struct vm_operations_struct ext4_dax_vm_ops = { 786 .fault = ext4_dax_fault, 787 .huge_fault = ext4_dax_huge_fault, 788 .page_mkwrite = ext4_dax_fault, 789 .pfn_mkwrite = ext4_dax_fault, 790 }; 791 #else 792 #define ext4_dax_vm_ops ext4_file_vm_ops 793 #endif 794 795 static const struct vm_operations_struct ext4_file_vm_ops = { 796 .fault = filemap_fault, 797 .map_pages = filemap_map_pages, 798 .page_mkwrite = ext4_page_mkwrite, 799 }; 800 801 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma) 802 { 803 struct inode *inode = file->f_mapping->host; 804 struct dax_device *dax_dev = EXT4_SB(inode->i_sb)->s_daxdev; 805 806 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 807 return -EIO; 808 809 /* 810 * We don't support synchronous mappings for non-DAX files and 811 * for DAX files if underneath dax_device is not synchronous. 812 */ 813 if (!daxdev_mapping_supported(vma, dax_dev)) 814 return -EOPNOTSUPP; 815 816 file_accessed(file); 817 if (IS_DAX(file_inode(file))) { 818 vma->vm_ops = &ext4_dax_vm_ops; 819 vm_flags_set(vma, VM_HUGEPAGE); 820 } else { 821 vma->vm_ops = &ext4_file_vm_ops; 822 } 823 return 0; 824 } 825 826 static int ext4_sample_last_mounted(struct super_block *sb, 827 struct vfsmount *mnt) 828 { 829 struct ext4_sb_info *sbi = EXT4_SB(sb); 830 struct path path; 831 char buf[64], *cp; 832 handle_t *handle; 833 int err; 834 835 if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED))) 836 return 0; 837 838 if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb)) 839 return 0; 840 841 ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED); 842 /* 843 * Sample where the filesystem has been mounted and 844 * store it in the superblock for sysadmin convenience 845 * when trying to sort through large numbers of block 846 * devices or filesystem images. 847 */ 848 memset(buf, 0, sizeof(buf)); 849 path.mnt = mnt; 850 path.dentry = mnt->mnt_root; 851 cp = d_path(&path, buf, sizeof(buf)); 852 err = 0; 853 if (IS_ERR(cp)) 854 goto out; 855 856 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); 857 err = PTR_ERR(handle); 858 if (IS_ERR(handle)) 859 goto out; 860 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 861 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh, 862 EXT4_JTR_NONE); 863 if (err) 864 goto out_journal; 865 lock_buffer(sbi->s_sbh); 866 strncpy(sbi->s_es->s_last_mounted, cp, 867 sizeof(sbi->s_es->s_last_mounted)); 868 ext4_superblock_csum_set(sb); 869 unlock_buffer(sbi->s_sbh); 870 ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); 871 out_journal: 872 ext4_journal_stop(handle); 873 out: 874 sb_end_intwrite(sb); 875 return err; 876 } 877 878 static int ext4_file_open(struct inode *inode, struct file *filp) 879 { 880 int ret; 881 882 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 883 return -EIO; 884 885 ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt); 886 if (ret) 887 return ret; 888 889 ret = fscrypt_file_open(inode, filp); 890 if (ret) 891 return ret; 892 893 ret = fsverity_file_open(inode, filp); 894 if (ret) 895 return ret; 896 897 /* 898 * Set up the jbd2_inode if we are opening the inode for 899 * writing and the journal is present 900 */ 901 if (filp->f_mode & FMODE_WRITE) { 902 ret = ext4_inode_attach_jinode(inode); 903 if (ret < 0) 904 return ret; 905 } 906 907 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | 908 FMODE_DIO_PARALLEL_WRITE; 909 return dquot_file_open(inode, filp); 910 } 911 912 /* 913 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values 914 * by calling generic_file_llseek_size() with the appropriate maxbytes 915 * value for each. 916 */ 917 loff_t ext4_llseek(struct file *file, loff_t offset, int whence) 918 { 919 struct inode *inode = file->f_mapping->host; 920 loff_t maxbytes; 921 922 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 923 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes; 924 else 925 maxbytes = inode->i_sb->s_maxbytes; 926 927 switch (whence) { 928 default: 929 return generic_file_llseek_size(file, offset, whence, 930 maxbytes, i_size_read(inode)); 931 case SEEK_HOLE: 932 inode_lock_shared(inode); 933 offset = iomap_seek_hole(inode, offset, 934 &ext4_iomap_report_ops); 935 inode_unlock_shared(inode); 936 break; 937 case SEEK_DATA: 938 inode_lock_shared(inode); 939 offset = iomap_seek_data(inode, offset, 940 &ext4_iomap_report_ops); 941 inode_unlock_shared(inode); 942 break; 943 } 944 945 if (offset < 0) 946 return offset; 947 return vfs_setpos(file, offset, maxbytes); 948 } 949 950 const struct file_operations ext4_file_operations = { 951 .llseek = ext4_llseek, 952 .read_iter = ext4_file_read_iter, 953 .write_iter = ext4_file_write_iter, 954 .iopoll = iocb_bio_iopoll, 955 .unlocked_ioctl = ext4_ioctl, 956 #ifdef CONFIG_COMPAT 957 .compat_ioctl = ext4_compat_ioctl, 958 #endif 959 .mmap = ext4_file_mmap, 960 .mmap_supported_flags = MAP_SYNC, 961 .open = ext4_file_open, 962 .release = ext4_release_file, 963 .fsync = ext4_sync_file, 964 .get_unmapped_area = thp_get_unmapped_area, 965 .splice_read = ext4_file_splice_read, 966 .splice_write = iter_file_splice_write, 967 .fallocate = ext4_fallocate, 968 }; 969 970 const struct inode_operations ext4_file_inode_operations = { 971 .setattr = ext4_setattr, 972 .getattr = ext4_file_getattr, 973 .listxattr = ext4_listxattr, 974 .get_inode_acl = ext4_get_acl, 975 .set_acl = ext4_set_acl, 976 .fiemap = ext4_fiemap, 977 .fileattr_get = ext4_fileattr_get, 978 .fileattr_set = ext4_fileattr_set, 979 }; 980