1/* 2 * linux/fs/ext4/file.c 3 * 4 * Copyright (C) 1992, 1993, 1994, 1995 5 * Remy Card (card@masi.ibp.fr) 6 * Laboratoire MASI - Institut Blaise Pascal 7 * Universite Pierre et Marie Curie (Paris VI) 8 * 9 * from 10 * 11 * linux/fs/minix/file.c 12 * 13 * Copyright (C) 1991, 1992 Linus Torvalds 14 * 15 * ext4 fs regular file handling primitives 16 * 17 * 64-bit file support on 64-bit platforms by Jakub Jelinek 18 * (jj@sunsite.ms.mff.cuni.cz) 19 */ 20 21#include <linux/time.h> 22#include <linux/fs.h> 23#include <linux/mount.h> 24#include <linux/path.h> 25#include <linux/quotaops.h> 26#include <linux/pagevec.h> 27#include <linux/uio.h> 28#include "ext4.h" 29#include "ext4_jbd2.h" 30#include "xattr.h" 31#include "acl.h" 32 33/* 34 * Called when an inode is released. Note that this is different 35 * from ext4_file_open: open gets called at every open, but release 36 * gets called only when /all/ the files are closed. 37 */ 38static int ext4_release_file(struct inode *inode, struct file *filp) 39{ 40 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) { 41 ext4_alloc_da_blocks(inode); 42 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); 43 } 44 /* if we are the last writer on the inode, drop the block reservation */ 45 if ((filp->f_mode & FMODE_WRITE) && 46 (atomic_read(&inode->i_writecount) == 1) && 47 !EXT4_I(inode)->i_reserved_data_blocks) 48 { 49 down_write(&EXT4_I(inode)->i_data_sem); 50 ext4_discard_preallocations(inode); 51 up_write(&EXT4_I(inode)->i_data_sem); 52 } 53 if (is_dx(inode) && filp->private_data) 54 ext4_htree_free_dir_info(filp->private_data); 55 56 return 0; 57} 58 59static void ext4_unwritten_wait(struct inode *inode) 60{ 61 wait_queue_head_t *wq = ext4_ioend_wq(inode); 62 63 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0)); 64} 65 66/* 67 * This tests whether the IO in question is block-aligned or not. 68 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they 69 * are converted to written only after the IO is complete. Until they are 70 * mapped, these blocks appear as holes, so dio_zero_block() will assume that 71 * it needs to zero out portions of the start and/or end block. If 2 AIO 72 * threads are at work on the same unwritten block, they must be synchronized 73 * or one thread will zero the other's data, causing corruption. 74 */ 75static int 76ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos) 77{ 78 struct super_block *sb = inode->i_sb; 79 int blockmask = sb->s_blocksize - 1; 80 81 if (pos >= i_size_read(inode)) 82 return 0; 83 84 if ((pos | iov_iter_alignment(from)) & blockmask) 85 return 1; 86 87 return 0; 88} 89 90static ssize_t 91ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 92{ 93 struct file *file = iocb->ki_filp; 94 struct inode *inode = file_inode(iocb->ki_filp); 95 struct mutex *aio_mutex = NULL; 96 struct blk_plug plug; 97 int o_direct = iocb->ki_flags & IOCB_DIRECT; 98 int overwrite = 0; 99 ssize_t ret; 100 101 /* 102 * Unaligned direct AIO must be serialized; see comment above 103 * In the case of O_APPEND, assume that we must always serialize 104 */ 105 if (o_direct && 106 ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) && 107 !is_sync_kiocb(iocb) && 108 (iocb->ki_flags & IOCB_APPEND || 109 ext4_unaligned_aio(inode, from, iocb->ki_pos))) { 110 aio_mutex = ext4_aio_mutex(inode); 111 mutex_lock(aio_mutex); 112 ext4_unwritten_wait(inode); 113 } 114 115 mutex_lock(&inode->i_mutex); 116 ret = generic_write_checks(iocb, from); 117 if (ret <= 0) 118 goto out; 119 120 /* 121 * If we have encountered a bitmap-format file, the size limit 122 * is smaller than s_maxbytes, which is for extent-mapped files. 123 */ 124 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { 125 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 126 127 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) { 128 ret = -EFBIG; 129 goto out; 130 } 131 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos); 132 } 133 134 iocb->private = &overwrite; 135 if (o_direct) { 136 size_t length = iov_iter_count(from); 137 loff_t pos = iocb->ki_pos; 138 blk_start_plug(&plug); 139 140 /* check whether we do a DIO overwrite or not */ 141 if (ext4_should_dioread_nolock(inode) && !aio_mutex && 142 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) { 143 struct ext4_map_blocks map; 144 unsigned int blkbits = inode->i_blkbits; 145 int err, len; 146 147 map.m_lblk = pos >> blkbits; 148 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits) 149 - map.m_lblk; 150 len = map.m_len; 151 152 err = ext4_map_blocks(NULL, inode, &map, 0); 153 /* 154 * 'err==len' means that all of blocks has 155 * been preallocated no matter they are 156 * initialized or not. For excluding 157 * unwritten extents, we need to check 158 * m_flags. There are two conditions that 159 * indicate for initialized extents. 1) If we 160 * hit extent cache, EXT4_MAP_MAPPED flag is 161 * returned; 2) If we do a real lookup, 162 * non-flags are returned. So we should check 163 * these two conditions. 164 */ 165 if (err == len && (map.m_flags & EXT4_MAP_MAPPED)) 166 overwrite = 1; 167 } 168 } 169 170 ret = __generic_file_write_iter(iocb, from); 171 mutex_unlock(&inode->i_mutex); 172 173 if (ret > 0) { 174 ssize_t err; 175 176 err = generic_write_sync(file, iocb->ki_pos - ret, ret); 177 if (err < 0) 178 ret = err; 179 } 180 if (o_direct) 181 blk_finish_plug(&plug); 182 183 if (aio_mutex) 184 mutex_unlock(aio_mutex); 185 return ret; 186 187out: 188 mutex_unlock(&inode->i_mutex); 189 if (aio_mutex) 190 mutex_unlock(aio_mutex); 191 return ret; 192} 193 194#ifdef CONFIG_FS_DAX 195static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate) 196{ 197 struct inode *inode = bh->b_assoc_map->host; 198 /* XXX: breaks on 32-bit > 16GB. Is that even supported? */ 199 loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits; 200 int err; 201 if (!uptodate) 202 return; 203 WARN_ON(!buffer_unwritten(bh)); 204 err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size); 205} 206 207static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 208{ 209 return dax_fault(vma, vmf, ext4_get_block, ext4_end_io_unwritten); 210 /* Is this the right get_block? */ 211} 212 213static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 214{ 215 return dax_mkwrite(vma, vmf, ext4_get_block, ext4_end_io_unwritten); 216} 217 218static const struct vm_operations_struct ext4_dax_vm_ops = { 219 .fault = ext4_dax_fault, 220 .page_mkwrite = ext4_dax_mkwrite, 221 .pfn_mkwrite = dax_pfn_mkwrite, 222}; 223#else 224#define ext4_dax_vm_ops ext4_file_vm_ops 225#endif 226 227static const struct vm_operations_struct ext4_file_vm_ops = { 228 .fault = ext4_filemap_fault, 229 .map_pages = filemap_map_pages, 230 .page_mkwrite = ext4_page_mkwrite, 231}; 232 233static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma) 234{ 235 struct inode *inode = file->f_mapping->host; 236 237 if (ext4_encrypted_inode(inode)) { 238 int err = ext4_generate_encryption_key(inode); 239 if (err) 240 return 0; 241 } 242 file_accessed(file); 243 if (IS_DAX(file_inode(file))) { 244 vma->vm_ops = &ext4_dax_vm_ops; 245 vma->vm_flags |= VM_MIXEDMAP; 246 } else { 247 vma->vm_ops = &ext4_file_vm_ops; 248 } 249 return 0; 250} 251 252static int ext4_file_open(struct inode * inode, struct file * filp) 253{ 254 struct super_block *sb = inode->i_sb; 255 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 256 struct vfsmount *mnt = filp->f_path.mnt; 257 struct path path; 258 char buf[64], *cp; 259 int ret; 260 261 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) && 262 !(sb->s_flags & MS_RDONLY))) { 263 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED; 264 /* 265 * Sample where the filesystem has been mounted and 266 * store it in the superblock for sysadmin convenience 267 * when trying to sort through large numbers of block 268 * devices or filesystem images. 269 */ 270 memset(buf, 0, sizeof(buf)); 271 path.mnt = mnt; 272 path.dentry = mnt->mnt_root; 273 cp = d_path(&path, buf, sizeof(buf)); 274 if (!IS_ERR(cp)) { 275 handle_t *handle; 276 int err; 277 278 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); 279 if (IS_ERR(handle)) 280 return PTR_ERR(handle); 281 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 282 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 283 if (err) { 284 ext4_journal_stop(handle); 285 return err; 286 } 287 strlcpy(sbi->s_es->s_last_mounted, cp, 288 sizeof(sbi->s_es->s_last_mounted)); 289 ext4_handle_dirty_super(handle, sb); 290 ext4_journal_stop(handle); 291 } 292 } 293 /* 294 * Set up the jbd2_inode if we are opening the inode for 295 * writing and the journal is present 296 */ 297 if (filp->f_mode & FMODE_WRITE) { 298 ret = ext4_inode_attach_jinode(inode); 299 if (ret < 0) 300 return ret; 301 } 302 ret = dquot_file_open(inode, filp); 303 if (!ret && ext4_encrypted_inode(inode)) { 304 ret = ext4_generate_encryption_key(inode); 305 if (ret) 306 ret = -EACCES; 307 } 308 return ret; 309} 310 311/* 312 * Here we use ext4_map_blocks() to get a block mapping for a extent-based 313 * file rather than ext4_ext_walk_space() because we can introduce 314 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same 315 * function. When extent status tree has been fully implemented, it will 316 * track all extent status for a file and we can directly use it to 317 * retrieve the offset for SEEK_DATA/SEEK_HOLE. 318 */ 319 320/* 321 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to 322 * lookup page cache to check whether or not there has some data between 323 * [startoff, endoff] because, if this range contains an unwritten extent, 324 * we determine this extent as a data or a hole according to whether the 325 * page cache has data or not. 326 */ 327static int ext4_find_unwritten_pgoff(struct inode *inode, 328 int whence, 329 struct ext4_map_blocks *map, 330 loff_t *offset) 331{ 332 struct pagevec pvec; 333 unsigned int blkbits; 334 pgoff_t index; 335 pgoff_t end; 336 loff_t endoff; 337 loff_t startoff; 338 loff_t lastoff; 339 int found = 0; 340 341 blkbits = inode->i_sb->s_blocksize_bits; 342 startoff = *offset; 343 lastoff = startoff; 344 endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits; 345 346 index = startoff >> PAGE_CACHE_SHIFT; 347 end = endoff >> PAGE_CACHE_SHIFT; 348 349 pagevec_init(&pvec, 0); 350 do { 351 int i, num; 352 unsigned long nr_pages; 353 354 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE); 355 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, 356 (pgoff_t)num); 357 if (nr_pages == 0) { 358 if (whence == SEEK_DATA) 359 break; 360 361 BUG_ON(whence != SEEK_HOLE); 362 /* 363 * If this is the first time to go into the loop and 364 * offset is not beyond the end offset, it will be a 365 * hole at this offset 366 */ 367 if (lastoff == startoff || lastoff < endoff) 368 found = 1; 369 break; 370 } 371 372 /* 373 * If this is the first time to go into the loop and 374 * offset is smaller than the first page offset, it will be a 375 * hole at this offset. 376 */ 377 if (lastoff == startoff && whence == SEEK_HOLE && 378 lastoff < page_offset(pvec.pages[0])) { 379 found = 1; 380 break; 381 } 382 383 for (i = 0; i < nr_pages; i++) { 384 struct page *page = pvec.pages[i]; 385 struct buffer_head *bh, *head; 386 387 /* 388 * If the current offset is not beyond the end of given 389 * range, it will be a hole. 390 */ 391 if (lastoff < endoff && whence == SEEK_HOLE && 392 page->index > end) { 393 found = 1; 394 *offset = lastoff; 395 goto out; 396 } 397 398 lock_page(page); 399 400 if (unlikely(page->mapping != inode->i_mapping)) { 401 unlock_page(page); 402 continue; 403 } 404 405 if (!page_has_buffers(page)) { 406 unlock_page(page); 407 continue; 408 } 409 410 if (page_has_buffers(page)) { 411 lastoff = page_offset(page); 412 bh = head = page_buffers(page); 413 do { 414 if (buffer_uptodate(bh) || 415 buffer_unwritten(bh)) { 416 if (whence == SEEK_DATA) 417 found = 1; 418 } else { 419 if (whence == SEEK_HOLE) 420 found = 1; 421 } 422 if (found) { 423 *offset = max_t(loff_t, 424 startoff, lastoff); 425 unlock_page(page); 426 goto out; 427 } 428 lastoff += bh->b_size; 429 bh = bh->b_this_page; 430 } while (bh != head); 431 } 432 433 lastoff = page_offset(page) + PAGE_SIZE; 434 unlock_page(page); 435 } 436 437 /* 438 * The no. of pages is less than our desired, that would be a 439 * hole in there. 440 */ 441 if (nr_pages < num && whence == SEEK_HOLE) { 442 found = 1; 443 *offset = lastoff; 444 break; 445 } 446 447 index = pvec.pages[i - 1]->index + 1; 448 pagevec_release(&pvec); 449 } while (index <= end); 450 451out: 452 pagevec_release(&pvec); 453 return found; 454} 455 456/* 457 * ext4_seek_data() retrieves the offset for SEEK_DATA. 458 */ 459static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize) 460{ 461 struct inode *inode = file->f_mapping->host; 462 struct ext4_map_blocks map; 463 struct extent_status es; 464 ext4_lblk_t start, last, end; 465 loff_t dataoff, isize; 466 int blkbits; 467 int ret = 0; 468 469 mutex_lock(&inode->i_mutex); 470 471 isize = i_size_read(inode); 472 if (offset >= isize) { 473 mutex_unlock(&inode->i_mutex); 474 return -ENXIO; 475 } 476 477 blkbits = inode->i_sb->s_blocksize_bits; 478 start = offset >> blkbits; 479 last = start; 480 end = isize >> blkbits; 481 dataoff = offset; 482 483 do { 484 map.m_lblk = last; 485 map.m_len = end - last + 1; 486 ret = ext4_map_blocks(NULL, inode, &map, 0); 487 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) { 488 if (last != start) 489 dataoff = (loff_t)last << blkbits; 490 break; 491 } 492 493 /* 494 * If there is a delay extent at this offset, 495 * it will be as a data. 496 */ 497 ext4_es_find_delayed_extent_range(inode, last, last, &es); 498 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) { 499 if (last != start) 500 dataoff = (loff_t)last << blkbits; 501 break; 502 } 503 504 /* 505 * If there is a unwritten extent at this offset, 506 * it will be as a data or a hole according to page 507 * cache that has data or not. 508 */ 509 if (map.m_flags & EXT4_MAP_UNWRITTEN) { 510 int unwritten; 511 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA, 512 &map, &dataoff); 513 if (unwritten) 514 break; 515 } 516 517 last++; 518 dataoff = (loff_t)last << blkbits; 519 } while (last <= end); 520 521 mutex_unlock(&inode->i_mutex); 522 523 if (dataoff > isize) 524 return -ENXIO; 525 526 return vfs_setpos(file, dataoff, maxsize); 527} 528 529/* 530 * ext4_seek_hole() retrieves the offset for SEEK_HOLE. 531 */ 532static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize) 533{ 534 struct inode *inode = file->f_mapping->host; 535 struct ext4_map_blocks map; 536 struct extent_status es; 537 ext4_lblk_t start, last, end; 538 loff_t holeoff, isize; 539 int blkbits; 540 int ret = 0; 541 542 mutex_lock(&inode->i_mutex); 543 544 isize = i_size_read(inode); 545 if (offset >= isize) { 546 mutex_unlock(&inode->i_mutex); 547 return -ENXIO; 548 } 549 550 blkbits = inode->i_sb->s_blocksize_bits; 551 start = offset >> blkbits; 552 last = start; 553 end = isize >> blkbits; 554 holeoff = offset; 555 556 do { 557 map.m_lblk = last; 558 map.m_len = end - last + 1; 559 ret = ext4_map_blocks(NULL, inode, &map, 0); 560 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) { 561 last += ret; 562 holeoff = (loff_t)last << blkbits; 563 continue; 564 } 565 566 /* 567 * If there is a delay extent at this offset, 568 * we will skip this extent. 569 */ 570 ext4_es_find_delayed_extent_range(inode, last, last, &es); 571 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) { 572 last = es.es_lblk + es.es_len; 573 holeoff = (loff_t)last << blkbits; 574 continue; 575 } 576 577 /* 578 * If there is a unwritten extent at this offset, 579 * it will be as a data or a hole according to page 580 * cache that has data or not. 581 */ 582 if (map.m_flags & EXT4_MAP_UNWRITTEN) { 583 int unwritten; 584 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE, 585 &map, &holeoff); 586 if (!unwritten) { 587 last += ret; 588 holeoff = (loff_t)last << blkbits; 589 continue; 590 } 591 } 592 593 /* find a hole */ 594 break; 595 } while (last <= end); 596 597 mutex_unlock(&inode->i_mutex); 598 599 if (holeoff > isize) 600 holeoff = isize; 601 602 return vfs_setpos(file, holeoff, maxsize); 603} 604 605/* 606 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values 607 * by calling generic_file_llseek_size() with the appropriate maxbytes 608 * value for each. 609 */ 610loff_t ext4_llseek(struct file *file, loff_t offset, int whence) 611{ 612 struct inode *inode = file->f_mapping->host; 613 loff_t maxbytes; 614 615 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 616 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes; 617 else 618 maxbytes = inode->i_sb->s_maxbytes; 619 620 switch (whence) { 621 case SEEK_SET: 622 case SEEK_CUR: 623 case SEEK_END: 624 return generic_file_llseek_size(file, offset, whence, 625 maxbytes, i_size_read(inode)); 626 case SEEK_DATA: 627 return ext4_seek_data(file, offset, maxbytes); 628 case SEEK_HOLE: 629 return ext4_seek_hole(file, offset, maxbytes); 630 } 631 632 return -EINVAL; 633} 634 635const struct file_operations ext4_file_operations = { 636 .llseek = ext4_llseek, 637 .read_iter = generic_file_read_iter, 638 .write_iter = ext4_file_write_iter, 639 .unlocked_ioctl = ext4_ioctl, 640#ifdef CONFIG_COMPAT 641 .compat_ioctl = ext4_compat_ioctl, 642#endif 643 .mmap = ext4_file_mmap, 644 .open = ext4_file_open, 645 .release = ext4_release_file, 646 .fsync = ext4_sync_file, 647 .splice_read = generic_file_splice_read, 648 .splice_write = iter_file_splice_write, 649 .fallocate = ext4_fallocate, 650}; 651 652const struct inode_operations ext4_file_inode_operations = { 653 .setattr = ext4_setattr, 654 .getattr = ext4_getattr, 655 .setxattr = generic_setxattr, 656 .getxattr = generic_getxattr, 657 .listxattr = ext4_listxattr, 658 .removexattr = generic_removexattr, 659 .get_acl = ext4_get_acl, 660 .set_acl = ext4_set_acl, 661 .fiemap = ext4_fiemap, 662}; 663 664