1 // SPDX-License-Identifier: GPL-2.0-or-later 1 2 /* 3 * Copyright (C) 2002, 2004 Oracle. All right 4 */ 5 6 #include <linux/fs.h> 7 #include <linux/slab.h> 8 #include <linux/highmem.h> 9 #include <linux/pagemap.h> 10 #include <asm/byteorder.h> 11 #include <linux/swap.h> 12 #include <linux/mpage.h> 13 #include <linux/quotaops.h> 14 #include <linux/blkdev.h> 15 #include <linux/uio.h> 16 #include <linux/mm.h> 17 18 #include <cluster/masklog.h> 19 20 #include "ocfs2.h" 21 22 #include "alloc.h" 23 #include "aops.h" 24 #include "dlmglue.h" 25 #include "extent_map.h" 26 #include "file.h" 27 #include "inode.h" 28 #include "journal.h" 29 #include "suballoc.h" 30 #include "super.h" 31 #include "symlink.h" 32 #include "refcounttree.h" 33 #include "ocfs2_trace.h" 34 35 #include "buffer_head_io.h" 36 #include "dir.h" 37 #include "namei.h" 38 #include "sysfile.h" 39 40 static int ocfs2_symlink_get_block(struct inod 41 struct buff 42 { 43 int err = -EIO; 44 int status; 45 struct ocfs2_dinode *fe = NULL; 46 struct buffer_head *bh = NULL; 47 struct buffer_head *buffer_cache_bh = 48 struct ocfs2_super *osb = OCFS2_SB(ino 49 void *kaddr; 50 51 trace_ocfs2_symlink_get_block( 52 (unsigned long long)OC 53 (unsigned long long)ib 54 55 BUG_ON(ocfs2_inode_is_fast_symlink(ino 56 57 if ((iblock << inode->i_sb->s_blocksiz 58 mlog(ML_ERROR, "block offset > 59 (unsigned long long)ibloc 60 goto bail; 61 } 62 63 status = ocfs2_read_inode_block(inode, 64 if (status < 0) { 65 mlog_errno(status); 66 goto bail; 67 } 68 fe = (struct ocfs2_dinode *) bh->b_dat 69 70 if ((u64)iblock >= ocfs2_clusters_to_b 71 72 err = -ENOMEM; 73 mlog(ML_ERROR, "block offset i 74 "%llu\n", (unsigned long 75 goto bail; 76 } 77 78 /* We don't use the page cache to crea 79 * need be, copy it over from the buff 80 if (!buffer_uptodate(bh_result) && ocf 81 u64 blkno = le64_to_cpu(fe->id 82 iblock; 83 buffer_cache_bh = sb_getblk(os 84 if (!buffer_cache_bh) { 85 err = -ENOMEM; 86 mlog(ML_ERROR, "couldn 87 goto bail; 88 } 89 90 /* we haven't locked out trans 91 * could've happened. Since we 92 * the bh, even if it commits 93 * copy, the data is still goo 94 if (buffer_jbd(buffer_cache_bh 95 && ocfs2_inode_is_new(inod 96 kaddr = kmap_atomic(bh 97 if (!kaddr) { 98 mlog(ML_ERROR, 99 goto bail; 100 } 101 memcpy(kaddr + (bh_res 102 buffer_cache_bh 103 bh_result->b_si 104 kunmap_atomic(kaddr); 105 set_buffer_uptodate(bh 106 } 107 brelse(buffer_cache_bh); 108 } 109 110 map_bh(bh_result, inode->i_sb, 111 le64_to_cpu(fe->id2.i_list.l_re 112 113 err = 0; 114 115 bail: 116 brelse(bh); 117 118 return err; 119 } 120 121 static int ocfs2_lock_get_block(struct inode * 122 struct buffer_head *bh_res 123 { 124 int ret = 0; 125 struct ocfs2_inode_info *oi = OCFS2_I( 126 127 down_read(&oi->ip_alloc_sem); 128 ret = ocfs2_get_block(inode, iblock, b 129 up_read(&oi->ip_alloc_sem); 130 131 return ret; 132 } 133 134 int ocfs2_get_block(struct inode *inode, secto 135 struct buffer_head *bh_res 136 { 137 int err = 0; 138 unsigned int ext_flags; 139 u64 max_blocks = bh_result->b_size >> 140 u64 p_blkno, count, past_eof; 141 struct ocfs2_super *osb = OCFS2_SB(ino 142 143 trace_ocfs2_get_block((unsigned long l 144 (unsigned long l 145 146 if (OCFS2_I(inode)->ip_flags & OCFS2_I 147 mlog(ML_NOTICE, "get_block on 148 inode, inode->i_ino); 149 150 if (S_ISLNK(inode->i_mode)) { 151 /* this always does I/O for so 152 err = ocfs2_symlink_get_block( 153 goto bail; 154 } 155 156 err = ocfs2_extent_map_get_blocks(inod 157 &ext 158 if (err) { 159 mlog(ML_ERROR, "Error %d from 160 "%llu, NULL)\n", err, ino 161 (unsigned long long)p_blk 162 goto bail; 163 } 164 165 if (max_blocks < count) 166 count = max_blocks; 167 168 /* 169 * ocfs2 never allocates in this funct 170 * need to use BH_New is when we're ex 171 * system which doesn't support holes, 172 * allows __block_write_begin() to zer 173 * 174 * If we see this on a sparse file sys 175 * raced us and removed the cluster. I 176 * the buffers dirty and uptodate bits 177 * ignore it as a hole. 178 */ 179 if (create && p_blkno == 0 && ocfs2_sp 180 clear_buffer_dirty(bh_result); 181 clear_buffer_uptodate(bh_resul 182 goto bail; 183 } 184 185 /* Treat the unwritten extent as a hol 186 if (p_blkno && !(ext_flags & OCFS2_EXT 187 map_bh(bh_result, inode->i_sb, 188 189 bh_result->b_size = count << inode->i_ 190 191 if (!ocfs2_sparse_alloc(osb)) { 192 if (p_blkno == 0) { 193 err = -EIO; 194 mlog(ML_ERROR, 195 "iblock = %llu p_ 196 (unsigned long lo 197 (unsigned long lo 198 (unsigned long lo 199 mlog(ML_ERROR, "Size % 200 dump_stack(); 201 goto bail; 202 } 203 } 204 205 past_eof = ocfs2_blocks_for_bytes(inod 206 207 trace_ocfs2_get_block_end((unsigned lo 208 (unsigned lo 209 if (create && (iblock >= past_eof)) 210 set_buffer_new(bh_result); 211 212 bail: 213 if (err < 0) 214 err = -EIO; 215 216 return err; 217 } 218 219 int ocfs2_read_inline_data(struct inode *inode 220 struct buffer_head 221 { 222 void *kaddr; 223 loff_t size; 224 struct ocfs2_dinode *di = (struct ocfs 225 226 if (!(le16_to_cpu(di->i_dyn_features) 227 ocfs2_error(inode->i_sb, "Inod 228 (unsigned long lon 229 return -EROFS; 230 } 231 232 size = i_size_read(inode); 233 234 if (size > PAGE_SIZE || 235 size > ocfs2_max_inline_data_with_ 236 ocfs2_error(inode->i_sb, 237 "Inode %llu has wi 238 (unsigned long lon 239 (unsigned long lon 240 return -EROFS; 241 } 242 243 kaddr = kmap_atomic(page); 244 if (size) 245 memcpy(kaddr, di->id2.i_data.i 246 /* Clear the remaining part of the pag 247 memset(kaddr + size, 0, PAGE_SIZE - si 248 flush_dcache_page(page); 249 kunmap_atomic(kaddr); 250 251 SetPageUptodate(page); 252 253 return 0; 254 } 255 256 static int ocfs2_readpage_inline(struct inode 257 { 258 int ret; 259 struct buffer_head *di_bh = NULL; 260 261 BUG_ON(!PageLocked(page)); 262 BUG_ON(!(OCFS2_I(inode)->ip_dyn_featur 263 264 ret = ocfs2_read_inode_block(inode, &d 265 if (ret) { 266 mlog_errno(ret); 267 goto out; 268 } 269 270 ret = ocfs2_read_inline_data(inode, pa 271 out: 272 unlock_page(page); 273 274 brelse(di_bh); 275 return ret; 276 } 277 278 static int ocfs2_read_folio(struct file *file, 279 { 280 struct inode *inode = folio->mapping-> 281 struct ocfs2_inode_info *oi = OCFS2_I( 282 loff_t start = folio_pos(folio); 283 int ret, unlock = 1; 284 285 trace_ocfs2_readpage((unsigned long lo 286 287 ret = ocfs2_inode_lock_with_page(inode 288 if (ret != 0) { 289 if (ret == AOP_TRUNCATED_PAGE) 290 unlock = 0; 291 mlog_errno(ret); 292 goto out; 293 } 294 295 if (down_read_trylock(&oi->ip_alloc_se 296 /* 297 * Unlock the folio and cycle 298 * busyloop waiting for ip_all 299 */ 300 ret = AOP_TRUNCATED_PAGE; 301 folio_unlock(folio); 302 unlock = 0; 303 down_read(&oi->ip_alloc_sem); 304 up_read(&oi->ip_alloc_sem); 305 goto out_inode_unlock; 306 } 307 308 /* 309 * i_size might have just been updated 310 * might now be discovering a truncate 311 * block_read_full_folio->get_block fr 312 * beyond the end of a file, so we che 313 * (generic_file_read, vm_ops->fault) 314 * and notice that the folio they just 315 * 316 * XXX sys_readahead() seems to get th 317 */ 318 if (start >= i_size_read(inode)) { 319 folio_zero_segment(folio, 0, f 320 folio_mark_uptodate(folio); 321 ret = 0; 322 goto out_alloc; 323 } 324 325 if (oi->ip_dyn_features & OCFS2_INLINE 326 ret = ocfs2_readpage_inline(in 327 else 328 ret = block_read_full_folio(fo 329 unlock = 0; 330 331 out_alloc: 332 up_read(&oi->ip_alloc_sem); 333 out_inode_unlock: 334 ocfs2_inode_unlock(inode, 0); 335 out: 336 if (unlock) 337 folio_unlock(folio); 338 return ret; 339 } 340 341 /* 342 * This is used only for read-ahead. Failures 343 * situations are safe to ignore. 344 * 345 * Right now, we don't bother with BH_Boundary 346 * are quite large (243 extents on 4k blocks), 347 * grow out to a tree. If need be, detecting b 348 * trivially be added in a future version of o 349 */ 350 static void ocfs2_readahead(struct readahead_c 351 { 352 int ret; 353 struct inode *inode = rac->mapping->ho 354 struct ocfs2_inode_info *oi = OCFS2_I( 355 356 /* 357 * Use the nonblocking flag for the dl 358 * lock inversion, but don't bother wi 359 */ 360 ret = ocfs2_inode_lock_full(inode, NUL 361 if (ret) 362 return; 363 364 if (down_read_trylock(&oi->ip_alloc_se 365 goto out_unlock; 366 367 /* 368 * Don't bother with inline-data. Ther 369 * to read-ahead in that case anyway.. 370 */ 371 if (oi->ip_dyn_features & OCFS2_INLINE 372 goto out_up; 373 374 /* 375 * Check whether a remote node truncat 376 * drop out in that case as it's not w 377 */ 378 if (readahead_pos(rac) >= i_size_read( 379 goto out_up; 380 381 mpage_readahead(rac, ocfs2_get_block); 382 383 out_up: 384 up_read(&oi->ip_alloc_sem); 385 out_unlock: 386 ocfs2_inode_unlock(inode, 0); 387 } 388 389 /* Note: Because we don't support holes, our a 390 * already happened (allocation writes zeros t 391 * so we don't have to worry about ordered wri 392 * ocfs2_writepages. 393 * 394 * ->writepages is called during the process o 395 * during blocked lock processing. It can't b 396 * to during block mapping. It's relying on t 397 * mapping can't have disappeared under the di 398 * being asked to write back. 399 */ 400 static int ocfs2_writepages(struct address_spa 401 struct writeback_control *wbc) 402 { 403 return mpage_writepages(mapping, wbc, 404 } 405 406 /* Taken from ext3. We don't necessarily need 407 * functionality yet, but IMHO it's better to 408 * thing so we can avoid introducing our own b 409 * their fixes when they happen) --Mark */ 410 int walk_page_buffers( handle_t *handle, 411 struct buffer_head *he 412 unsigned from, 413 unsigned to, 414 int *partial, 415 int (*fn)( handle 416 struct 417 { 418 struct buffer_head *bh; 419 unsigned block_start, block_end; 420 unsigned blocksize = head->b_size; 421 int err, ret = 0; 422 struct buffer_head *next; 423 424 for ( bh = head, block_start = 0; 425 ret == 0 && (bh != head || !bl 426 block_start = block_end, bh = 427 { 428 next = bh->b_this_page; 429 block_end = block_start + bloc 430 if (block_end <= from || block 431 if (partial && !buffer 432 *partial = 1; 433 continue; 434 } 435 err = (*fn)(handle, bh); 436 if (!ret) 437 ret = err; 438 } 439 return ret; 440 } 441 442 static sector_t ocfs2_bmap(struct address_spac 443 { 444 sector_t status; 445 u64 p_blkno = 0; 446 int err = 0; 447 struct inode *inode = mapping->host; 448 449 trace_ocfs2_bmap((unsigned long long)O 450 (unsigned long long)b 451 452 /* 453 * The swap code (ab-)uses ->bmap to g 454 * bypasseѕ the file system for actua 455 * that on refcounted inodes, so we ha 456 * 0 is the magic code for a bmap erro 457 */ 458 if (ocfs2_is_refcount_inode(inode)) 459 return 0; 460 461 /* We don't need to lock journal syste 462 * accessed concurrently from multiple 463 */ 464 if (!INODE_JOURNAL(inode)) { 465 err = ocfs2_inode_lock(inode, 466 if (err) { 467 if (err != -ENOENT) 468 mlog_errno(err 469 goto bail; 470 } 471 down_read(&OCFS2_I(inode)->ip_ 472 } 473 474 if (!(OCFS2_I(inode)->ip_dyn_features 475 err = ocfs2_extent_map_get_blo 476 477 478 if (!INODE_JOURNAL(inode)) { 479 up_read(&OCFS2_I(inode)->ip_al 480 ocfs2_inode_unlock(inode, 0); 481 } 482 483 if (err) { 484 mlog(ML_ERROR, "get_blocks() f 485 (unsigned long long)block 486 mlog_errno(err); 487 goto bail; 488 } 489 490 bail: 491 status = err ? 0 : p_blkno; 492 493 return status; 494 } 495 496 static bool ocfs2_release_folio(struct folio * 497 { 498 if (!folio_buffers(folio)) 499 return false; 500 return try_to_free_buffers(folio); 501 } 502 503 static void ocfs2_figure_cluster_boundaries(st 504 u3 505 un 506 un 507 { 508 unsigned int cluster_start = 0, cluste 509 510 if (unlikely(PAGE_SHIFT > osb->s_clust 511 unsigned int cpp; 512 513 cpp = 1 << (PAGE_SHIFT - osb-> 514 515 cluster_start = cpos % cpp; 516 cluster_start = cluster_start 517 518 cluster_end = cluster_start + 519 } 520 521 BUG_ON(cluster_start > PAGE_SIZE); 522 BUG_ON(cluster_end > PAGE_SIZE); 523 524 if (start) 525 *start = cluster_start; 526 if (end) 527 *end = cluster_end; 528 } 529 530 /* 531 * 'from' and 'to' are the region in the page 532 * 533 * If pagesize > clustersize, this function wi 534 * of the cluster boundary. 535 * 536 * from == to == 0 is code for "zero the entir 537 */ 538 static void ocfs2_clear_page_regions(struct pa 539 struct oc 540 unsigned 541 { 542 void *kaddr; 543 unsigned int cluster_start, cluster_en 544 545 ocfs2_figure_cluster_boundaries(osb, c 546 547 kaddr = kmap_atomic(page); 548 549 if (from || to) { 550 if (from > cluster_start) 551 memset(kaddr + cluster 552 if (to < cluster_end) 553 memset(kaddr + to, 0, 554 } else { 555 memset(kaddr + cluster_start, 556 } 557 558 kunmap_atomic(kaddr); 559 } 560 561 /* 562 * Nonsparse file systems fully allocate befor 563 * code. This prevents ocfs2_write() from tagg 564 * allocating one, which means ocfs2_map_page_ 565 * read-in the blocks at the tail of our file. 566 * testing i_size against each block offset. 567 */ 568 static int ocfs2_should_read_blk(struct inode 569 unsigned int 570 { 571 u64 offset = folio_pos(folio) + block_ 572 573 if (ocfs2_sparse_alloc(OCFS2_SB(inode- 574 return 1; 575 576 if (i_size_read(inode) > offset) 577 return 1; 578 579 return 0; 580 } 581 582 /* 583 * Some of this taken from __block_write_begin 584 * mapping by now though, and the entire write 585 * it won't, so not much need to use BH_New. 586 * 587 * This will also skip zeroing, which is handl 588 */ 589 int ocfs2_map_page_blocks(struct page *page, u 590 struct inode *inode, 591 unsigned int to, int 592 { 593 struct folio *folio = page_folio(page) 594 int ret = 0; 595 struct buffer_head *head, *bh, *wait[2 596 unsigned int block_end, block_start; 597 unsigned int bsize = i_blocksize(inode 598 599 head = folio_buffers(folio); 600 if (!head) 601 head = create_empty_buffers(fo 602 603 for (bh = head, block_start = 0; bh != 604 bh = bh->b_this_page, block_start 605 block_end = block_start + bsiz 606 607 clear_buffer_new(bh); 608 609 /* 610 * Ignore blocks outside of ou 611 * they may belong to unalloca 612 */ 613 if (block_start >= to || block 614 if (folio_test_uptodat 615 set_buffer_upt 616 continue; 617 } 618 619 /* 620 * For an allocating write wit 621 * size, we always write the e 622 */ 623 if (new) 624 set_buffer_new(bh); 625 626 if (!buffer_mapped(bh)) { 627 map_bh(bh, inode->i_sb 628 clean_bdev_bh_alias(bh 629 } 630 631 if (folio_test_uptodate(folio) 632 set_buffer_uptodate(bh 633 } else if (!buffer_uptodate(bh 634 !buffer_new(bh) && 635 ocfs2_should_read_b 636 (block_start < from 637 bh_read_nowait(bh, 0); 638 *wait_bh++=bh; 639 } 640 641 *p_blkno = *p_blkno + 1; 642 } 643 644 /* 645 * If we issued read requests - let th 646 */ 647 while(wait_bh > wait) { 648 wait_on_buffer(*--wait_bh); 649 if (!buffer_uptodate(*wait_bh) 650 ret = -EIO; 651 } 652 653 if (ret == 0 || !new) 654 return ret; 655 656 /* 657 * If we get -EIO above, zero out any 658 * to avoid exposing stale data. 659 */ 660 bh = head; 661 block_start = 0; 662 do { 663 block_end = block_start + bsiz 664 if (block_end <= from) 665 goto next_bh; 666 if (block_start >= to) 667 break; 668 669 folio_zero_range(folio, block_ 670 set_buffer_uptodate(bh); 671 mark_buffer_dirty(bh); 672 673 next_bh: 674 block_start = block_end; 675 bh = bh->b_this_page; 676 } while (bh != head); 677 678 return ret; 679 } 680 681 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 682 #define OCFS2_MAX_CTXT_PAGES 1 683 #else 684 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLU 685 #endif 686 687 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_ 688 689 struct ocfs2_unwritten_extent { 690 struct list_head ue_node; 691 struct list_head ue_ip_node; 692 u32 ue_cpos; 693 u32 ue_phys; 694 }; 695 696 /* 697 * Describe the state of a single cluster to b 698 */ 699 struct ocfs2_write_cluster_desc { 700 u32 c_cpos; 701 u32 c_phys; 702 /* 703 * Give this a unique field because c_ 704 * filled. 705 */ 706 unsigned c_new; 707 unsigned c_clear_unwritten; 708 unsigned c_needs_zero; 709 }; 710 711 struct ocfs2_write_ctxt { 712 /* Logical cluster position / len of w 713 u32 w_cpos 714 u32 w_clen 715 716 /* First cluster allocated in a nonspa 717 u32 w_firs 718 719 /* Type of caller. Must be one of buff 720 ocfs2_write_type_t w_type 721 722 struct ocfs2_write_cluster_desc w_desc 723 724 /* 725 * This is true if page_size > cluster 726 * 727 * It triggers a set of special cases 728 * have to deal with allocating writes 729 */ 730 unsigned int w_larg 731 732 /* 733 * Pages involved in this write. 734 * 735 * w_target_page is the page being wri 736 * 737 * w_pages is an array of pages which 738 * w_target_page, and in the case of a 739 * page_size < cluster size, it will c 740 * pages adjacent to w_target_page whi 741 * out in so that future reads from th 742 * zero's. 743 */ 744 unsigned int w_num_ 745 struct page *w_pag 746 struct page *w_tar 747 748 /* 749 * w_target_locked is used for page_mk 750 * against w_target_page in ocfs2_writ 751 */ 752 unsigned int w_targ 753 754 /* 755 * ocfs2_write_end() uses this to know 756 * write in the target should be. 757 */ 758 unsigned int w_targ 759 unsigned int w_targ 760 761 /* 762 * We could use journal_current_handle 763 * IMHO -Mark 764 */ 765 handle_t *w_han 766 767 struct buffer_head *w_di_ 768 769 struct ocfs2_cached_dealloc_ctxt w_dea 770 771 struct list_head w_unwr 772 unsigned int w_unwr 773 }; 774 775 void ocfs2_unlock_and_free_pages(struct page * 776 { 777 int i; 778 779 for(i = 0; i < num_pages; i++) { 780 if (pages[i]) { 781 unlock_page(pages[i]); 782 mark_page_accessed(pag 783 put_page(pages[i]); 784 } 785 } 786 } 787 788 static void ocfs2_unlock_pages(struct ocfs2_wr 789 { 790 int i; 791 792 /* 793 * w_target_locked is only set to true 794 * The intent is to allow us to lock t 795 * to write_end(). The caller must hol 796 */ 797 if (wc->w_target_locked) { 798 BUG_ON(!wc->w_target_page); 799 for (i = 0; i < wc->w_num_page 800 if (wc->w_target_page 801 wc->w_pages[i] 802 break; 803 } 804 } 805 mark_page_accessed(wc->w_targe 806 put_page(wc->w_target_page); 807 } 808 ocfs2_unlock_and_free_pages(wc->w_page 809 } 810 811 static void ocfs2_free_unwritten_list(struct i 812 struct list_h 813 { 814 struct ocfs2_inode_info *oi = OCFS2_I( 815 struct ocfs2_unwritten_extent *ue = NU 816 817 list_for_each_entry_safe(ue, tmp, head 818 list_del(&ue->ue_node); 819 spin_lock(&oi->ip_lock); 820 list_del(&ue->ue_ip_node); 821 spin_unlock(&oi->ip_lock); 822 kfree(ue); 823 } 824 } 825 826 static void ocfs2_free_write_ctxt(struct inode 827 struct ocfs2 828 { 829 ocfs2_free_unwritten_list(inode, &wc-> 830 ocfs2_unlock_pages(wc); 831 brelse(wc->w_di_bh); 832 kfree(wc); 833 } 834 835 static int ocfs2_alloc_write_ctxt(struct ocfs2 836 struct ocfs2 837 unsigned len 838 struct buffe 839 { 840 u32 cend; 841 struct ocfs2_write_ctxt *wc; 842 843 wc = kzalloc(sizeof(struct ocfs2_write 844 if (!wc) 845 return -ENOMEM; 846 847 wc->w_cpos = pos >> osb->s_clustersize 848 wc->w_first_new_cpos = UINT_MAX; 849 cend = (pos + len - 1) >> osb->s_clust 850 wc->w_clen = cend - wc->w_cpos + 1; 851 get_bh(di_bh); 852 wc->w_di_bh = di_bh; 853 wc->w_type = type; 854 855 if (unlikely(PAGE_SHIFT > osb->s_clust 856 wc->w_large_pages = 1; 857 else 858 wc->w_large_pages = 0; 859 860 ocfs2_init_dealloc_ctxt(&wc->w_dealloc 861 INIT_LIST_HEAD(&wc->w_unwritten_list); 862 863 *wcp = wc; 864 865 return 0; 866 } 867 868 /* 869 * If a page has any new buffers, zero them ou 870 * and dirty so they'll be written out (in ord 871 * block data from leaking). And clear the new 872 */ 873 static void ocfs2_zero_new_buffers(struct page 874 { 875 unsigned int block_start, block_end; 876 struct buffer_head *head, *bh; 877 878 BUG_ON(!PageLocked(page)); 879 if (!page_has_buffers(page)) 880 return; 881 882 bh = head = page_buffers(page); 883 block_start = 0; 884 do { 885 block_end = block_start + bh-> 886 887 if (buffer_new(bh)) { 888 if (block_end > from & 889 if (!PageUptod 890 unsign 891 892 start 893 end = 894 895 zero_u 896 set_bu 897 } 898 899 clear_buffer_n 900 mark_buffer_di 901 } 902 } 903 904 block_start = block_end; 905 bh = bh->b_this_page; 906 } while (bh != head); 907 } 908 909 /* 910 * Only called when we have a failure during a 911 * zero's to the newly allocated region. 912 */ 913 static void ocfs2_write_failure(struct inode * 914 struct ocfs2_w 915 loff_t user_po 916 { 917 int i; 918 unsigned from = user_pos & (PAGE_SIZE 919 to = user_pos + user_len; 920 struct page *tmppage; 921 922 if (wc->w_target_page) 923 ocfs2_zero_new_buffers(wc->w_t 924 925 for(i = 0; i < wc->w_num_pages; i++) { 926 tmppage = wc->w_pages[i]; 927 928 if (tmppage && page_has_buffer 929 if (ocfs2_should_order 930 ocfs2_jbd2_ino 931 932 933 block_commit_write(tmp 934 } 935 } 936 } 937 938 static int ocfs2_prepare_page_for_write(struct 939 struct 940 struct 941 loff_t 942 int ne 943 { 944 int ret; 945 unsigned int map_from = 0, map_to = 0; 946 unsigned int cluster_start, cluster_en 947 unsigned int user_data_from = 0, user_ 948 949 ocfs2_figure_cluster_boundaries(OCFS2_ 950 &clust 951 952 /* treat the write as new if the a hol 953 * the page boundary. 954 */ 955 new = new | ((i_size_read(inode) <= pa 956 (page_offset(page) <= 957 958 if (page == wc->w_target_page) { 959 map_from = user_pos & (PAGE_SI 960 map_to = map_from + user_len; 961 962 if (new) 963 ret = ocfs2_map_page_b 964 965 966 else 967 ret = ocfs2_map_page_b 968 969 if (ret) { 970 mlog_errno(ret); 971 goto out; 972 } 973 974 user_data_from = map_from; 975 user_data_to = map_to; 976 if (new) { 977 map_from = cluster_sta 978 map_to = cluster_end; 979 } 980 } else { 981 /* 982 * If we haven't allocated the 983 * shouldn't be writing it out 984 * data. This is likely a math 985 */ 986 BUG_ON(!new); 987 988 map_from = cluster_start; 989 map_to = cluster_end; 990 991 ret = ocfs2_map_page_blocks(pa 992 cl 993 if (ret) { 994 mlog_errno(ret); 995 goto out; 996 } 997 } 998 999 /* 1000 * Parts of newly allocated pages nee 1001 * 1002 * Above, we have also rewritten 'to' 1003 * the rest of the function is concer 1004 * range inside of a page needs to be 1005 * 1006 * We can skip this if the page is up 1007 * been zero'd from being read in as 1008 */ 1009 if (new && !PageUptodate(page)) 1010 ocfs2_clear_page_regions(page 1011 cpos 1012 1013 flush_dcache_page(page); 1014 1015 out: 1016 return ret; 1017 } 1018 1019 /* 1020 * This function will only grab one clusters 1021 */ 1022 static int ocfs2_grab_pages_for_write(struct 1023 struct 1024 u32 cpo 1025 unsigne 1026 struct 1027 { 1028 int ret = 0, i; 1029 unsigned long start, target_index, en 1030 struct inode *inode = mapping->host; 1031 loff_t last_byte; 1032 1033 target_index = user_pos >> PAGE_SHIFT 1034 1035 /* 1036 * Figure out how many pages we'll be 1037 * non allocating write, we just chan 1038 * page. Otherwise, we'll need a whol 1039 * writing past i_size, we only need 1040 * last page of the write. 1041 */ 1042 if (new) { 1043 wc->w_num_pages = ocfs2_pages 1044 start = ocfs2_align_clusters_ 1045 /* 1046 * We need the index *past* t 1047 * touch. This is the page p 1048 * i_size, whichever is great 1049 */ 1050 last_byte = max(user_pos + us 1051 BUG_ON(last_byte < 1); 1052 end_index = ((last_byte - 1) 1053 if ((start + wc->w_num_pages) 1054 wc->w_num_pages = end 1055 } else { 1056 wc->w_num_pages = 1; 1057 start = target_index; 1058 } 1059 end_index = (user_pos + user_len - 1) 1060 1061 for(i = 0; i < wc->w_num_pages; i++) 1062 index = start + i; 1063 1064 if (index >= target_index && 1065 wc->w_type == OCFS2_WRITE 1066 /* 1067 * ocfs2_pagemkwrite( 1068 * and wants us to di 1069 * passed in. 1070 */ 1071 lock_page(mmap_page); 1072 1073 /* Exit and let the c 1074 if (mmap_page->mappin 1075 WARN_ON(mmap_ 1076 unlock_page(m 1077 ret = -EAGAIN 1078 goto out; 1079 } 1080 1081 get_page(mmap_page); 1082 wc->w_pages[i] = mmap 1083 wc->w_target_locked = 1084 } else if (index >= target_in 1085 wc->w_type == OCFS 1086 /* Direct write has n 1087 wc->w_pages[i] = NULL 1088 continue; 1089 } else { 1090 wc->w_pages[i] = find 1091 1092 if (!wc->w_pages[i]) 1093 ret = -ENOMEM 1094 mlog_errno(re 1095 goto out; 1096 } 1097 } 1098 wait_for_stable_page(wc->w_pa 1099 1100 if (index == target_index) 1101 wc->w_target_page = w 1102 } 1103 out: 1104 if (ret) 1105 wc->w_target_locked = false; 1106 return ret; 1107 } 1108 1109 /* 1110 * Prepare a single cluster for write one clu 1111 */ 1112 static int ocfs2_write_cluster(struct address 1113 u32 *phys, uns 1114 unsigned int c 1115 unsigned int s 1116 struct ocfs2_a 1117 struct ocfs2_a 1118 struct ocfs2_w 1119 loff_t user_po 1120 { 1121 int ret, i; 1122 u64 p_blkno; 1123 struct inode *inode = mapping->host; 1124 struct ocfs2_extent_tree et; 1125 int bpc = ocfs2_clusters_to_blocks(in 1126 1127 if (new) { 1128 u32 tmp_pos; 1129 1130 /* 1131 * This is safe to call with 1132 * any additional semaphores 1133 */ 1134 tmp_pos = cpos; 1135 ret = ocfs2_add_inode_data(OC 1136 &t 1137 wc 1138 da 1139 /* 1140 * This shouldn't happen beca 1141 * calculated the correct met 1142 * internal tree allocation c 1143 * transaction credits itself 1144 * 1145 * If need be, we could handl 1146 * RESTART_TRANS here. 1147 */ 1148 mlog_bug_on_msg(ret == -EAGAI 1149 "Inode %llu: 1150 (unsigned lon 1151 if (ret < 0) { 1152 mlog_errno(ret); 1153 goto out; 1154 } 1155 } else if (clear_unwritten) { 1156 ocfs2_init_dinode_extent_tree 1157 1158 ret = ocfs2_mark_extent_writt 1159 1160 1161 if (ret < 0) { 1162 mlog_errno(ret); 1163 goto out; 1164 } 1165 } 1166 1167 /* 1168 * The only reason this should fail i 1169 * find the extent added. 1170 */ 1171 ret = ocfs2_get_clusters(inode, cpos, 1172 if (ret < 0) { 1173 mlog(ML_ERROR, "Get physical 1174 "at logical clust 1175 (unsigned long lo 1176 goto out; 1177 } 1178 1179 BUG_ON(*phys == 0); 1180 1181 p_blkno = ocfs2_clusters_to_blocks(in 1182 if (!should_zero) 1183 p_blkno += (user_pos >> inode 1184 1185 for(i = 0; i < wc->w_num_pages; i++) 1186 int tmpret; 1187 1188 /* This is the direct io targ 1189 if (wc->w_pages[i] == NULL) { 1190 p_blkno++; 1191 continue; 1192 } 1193 1194 tmpret = ocfs2_prepare_page_f 1195 1196 1197 1198 if (tmpret) { 1199 mlog_errno(tmpret); 1200 if (ret == 0) 1201 ret = tmpret; 1202 } 1203 } 1204 1205 /* 1206 * We only have cleanup to do in case 1207 */ 1208 if (ret && new) 1209 ocfs2_write_failure(inode, wc 1210 1211 out: 1212 1213 return ret; 1214 } 1215 1216 static int ocfs2_write_cluster_by_desc(struct 1217 struct 1218 struct 1219 struct 1220 loff_t 1221 { 1222 int ret, i; 1223 loff_t cluster_off; 1224 unsigned int local_len = len; 1225 struct ocfs2_write_cluster_desc *desc 1226 struct ocfs2_super *osb = OCFS2_SB(ma 1227 1228 for (i = 0; i < wc->w_clen; i++) { 1229 desc = &wc->w_desc[i]; 1230 1231 /* 1232 * We have to make sure that 1233 * doesn't extend past a sing 1234 */ 1235 local_len = len; 1236 cluster_off = pos & (osb->s_c 1237 if ((cluster_off + local_len) 1238 local_len = osb->s_cl 1239 1240 ret = ocfs2_write_cluster(map 1241 des 1242 des 1243 des 1244 dat 1245 wc, 1246 if (ret) { 1247 mlog_errno(ret); 1248 goto out; 1249 } 1250 1251 len -= local_len; 1252 pos += local_len; 1253 } 1254 1255 ret = 0; 1256 out: 1257 return ret; 1258 } 1259 1260 /* 1261 * ocfs2_write_end() wants to know which part 1262 * should complete the write on. It's easiest 1263 * time when a more complete view of the writ 1264 */ 1265 static void ocfs2_set_target_boundaries(struc 1266 struc 1267 loff_ 1268 { 1269 struct ocfs2_write_cluster_desc *desc 1270 1271 wc->w_target_from = pos & (PAGE_SIZE 1272 wc->w_target_to = wc->w_target_from + 1273 1274 if (alloc == 0) 1275 return; 1276 1277 /* 1278 * Allocating write - we may have dif 1279 * on page size and cluster size. 1280 * 1281 * NOTE: We can no longer compute one 1282 * the actual write length and user p 1283 * different. 1284 */ 1285 1286 if (wc->w_large_pages) { 1287 /* 1288 * We only care about the 1st 1289 * our range and whether they 1290 * value may be extended out 1291 * newly allocated cluster. 1292 */ 1293 desc = &wc->w_desc[0]; 1294 if (desc->c_needs_zero) 1295 ocfs2_figure_cluster_ 1296 1297 1298 1299 1300 desc = &wc->w_desc[wc->w_clen 1301 if (desc->c_needs_zero) 1302 ocfs2_figure_cluster_ 1303 1304 1305 1306 } else { 1307 wc->w_target_from = 0; 1308 wc->w_target_to = PAGE_SIZE; 1309 } 1310 } 1311 1312 /* 1313 * Check if this extent is marked UNWRITTEN b 1314 * do the zero work. And should not to clear 1315 * by the direct io procedure. 1316 * If this is a new extent that allocated by 1317 * the ip_unwritten_list. 1318 */ 1319 static int ocfs2_unwritten_check(struct inode 1320 struct ocfs2 1321 struct ocfs2 1322 { 1323 struct ocfs2_inode_info *oi = OCFS2_I 1324 struct ocfs2_unwritten_extent *ue = N 1325 int ret = 0; 1326 1327 if (!desc->c_needs_zero) 1328 return 0; 1329 1330 retry: 1331 spin_lock(&oi->ip_lock); 1332 /* Needs not to zero no metter buffer 1333 * the cluster is doing zero. And he 1334 * cluster io finished. */ 1335 list_for_each_entry(ue, &oi->ip_unwri 1336 if (desc->c_cpos == ue->ue_cp 1337 BUG_ON(desc->c_new); 1338 desc->c_needs_zero = 1339 desc->c_clear_unwritt 1340 goto unlock; 1341 } 1342 } 1343 1344 if (wc->w_type != OCFS2_WRITE_DIRECT) 1345 goto unlock; 1346 1347 if (new == NULL) { 1348 spin_unlock(&oi->ip_lock); 1349 new = kmalloc(sizeof(struct o 1350 GFP_NOFS); 1351 if (new == NULL) { 1352 ret = -ENOMEM; 1353 goto out; 1354 } 1355 goto retry; 1356 } 1357 /* This direct write will doing zero. 1358 new->ue_cpos = desc->c_cpos; 1359 new->ue_phys = desc->c_phys; 1360 desc->c_clear_unwritten = 0; 1361 list_add_tail(&new->ue_ip_node, &oi-> 1362 list_add_tail(&new->ue_node, &wc->w_u 1363 wc->w_unwritten_count++; 1364 new = NULL; 1365 unlock: 1366 spin_unlock(&oi->ip_lock); 1367 out: 1368 kfree(new); 1369 return ret; 1370 } 1371 1372 /* 1373 * Populate each single-cluster write descrip 1374 * with information about the i/o to be done. 1375 * 1376 * Returns the number of clusters that will h 1377 * well as a worst case estimate of the numbe 1378 * would have to be created during a write to 1379 */ 1380 static int ocfs2_populate_write_desc(struct i 1381 struct o 1382 unsigned 1383 unsigned 1384 { 1385 int ret; 1386 struct ocfs2_write_cluster_desc *desc 1387 unsigned int num_clusters = 0; 1388 unsigned int ext_flags = 0; 1389 u32 phys = 0; 1390 int i; 1391 1392 *clusters_to_alloc = 0; 1393 *extents_to_split = 0; 1394 1395 for (i = 0; i < wc->w_clen; i++) { 1396 desc = &wc->w_desc[i]; 1397 desc->c_cpos = wc->w_cpos + i 1398 1399 if (num_clusters == 0) { 1400 /* 1401 * Need to look up th 1402 */ 1403 ret = ocfs2_get_clust 1404 1405 if (ret) { 1406 mlog_errno(re 1407 goto out; 1408 } 1409 1410 /* We should already 1411 BUG_ON(ext_flags & OC 1412 1413 /* 1414 * Assume worst case 1415 * the middle of the 1416 * 1417 * We can assume that 1418 * left to right, in 1419 * insert code is sma 1420 * next splits into t 1421 */ 1422 if (ext_flags & OCFS2 1423 *extents_to_s 1424 } else if (phys) { 1425 /* 1426 * Only increment phy 1427 * a hole. 1428 */ 1429 phys++; 1430 } 1431 1432 /* 1433 * If w_first_new_cpos is < U 1434 * file that got extended. w 1435 * where the newly allocated 1436 * zero them. 1437 */ 1438 if (desc->c_cpos >= wc->w_fir 1439 BUG_ON(phys == 0); 1440 desc->c_needs_zero = 1441 } 1442 1443 desc->c_phys = phys; 1444 if (phys == 0) { 1445 desc->c_new = 1; 1446 desc->c_needs_zero = 1447 desc->c_clear_unwritt 1448 *clusters_to_alloc = 1449 } 1450 1451 if (ext_flags & OCFS2_EXT_UNW 1452 desc->c_clear_unwritt 1453 desc->c_needs_zero = 1454 } 1455 1456 ret = ocfs2_unwritten_check(i 1457 if (ret) { 1458 mlog_errno(ret); 1459 goto out; 1460 } 1461 1462 num_clusters--; 1463 } 1464 1465 ret = 0; 1466 out: 1467 return ret; 1468 } 1469 1470 static int ocfs2_write_begin_inline(struct ad 1471 struct in 1472 struct oc 1473 { 1474 int ret; 1475 struct ocfs2_super *osb = OCFS2_SB(in 1476 struct page *page; 1477 handle_t *handle; 1478 struct ocfs2_dinode *di = (struct ocf 1479 1480 handle = ocfs2_start_trans(osb, OCFS2 1481 if (IS_ERR(handle)) { 1482 ret = PTR_ERR(handle); 1483 mlog_errno(ret); 1484 goto out; 1485 } 1486 1487 page = find_or_create_page(mapping, 0 1488 if (!page) { 1489 ocfs2_commit_trans(osb, handl 1490 ret = -ENOMEM; 1491 mlog_errno(ret); 1492 goto out; 1493 } 1494 /* 1495 * If we don't set w_num_pages then t 1496 * and freed on cleanup of the write 1497 */ 1498 wc->w_pages[0] = wc->w_target_page = 1499 wc->w_num_pages = 1; 1500 1501 ret = ocfs2_journal_access_di(handle, 1502 OCFS2_J 1503 if (ret) { 1504 ocfs2_commit_trans(osb, handl 1505 1506 mlog_errno(ret); 1507 goto out; 1508 } 1509 1510 if (!(OCFS2_I(inode)->ip_dyn_features 1511 ocfs2_set_inode_data_inline(i 1512 1513 if (!PageUptodate(page)) { 1514 ret = ocfs2_read_inline_data( 1515 if (ret) { 1516 ocfs2_commit_trans(os 1517 1518 goto out; 1519 } 1520 } 1521 1522 wc->w_handle = handle; 1523 out: 1524 return ret; 1525 } 1526 1527 int ocfs2_size_fits_inline_data(struct buffer 1528 { 1529 struct ocfs2_dinode *di = (struct ocf 1530 1531 if (new_size <= le16_to_cpu(di->id2.i 1532 return 1; 1533 return 0; 1534 } 1535 1536 static int ocfs2_try_to_write_inline_data(str 1537 str 1538 uns 1539 str 1540 { 1541 int ret, written = 0; 1542 loff_t end = pos + len; 1543 struct ocfs2_inode_info *oi = OCFS2_I 1544 struct ocfs2_dinode *di = NULL; 1545 1546 trace_ocfs2_try_to_write_inline_data( 1547 1548 1549 1550 /* 1551 * Handle inodes which already have i 1552 */ 1553 if (oi->ip_dyn_features & OCFS2_INLIN 1554 if (mmap_page == NULL && 1555 ocfs2_size_fits_inline_da 1556 goto do_inline_write; 1557 1558 /* 1559 * The write won't fit - we h 1560 * inline extent list now. 1561 */ 1562 ret = ocfs2_convert_inline_da 1563 if (ret) 1564 mlog_errno(ret); 1565 goto out; 1566 } 1567 1568 /* 1569 * Check whether the inode can accept 1570 */ 1571 if (oi->ip_clusters != 0 || i_size_re 1572 return 0; 1573 1574 /* 1575 * Check whether the write can fit. 1576 */ 1577 di = (struct ocfs2_dinode *)wc->w_di_ 1578 if (mmap_page || 1579 end > ocfs2_max_inline_data_with_ 1580 return 0; 1581 1582 do_inline_write: 1583 ret = ocfs2_write_begin_inline(mappin 1584 if (ret) { 1585 mlog_errno(ret); 1586 goto out; 1587 } 1588 1589 /* 1590 * This signals to the caller that th 1591 * inline. 1592 */ 1593 written = 1; 1594 out: 1595 return written ? written : ret; 1596 } 1597 1598 /* 1599 * This function only does anything for file 1600 * handle sparse files. 1601 * 1602 * What we want to do here is fill in any hol 1603 * of allocation and the end of our write. Th 1604 * write path can treat it as an non-allocati 1605 * special case code for sparse/nonsparse fil 1606 */ 1607 static int ocfs2_expand_nonsparse_inode(struc 1608 struc 1609 loff_ 1610 struc 1611 { 1612 int ret; 1613 loff_t newsize = pos + len; 1614 1615 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(in 1616 1617 if (newsize <= i_size_read(inode)) 1618 return 0; 1619 1620 ret = ocfs2_extend_no_holes(inode, di 1621 if (ret) 1622 mlog_errno(ret); 1623 1624 /* There is no wc if this is call fro 1625 if (wc) 1626 wc->w_first_new_cpos = 1627 ocfs2_clusters_for_by 1628 1629 return ret; 1630 } 1631 1632 static int ocfs2_zero_tail(struct inode *inod 1633 loff_t pos) 1634 { 1635 int ret = 0; 1636 1637 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(i 1638 if (pos > i_size_read(inode)) 1639 ret = ocfs2_zero_extend(inode 1640 1641 return ret; 1642 } 1643 1644 int ocfs2_write_begin_nolock(struct address_s 1645 loff_t pos, unsi 1646 struct page **pa 1647 struct buffer_he 1648 { 1649 int ret, cluster_of_pages, credits = 1650 unsigned int clusters_to_alloc, exten 1651 struct ocfs2_write_ctxt *wc; 1652 struct inode *inode = mapping->host; 1653 struct ocfs2_super *osb = OCFS2_SB(in 1654 struct ocfs2_dinode *di; 1655 struct ocfs2_alloc_context *data_ac = 1656 struct ocfs2_alloc_context *meta_ac = 1657 handle_t *handle; 1658 struct ocfs2_extent_tree et; 1659 int try_free = 1, ret1; 1660 1661 try_again: 1662 ret = ocfs2_alloc_write_ctxt(&wc, osb 1663 if (ret) { 1664 mlog_errno(ret); 1665 return ret; 1666 } 1667 1668 if (ocfs2_supports_inline_data(osb)) 1669 ret = ocfs2_try_to_write_inli 1670 1671 if (ret == 1) { 1672 ret = 0; 1673 goto success; 1674 } 1675 if (ret < 0) { 1676 mlog_errno(ret); 1677 goto out; 1678 } 1679 } 1680 1681 /* Direct io change i_size late, shou 1682 if (type != OCFS2_WRITE_DIRECT) { 1683 if (ocfs2_sparse_alloc(osb)) 1684 ret = ocfs2_zero_tail 1685 else 1686 ret = ocfs2_expand_no 1687 1688 if (ret) { 1689 mlog_errno(ret); 1690 goto out; 1691 } 1692 } 1693 1694 ret = ocfs2_check_range_for_refcount( 1695 if (ret < 0) { 1696 mlog_errno(ret); 1697 goto out; 1698 } else if (ret == 1) { 1699 clusters_need = wc->w_clen; 1700 ret = ocfs2_refcount_cow(inod 1701 wc-> 1702 if (ret) { 1703 mlog_errno(ret); 1704 goto out; 1705 } 1706 } 1707 1708 ret = ocfs2_populate_write_desc(inode 1709 &exte 1710 if (ret) { 1711 mlog_errno(ret); 1712 goto out; 1713 } 1714 clusters_need += clusters_to_alloc; 1715 1716 di = (struct ocfs2_dinode *)wc->w_di_ 1717 1718 trace_ocfs2_write_begin_nolock( 1719 (unsigned long long)O 1720 (long long)i_size_rea 1721 le32_to_cpu(di->i_clu 1722 pos, len, type, mmap_ 1723 clusters_to_alloc, ex 1724 1725 /* 1726 * We set w_target_from, w_target_to 1727 * ocfs2_write_end() knows which rang 1728 * write out. An allocation requires 1729 * cluster range. 1730 */ 1731 if (clusters_to_alloc || extents_to_s 1732 /* 1733 * XXX: We are stretching the 1734 * ocfs2_lock_allocators(). I 1735 * the work to be done. 1736 */ 1737 ocfs2_init_dinode_extent_tree 1738 1739 ret = ocfs2_lock_allocators(i 1740 c 1741 & 1742 if (ret) { 1743 mlog_errno(ret); 1744 goto out; 1745 } 1746 1747 if (data_ac) 1748 data_ac->ac_resv = &O 1749 1750 credits = ocfs2_calc_extend_c 1751 1752 } else if (type == OCFS2_WRITE_DIRECT 1753 /* direct write needs not to 1754 goto success; 1755 1756 /* 1757 * We have to zero sparse allocated c 1758 * and non-sparse clusters we just ex 1759 * we know zeros will only be needed 1760 */ 1761 if (wc->w_clen && (wc->w_desc[0].c_ne 1762 wc->w_desc[wc->w_c 1763 cluster_of_pages = 1; 1764 else 1765 cluster_of_pages = 0; 1766 1767 ocfs2_set_target_boundaries(osb, wc, 1768 1769 handle = ocfs2_start_trans(osb, credi 1770 if (IS_ERR(handle)) { 1771 ret = PTR_ERR(handle); 1772 mlog_errno(ret); 1773 goto out; 1774 } 1775 1776 wc->w_handle = handle; 1777 1778 if (clusters_to_alloc) { 1779 ret = dquot_alloc_space_nodir 1780 ocfs2_clusters_to_byt 1781 if (ret) 1782 goto out_commit; 1783 } 1784 1785 ret = ocfs2_journal_access_di(handle, 1786 OCFS2_J 1787 if (ret) { 1788 mlog_errno(ret); 1789 goto out_quota; 1790 } 1791 1792 /* 1793 * Fill our page array first. That wa 1794 * that we can zero and flush if we e 1795 * extent. 1796 */ 1797 ret = ocfs2_grab_pages_for_write(mapp 1798 clus 1799 if (ret) { 1800 /* 1801 * ocfs2_grab_pages_for_write 1802 * the target page. In this c 1803 * page. This will trigger th 1804 * the operation. 1805 */ 1806 if (type == OCFS2_WRITE_MMAP 1807 BUG_ON(wc->w_target_p 1808 ret = 0; 1809 goto out_quota; 1810 } 1811 1812 mlog_errno(ret); 1813 goto out_quota; 1814 } 1815 1816 ret = ocfs2_write_cluster_by_desc(map 1817 len 1818 if (ret) { 1819 mlog_errno(ret); 1820 goto out_quota; 1821 } 1822 1823 if (data_ac) 1824 ocfs2_free_alloc_context(data 1825 if (meta_ac) 1826 ocfs2_free_alloc_context(meta 1827 1828 success: 1829 if (pagep) 1830 *pagep = wc->w_target_page; 1831 *fsdata = wc; 1832 return 0; 1833 out_quota: 1834 if (clusters_to_alloc) 1835 dquot_free_space(inode, 1836 ocfs2_clusters_to_b 1837 out_commit: 1838 ocfs2_commit_trans(osb, handle); 1839 1840 out: 1841 /* 1842 * The mmapped page won't be unlocked 1843 * even in case of error here like EN 1844 * to unlock the target page manually 1845 * retrying again on ENOSPC, or when 1846 * to VM code. 1847 */ 1848 if (wc->w_target_locked) 1849 unlock_page(mmap_page); 1850 1851 ocfs2_free_write_ctxt(inode, wc); 1852 1853 if (data_ac) { 1854 ocfs2_free_alloc_context(data 1855 data_ac = NULL; 1856 } 1857 if (meta_ac) { 1858 ocfs2_free_alloc_context(meta 1859 meta_ac = NULL; 1860 } 1861 1862 if (ret == -ENOSPC && try_free) { 1863 /* 1864 * Try to free some truncate 1865 * clusters to allocate. 1866 */ 1867 try_free = 0; 1868 1869 ret1 = ocfs2_try_to_free_trun 1870 if (ret1 == 1) 1871 goto try_again; 1872 1873 if (ret1 < 0) 1874 mlog_errno(ret1); 1875 } 1876 1877 return ret; 1878 } 1879 1880 static int ocfs2_write_begin(struct file *fil 1881 loff_t pos, unsi 1882 struct page **pa 1883 { 1884 int ret; 1885 struct buffer_head *di_bh = NULL; 1886 struct inode *inode = mapping->host; 1887 1888 ret = ocfs2_inode_lock(inode, &di_bh, 1889 if (ret) { 1890 mlog_errno(ret); 1891 return ret; 1892 } 1893 1894 /* 1895 * Take alloc sem here to prevent con 1896 * the mapping, zeroing and tree mani 1897 * ocfs2_write() will be safe against 1898 * should also serve to lock out allo 1899 * writeable region. 1900 */ 1901 down_write(&OCFS2_I(inode)->ip_alloc_ 1902 1903 ret = ocfs2_write_begin_nolock(mappin 1904 pagep, 1905 if (ret) { 1906 mlog_errno(ret); 1907 goto out_fail; 1908 } 1909 1910 brelse(di_bh); 1911 1912 return 0; 1913 1914 out_fail: 1915 up_write(&OCFS2_I(inode)->ip_alloc_se 1916 1917 brelse(di_bh); 1918 ocfs2_inode_unlock(inode, 1); 1919 1920 return ret; 1921 } 1922 1923 static void ocfs2_write_end_inline(struct ino 1924 unsigned l 1925 struct ocf 1926 struct ocf 1927 { 1928 void *kaddr; 1929 1930 if (unlikely(*copied < len)) { 1931 if (!PageUptodate(wc->w_targe 1932 *copied = 0; 1933 return; 1934 } 1935 } 1936 1937 kaddr = kmap_atomic(wc->w_target_page 1938 memcpy(di->id2.i_data.id_data + pos, 1939 kunmap_atomic(kaddr); 1940 1941 trace_ocfs2_write_end_inline( 1942 (unsigned long long)OCFS2_I(inod 1943 (unsigned long long)pos, *copied 1944 le16_to_cpu(di->id2.i_data.id_co 1945 le16_to_cpu(di->i_dyn_features)) 1946 } 1947 1948 int ocfs2_write_end_nolock(struct address_spa 1949 loff_t pos, unsign 1950 { 1951 int i, ret; 1952 unsigned from, to, start = pos & (PAG 1953 struct inode *inode = mapping->host; 1954 struct ocfs2_super *osb = OCFS2_SB(in 1955 struct ocfs2_write_ctxt *wc = fsdata; 1956 struct ocfs2_dinode *di = (struct ocf 1957 handle_t *handle = wc->w_handle; 1958 struct page *tmppage; 1959 1960 BUG_ON(!list_empty(&wc->w_unwritten_l 1961 1962 if (handle) { 1963 ret = ocfs2_journal_access_di 1964 wc->w_di_bh, 1965 if (ret) { 1966 copied = ret; 1967 mlog_errno(ret); 1968 goto out; 1969 } 1970 } 1971 1972 if (OCFS2_I(inode)->ip_dyn_features & 1973 ocfs2_write_end_inline(inode, 1974 goto out_write_size; 1975 } 1976 1977 if (unlikely(copied < len) && wc->w_t 1978 loff_t new_isize; 1979 1980 if (!PageUptodate(wc->w_targe 1981 copied = 0; 1982 1983 new_isize = max_t(loff_t, i_s 1984 if (new_isize > page_offset(w 1985 ocfs2_zero_new_buffer 1986 1987 else { 1988 /* 1989 * When page is fully 1990 * failed), do not bo 1991 * it instead so that 1992 * put page & buffer 1993 * state. 1994 */ 1995 block_invalidate_foli 1996 1997 } 1998 } 1999 if (wc->w_target_page) 2000 flush_dcache_page(wc->w_targe 2001 2002 for(i = 0; i < wc->w_num_pages; i++) 2003 tmppage = wc->w_pages[i]; 2004 2005 /* This is the direct io targ 2006 if (tmppage == NULL) 2007 continue; 2008 2009 if (tmppage == wc->w_target_p 2010 from = wc->w_target_f 2011 to = wc->w_target_to; 2012 2013 BUG_ON(from > PAGE_SI 2014 to > PAGE_SIZE 2015 to < from); 2016 } else { 2017 /* 2018 * Pages adjacent to 2019 * a hole-filling wri 2020 * to flush their ent 2021 */ 2022 from = 0; 2023 to = PAGE_SIZE; 2024 } 2025 2026 if (page_has_buffers(tmppage) 2027 if (handle && ocfs2_s 2028 loff_t start_ 2029 ((lof 2030 from; 2031 loff_t length 2032 ocfs2_jbd2_in 2033 2034 } 2035 block_commit_write(tm 2036 } 2037 } 2038 2039 out_write_size: 2040 /* Direct io do not update i_size her 2041 if (wc->w_type != OCFS2_WRITE_DIRECT) 2042 pos += copied; 2043 if (pos > i_size_read(inode)) 2044 i_size_write(inode, p 2045 mark_inode_dirty(inod 2046 } 2047 inode->i_blocks = ocfs2_inode 2048 di->i_size = cpu_to_le64((u64 2049 inode_set_mtime_to_ts(inode, 2050 di->i_mtime = di->i_ctime = c 2051 di->i_mtime_nsec = di->i_ctim 2052 if (handle) 2053 ocfs2_update_inode_fs 2054 } 2055 if (handle) 2056 ocfs2_journal_dirty(handle, w 2057 2058 out: 2059 /* unlock pages before dealloc since 2060 * lock, or it will cause a deadlock 2061 * this lock and will ask for the pag 2062 * put it here to preserve the unlock 2063 */ 2064 ocfs2_unlock_pages(wc); 2065 2066 if (handle) 2067 ocfs2_commit_trans(osb, handl 2068 2069 ocfs2_run_deallocs(osb, &wc->w_deallo 2070 2071 brelse(wc->w_di_bh); 2072 kfree(wc); 2073 2074 return copied; 2075 } 2076 2077 static int ocfs2_write_end(struct file *file, 2078 loff_t pos, unsign 2079 struct page *page, 2080 { 2081 int ret; 2082 struct inode *inode = mapping->host; 2083 2084 ret = ocfs2_write_end_nolock(mapping, 2085 2086 up_write(&OCFS2_I(inode)->ip_alloc_se 2087 ocfs2_inode_unlock(inode, 1); 2088 2089 return ret; 2090 } 2091 2092 struct ocfs2_dio_write_ctxt { 2093 struct list_head dw_zero_list; 2094 unsigned dw_zero_count 2095 int dw_orphaned; 2096 pid_t dw_writer_pid 2097 }; 2098 2099 static struct ocfs2_dio_write_ctxt * 2100 ocfs2_dio_alloc_write_ctx(struct buffer_head 2101 { 2102 struct ocfs2_dio_write_ctxt *dwc = NU 2103 2104 if (bh->b_private) 2105 return bh->b_private; 2106 2107 dwc = kmalloc(sizeof(struct ocfs2_dio 2108 if (dwc == NULL) 2109 return NULL; 2110 INIT_LIST_HEAD(&dwc->dw_zero_list); 2111 dwc->dw_zero_count = 0; 2112 dwc->dw_orphaned = 0; 2113 dwc->dw_writer_pid = task_pid_nr(curr 2114 bh->b_private = dwc; 2115 *alloc = 1; 2116 2117 return dwc; 2118 } 2119 2120 static void ocfs2_dio_free_write_ctx(struct i 2121 struct o 2122 { 2123 ocfs2_free_unwritten_list(inode, &dwc 2124 kfree(dwc); 2125 } 2126 2127 /* 2128 * TODO: Make this into a generic get_blocks 2129 * 2130 * From do_direct_io in direct-io.c: 2131 * "So what we do is to permit the ->get_blo 2132 * bh.b_size with the size of IO which is p 2133 * this i_blkbits." 2134 * 2135 * This function is called directly from get_ 2136 * 2137 * called like this: dio->get_blocks(dio->ino 2138 * fs_co 2139 */ 2140 static int ocfs2_dio_wr_get_block(struct inod 2141 struct buffer_ 2142 { 2143 struct ocfs2_super *osb = OCFS2_SB(in 2144 struct ocfs2_inode_info *oi = OCFS2_I 2145 struct ocfs2_write_ctxt *wc; 2146 struct ocfs2_write_cluster_desc *desc 2147 struct ocfs2_dio_write_ctxt *dwc = NU 2148 struct buffer_head *di_bh = NULL; 2149 u64 p_blkno; 2150 unsigned int i_blkbits = inode->i_sb- 2151 loff_t pos = iblock << i_blkbits; 2152 sector_t endblk = (i_size_read(inode) 2153 unsigned len, total_len = bh_result-> 2154 int ret = 0, first_get_block = 0; 2155 2156 len = osb->s_clustersize - (pos & (os 2157 len = min(total_len, len); 2158 2159 /* 2160 * bh_result->b_size is count in get_ 2161 * "pos" and "end", we need map twice 2162 * 1. area in file size, not set NEW; 2163 * 2. area out file size, set NEW. 2164 * 2165 * iblock endblk 2166 * |--------|---------|---------|---- 2167 * |<-------area in file------->| 2168 */ 2169 2170 if ((iblock <= endblk) && 2171 ((iblock + ((len - 1) >> i_blkbit 2172 len = (endblk - iblock + 1) < 2173 2174 mlog(0, "get block of %lu at %llu:%u 2175 inode->i_ino, pos, le 2176 2177 /* 2178 * Because we need to change file siz 2179 * we may need to add it to orphan di 2180 * while file size will be changed. 2181 */ 2182 if (pos + total_len <= i_size_read(in 2183 2184 /* This is the fast path for 2185 ret = ocfs2_lock_get_block(in 2186 if (buffer_mapped(bh_result) 2187 !buffer_new(bh_result) && 2188 ret == 0) 2189 goto out; 2190 2191 /* Clear state set by ocfs2_g 2192 bh_result->b_state = 0; 2193 } 2194 2195 dwc = ocfs2_dio_alloc_write_ctx(bh_re 2196 if (unlikely(dwc == NULL)) { 2197 ret = -ENOMEM; 2198 mlog_errno(ret); 2199 goto out; 2200 } 2201 2202 if (ocfs2_clusters_for_bytes(inode->i 2203 ocfs2_clusters_for_bytes(inode->i 2204 !dwc->dw_orphaned) { 2205 /* 2206 * when we are going to alloc 2207 * inode to orphan dir, so we 2208 * system crashed during writ 2209 */ 2210 ret = ocfs2_add_inode_to_orph 2211 if (ret < 0) { 2212 mlog_errno(ret); 2213 goto out; 2214 } 2215 dwc->dw_orphaned = 1; 2216 } 2217 2218 ret = ocfs2_inode_lock(inode, &di_bh, 2219 if (ret) { 2220 mlog_errno(ret); 2221 goto out; 2222 } 2223 2224 down_write(&oi->ip_alloc_sem); 2225 2226 if (first_get_block) { 2227 if (ocfs2_sparse_alloc(osb)) 2228 ret = ocfs2_zero_tail 2229 else 2230 ret = ocfs2_expand_no 2231 2232 if (ret < 0) { 2233 mlog_errno(ret); 2234 goto unlock; 2235 } 2236 } 2237 2238 ret = ocfs2_write_begin_nolock(inode- 2239 OCFS2_ 2240 (void 2241 if (ret) { 2242 mlog_errno(ret); 2243 goto unlock; 2244 } 2245 2246 desc = &wc->w_desc[0]; 2247 2248 p_blkno = ocfs2_clusters_to_blocks(in 2249 BUG_ON(p_blkno == 0); 2250 p_blkno += iblock & (u64)(ocfs2_clust 2251 2252 map_bh(bh_result, inode->i_sb, p_blkn 2253 bh_result->b_size = len; 2254 if (desc->c_needs_zero) 2255 set_buffer_new(bh_result); 2256 2257 if (iblock > endblk) 2258 set_buffer_new(bh_result); 2259 2260 /* May sleep in end_io. It should not 2261 * it to dio work queue. */ 2262 set_buffer_defer_completion(bh_result 2263 2264 if (!list_empty(&wc->w_unwritten_list 2265 struct ocfs2_unwritten_extent 2266 2267 ue = list_first_entry(&wc->w_ 2268 struct 2269 ue_node 2270 BUG_ON(ue->ue_cpos != desc->c 2271 /* The physical address may b 2272 ue->ue_phys = desc->c_phys; 2273 2274 list_splice_tail_init(&wc->w_ 2275 dwc->dw_zero_count += wc->w_u 2276 } 2277 2278 ret = ocfs2_write_end_nolock(inode->i 2279 BUG_ON(ret != len); 2280 ret = 0; 2281 unlock: 2282 up_write(&oi->ip_alloc_sem); 2283 ocfs2_inode_unlock(inode, 1); 2284 brelse(di_bh); 2285 out: 2286 return ret; 2287 } 2288 2289 static int ocfs2_dio_end_io_write(struct inod 2290 struct ocfs 2291 loff_t offs 2292 ssize_t byt 2293 { 2294 struct ocfs2_cached_dealloc_ctxt deal 2295 struct ocfs2_extent_tree et; 2296 struct ocfs2_super *osb = OCFS2_SB(in 2297 struct ocfs2_inode_info *oi = OCFS2_I 2298 struct ocfs2_unwritten_extent *ue = N 2299 struct buffer_head *di_bh = NULL; 2300 struct ocfs2_dinode *di; 2301 struct ocfs2_alloc_context *data_ac = 2302 struct ocfs2_alloc_context *meta_ac = 2303 handle_t *handle = NULL; 2304 loff_t end = offset + bytes; 2305 int ret = 0, credits = 0; 2306 2307 ocfs2_init_dealloc_ctxt(&dealloc); 2308 2309 /* We do clear unwritten, delete orph 2310 * of these happen, we can skip all t 2311 if (list_empty(&dwc->dw_zero_list) && 2312 end <= i_size_read(inode) && 2313 !dwc->dw_orphaned) 2314 goto out; 2315 2316 ret = ocfs2_inode_lock(inode, &di_bh, 2317 if (ret < 0) { 2318 mlog_errno(ret); 2319 goto out; 2320 } 2321 2322 down_write(&oi->ip_alloc_sem); 2323 2324 /* Delete orphan before acquire i_rws 2325 if (dwc->dw_orphaned) { 2326 BUG_ON(dwc->dw_writer_pid != 2327 2328 end = end > i_size_read(inode 2329 2330 ret = ocfs2_del_inode_from_or 2331 !!end, end); 2332 if (ret < 0) 2333 mlog_errno(ret); 2334 } 2335 2336 di = (struct ocfs2_dinode *)di_bh->b_ 2337 2338 ocfs2_init_dinode_extent_tree(&et, IN 2339 2340 /* Attach dealloc with extent tree in 2341 * which are already unlinked from cu 2342 * rotation and merging. 2343 */ 2344 et.et_dealloc = &dealloc; 2345 2346 ret = ocfs2_lock_allocators(inode, &e 2347 &data_ac, 2348 if (ret) { 2349 mlog_errno(ret); 2350 goto unlock; 2351 } 2352 2353 credits = ocfs2_calc_extend_credits(i 2354 2355 handle = ocfs2_start_trans(osb, credi 2356 if (IS_ERR(handle)) { 2357 ret = PTR_ERR(handle); 2358 mlog_errno(ret); 2359 goto unlock; 2360 } 2361 ret = ocfs2_journal_access_di(handle, 2362 OCFS2_J 2363 if (ret) { 2364 mlog_errno(ret); 2365 goto commit; 2366 } 2367 2368 list_for_each_entry(ue, &dwc->dw_zero 2369 ret = ocfs2_assure_trans_cred 2370 if (ret < 0) { 2371 mlog_errno(ret); 2372 break; 2373 } 2374 ret = ocfs2_mark_extent_writt 2375 2376 2377 2378 if (ret < 0) { 2379 mlog_errno(ret); 2380 break; 2381 } 2382 } 2383 2384 if (end > i_size_read(inode)) { 2385 ret = ocfs2_set_inode_size(ha 2386 if (ret < 0) 2387 mlog_errno(ret); 2388 } 2389 commit: 2390 ocfs2_commit_trans(osb, handle); 2391 unlock: 2392 up_write(&oi->ip_alloc_sem); 2393 ocfs2_inode_unlock(inode, 1); 2394 brelse(di_bh); 2395 out: 2396 if (data_ac) 2397 ocfs2_free_alloc_context(data 2398 if (meta_ac) 2399 ocfs2_free_alloc_context(meta 2400 ocfs2_run_deallocs(osb, &dealloc); 2401 ocfs2_dio_free_write_ctx(inode, dwc); 2402 2403 return ret; 2404 } 2405 2406 /* 2407 * ocfs2_dio_end_io is called by the dio core 2408 * particularly interested in the aio/dio cas 2409 * to protect io on one node from truncation 2410 */ 2411 static int ocfs2_dio_end_io(struct kiocb *ioc 2412 loff_t offset, 2413 ssize_t bytes, 2414 void *private) 2415 { 2416 struct inode *inode = file_inode(iocb 2417 int level; 2418 int ret = 0; 2419 2420 /* this io's submitter should not hav 2421 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb) 2422 2423 if (bytes <= 0) 2424 mlog_ratelimited(ML_ERROR, "D 2425 (long long)b 2426 if (private) { 2427 if (bytes > 0) 2428 ret = ocfs2_dio_end_i 2429 2430 else 2431 ocfs2_dio_free_write_ 2432 } 2433 2434 ocfs2_iocb_clear_rw_locked(iocb); 2435 2436 level = ocfs2_iocb_rw_locked_level(io 2437 ocfs2_rw_unlock(inode, level); 2438 return ret; 2439 } 2440 2441 static ssize_t ocfs2_direct_IO(struct kiocb * 2442 { 2443 struct file *file = iocb->ki_filp; 2444 struct inode *inode = file->f_mapping 2445 struct ocfs2_super *osb = OCFS2_SB(in 2446 get_block_t *get_block; 2447 2448 /* 2449 * Fallback to buffered I/O if we see 2450 * extents. 2451 */ 2452 if (OCFS2_I(inode)->ip_dyn_features & 2453 return 0; 2454 2455 /* Fallback to buffered I/O if we do 2456 if (iocb->ki_pos + iter->count > i_si 2457 !ocfs2_supports_append_dio(osb)) 2458 return 0; 2459 2460 if (iov_iter_rw(iter) == READ) 2461 get_block = ocfs2_lock_get_bl 2462 else 2463 get_block = ocfs2_dio_wr_get_ 2464 2465 return __blockdev_direct_IO(iocb, ino 2466 iter, get 2467 ocfs2_dio 2468 } 2469 2470 const struct address_space_operations ocfs2_a 2471 .dirty_folio = block_dirty 2472 .read_folio = ocfs2_read_ 2473 .readahead = ocfs2_reada 2474 .writepages = ocfs2_write 2475 .write_begin = ocfs2_write 2476 .write_end = ocfs2_write 2477 .bmap = ocfs2_bmap, 2478 .direct_IO = ocfs2_direc 2479 .invalidate_folio = block_inval 2480 .release_folio = ocfs2_relea 2481 .migrate_folio = buffer_migr 2482 .is_partially_uptodate = block_is_pa 2483 .error_remove_folio = generic_err 2484 }; 2485
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.