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